IEEE Communications Magazine â€¢ March 2011 Volume 49 Issue 3

A Publication of the IEEE Communications Society®

•Dynamic Spectrum Access•Cognitive Radio Networks•Future Media Internet•Network Testing

FreeComSocTutorial

LTE-Advanced

SeePage

9

IEEE

M A G A Z I N E

March 2011, Vol. 49, No. 3

www.comsoc.org

March 2011 Cover 1 2/22/11 5:25 PM

Director of MagazinesAndrzej Jajszczyk, AGH U. of Sci. & Tech. (Poland)

Editor-in-ChiefSteve Gorshe, PMC-Sierra, Inc. (USA)

Associate Editor-in-ChiefSean Moore, Centripetal Networks (USA)

Senior Technical EditorsTom Chen, Swansea University (UK)

Nim Cheung, ASTRI (China)Nelson Fonseca, State Univ. of Campinas (Brazil)

Torleiv Maseng, Norwegian Def. Res. Est. (Norway)Peter T. S. Yum, The Chinese U. Hong Kong (China)

Technical EditorsSonia Aissa, Univ. of Quebec (Canada)

Mohammed Atiquzzaman, U. of Oklahoma (USA)Paolo Bellavista, DEIS (Italy)

Tee-Hiang Cheng, Nanyang Tech. U. (Rep. Singapore)Jacek Chrostowski, Scheelite Techn. LLC (USA)Sudhir S. Dixit, Nokia Siemens Networks (USA)

Stefano Galli, Panasonic R&D Co. of America (USA)Joan Garcia-Haro, Poly. U. of Cartagena (Spain)Vimal K. Khanna, mCalibre Technologies (India)

Janusz Konrad, Boston University (USA)Abbas Jamalipour, U. of Sydney (Australia)

Deep Medhi, Univ. of Missouri-Kansas City (USA)Nader F. Mir, San Jose State Univ. (USA)

Amitabh Mishra, Johns Hopkins University (USA)Sedat Ölçer, IBM (Switzerland)

Glenn Parsons, Ericsson Canada (Canada)Harry Rudin, IBM Zurich Res.Lab. (Switzerland)Hady Salloum, Stevens Institute of Tech. (USA)Antonio Sánchez Esguevillas, Telefonica (Spain)

Heinrich J. Stüttgen, NEC Europe Ltd. (Germany)Dan Keun Sung, Korea Adv. Inst. Sci. & Tech. (Korea)Danny Tsang, Hong Kong U. of Sci. & Tech. (Japan)

Series EditorsAd Hoc and Sensor Networks

Edoardo Biagioni, U. of Hawaii, Manoa (USA)Silvia Giordano, Univ. of App. Sci. (Switzerland)

Automotive Networking and ApplicationsWai Chen, Telcordia Technologies, Inc (USA)

Luca Delgrossi, Mercedes-Benz R&D N.A. (USA)Timo Kosch, BMW Group (Germany)

Tadao Saito, University of Tokyo (Japan)Consumer Communicatons and Networking

Madjid Merabti, Liverpool John Moores U. (UK)Mario Kolberg, University of Sterling (UK)

Stan Moyer, Telcordia (USA)Design & Implementation

Sean Moore, Avaya (USA)Salvatore Loreto, Ericsson Research (Finland)

Integrated Circuits for CommunicationsCharles Chien (USA)

Zhiwei Xu, SST Communication Inc. (USA)Stephen Molloy, Qualcomm (USA)

Network and Service Management SeriesGeorge Pavlou, U. of Surrey (UK)

Aiko Pras, U. of Twente (The Netherlands)Networking Testing Series

Yingdar Lin, National Chiao Tung University (Taiwan)Erica Johnson, University of New Hampshire (USA)Tom McBeath, Spirent Communications Inc. (USA)

Eduardo Joo, Empirix Inc. (USA)Topics in Optical Communications

Hideo Kuwahara, Fujitsu Laboratories, Ltd. (Japan)Osman Gebizlioglu, Telcordia Technologies (USA)

John Spencer, Optelian (USA)Vijay Jain, Verizon (USA)

Topics in Radio CommunicationsJoseph B. Evans, U. of Kansas (USA)

Zoran Zvonar, MediaTek (USA)Standards

Yoichi Maeda, NTT Adv. Tech. Corp. (Japan)Mostafa Hashem Sherif, AT&T (USA)

ColumnsBook Reviews

Andrzej Jajszczyk, AGH U. of Sci. & Tech. (Poland)History of Communications

Mischa Schwartz, Columbia U. (USA)Regulatory and Policy Issues

J. Scott Marcus, WIK (Germany)Jon M. Peha, Carnegie Mellon U. (USA)

Technology Leaders' ForumSteve Weinstein (USA)

Very Large ProjectsKen Young, Telcordia Technologies (USA)

Publications StaffJoseph Milizzo, Assistant Publisher

Eric Levine, Associate PublisherSusan Lange, Online Production ManagerJennifer Porcello, Publications Specialist

Catherine Kemelmacher, Associate Editor

2 IEEE Communications Magazine • March 2011

IEEE

M A G A Z I N EMarch 2011, Vol. 49, No. 3

www.comsoc.org/~ci

TOPICS IN RADIO COMMUNICATIONSSERIES EDITORS: JOSEPH EVANS AND ZORAN ZVONAR

GUEST EDITORIAL: THE MATURATION OF DYNAMIC SPECTRUM ACCESS FROM AFUTURE TECHNOLOGY, TO AN ESSENTIAL SOLUTION TO IMMEDIATE CHALLENGESGUEST EDITOR: PRESTON MARSHALL

UNDERSTANDING CONDITIONS THAT LEAD TO EMULATION ATTACKS IN DYNAMICSPECTRUM ACCESSThe threat of emulation attacks, in which users pretend to be of a type they are not in order to gain unauthorized access to spectrum, has the potential to severely degrade the expected performance of the system. The authors analyze this problem within a Bayesian game framework, in which users are unsure of the legitimacy of the claimed type of other users. RYAN W. THOMAS, BRETT J. BORGHETTI, RAMAKANT S. KOMALI, AND PETRI MÄHÖNEN

DYNAMIC SPECTRUM ACCESS OPERATIONAL PARAMETERS WITH WIRELESSMICROPHONESThe authors provide a comprehensive analysis of dynamic spectrum access operational parameters in a typical hidden node scenario with protected wireless microphones in the TV white space.TUGBA ERPEK, MARK A. MCHENRY, AND ANDREW STIRLING

THE VIABILITY OF SPECTRUM TRADING MARKETSThe authors focus on determining the conditions for viability of spectrum trading markets. They make use of agent-based computational economics to analyze different market scenarios and the behaviors of their participants.CARLOS E. CAICEDO AND MARTIN B. H. WEISS

UNIFIED SPACE-TIME METRICS TO EVALUATE SPECTRUM SENSINGThe authors present a unified framework in which the natural ROC curve correctly captures the two features desired from a spectrum sensing system: safety to primary users and performance for the secondary users.RAHUL TANDRA, ANANT SAHAI, AND VENUGOPAL VEERAVALLI

ADVANCES IN STANDARDS AND TESTBEDS FORCOGNITIVE RADIO NETWORKS: PART II

GUEST EDITORS: EDWARD K. AU, DAVE CAVALCANTI, GEOFFREY YE LI, WINSTON CALDWELL, AND KHALED BEN LETAIEF

GUEST EDITORIAL

WIRELESS SERVICE PROVISION IN TV WHITE SPACE WITH COGNITIVE RADIOTECHNOLOGY: A TELECOM OPERATOR’S PERSPECTIVE AND EXPERIENCEThere is a fundamental change happening in spectrum regulation: the enabling of spectrum sharing, where primary (licensed) users of the spectrum are forced to allow sharing with secondary users, who use license-exempt equipment.MICHAEL FITCH, MAZIAR NEKOVEE, SANTOSH KAWADE, KEITH BRIGGS, AND RICHARD MACKENZIE

EMERGING COGNITIVE RADIO APPLICATIONS: A SURVEYThere are new opportunities for cognitive radio to enable a variety of emerging applications. The authors present a high-level view of how cognitive radio would support such applicationsJIANFENG WANG, MONISHA GHOSH, AND KIRAN CHALLAPALI

INTERNATIONAL STANDARDIZATION OF COGNITIVE RADIO SYSTEMSThe authors describe the current concept of the CRS and show the big picture of international standardization of the CRS. Understanding these standardization activities is important for both academia and industry.STANISLAV FILIN, HIROSHI HARADA, HOMARE MURAKAMI, AND KENTARO ISHIZU

COGNITIVE RADIO: TEN YEARS OF EXPERIMENTATION AND DEVELOPMENTAlthough theoretical research is blooming, hardware and system development for CR is progressing at a slower pace. The authors provide synopses of the commonly used platforms and testbeds, examine what has been achieved in the last decade of experimentation and trials, and draw several perhaps surprising conclusions.PRZEMYSLAW PAWELCZAK, KEITH NOLAN, LINDA DOYLE, SER WAH OH, AND DANIJELA CABRIC

30

32

38

46

54

6264

74

82

90

®

LYT-TOC-MAR 3/5/11 12:58 AM

2011 Communications SocietyElected Officers

Byeong Gi Lee, PresidentVijay Bhargava, President-Elect

Mark Karol, VP–Technical ActivitiesKhaled B. Letaief, VP–Conferences

Sergio Benedetto, VP–Member RelationsLeonard Cimini, VP–Publications

Members-at-LargeClass of 2011

Robert Fish, Joseph EvansNelson Fonseca, Michele Zorzi

Class of 2012Stefano Bregni, V. Chan

Iwao Sasase, Sarah K. WilsonClass of 2013

Gerhard Fettweis, Stefano GalliRobert Shapiro, Moe Win

2011 IEEE OfficersMoshe Kam, President

Gordon W. Day, President-ElectRoger D. Pollard, Secretary

Harold L. Flescher, TreasurerPedro A. Ray, Past-President

E. James Prendergast, Executive DirectorNim Cheung, Director, Division III

IEEE COMMUNICATIONS MAGAZINE (ISSN 0163-6804) is published monthly by The Institute ofElectrical and Electronics Engineers, Inc.Headquarters address: IEEE, 3 Park Avenue, 17thFloor, New York, NY 10016-5997, USA; tel: +1-212-705-8900; http://www.comsoc.org/ci. Responsibility forthe contents rests upon authors of signed articles andnot the IEEE or its members. Unless otherwise speci-fied, the IEEE neither endorses nor sanctions any posi-tions or actions espoused in IEEE CommunicationsMagazine.

ANNUAL SUBSCRIPTION: $27 per year print subscrip-tion. $16 per year digital subscription. Non-member printsubscription: $400. Single copy price is $25.

EDITORIAL CORRESPONDENCE: Address to: Editor-in-Chief, Steve Gorshe, PMC-Sierra, Inc., 10565 S.W.Nimbus Avenue, Portland, OR 97223; tel: +(503) 431-7440, e-mail: [email protected].

COPYRIGHT AND REPRINT PERMISSIONS:Abstracting is permitted with credit to the source. Librariesare permitted to photocopy beyond the limits of U.S.Copyright law for private use of patrons: those post-1977articles that carry a code on the bottom of the first page pro-vided the per copy fee indicated in the code is paid throughthe Copyright Clearance Center, 222 Rosewood Drive,Danvers, MA 01923. For other copying, reprint, or republi-cation permission, write to Director, Publishing Services,at IEEE Headquarters. All rights reserved. Copyright © 2011by The Institute of Electrical and Electronics Engineers, Inc.

POSTMASTER: Send address changes to IEEECommunications Magazine, IEEE, 445 Hoes Lane,Piscataway, NJ 08855-1331. GST Registration No.125634188. Printed in USA. Periodicals postage paid at NewYork, NY and at additional mailing offices. Canadian PostInternational Publications Mail (Canadian Distribution)Sales Agreement No. 40030962. Return undeliverableCanadian addresses to: Frontier, PO Box 1051, 1031 HelenaStreet, Fort Eire, ON L2A 6C7

SUBSCRIPTIONS, orders, address changes — IEEEService Center, 445 Hoes Lane, Piscataway, NJ08855-1331, USA; tel: +1-732-981-0060; e-mail:[email protected].

ADVERTISING: Advertising is accepted at the dis-cretion of the publisher. Address correspondence to:Advertising Manager, IEEE Communications Magazine,3 Park Avenue, 17th Floor, New York, NY 10016.

SUBMISSIONS: The magazine welcomes tutorial orsurvey articles that span the breadth of communica-tions. Submissions will normally be approximately 4500words, with few mathematical formulas, accompaniedby up to six figures and/or tables, with up to 10 careful-ly selected references. Electronic submissions are pre-ferred, and should be sumitted through ManuscriptCentral http://mc.manuscriptcentral.com/commag-ieee.Instructions can be found at the following: http://dl.com-soc.org/livepubs/ci1/info/sub_guidelines.html. For furtherinformation contact Sean Moore, Associate Editor-in-Chief ([email protected]). All submissions will bepeer reviewed.


SPIDERRADIO: A COGNITIVE RADIO NETWORK WITH COMMODITY HARDWARE ANDOPEN SOURCE SOFTWAREThe authors present SpiderRadio, a cognitive radio prototype for dynamic spectrum access networking. SpiderRadio is built using commodity IEEE 802.11a/b/g hardware and the open source MadWiFi driver.S. SENGUPTA, K. HONG, R. CHANDRAMOULI, AND K. P. SUBBALAKSHMI

FUTURE MEDIA INTERNETGUEST EDITORS: THEODORE ZAHARIADIS, GIOVANNI PAU, AND GONZALO CAMARILO

GUEST EDITORIAL

CURLING: CONTENT-UBIQUITOUS RESOLUTION AND DELIVERY INFRASTRUCTURE FORNEXT-GENERATION SERVICESCURLING aims to enable a future content-centric Internet that will overcome the current intrinsic constraints by efficiently diffusing media content of massive scale.WEI KOONG CHAI, NING WANG, IOANNIS PSARAS, GEORGE PAVLOU, CHAOJIONG WANG, GERARDO GARCIA DE BLAS, FRANCISCO JAVIER RAMON SALGUERO, LEI LIANG, SPIROS SPIROU, ANDRZEJ BEBEN, AND ELEFTHERIA HADJIOANNOU

A SURVEY ON CONTENT-ORIENTED NETWORKING FOR EFFICIENT CONTENT DELIVERYThe authors present a comprehensive survey of content naming and name-based routing, and discuss further research issues in CON.JAEYOUNG CHOI, JINYOUNG HAN, EUNSANG CHO, TED “TAEKYOUNG” KWON, AND YANGHEE CHOI

PEER-TO-PEER STREAMING OF SCALABLE VIDEO IN FUTURE INTERNET APPLICATIONSIf video is encoded in a scalable way, it can be adapted to any required spatio-temporal resolution and quality in the compressed domain, according to a peer bandwidth and other peers’ context requirements. NAEEM RAMZAN, EMANUELE QUACCHIO, TONI ZGALJIC, STEFANO ASIOLI, LUCA CELETTO, EBROUL IZQUIERDO, AND FABRIZIO ROVATI

IMPROVING END-TO-END QOE VIA CLOSE COOPERATION BETWEEN APPLICATIONSAND ISPSThe authors present an architecture to enable cooperation between the application providers, the users, and the communications networks so that the quality of experience of the users of the application is improved and network traffic optimized.BERTRAND MATHIEU, SELIM ELLOUZE, NICO SCHWAN, DAVID GRIFFIN, ELENI MYKONIATI, TOUFIK AHMED, AND ORIOL RIBERA PRATS

SYSTEM ARCHITECTURE FOR ENRICHED SEMANTIC PERSONALIZED MEDIA SEARCHAND RETRIEVAL IN THE FUTURE MEDIA INTERNETThe authors describe a novel system and its architecture to handle, process, deliver, personalize, and find digital media, based on continuous enrichment of the media objects through the intrinsic operation within a content oriented architecture.MARIA ALDUAN, FAUSTINO SANCHEZ, FEDERICO ÁLVAREZ, DAVID JIMÉNEZ, JOSÉ MANUEL MENÉNDEZ, AND CAROLINA CEBRECOS

AUTOMATIC CREATION OF 3D ENVIRONMENTS FROM A SINGLE SKETCH USINGCONTENT-CENTRIC NETWORKSThe authors present a complete and innovative system for automatic creation of 3D environments from multimedia content available in the network.THEODOROS SEMERTZIDIS, PETROS DARAS, PAUL MOORE, LAMBROS MAKRIS, AND MICHAEL G. STRINTZIS

TOPICS IN NETWORK TESTINGSERIES EDITORS: YING-DAR LIN, ERICA JOHNSON, AND EDUARDO JOO

SERIES EDITORIAL

ADJACENT CHANNEL INTERFERENCE IN 802.11A IS HARMFUL: TESTBED VALIDATIONOF A SIMPLE QUANTIFICATION MODELThe authors report results that show clear throughput degradation because of ACI in 802.11a, the magnitude of which depends on the interfering data rates, packet sizes, and utilization of the medium.VANGELIS ANGELAKIS, STEFANOS PAPADAKIS, VASILIOS A. SIRIS, APOSTOLOS TRAGANITIS

EMERGING TESTING TRENDS AND THE PANLAB ENABLING INFRASTRUCTUREThe authors address a number of fundamental principles and their corresponding technology implementations that enable the provisioning of large-scale testbeds fortesting and experimentation as well as deploying future Internet platforms for piloting novel applications. SEBASTIAN WAHLE, CHRISTOS TRANORIS, SPYROS DENAZIS, ANASTASIUS GAVRAS, KONSTANTINOS KOUTSOPOULOS, THOMAS MAGEDANZ, AND SPYROS TOMPROS

101

110112

121

128

167

160

136

144

152

158

President’s Page 6Certification Corner 12Conference Report/CCNC 14Conference Report/GLOBECOM 16Conference Preview/GrrenCom 20

Product Spotlights 23New Products 24Global Communications Newsletter 25Advertisers’ Index 176

LYT-TOC-MAR 3/5/11 12:58 AM Page 4

IEEE Communications Magazine • March 20116

THE PRESIDENT’S PAGE

ICT SERVICES - “COMMUNICATIONS SOCIETY ON-LINE”

he IEEE Communications Society(ComSoc), being geographically diverse,

relies heavily on state of the art Informationand Communication Technology (ICT) infra-structure and services for its governance,operations, and to host its community andgroup activities. All ComSoc activities involveparticipation from volunteers who live on var-ious continents and in different time zones.So ICT infrastructure is essential in keepingsuch a global Society operational. The Com-munications Society’s ICT infrastructure andservices are the technological foundation forour “Communications Society On-Line.”

The success of ComSoc in serving its mem-bership, our profession, industry, and humani-ty depends on ingenuity, pro-fessional interests, and the enthusi-astic volunteerism of its memberswho apply those virtues toward thecreation and maintenance of Com-Soc products and services. The on-line mechanisms provided by theICT infrastructure aid in organizingconferences, running magazinesand journals, conducting meetings,and socializing ideas. These mecha-nisms are critical to volunteer pro-ductivity and the effective use ofeach volunteer’s precious time.

Serving the ICT needs of thesociety, including providing on-linecontent, is the responsibility of theChief Information Officer (CIO),Alexander Gelman, who reports tothe ComSoc President, and theDirector of On-Line Content,James Hong, who reports to theVP-Publications. Staff support isprovided by the ICT Department,headed by David Alvarez, based inComSoc’s New York office. I sharethis month’s President’s Page withAlex Gelman, James Hong, DavidAlvarez, and Fred Bauer, whochairs ComSoc’s ad hoc “Smartpho-nomics” Committee.

Alexander D. Gelman receivedhis M.E. and Ph.D. in electricalengineering from CUNY. Presently he is CTO of the NETovations consulting group. During 1998-2007 Alex wasthe Chief Scientist of the Panasonic Princeton Laboratory.During 1984-1998 he worked at Bellcore as Director of theInternet Access Architectures Research group. Alex hasserved on several IEEE, ComSoc, and IEEE-SA (IEEE Stan-dards Association) committees; he has worked on publicationsand conferences; served three terms as a ComSoc Vice Presi-dent; and served on the ComSoc Board of Governors (BoG)and IEEE-SA BoG. Currently, he is ComSoc’s Chief Infor-mation Officer, Vice Chair of the ComSoc Standards Board,and a member of the IEEE-SA Standards Board. In recogni-

tion of his volunteer contributions, Alexreceived the 2006 Donald W. McLellan Meri-torious Service Award of ComSoc.

James Won-Ki Hong is Professor andHead of the Division of IT ConvergenceEngineering and Dean of the GraduateSchool for Information Technology,POSTECH, Pohang, Korea. He received hisPh.D. from the University of Waterloo, Cana-da in 1991. His research interests include net-work management, network monitoring andanalysis, and convergence engineering. Dur-ing 2005-2009, James served as chair of theIEEE ComSoc Committee on Network Oper-ations and Management (CNOM). Currentlyhe serves on ComSoc’s BoG as Director of

On-Line Content, and is an activemember of the NOMS/IM andAPNOMS steering committees. Hewas also General Co-Chair of the2010 IEEE/IFIP Network Opera-tions and Management Symposium(NOMS 2010). In addition, he is aneditorial advisory board memberfor JNSM, IJNM, JTM, and TNSM.

Fred Bauer works at Cisco Sys-tems as a Technical Leader in theSmart Grids Group. Previously hewas at Nokia, PacketHop, SRI, andIntel. He received his Ph.D. in com-puter engineering from the Univer-sity of California at Santa Cruz in1996. His research interests includemulticast routing, wireless networks,and mesh routing. He just complet-ed a term as a Member-at-Large ofComSoc’s BoG (2008-2010).Cur-rently he chairs the ComSoc Gover-nance Committee as well as the adhoc Smartphonomics Committee.He serves as a member of the IEEEINFOCOM and the SECON steer-ing committees and as a member ofthe IEEE Conferences Board. He isthe chair of the Society’s TechnicalProgram Integrity Initiative Com-mittee (TPII).

David Alvarez is a graduate ofthe University of Florida and has

been in the ICT field for 24 years. Prior to his employmentwith IEEE ComSoc, he worked in development and ICT man-agement for T2 Medical and Coram Healthcare in Atlanta,Georgia. He has been serving as ComSoc’s Director of Infor-mation Technology since 1997.

COMSOC’S ICT INFRASTRUCTUREComSoc’s ICT Department employs the following staff

support team:•Director – David Alvarez. David works with the ComSoc

CIO to determine ICT strategy, and manage information sys-tems and employees.

T

BYEONG GI LEE

ALEX GELMAN FRED BAUER

JAMES HONG DAVID ALVAREZ

LYT-PRES PAGE-MAR 2/22/11 2:15 PM Page 6

IEEE Communications Magazine • March 2011 7


•Manager of On-Line Content andDesign – Natasha Simonovski. Natashamanages all ComSoc web sites and on-linecontent. She is responsible for design andlayout of ComSoc web page templates.

•Program Manager – Joel Basco. Joelis responsible for ongoing management ofComSoc products and services. Currentlyhis responsibilities include support of theComSoc Digital Library and Webcastingprograms.

•Programmer/Developer – Matt Sielski.Matt is responsible for development ofapplications on the ComSoc digital plat-form (Drupal CMS) and related tasks.

•Systems Administrator – Tony Ruiz.Tony is responsible for supporting all ofComSoc’s systems and networks. He worksclosely with the IEEE IT group on networking, email andsecurity services.

•PC Technician – John Porter. John supports staff equip-ment, including desktops, printers, mobile devices, and soft-ware. He also assists the Systems Administrator and workswith the Program Manager on video projects.

•Web Content Administrator – Tammy Hung. Tammyworks on keeping on-line content current. She also manages ourSocial Networking sites. She works with the Program Managerto produce and maintain the Society’s Tutorials and Webinars.

These are the products supported by the ICT team:•ComSoc publications•ComSoc Digital Library•ComSoc webcasts•ComSoc training•Advertising•ComSoc conferences

The following services are currently offered:•ComSoc website•Conference websites•Community website•Chapter and Technical Committee websites•ComSoc member aliases•Email lists•ComSoc E-News•ComSoc Community Booklet

In 2010, ComSoc completed the “virtualization” of its ICTinfrastructure. Now ComSoc ICT relies on production serversin the IEEE Datacenter for support while still having the abil-ity to create and deploy its own applications. Some applica-tions are also a virtual overlay on IEEE’s applications services.

COMSOC ICT SERVICESComSoc Web Presence

Some years ago, the Meetings and Conferences (M&C),Marketing, and ICT staff members selected a commercialcontent management system, Eprise, to improve workflow andallow assistance with website updates from M&C staff andvolunteers. Eprise has since served as the major platform forconference, chapter, and technical committee web pages. It isrelatively easy to use and many volunteer groups have alreadytaken advantage of the tool.

Meetings and Conferences ServicesAmong the major activities in ComSoc, M&C requires

very intensive and responsive ICT support. Our workshops,symposia, and conferences are in need of creating andmaintaining pages for ComSoc events. The websites of

those events may be maintained by our ICT team, in whichcase quick responses from the ICT team are essential, orthe websites may be created and maintained by volunteerseither on their own or on the ComSoc ICT infrastructure.Both approaches require support by the ComSoc ICTteam.

In 2010, we began webcasting specific program elementsfrom several conferences. Recording takes place during theconference and is then made available on-line. The scenariofor this service includes the following steps:•Identification of appropriate sessions for recording•Obtaining presenters’ consent•Real-time webcast as well as recording utilizing voice over

Power Point (VoPP) •Questions and answers•Conversion of recordings to Flash format•Employing a content delivery network (CDN) for distribu-

tionDetails of this program can be found at http://www.com-

soc.org/webcasts.We are now pursuing an experiment that may lead to

offering virtual conferences. ComSoc is in the process of orga-nizing a completely on-line conference using the Cisco WebExplatform. The experimental conference, itself “green,” is ongreen communications, IEEE GreenCom 2011. The confer-ence will feature parallel on-line sessions. For further infor-mation, please visit http://www.ieee-greencom.org/.

ComSoc Email Alias: [email protected] has been a long-term goal of our volunteer leaders

to make ComSoc email aliases available to all ComSocmembers. Toward the end of last year, we finally imple-mented the mechanism that allows its members to proudlypossess an email address that carries the name of ourbeloved society by displaying “@comsoc.org.” You mayobtain a ComSoc email alias at: http://cmsc-ems.ieee.org.We encourage all ComSoc members to take advantage ofthis opportunity.

ComSoc Mailing Lists To support group activities, the ComSoc ICT team, led by

Tony Ruiz, Systems Administrator, and John Porter, TechSupport Specialist, in partnership with their IEEE col-leagues, has implemented LISTSERV support for emailapplications hosted on the IEEE Data Center servers.Group leaders may now create and manage mailing listsusing the LISTSERV website: http://listserv.ieee.org/request/add-listserv.html.

ComSoc Website www.comsoc.org

Digital Library dl.comsoc.org/comsocdl

Community community.comsoc.org

ComSoc Webcasts www.comsoc.org/webcasts

ComSoc Online Training www.comsoc.org/training/online-training

ComSoc Member Aliases cmsc-ems.ieee.org/

ComSoc Email Lists comsoc-listserv.ieee.org/cgi-bin/wa?HOME

Community Booklet www.comsoc.org/about/ComSocCommunityBooklet

Facebook www.facebook.com/IEEEComSoc

LinkedIn www.linkedin.com/groups?mostPopular=&gid=81475

Twitter twitter.com/#!/comsoc

ComSoc ICT Products and Services


LEVERAGING OPEN SOURCE SOFTWARE

During 2008-2009 the ComSoc ICT team made the deci-sion to deploy open source software for support of ComSocweb applications. This initiative was started by the previousCIO, Harvey Freeman, supported by Fred Bauer. The majorobjective of this initiative was to deploy a wide range of appli-cations and to improve the ComSoc ICT Department’s abilityto respond to the ICT needs of our constituency in a timelyfashion. The team chose Drupal as the platform that wouldbest meet these needs.

Drupal is an open source content management platformthat is widely accepted around the world. Its applicationsrange from personal blogs to enterprise solutions. Drupal isfree, flexible, robust, and constantly being improved by hun-dreds of thousands of enthusiastic professionals from all overthe world. ComSoc’s ICT team has joined this community. Wenow create our own modules and contribute them to the Dru-pal community, as well as benefit from the contributions ofother community members.

The Drupal platform was, and is, ideal for website develop-ment, thus leading naturally to ICT replicating comsoc.org onthe Drupal platform. The new site, available “24/7,” went livein 2009. New features developed on the Drupal platforminclude:•A mobile version of the comsoc.org website (http://m.com-

soc.org), the first in IEEE.•RSS feeds for ComSoc news, conference events and publi-

cations into social media pages setup on Facebook,LinkedIn, and Twitter.

•A robust Webcast Storefront for the new ConferencesWebcast program.

•Marketing blog for promotion of the ComSoc websitewith RSS feeds into Social Media pages.ComSoc’s CIO, Alex Gelman, and the ICT team call on

you, the ComSoc volunteers, to join the ComSoc Drupaldevelopment team and contribute your work to ComSoc andto the entire Drupal community. If you are interested in thedevelopment of this feature-rich content management plat-form, please e-mail Alex at [email protected].

SERVING MOBILE USERS: SMARTPHONOMICSUnder the leadership of ComSoc’s Vice President of Publi-

cations, Len Cimini, our Director of On-Line Content, JamesHong, proposed a new initiative aimed at providing ComSocservices to members carrying smart phones, or any form ofsmart devices. James already had his students doing researchon the new field of “smartphonomics” (in fact, this may havebeen the first use of that term). To facilitate the initiative, wecreated an ad hoc committee on “Smartphonomics,” withFred Bauer as its Chair. This committee, involving both theComSoc ICT team and a range of volunteers, has alreadyachieved some goals, including mobile web access tocomsoc.org and a ComSoc “app” for iPhones.

There is much to be done in this smartphonomics area asan increasing number of members will be using mobile devicessuch as smart phones and tablet computers to access a grow-ing number of resources, including those of ComSoc. The ideaof the smartphonomics initiative is to give our members accessto ComSoc material and resources directly from their mobiledevices wherever they might be. There are two parts to thisproblem: 1) serving ComSoc webpages to mobile devices and2) providing new and interesting content regularly to thosemobile devices.

The smartphonomics team started with the problem ofwebpage delivery. Last year, as described above, the ComSoc

ICT team members, Natasha Simonovski and Matt Sielski,enabled the www.comsoc.org site to recognize smartphonebrowsers and serve mobile content from the website m.comsoc.org. Currently, ComSoc Drupal-based content isautomatically processed for mobile use. This new mechanismallows ComSoc to quickly serve many types of mobile devicesconveniently with content stored on m.comsoc.org. The nextlogical step is to support applications native to the most popu-lar mobile platforms.

James Hong and his students at POSTECH graciously builtand provided an iPhone application for ComSoc memberswith an Apple iOS device such as the iPhone, iPad, or iPodTouch. This application was demonstrated during December’sComSoc committee meetings, showing how ComSoc membersmight soon be able to access all of the Communications Soci-ety via their smart devices. Smartphonomics service is nowavailable to our members from the Apple App Store for free(http://itunes.apple.com/us/app/comsoc/id413046307). We planto provide similar native smart phone applications forAndroid-based mobile devices soon, and possibly others lateron.

The second problem to be addressed by the ad hoc Smart-phonomics Committee is what content to provide on a regularbasis that would be of interest to our members. For this, wedraw inspiration from IEEE’s newly created IEEE Technolo-gy News website: http://www.ieeetechnews.org. This IEEEwebsite aggregates and summarizes a subset of articles fromIEEE periodicals, making the summaries accessible to thegeneral public. The ad hoc Smartphonomics Committee isworking with a number of similar groups within ComSoc toidentify which regularly updated content we can provide thatwould interest our members with smart devices. As always, wewelcome input from you the reader on what you would like tosee on ComSoc’s mobile website. Please direct your commentsto Fred at [email protected].

COMMUNITY AND SOCIAL NETWORKINGComSoc Community Sites

The new community site (community.comsoc.org) featur-ing groups, blogs, forums and news feeds, went on-line inearly 2009. The community site was open to all communica-tions professionals, not just IEEE members. Groups were themost popular feature and many ComSoc committees beganusing them to communicate, share information and files, listevents, create discussions, and conduct the day to day businessof the groups.

Presently, ComSoc hosts several community groups. Eachgroup has its own policies for joining and operation. Weencourage all ComSoc communities and groups, including alltechnical committees, chapters, organizing committees andother groups, to create community sites and use them fortheir ComSoc activities.

ComSoc Blog Sites The ComSoc blog was set up by ComSoc’s ICT Depart-

ment for the Marketing Department to blog updates on pro-motions, conferences, etc. All entries are automatically fedinto social media sites. This site has attracted quite a few fol-lowers on our social media pages and has created significantinterest in ComSoc promotions and services. The ComSocblog is updated by the IEEE Marketing staff, Ting Qian andMax Loskutnikov.

Also a blog page has been created on the ComSoc Com-munity site: http://community.comsoc.org/blogs. It is moderat-ed by a prominent volunteer, Alan Weissberger, who is also



(Continued on page 10)


10



the Chair of the ComSoc Santa Clara Valley Chapter. Alan’stireless efforts helped to establish, debug, and maintain thissite.

ComSoc in Social MediaComSoc now has a presence on Facebook

(http://www.facebook.com/IEEEComSoc) and LinkedIn(http://www. linkedin.com/groups?mostPopular=&gid=81475). ComSoc e-News is also available on Facebook. Toreceive timely messages from ComSoc, please follow us ontwitter: http:// twitter.com/comsoc. In addition, there is aComSoc channel on YouTube: http://www.youtube.com/user/ieeecomsoc. The very talented channel creator and artisticdirector is Max Loskutnikov.

On-Line ContentThe cornerstone of the Society’s on-line content is the

ComSoc Digital Library, http://dl.comsoc.org, which containsjournals, magazines, conference proceedings, newsletters, on-line tutorials, and other resources such as distinguished lec-tures, webinars, and videos. It also contains material relatedto the IEEE Wireless Communication Engineering Technolo-gies (WCET) certification program.

All ComSoc-sponsored journals, magazines, and confer-ence publications can be found in the ComSoc Digital Library.ComSoc newsletters include e-News (http://e-news.comsoc.org), the Global Communications Newsletter(http://dl.comsoc.org/gcn), and the IEEE Wireless Communi-cations Professional Newsletter (http://www.ieee-wcet.org/).ComSoc’s on-line tutorial program, called Tutorials Now, pro-vides a collection of recent tutorials given at ComSoc-spon-sored conferences – GLOBECOM/ICC, INFOCOM,NOMS/IM, WCNC, CCNC, ENTNET, SECON, PIMRC andMILCOM. Each tutorial features expert presenters whoreview current topics in communications, with the aim ofbringing newcomers up to speed and keeping others current.Available tutorials, which are 2.5 to 5 hours in length, containthe original visuals and voice-over by the presenter.

Distinguished Lectures On-Line provides recorded lecturesof distinguished lecturers (DLs) selected by the ComSoc Dis-tinguished Lecturer Selection Committee. DLs are typicallyinvited by ComSoc chapters and sister societies around theworld. Having these available on line greatly increases theirreach.

Live ComSoc Webinars and on-line panel sessions areopen to all. Viewers and panelists participate from the conve-nience of their desks. Webinars focus on technologies, systemsand services of current interest to communications engineersand scientists (http://www.comsoc.org/webinars). ComSocVideos provides video recordings of notable keynote speechesat our major conferences such as the keynote speech on thehistory of the Internet given by Leonard Kleinrock atGLOBECOM 2007.

MORE SERVICES TO COMEThe ComSoc ICT team has aggressive plans for the

future. Among them is to enable all ComSoc groups, includ-ing chapters and our sister societies, to deploy their owncommunity web sites, conduct electronic meetings, and tocreate and maintain their own mailing lists using LIST-SERV. Also in the plan is to enable all conferences to offerwebcasts and recorded sessions as well as to make totallyon-line conferences a reality. The latter is important forreaching out to those who may not be able to attend confer-ences in person. We also recognize that many colleagueswill still seek the rich experience of attending a conferencein person, which is something that current technology is notyet able to replicate.

An important task in the works is to change the presenta-tion format for the electronic version of IEEE Communica-tions Magazine, which is included with Society membership.The technology under consideration is FlipBook. The newapproach will allow all the features now available with PDFformat and more. Tammy Hung is working closely with Jen-nifer Porcello of ComSoc’s Publications Department to deliv-er this new format.

More features are in the planning stages to leverage theDrupal Content Management System, such as automatic gen-eration of mobile web pages and RSS feeds. In view of Dru-pal’s power and flexibility, the ComSoc ICT team plans tocreate a robust digital platform and leverage it to deliver valu-able current and future products and services.

As presented so far, Communications Society On-Line, acollection of on-line content, products, and services beingoffered through our ICT platform, has grown rapidly to keepup with the information-sharing needs of our members andthe broader community. All those involved with creating our“anytime, anywhere” cyberspace environment, including ourICT team, will continue to develop these capabilities even fur-ther. We welcome your input.

(Continued from page 8)


12

In a fast changing field such as wire-less communications, technology and theknowledge and skills that go with it canquickly become outdated. WCET certifi-cation represents, in a sense, a snapshotin time. The WCP credential indicatesthat the holder has demonstrated masteryof the field “today.” For those whoearned the credential two or three yearsago, that “today” is becoming “yester-day.” To maintain the credential in goodstanding, they need to recertify theirknowledge and skills. As stated in the2011 Candidate’s Handbook (“Recertifi-cation,” page 35):

“… passing the examination is onlyone portion of certification. The wirelesscommunication field is constantly chang-ing and requires that wireless communica-tion professionals keep current withchanges in the profession. Maintaining anactive certification status through recerti-fication is the way in which certified pro-fessionals demonstrate their currency andpreserve their professional edge. Recerti-fication is required every five years, deter-

mined by the expiration date of your cur-rent certification …”

The first professionals to earn theWCP credential passed the exam in theFall of 2008; they will need to recertify in2013. That seems a long way off, but inreality, preparation for recertificationneeds to start well in advance. The mostobvious way to recertify, of course, is toretake the exam. Passing the 2013 exam,which will have been significantly changedand updated in the intervening five years,will clearly demonstrate that a credentialholder has kept up with advances in wire-less communications technology.

However, as with many other certifica-

tion programs that require recertification,ComSoc is working to provide alternativemeans for people to demonstrate thatthey have kept their skills and knowledgecurrent with changes in technology. Theformal recertification program is underdevelopment, and it is ComSoc’s intent toroll it out later this year. It will includeoptions for recertifying by earning Person-al Development Points (PDPs). Amongthe options under discussion for acquiringPDPs are:

•Working in one or more of the seventechnical areas covered by the WCETexam, performing tasks and holdingresponsibilities at the professional level.

•Taking training courses from Com-Soc and other providers that are specificto technology developments in one ormore of the seven technical areas.

•Attending professional conferencesand workshops that address technicaladvances in one or more of the seventechnical areas.

•Participating in relevant sessions oflocal ComSoc chapters or communities,ranging from attending an educationalsession to being a featured speaker atsuch a session.

•Authoring papers or articles in recog-nized industry publications on a topic ortopics relevant to the technical areas cov-ered by the WCET exam.

•Conducting self-directed study viaweb-based programs, coaching or mentor-ing with peers, or individual study leadingto a demonstrated increase in job skillsand responsibilities.

For any of these activities, of course,evidence of completion and success mustbe compiled and presented for evaluationand validation by a committee of wirelessprofessionals. Accumulation of sufficientPDPs through a combination of variedactivities – not just a single activity –would be the basis for authorizing recerti-fication of the WCP credential for anoth-er five year period.

As noted, the recertification programis currently under development. Com-ments or suggestions regarding the pro-gram (e.g., additional activities that mightqualify for PDPs, or relative importanceof activities) are welcomed via email [email protected].


CERTIFICATION CORNER

THE NEED FOR RECERTIFICATIONBY ROLF FRANTZ

IENYCM2854.indd 1 2/16/11 3:25:05 PM

LYT-CERTIFICATION-MAR 2/18/11 3:36 PM Page 12

14

The IEEE Consumer Communicationsand Networking Conference, CCNC 2011,recently completed its 8th annual event inLas Vegas with hundreds of internationalconsumer communications and networkingexperts exploring the next generation oftechnologies designed to provide on-demand, anytime access to entertainmentand information anywhere in the world.

Held in conjunction with the 2011 Inter-national Consumer Electronics Show(CES), IEEE CCNC 2011 hosted severalkeynotes, panels, workshops, tutorials andresearch prototype demonstrations as well as nearly 350 technicalpapers highlighting the latest advances in industrial and academicresearch in a wide range of technologies related to consumercommunications and networking: home and wide area networks,wireless and wireline communications, cognitive networks, peer-to-peer networking, middleware and other applications enablingtechnologies, information and applications security, etc.

Prior to IEEE CCNC 2011, the conference began showcasingresearch prototypes at the IEEE Communications Society (Com-Soc) booth at CES located in the Las Vegas Convention Center.Throughout CES and then again at IEEE CCNC 2011, Telcordiademonstrated a research prototype of a real-time search tool thatcan enable users to save time, money, and bandwidth by preview-ing videos from different perspectives, rapidly determine theirappropriateness prior to launching streaming applications.

In addition, Drakontas, a New Jersey-based provider of geospa-tial tools, offered live demonstrations of its new SMYLE researchprototype, which allows user groups to share locations, text mes-sages, photos and graphical annotations in real-time environments.Additional advantages are the ability to collaborate on meetingtimes and experiences almost anywhere within a building complex,such as a theme park, through the use of the SMYLE mobilephone application or a standard PC web browser.

IEEE CCNC 2011 then officially commenced on Sunday, Jan-uary 9th with a full day of workshops dedicated to topics such as“Vehicular Communications System,” “Personalized Networks,Digital Rights Management Impact on Consumer Communica-tions,” “Social TV: the Next Wave, Social Networking (SocNets)”and “Consumer eHealth Platforms, Services and Applications.”

On the following morning, Dr. Simon Gibbs, IEEE CCNC2011 General Co-Chair and Principal Engineer at Samsung, gra-ciously welcomed all attendees. He thanked the conference’sorganizing committee members as well as conference patrons –Samsung, Nokia, HD-PLC, Telcordia, and Drakontas – for help-

ing to make IEEE CCNC the flagship Con-sumer Communications and Networkingconference. Afterwards, Dr. Kari Pulli ofthe Nokia Research Center in Palo Alto,California, addressed the forum about the“demise of film cameras and ascent of digi-tal cameras” during his keynote on “MobileComputational Photography.” As high-lighted by Dr. Pulli, this is a marketplacethat will continue to grow over the nextfive years with more than 150 million digi-tal cameras shipped and nearly 1.1 billioncamera phones sold internationally by

2014. Spurred by the introduction of real-time computationaltechnology that allows “instant image processing,” future digitalcameras will introduce numerous features that provide “full, low-level control of all camera parameters” as well as the ability to“rewrite the autofocus” function so as to combine images to cre-ate cleaner, clearer and brighter photos in seconds.

Following Dr. Pulli’s speech, IEEE CCNC 2011 then launchedthe first of two days of technical sessions, technology and applica-tions panels, and research prototype demonstrations in variousConsumer Communications and Networking areas including securi-ty and content protection, entertainment networking, automotivemultimedia, multiplayer networked gaming, next generation IPTV,social media, and personal broadcasting. Specific topics addressedthe “Dissemination of Information in Vehicular Networks,” SmartGrid Emerging Services,” Ecological Home Networks” and “Smart-phone Location-Aware Technologies.”

On Wednesday, the conference program had another full dayof keynotes, technical sessions and research prototype demon-strations. In the morning, Dr. Kiho Kim, Executive Vice Presi-dent and Head of Digital Media & Communication R&D Center,Samsung Electronics, began the proceedings with his addresstitled “A Future Mobile Outlook: Infrastructure, Device and Ser-vice.” In his speech, Dr. Kim mentioned that “In the past, theICT (Information and Communications Technology) industry’smegatrends - “being digital,” “being networked” and “beingmobile” - have led us to paradigm shifts such as the “Internetrevolution” of today.” He also said that “In the near future, nolater than 2020, new technology enablers in mobile devices andwireless access infrastructures will initiate another paradigm shift.New life care services, in addition to the legacy infotainment ser-vices, will be delivered via “intelligent, not just smart,” mobiledevices through an enhanced network that seamlessly utilizeslocal and personal networks.”

Later that evening at the conference’s Awards Banquet, Dr.


CONFERENCE REPORT

IEEE CCNC 2011 HIGHLIGHTS LATEST ADVANCES INANYTIME, ANYWHERE CONSUMER COMMUNICATIONS

Dr. Kiho Kim, Executive Vice President and Head of DigitalMedia & Communication R&D Center of Samsung Electron-ics, addressed attendees at the Wednesday Keynote Session.

Ben Falchuk demonstrated Telcordia’s latest multimedia visu-al search technology.

LYT-CONF REP-CCNC-MAR 2/18/11 3:30 PM Page 14


Monica Lam, a Professor of Computer Sciences at Stanford Uni-versity and the co-author of the “Dragon” book or as it is formallyknown, “Compliers, Principles, Techniques and Tools,” continuedthe discussion on future trends, while addressing “In Situ OnlineSocial Networking.” In her presentation, Dr. Lam urged attendeesto “ride the mobile computing wave” as she described mobilephones as “the perfect devices for changing the way we network”and “creating more peer-to-peer personal communications.” Dr.Lam also explored the development of new instantaneous groupcommunication systems based on existing email technologies,which will soon create “open federated social networks” and facil-itate the sharing of information “among pockets of friends” with-out the need to utilize third party proprietary infrastructures.

The second banquet keynote presentation was by Jean-Philippe Faure, who is the CEO of Progilon and affiliated withHD-PLC consortium. Jean-Philippe is the Chairman of theBroadband over Power Line (BPL) IEEE P1901 Working Group.The IEEE 1901 standard has been recently successfully complet-ed. The presentation focused on merits of the BPL technologythat enables Internet Access services as well as in-home, in-vehi-cle and in-airplane networking over power lines for support of abroad range of applications that include Smart Grid communica-tions, Web browsing, and entertainment. After the keynote pre-sentation, Jean-Philippe Foure was presented a plaque for hisachievements as the founding chair of the IEEE P1901 WorkingGroup and for successful completion of the Broadband overPower Line Standard IEEE 1901. The plaque was presented byAlex Gelman, who initiated this standardization project, onbehalf of the IEEE Communications Society President ByeongGi Lee, VP-Technical Activities Mark Karol, and ComSoc Direc-tor of Standards Curtis Siller.

The banquet also featured numerous honors. This included thepresentation of the Best Paper Award to Hosub Lee of SamsungElectronics for his paper on “An Adaptive User Interface Based onSpatiotemporal Structure Learning” and Special Mention Honor toGreg Elliott of MIT Media Lab for his paper on “TakeoverTV:Facilitating the Social Negotiation of Television Content in PublicSpaces.” The overall Best Student Paper Award was also presentedto Peter Vingelmann and Hassan Charaf of the Budapest Universi-ty of Technology and Economics as well as Frank H.P. Fitzek,Morten Videbæk Pedersen and Janus Heide of the Aalborg Uni-versity in Denmark for their paper on “Synchronized MultimediaStreaming on the iPhone Platform with Network Coding.”

IEEE CCNC 2011 concluded on Wednesday, January 12with a complete schedule of tutorials offering insights into sub-jects such as “State of the Art Research Challenges and P2PNetworking,” “Cognitive Radio Networks,” “4G - Next Genera-tion Mobile Applications,” “Wireless Mesh NetworkingAdvances and Architectures,” “Consumer Network Standardiza-tion,” and “Technologies and Applications for Connecting AllYour Electronic Devices with Personal Networks.”

As Dr. Robert Fish, CCNC Steering Committee Chairman,mentioned at the banquet, next year, the 9th Annual IEEEConsumer Communications Networking Conference will beginonce again with a preview of its comprehensive research demon-strations at the IEEE booth located within CES 2012. In addi-tion, the IEEE CCNC 2012 “Call for Papers” has already beenannounced with all interested parties urged to visithttp://www.ieee-ccnc.org/2012 for submission details. Ongoingconference updates can also be obtained via Twitter @IEEEC-CNC or by contacting Heather Ann Sweeney of IEEE ComSocat 212-705-8938 or [email protected].

CONFERENCE REPORT

Smart Measurement Solutions

Get to know your power supply!

Combining OMICRON Lab’s Bode 100, Vector Network Analyzer with the new Picotest Signal Injectorsenables you to perform high-fidelity measurements of:

Non-Invasive & traditional Stability

PSRR

Input & Output Impedance

Reverse Transfer

... and many other important power supply parameters in the range from 1 Hz - 40 MHz

For FREE application notes and more, please visit:www.omicron-lab.com & www.picotest.com/blog

Smart Measurement Solutions

IENYCM2857_1.indd 1 2/16/11 6:34:21 PM

LYT-CONF REP-CCNC-MAR 2/18/11 3:30 PM Page 15

16

IEEE GLOBECOM 2010 recentlyheld its most successful meeting in theconference’s 53-year history. Held from 6– 10 December 2010 in Miami, Florida,this annual flagship event of the IEEECommunications Society (ComSoc) setnumerous milestones including number ofattendees (2,500+); symposia and workshop paper submissions(4,614); lowest tutorial acceptance ratio (8%); highest numberof tutorial registrations with over 300 attendees per session;largest team of volunteers (8,500+) contributing to the confer-ence’s success; and the largest number of high-profile invitedspeakers (100+) delivering keynote and plenary sessions, techni-cal symposia, workshops, panels and tutorials (469).

Now considered one of the world’s premier communicationsconferences, IEEE GLOBECOM 2010 initiated its five-day pro-gram on Monday, December 6 with the first of two full days oftutorials and a record-setting 21 workshops. For the first time inconference history, each tutorial session was free-of-charge to allconference attendees, who were extremely pleased with the highdegree of diversity in the agenda and high-quality of presenta-tions. Dedicated to the most recent industry and academicadvancements, session topics ranged from adaptive wireless com-munications, wireless MediaNets and biologically-inspired com-munications to Femtocell networking, broadband wireless access,pervasive group communications and smart communications.

The conference then officially commenced on Tuesday, whenExecutive General Chair Dr. Kia Makki was presented anAmerican flag previously flown at the Florida state capitol build-ing in Tallahassee during the morning introductions. Immedi-

ately afterwards, IEEE ComSoc PresidentDr. Byeong Gi Lee highlighted the con-ference’s general theme of “Mobile Inter-activity,” while thanking the ongoingefforts of the society’s 48,000 global mem-bers, who are “constantly researching,advancing and working to implement

global communications technologies that enhance humanity andeveryday life.” Dr. Haohong Wang, the conference’s technicalprogram chair, followed these comments by reviewing the agen-da of the largest technical program ever presented at IEEEGLOBECOM since its founding in 1957.

Following these introductory remarks, Dr. Regina E. Dugan,Director of Defense Advanced Research Projects Agency(DARPA) then challenged all IEEE GLOBECOM 2010 partici-pants to overcome the fear of failure, while reminding everyonethat one person can forever change the quality of life for 6.9 bil-lion people worldwide. She also continually reinforced this mis-sion by stating that life is a miracle and the magic that can makeall our lives better is needed more now than ever before.

During her keynote, Dr. Dugan also noted how “diverse per-spectives can lead to innovations that make difficult solutionsseem easy.” This includes the use of “cognitive clouds” andsocial networks that can help surmount dilemmas — for in theend diversity of thought will always surpass ability.

After this speech, IEEE GLOBECOM 2010 then proceededwith a three-day program filled with plenary addresses, techni-cal symposia, business & technology forums, awards andexhibits designed to explore the entire range of communica-tions technologies. This included a new plenary format thatoffered the visionary thoughts of 34 noted scientists and indus-try executives; new demo program featuring 20 demonstrationsshowcasing the latest research achievements of internationalexperts working within a wide range of communications fields;new funding forum with invited speakers representing severalimportant government funding agencies such as NSF, DARPA,ARO, ONR, AFOSR and DHS; and new early-bird StudentAward encouraging productive student authors to submit highquality technical papers.

Among the highlights of Tuesday’s agenda were also severalhigh-level forums and presentations detailing the latest researchand application advancements in areas such as “Faster, Greenerand More Frugal Network Transport,” “The Channel AccessConundrum,” “Dynamic Spectrum Access,” “An Executive’s Per-spective of Tomorrow’s Technology” and “Next GenerationInternet.” For instance, the event’s Cloud Computing Forumreviewed the latest services and processes for not only enhancingthe consistency of healthcare delivery systems throughout theworld, but also the implementation of technologies that turn cap-ital expenditures to operational expenses, greatly increase com-petencies and simplify the manageability and monitorability ofinternal infrastructures. In addition, the day’s Wireless NetworkForum also explored “New Frontiers for Wireless” including theexplosive growth of mobile video traffic, which is expected toexpand to 500,000 worldwide subscribers by 2015.

Additionally, the first of three days of technical symposia alsobegan on Tuesday with a complete array of sessions dedicated toenergy savings and power control protection protocols, privacyprotection, Internet security, network coding and attacks, cogni-tive radio, sensing, estimation and communication feedback. Inall, this three-day technical symposia program stretching fromTuesday through Thursday would include approximately 300 ses-sions composed of nearly 1,300 paper presentations ranging in


CONFERENCE REPORT

IEEE GLOBECOM 2010 COMPLETESMOST SUCCESSFUL MEETING IN 53-YEAR HISTORY

Poster sessions provided networking and educational opportu-nities.

Business & Tech Forums were well received by attendees.

LYT-CONF REP-GLOBECOM-MAR 2/24/11 12:34 PM Page 16


subject from the latest routing protocols to resource manage-ment and peer-to-peer technologies for communication services.

Other Tuesday highlights were delivered by the AnnualAwards Luncheon, which honored the career and service con-tributions of IEEE ComSoc members and volunteers, and theWelcome Reception & Expo Opening held in the Hyatt Regen-cy’s Riverfront Hall later that night. During the luncheon’s cer-emonies, IEEE ComSoc Awards Chair H. Vincent Poor andByeong Gi Lee presented to deserving honorees such asLeonard J. Cimini, Jr. of the University of Delaware, whoreceived the IEEE Communications Society Donald W. McLel-lan Meritorious Service Award;” Abbas Jamalipour of the Uni-versity of Sydney, who was presented the IEEECommunications Society Harold Sobol Award for ExemplaryService to Meetings & Conferences; Larry Greenstein of Win-Lab, Rutgers, who received the IEEE Communications SocietyJoseph LoCicero Award for Exemplary Service to Publications;Roberto Saracco of Telecom Italia, who earned the IEEEComSoc/KICS Exemplary Global Service Award; and TadashiMatsumoto, Milicia Stojanovic, Zoran Zvonar and Wei Su, whoall were presented 2010 IEEE Fellow Awards.

IEEE GLOBECOM 2010 then concluded on Tuesdayevening with the Welcome Reception, which began when Execu-tive General Chair Kia Makki and IEEE ComSoc PresidentByeong Gi Lee cut the traditional gateway ribbon. Once insidethe hall, hundreds of conference attendees feasted on a broadselection of local delicacies, desserts and beverages, while con-necting with old friends and making new ones. Other activitiesincluded browsing the numerous booths of exhibitors such asTelcordia, Cambridge, Elsevier, Wiley and Springer as live reg-gae and calypso music swirled through the hall. The IEEE Com-Soc Pavilion also offered a host of society information includingmembership and networking benefits as well as details onupcoming IEEE ICC and IEEE GLOBECOM events.

On the following morning, the IEEE GLOBECOM 2010agenda renewed with the keynote presentation of YoshihiroObata, Executive Vice President and Chief Technology Officerof eAcess Ltd. and Executive Vice President of eMOBILE Ltd.Introduced by Dr. John Thompson, the conference’s technicalprogram co-chair, Yoshihiro proceeded to highlight the historyof Japan’s broadband industry and his own company’s growth,which rose from a two-person business in 1999 to a nationalenterprise that currently has nearly 3,000 individuals and $2.5billion in sales.

Beset within a deeply competitive environment, Yoshihirodetailed the market’s challenges and his company’s successfulgrowth strategies in the world’s second largest telecom market-place. Realizing Japan’s decreased dependence on fixed telecomservices, eAccess and eMOBILE has invested heavily in mobile

broadband delivery with the goal of offering 24-hour-a-day accessto 80 to 90 percent of the population in the near future. Whilethere were only 300,000 corporate users in the entire nation 10years ago, today Japan’s mobile broadband industry generates$17 billion in revenue and represents 3.4 percent of the country’sGDP. By 2016, estimates also predict that Japan’s mobile broad-band market will amass approximately 16 million subscribers.

In addition, Wednesday’s well-attended IPv6 Forum beganwith the very real fact that the entire spate of global IPv4 Inter-net addresses will be exhausted within the next two years and thetransition to IPv6 has “already taken five years too long.” As aresult, many of the session’s speakers concurred that “if you donot transition to IPv6, you will not be a player in the Internet.”

Subsequently, the effort to integrate IPv6 into existing andnew infrastructures has seen “more activity in the past year thanthe previous 10 years combined.” Although in the early stages,this includes providing consumers with IPv6 connectivity optionsin the very near future.

Furthermore, the forum’s distinguished panel of expertsoffered advice to any and all service and content providers deal-ing with IPv4 issues. First, “Don’t wait for customers to ask forIPv6. You won’t have the time to react.” Next, “Never pay forIPv6 beyond the service for IPv4. Run from vendors that want tocharge extra.” And, education is key. Dispel IPv6 myths andfears by investing in training and classes that inform all interestedparties on the differences between IPv4 and IPv6, while advisingon their similarities.

Other sessions such as the “Wireless CommunicationsForum” dealt with the challenges of servicing the world’s mobiledevice users, which already includes 4.1 billion subscriptions and

CONFERENCE REPORT

Participants experienced live demos for the first time in con-ference history.

Executive General Chair Kia Makki was presented an Ameri-can flag previously flown at the Florida state capital building.

Executive General Chair Kia Makki and ComSoc PresidentByeong Gi Lee opened the Welcome Reception.


the “Smart Grid and Green Technology Forum” that offered ahost of options for enabling cost savings and lowering energyuse through the implementation of smart appliances and meterson a global scale. Additional Wednesday highlights included ple-nary addresses on highly sensitive areas like “Energy-EfficientWireless Applications,” “Multidimensional Convergence ofBroadband Access Technologies,” “Physical Layer Security inWireless Networks” and “Smart Phones: A Revolution inMobile Computing.”

On Thursday, Dr. Niki Pissinou, the conference’s operationchair, introduced Dr. Frederica Darema, the director of the U.S.Air Force Office of Scientific Research, to open the day’s pro-ceedings. Dr. Darema’s keynote detailed the U.S. Air Force’sconcept of network system science and the fundamental aspectsof its practical impact. Lauded for her comprehensive mathemat-ical approach to fostering synergism across widespread scientificdisciplines, Dr. Darema’s complex network of sensoring andinformational gathering has helped create encompassingadvances in tornado monitoring, aircraft design, oil exploration,semiconductor manufacturing and electrical power grid enhance-ments, among others, over the past five years.

Throughout the day, IEEE GLOBECOM 2010 participantswere then feted to another full schedule of technical symposia,plenary speeches and informational forums highlighting the areasof design & development, multimedia communications, wirelessnetworking, intelligent transportation and 4G architecturaladvancements. For example, presenters within the QoS Multime-dia Forum offered insights and research findings related to thedevelopment of processing techniques that will help reduce traf-fic congestion as well as enhance the transmission and storage ofreal-time multimedia experiences. The Multimedia Communica-

tions Plenary then offered attendees a glimpse of next steps,which are in progress and starting to digitally teleport individualsand groups of people to virtual meeting arenas, while reconnect-ing families, delivering healthcare to remote regions and savingenergy and commuting costs.

Several other popular forums also featured the latest tech-niques for furthering the latency, spectral efficiency, antenna sup-port and ultimately the performance of next generation mobilebroadband wireless networks as well as the initial thrusts todevelop Intelligent Transportation Systems (ITS) that shortendriving times, deliver quicker medical aid and reduce road-relat-ed injuries. According to session speakers, the implementation ofvehicular communications should be a governmental and corpo-rate mandate worldwide due to its ability to save hundreds ofthousands of lives annually in addition to reducing drive timeand costs, and delivering another full array of fee-based servicesto commuters.

IEEE GLOBECOM 2010 then completed the most success-ful conference in its history with another full agenda of work-shops on Friday, December 10. Notable technology expertsrepresenting nearly every phase of voice, data, image, and multi-media communications supervised learning sessions in numerousareas that included separation and overlay networks, P2P livestreaming systems, self-managed future Internets, virtualmachine migration, multimedia computing and real-time Internettrafficking.

Buoyed by the tremendous response to last year’s meeting,IEEE GLOBECOM 2011 planning is already underway. Forinformation on this global event, which will be held 5 – 9 Decem-ber 2011 in Houston, Texas, please visit the conference web siteat www.ieee-globecom.org/2011.


CONFERENCE REPORT

Remcom Consulting is an ideal solution for those organizations that need electromagnetic expertise without an ongoing need for in-house EM simulation software. Remcom provides EM modeling experts, software developers, and high performance computational resources to meet your most challenging EM simulation analysis needs.

Commercial Consulting Government Contracting Custom Software Development

Learn more at:www.remcom.com/consulting

Are You Missing Some

Let Remcom Consulting Complete It

Pieces to the Puzzle?

+1.888.7.REMCOM (US/CAN)+1.814.861.1299www.remcom.com

Electromagnetic Simulation Solutions

IENYCM2871.indd 1 2/22/11 10:29:26 PM


20

The IEEE Communications Society(ComSoc), the leading worldwide pro-fessional organization dedicated to theadvancement of communications tech-nologies, will emphasize the ongoingneed to reduce global greenhouse gasemissions by hosting the first annual IEEE Online Conferenceon Green Communications (GreenCom) totally online fromSeptember 26 – 29, 2011. Dedicated to the latest advances inenergy-efficient communications and green technologies,IEEE GreenCom’11 will enable attendees from around theworld to engage in discussions on the newest networking,energy management and smart grid communications solutionswithout travel and from the comfort of their own home and/orwork environments.

Webcast to international attendeesby IEEE ComSoc and then publishedat IEEE Xplore, IEEE GreenCom’11was specifically designed to addressglobal warming developments and itssocietal impact in an alternative, eco-logical conferencing model that reach-es broader audiences, offerstime-flexible participation and providesnear-physical experiences in a power-ful, virtual forum where “energy effi-ciency is discussed energy-efficiently.”Other distinct features include theability of speakers to present researchlive and then answer audience ques-tions with the aid of moderators.

For more information on IEEEGreenCom’11, including “Call for

Paper” details, interested researchers,academics and industry experts areurged to visit http://www.ieee-green-com.org.

With a deadline of March 20, 2011,the conference’s peer-review board is

currently accepting paper submissions on a wide range of top-ics covering energy-efficient fixed and wireless communica-tions and networking, communications technologies forgreen solutions, and smart grid communications. Specific top-ics of interest include, but are not limited to:•Energy-efficient protocols, extensions, transmission and net-

working technologies.•Energy-efficient communications management.

•Energy-efficiency in mobile, home,sensor, and vehicular networks, andin data centers.

•Green communication architecturesand frameworks.

•Solutions for energy-efficient transport,logistics, industries, and buildings.

•Communication networks for SmartGrids and smart metering.In addition, anyone interested in

networking with colleagues or otherattendees via Twitter, Facebook,LinkedIn as well as receiving confer-ence updates should visithttp://www.ieee-greencom.org on aregular basis or contact Heather AnnSweeney of the IEEE CommunicationsSociety at 212-705-8938 or [email protected].


CONFERENCE PREVIEW

2011 IEEE ONLINE CONFERENCE ON GREEN COMMUNICATIONSSEPTEMBER 26 – 29, 2011

IEEE COMSOC EMPHASIZES NEED TO REDUCE GLOBAL GREENHOUSE GAS EMISSIONS BY HOSTINGTHREE-DAY EVENT ON ENERGY-EFFICIENT COMMUNICATIONS & GREEN TECHNOLOGIES TOTALLY ONLINE

Efficient and FlexibleLive and recorded sessions

to fit your schedule

EcologicalReduce global

greenhouse gas emissions

Cost-EfficientLow registration fees,

no travel costs

High-Visibility and High-QualityPeer-reviewed papers

published in IEEE Xplore®

LYT-CONF REP-GREENCOM-MAR 2/18/11 3:32 PM Page 20


PRODUCT SPOTLIGHTS

Silicon LabsLearn how to simplify your timingdesign using glitch-free frequencyshifting to address low-power designchallenges and the complexity of gen-erating a wide range of frequencies inconsumer electronics applicationssuch as audio, video, computing orany application that requires multiplefrequencies. Download this in-depthwhite paper from Silicon Labs.

http://www.silabs.com/frequency-shifting

RemcomRemcom’s Wireless InSite® is site-specific radio propagation software forthe analysis and design of wirelesscommunication systems. It providesaccurate predictions of propagationand communication channel charac-teristics in complex urban, indoor,rural and mixed path environments.Applications include wireless links,antenna coverage optimization, andjammer effectiveness.

http://www.remcom.com/wireless-insite

SynopsysDesigners today need high-level syn-thesis optimization technologies thatdeliver high quality of results forFPGA and ASIC while enablingrapid exploration of performance,power, and area. Synphony High-Level Synthesis (HLS) tools providean efficient path from algorithm con-cept to silicon and enable greaterdesign and verification productivity.

http://www.synopsys.com

JFW IndustriesSince 1979 JFW Industries has beena leader in the engineering andmanufacturing of attenuators, RFswitches, power dividers, and RFtest systems. We deliver customsolutions at catalog prices. Withmore than 15,000 existing designs,the company can provide applica-tion specific devices to solve almostany RF attenuation and switchingproblem.

http://www.jfwindustries.com

GL CommunicationsGL’s PacketExpert™ is a portablequad port (electrical and optical)tester that can perform independentEthernet/IP testing at wirespeed. Itincludes bi-directional RFC 2544 test-ing, wirespeed BERT, frame capture,and loopback. It takes the confusion

out of Ethernet testing at all protocol layers, from raw Ethernet frames toIP/UDP packets. It can be used as a general purpose Ethernet performanceanalysis tool for 10, 100, and 1 Gbps Ethernet local area networks. Key perfor-mance indicators include bit error count and rate, frame loss, sync loss and errorfree count/seconds, throughput, latency, frame count/rate and more. PacketEx-pert™ further supports IPV4, IPV6, smart loopback, user-defined loopback, con-figurable headers for MAC/IP/UDP, error insertion, sequence numbergeneration, and detail reports in PDF format.

http://www.gl.com/packetexpert

Anritsu CompanyAnritsu Company introduces theMS272xC Spectrum Master series thatprovides the broadest frequency rangeever available in a handheld spectrumanalyzer. Providing frequency cover-age up to 43 GHz in an instrumentthat weighs less than 8 lbs., the

MS272xC series is also designed with an assortment of applications to test theRF physical layer, making it easier than ever for field technicians, monitoringagencies and engineers to monitor over-the-air signals, locate interferers, anddetect hidden transmitters.

http://www.us.anritsu.com

SPOT PAGE-11-03 2/18/11 3:41 PM Page 23


NEXT-GENERATION BASE STATIONANALYZER

Anritsu Company

The MT8222B BTS Master is Anrit-su’s next-generation handheld base sta-tion analyzer that supports 4Gstandards now being deployed, as wellas installed 2G/3G networks. Combin-ing the inherent advantages of the BTSMaster platform with new measurementcapability, the MT8222B provides fieldengineers and technicians with alightweight, handheld analyzer that canaccurately and quickly measure all thekey wireless standards, including LTE,WiMAX, WCDMA/HSDPA, CDMA/EV-DO and GSM/EDGE.

A 20 MHz demodulation capabilityhas been designed into the platform ofthe MT8222B BTS Master, allowing theinstrument to measure LTE andWiMAX signals. Additionally, theMT8222B features a 30 MHz Zero-Span IF output for external demodula-tion of virtually any other widebandsignal. For comprehensive receiver test-ing, a Vector Signal Generator optionis available that covers 400 MHz to 6GHz and can generate two simultane-ously modulated signals, plus noise.

Accurate two-port cable and anten-na analysis from 400 MHz to 6 GHzcan be conducted with the MT8222BBTS Master. All key measurements –including return loss, cable loss, VSWRand distance-to-fault (DTF) – can bemade with the compact analyzer. It canalso measure gain, isolation, and inser-tion loss, to verify sector-to-antennaisolation.

The MT8222B BTS Master also hasspectrum analysis capability typicallyfound in a bench-top instrument. Inspectrum analyzer mode, the instru-ment has a wide frequency range of 150kHz to 7.1 GHz, low phase noise of typ-ically -100 dBc/Hz @ 10 kHz offset, lowdisplayed average noise level (DANL)of typically -163 dBm in 1 Hz RBW,and wide dynamic range of >95 dB in 1 Hz RBW.

A number of options are available toconfigure the MT8222B BTS Master tosuit the specific measurement require-ments of any field test application.Among the options are a 140 MHz ZeroSpan IF output with 30 MHz IF band-width, a GPS receiver that works withthe analyzer’s standard 3.3/5 V BiasTee that enables connection to BTS siteGPS antennas, and frequency outputand input with 25 ppb accuracy.

The MT8222B BTS Master featuresAnritsu’s Master Software Tools (MST)and Line Sweep Tools (LST) compre-hensive data management and analysis

software that save significant time whengenerating reports.

Designed specifically for the field,the MT8222B BTS Master measuresonly 315 x 211 x 94 (mm), weighs 4.9kg, and is battery operated. It incorpo-rates Anritsu’s field-proven design, sothe MT8222B BTS Master can with-stand the harsh environments in whichit is used. http://www.anritsu.com

SKYWORKS DEBUTS 6- AND 7-BITDIGITAL ATTENUATORS

Skyworks

Skyworks has introduced two digitalattenuators with superior attenuationaccuracy for base station, cellular-headend, repeater, test equipment, and fem-tocell manufacturers. The 6-bit device

has an 0.5 dB least significant bit (LSB),and the 7-bit device has a 0.25 dB LSB.These attenuators provide precise con-trol over multi-standard radio transmit-ters and receivers, allow the user toconfigure the device for serial or paral-lel control, and do not utilize blockingcapacitors so that low frequency opera-tion is possible.

http://www.skyworksinc.com

WIRESPEED ETHERNET/PACKET TESTER

GL Communications Inc.

GL Communications has released itsenhanced PacketExpert™, a quad portwirespeed Ethernet/packet tester. Packe-tExpert™ supports four electrical Ether-net ports (10/100/1000 Mbps) and twpoptical ports (1000 Mbps). It connects tothe PC through a USB 2.0 interface. EachGigE port provides independent Ether-net/IP testing at wirespeed for applica-tions such as BERT, RFC 2544, andmany more. The application has beenenhanced with the following features:

•User-defined or auto-negotiatedelectrical ports that can operate at10/100/1000 Mbps line rates in fullduplex mode; optical ports can operateat 1000 Mbps line rate in full duplexmode only.

•Resolve utility to easily configureMAC address (through ARP).

•PING utility to check remote IPaddress availability.

The application takes confusion outof Ethernet testing at all protocol lay-ers, from raw Ethernet to IP/UDP pack-ets. It can be used as a general purposeEthernet performance analysis for10/100 Mbps or 1 Gbps Ethernet localarea networks. Two of the four portshave both electrical and optical inter-faces, enabling testing on optical fiberlinks also. http://www.gl.com

INSTRUMENT-GRADE BROADBANDMICROWAVE POWER AMPLIFIERS

Giga-tronics Incorporated

The GT-1020A and GT-1040A instru-ment-grade broadband microwave poweramplifiers from Giga-tronics cover 100MHz to 20 GHz and 10 MHz to 40 GHzrespectively, with flat frequencyresponse, low noise figure, and low har-monics. Designed using broadbandMMIC technology, these amplifiers typi-

cally provide 1/2 Watt (+27 dBm) at 20GHz and 1/4 Watt (+24 dBm) at 40GHz with > 25 dB gain and < 6 dBnoise figure. Gain flatness is typically +/-2.5 dB over the full frequency range.

The Giga-tronics GT-1020A, 20 GHzand GT-1040A, 40 GHz amplifiers weredesigned in response to customerrequests for higher power with flat fre-quency response and low noise. Theamplifiers are easily used in R&D labapplications and manufacturing auto-mated test systems to overcome powerlosses from a signal generator or when-ever higher power is required. Thesmall size and light weight make themideal for Lab bench applications, whilethe ability to place the amplifier close tothe device under test (DUT) minimizescable loss for more optimal testing.

The Giga-tronics GT-1020A andGT-1040A microwave power amplifiersare ideal companions to the Giga-tron-ics 20 GHz and 40 GHz Microwave Sig-nal Generators. The amplifiers alsofeature high reverse isolation, excellentinput and output match and the longlife and reliability of solid-state technol-ogy. http://www.gigatronics.com

NEW PRODUCTS

LYT-PRODUCTS-MAR 2/18/11 3:40 PM Page 24

Global Communications Newsletter • March 2011

G l o b a l

N e w s l e t t e rMarch 2011

1

Over 150 delegates from 29 countries representing about80 different organizations met for the 14th ICIN conferenceto discuss how new Internet and telecommunications tech-nologies blend to deliver rich new services worldwide. ICIN2010 took place in Berlin, Germany, on 11–14 October, withthe technical co-sponsorship of the IEEE and IEEE Commu-nications Society, and the support of many patrons includingDeutsche Telekom Laboratories, Ericsson, Nokia SiemensNetworks, Huawei, Alcatel-Lucent, EICT, Orange, and BerlinPartner. ICIN has a 21-year history of anticipating the keytrends in the telecommunications services industry, and show-casing technologies and architectures that have become vitalto the delivery of new services.

The conference began with two tutorials, one focusing onidentity management in web 2.0 and telecommunications, and theother covering service enabling technologies in fixed and mobilenetworks. The tutorials laid the foundation for key aspects ofnext-generation networks discussed throughout the week.

The conference was formally opened by Max Michel (FranceTelecom), Chair of the ICIN 2010 Technical Programme Com-mittee, and Heinrich Arnold representing Deutsche TelekomLaboratories, the local host. A series of keynotes accented thetheme for the conference. Malcolm Johnson of the ITU deliv-ered a video presentation concerning standards needed toachieve consistent user experience. Philip Kelley of Alcatel-Lucent presented a keynote on the market impact of theiPhone and similar devices. Bengt Nordström of Northstreampresented a keynote on the critical issue of increasing operatorrevenue to support the cost of handling the rapid growth ofmobile data traffic, given the impact of Internet players oncommunications. Thomas Michael Bohnert of SAP presented akeynote highlighting recent research on a next-generation Inter-net and the need for industry cooperation.

The second keynote session focused on host country per-spectives. Thomas Aidan Curran of Deutsche Telekom sug-gested that carriers should become more like softwarecompanies. Sigurd Schuster from Nokia Siemens Networkspresented the challenges of managing user identities, highlight-ing the role carriers can play. In a final keynote, Felix Zhangof Huawei discussed the migration of applications to cloudcomputing and the role that the network can play in enhancinguser experience. The keynote sessions were capped by a recep-tion (sponsored by Deutsche Telekom Laboratories and BerlinPartner) attended by both ICIN attendees and leading Internetand communication entrepreneurs from the Berlin area.

The main body of the conference took place on Tuesdayand Wednesday at the Park Inn Berlin-Alexanderplatz, and

included 55 presentations in 14 sessions in two parallel tracks,plus eight demonstrations and poster presentations availablefor viewing both days and during an evening reception onTuesday.

The conference programme included a session on theindustry impact of marketplaces and app stores, proposingthat applications are much more important in motivatingdevice purchases and service subscriptions than generatingnew revenue, and that they have the potential to profoundlychange the user experience with networks. Another sessionpresented three practical applications of IP Multimedia Sub-sytem (IMS) and discussed the application of IMS principlesto all-IP networking going forward. A session on content dis-tribution addressed the business and technical challenges ofcontent delivery, including both new solutions to the problemof efficient content delivery to many types of devices and newchallenges resulting from content sharing in social networking.A session on privacy highlighted the need for greater usercontrol over private information, and for great care by net-work operators and application builders to avoid inadvertentrelease of private user information.

A session on home networking discussed home networksecurity challenges, secure sharing across home networks, andnew types of home networking services. A session on the keyservice enabler context management highlighted the need toprovide common access to the many sources of context infor-mation to enhance user experience. A session on service com-position and several of the demonstrations illustrated theprogress of service composition from theory to practice tobecome important in generating new services revenue. A ses-sion on social networking described the combination of socialnetworking and multimedia, and the use of social networkingas a viral marketing tool.

On Wednesday the conference opened with sessions on “Xas a service,” and intersystem and interdevice mobility. TheXaaS session covered the migration of many applications to“cloud”-based implementations, and the natural role of thenetwork in enhancing service implementation and delivery,while the mobility session focused on the Third GenerationPartnership Project (3GPP) Enhanced Packet Core (EPC)architecture, identifying both strengths and gaps in supportingservice and device migration. Sessions on content delivery andsensor networks filled out the morning. The content deliverysession addressed some novel aspects of content delivery,including delivery of content-associated advertising and pre-senting user content recommendations. The sensor network

ICIN 2010 — Weaving Applications into the Network Fabric:The Transformational Challenge

By Warren Montgomery, Insight Research, and Chet Mc Quaide, StraDis Consulting, USA

(Continued on Newsletter page 4)

LYT-NEWSLETTER-MAR 2/18/11 3:39 PM Page 25


RNDM 2010, the Second International Workshop on Reli-able Networks Design and Modeling, held in Moscow, Russia,19–20 October, 2010, was a two-day event organized by GdanskUniversity of Technology, Poland, in cooperation with the Uni-versity of Pittsburgh, United States, and Wroclaw University ofTechnology, Poland. The workshop was technically co-spon-sored by IEEE Communications Society and IFIP TC6 WG6.10 (Photonic Networking Group). It was collocated with the2nd International Congress on Ultra Modern Telecommuni-cations and Control Systems (ICUMT 2010).

RNDM ’10 followed up on a very successful first editionthat took place last year in St. Petersburg, Russia. The objectivewas to provide a forum for researchers from both academia andindustry to present high-quality results in the area of reliablenetworks design and modeling. Special attention was paid tonetwork survivability.

Submitted papers were extensively reviewed by 45 membersof the TPC and 18 external reviewers. The 23 accepted paperswere organized into six technical sessions: Fault Managementand Control in Survivable Networks; Survivability of Anycast,Multicast, and Overlay Networks; Fast Service Recovery; Meth-ods for Measurement, Evaluation, or Validation of Survivabili-ty; Design of Dedicated/Shared Backup Paths; and Models andAlgorithms of Survivable Networks Design and Modeling.

The workshop program was enriched by the keynote talks oftwo speakers: Professor Tibor Cinkler (Budapest University ofTechnology and Economics, Hungary) and Professor James P. G.Sterbenz (University of Kansas, United States, and Lancaster

University, United Kingdom). A panel discussion entitled “FutureResearch in Reliable Networks” chaired by Professor MauriceGagnaire (Telecom ParisTech, France) concluded the event.

Conference papers were published in printed proceedingsand are also available at IEEE Xplore. Authors of the toppapers have been invited to publish extended versions of theircontributions in a special issue of Telecommunication SystemsJournal (Springer). The next edition will be held in Budapest,Hungary, on 5–7 October, 2011. More information on theworkshop may be found at http://www.rndm.pl

IEEE RNDM 2010 Workshop, Moscow, RussiaBy Jacek Rak, Gdansk University of Technology, Poland, David Tipper, University of Pittsburgh, USA,

and Krzysztof Walkowiak, Wroclaw University of Technology, Poland

Broadcasters and government officials pledged to ensure asmooth digital transition in southeast Europe, and called onEuropean institutions for financial support in the interests ofEuropean cohesion. This was a key message of a two-dayroundtable conference entitled “Addressing Europe’s DigitalDivide: Towards Sustainable Public Service Media in SouthEast Europe.” Organized by the Regional Cooperation Council(RCC) Secretariat and the Geneva-based European Broadcast-ing Union (EBU), the meeting was held in Sarajevo, Bosniaand Herzegovina, on 14–15 October, 2010. It was attended bymore than 50 general directors, media experts, and govern-ment, broadcast and regulatory officials.

In a “way forward to 2020” signed by EBU Vice-PresidentClaudio Cappon and RCC Secretary General Hido Biscevic,the conference participants agreed:

•To establish an enduring cooperation to ensure the sustainabil-ity of all public service broadcasters in southeast Europe by 2020

•To promote the values and principles of public service media,as recognized by the Council of Europe and the European Union

•To call on the European Union to support these goals politi-cally and financially under its aim to guarantee European cohesion

The conference highlighted the need for the public broadcast-ers, relevant ministries, and regulators in southeast Europe to joinforces to ensure that the region does not lag behind the rest ofEurope in meeting deadlines for digitalization and analogueswitch-off. The participants agreed that the region’s public broad-casters in the digital era need to remain key actors in the evolvingknowledge society, provide reliable information, quality educa-

tional, cultural and entertainment programmes, and be motors forregional development and investment in the creative industries.

Speaking at the conference, Minister of Communicationsand Transport of Bosnia and Herzegovina, Rudo Vidovic, toldthe conference that digitalization would create a “free andopen media market. This momentum needs to be seized.” Mr.Biscevic said sustainable public service media were vital for thecountries of southeast Europe as a whole, and for the Euro-pean Union they hope to join. But he said their role was threat-ened by a lack of investment in infrastructures “and also inhuman capital.” In an opening keynote speech, EBU DirectorGeneral Ingrid Deltenre said public service broadcasters need-ed to play a leading role in the digitalization process in south-east Europe as they have played elsewhere on the continent.

Conference speakers included Peter Karanakov, ExecutiveDirector, MKRT; Natasa Vuckovic Lesendric, Assistant Minis-ter for the Media, Ministry of Culture of Serbia; Marija Nem-cic, Deputy HRT General Manager for International Relations;Bledar Meniku, Ministry for Innovation, Information Technol-ogy and Communication of Albania; Maria Luisa FernandezEsteban, Directorate General for the Information Society andthe Media, European Commission; and Oliver Vujovic, Secre-tary General of the South East Europe Media Organisation.

The participants comprised senior representatives of govern-ments, broadcasters, and regulators from Albania, Bosnia andHerzegovina, Croatia, Greece, Moldova, Montenegro, Serbia,Slovenia, the Former Yugoslav Republic of Macedonia, andTurkey, as well as the EU and other relevant institutions.

Addressing Europe’s Digital Divide:Toward Sustainable Public Service Media in South East Europe

Roundtable International Conference, Sarajevo, 14-15 October 2010By Dinka Zivalj, Regional Cooperation Council, and Kerim Kalamujic, University of Sarajevo, Bosnia and Herzegovina

Presentation of the best paper award (left to right: Jacek Rak,Gdansk University of Technology, Poland, and Wouter Tav-ernier, IBBT, Ghent University, Belgium).


Global Communications Newsletter • March 2011 3

The 3rd International Conference on Advanced InfocommTechnology (ICAIT 2010) was held from 20–23 July, 2010,Hainan, China.

ICAIT is a three-day event comprising keynote speeches byboth leading academic researchers and industrial experts,focused symposia, technical oral presentations, and updates byindustrial presenters. This event provides a platform to intro-duce advanced infocomm technologies that will shape the nextgeneration of information and communication systems andtechnology platforms.

The conference was hosted by Hainan University in collab-oration with Guangxi University, HoChiMinh City Universityof Technology, Huazhong University of Science and Technol-ogy, and Ningbo University. IEEE Communications SocietyShanghai Chapter was the technical co-sponsor.

ICAIT has been a yearly event since 2008 to bring togetherresearchers, scientists, engineers, academicians, and studentsall around the world to share the latest updates on new tech-nologies that will enhance and facilitate the next generation ofinformation and communications systems and technologies.ICAIT forges links between nations, and builds bridgesbetween the academic and non-academic communities, andbetween government and non-government organizations.

Submissions from about 20 countries for technical presen-tations were received for ICAIT 2010. The papers wentthrough peer reviews, nominally two or more reviews perpaper, to ensure quality. We were highly honored to havetwo renowned keynote speakers, Prof. Muriel Medard fromMIT, United States, and Prof. Anthony TS Ho from SurreyUniversity, United Kingdom, to share with us their insight andperspectives on current trends, convergence technologies, andthe strategic updates.

ICAIT 2010’s venue, Hainan, the second largest oceanisland and smallest land province in China, is located at the

Report on the 3rd International Conference onAdvanced Infocomm Technology (ICAIT 2010)

20 – 23 July 2010, Hainan, ChinaBy Xinwan Li and Yikai Su, Shanghai Jiaotong University, China

Jitel and Telecom I+D: Spanish Rendezvous Points for TLC ProfessionalsBy Pilar Manzanares-Lopez and Josemaría Malgosa-Sanahuja, Spain

The 9th Conference on Telematic Engineering (IX Jornadasde Ingeniería Telemática, JITEL 2010) took place from 29September to 1 October in the University of Valladolid, Spain.This conference tries to be a propitious forum for the Spanishresearch groups on networking and telematic services. It tries toencourage both the interchange of experiences and results andthe cooperation among the researching groups working in thisfield of knowledge.

The origin of this conference dates back to 1997 when, coin-ciding with the centenary of the Telecommunication Faculty ofBilbao, it took place in the University of the Basque Country.Since then, JITEL was organized biannually until 2005 and, dueto its great success, it has been held annually since 2006. Eachedition takes place in a different city, at Spanish universities allaround the country.

JITEL is co-sponsored by the IEEE Spanish Section. Tradi-tionally, the best papers are published in IEEE America LatinaTransactions Magazine. In addition, the best two papers relatedto telematic engineering education will be published in a bookcalled TIC Aplicadas a la Ingeniería (TICAI), an initiative ofthe Spanish IEEE Education Society.

In this last edition of JITEL, with the challenge of adaptingthe current studies to the new European Higher EducationArea, the 1st Symposium on Educational Innovation on Telem-atic Engineering (I Jornadas de Innovacion Educativa en Inge-nieria Telematica, JIE 2010) was created. This is a new forumwhere university staff can meet and exchange experiences ininnovations in networking and telematic teaching, and also inthe use of information and communication technologies in high-er education in general and this technical area in particular.

JITEL 2010’s schedule was coordinated with the XX Tele-com I+D conference, with part of the session program coincid-ing with it. Telecom I+D is a meeting point for members of theacademic world (universities and the main research centers)and members of the business world (national and internationalleading companies in this sector) to advance the interchange ofexperience and dissemination of knowledge, focused on thepromotion of technological innovation. Under the lemma “20Years Leading the Innovation to Change the Future,” TelecomI+D 2010 was focused on the commitment of ICT to society.Social networks, the television of the future, vehicular network



ICAIT 2010 Conference Opening: (from left to right) Prof. DuWencai from Hainan University; Prof. Muriel Medard from MIT;Prof. Fu Guohua, Vice president from Hainan University; andProf. Anthony TS Ho from Surrey University.

Student volunteers from Hainan University.


4

STEFANO BREGNI

EditorPolitecnico di Milano - Dept. of Electronics and Information

Piazza Leonardo da Vinci 32, 20133 MILANO MI, ItalyPh.: +39-02-2399.3503 - Fax: +39-02-2399.3413Email: [email protected], [email protected]

IEEE COMMUNICATIONS SOCIETY

KHALED B. LETAIEF, VICE-PRESIDENT CONFERENCESSERGIO BENEDETTO, VICE-PRESIDENT MEMBER RELATIONSJOSÉ-DAVID CELY, DIRECTOR OF LA REGIONGABE JAKOBSON, DIRECTOR OF NA REGIONTARIQ DURRANI, DIRECTOR OF EAME REGIONNAOAKI YAMANAKA, DIRECTOR OF AP REGIONROBERTO SARACCO, DIRECTOR OF SISTER AND RELATED SOCIETIES

REGIONAL CORRESPONDENTS WHO CONTRIBUTED TO THIS ISSUE

THOMAS M. BOHNERT, SWITZERLAND([email protected])KERIM KALAMUJIC, BOSNIA ([email protected])JOSEMARIA MALGOSA SANAHUJA, SPAIN ([email protected])EWELL TAN, SINGAPORE ([email protected])

®�

A publication of the IEEE Communications Society

G l o b a l

N e w s l e t t e rwww.comsoc.org/pubs/gcn


ICAIT 2010/continued from page 3

JITEL AND TELECOM I+D/continued from page 3

technology, the digital home, and energy efficiency are fields ofinterest that were analyzed in different workshops and roundtables. The strategic contribution of ICT to public education,public health, ande-administration were covered by discussionsabout social communications, smart health, and smart govern-ment, and emergency and disaster support.

JITEL and Telecom I+D coordination aims to strengthenthe links among the different agents of R&D&I in the scope ofnetworking and telematic engineering. The success of participa-tion and positive assessments indicates that, to participants andorganizing committee members, it is desirable that this coordi-nated organization and collaboration be maintained in thefuture for the sake of progress in periods of both crisis and eco-nomic and social growth.

ICIN 2010/continued from page 1

session described several applications and illustrated theimportance of interfacing sensors to communications net-works, especially in machine-to-machine communication.

The final two sessions in ICIN covered business opportunitiesin networks and future trends. The business opportunities sessionexplored the impact on communication business of convergencewith the Internet, necessitating operator strategy adjustments.The future opportunities session presented a series of new con-cepts including application of peer-to-peer networking conceptsto increase network flexibility, techniques for using context, andstrategies for managing data volume increases required to delivermultimedia services with high quality of service (QoS).

At the closing session, TPC Chair Max Michel reviewedhighlights of the conference, and Roberto Minerva (TelecomItalia Laboratories) discussed a recent ConnectWorld article on“deperimeterization.” Delegate feedback on ICIN wasextremely positive.

On the final day, two workshops were provided in conjunc-tion with ICIN: The Second International Workshop on Busi-ness Models for Mobile Platforms, chaired by Pieter Ballon ofVrieje Universitet Brussel, Belgium; and Telecom Transforma-tion: Are You Ready? delivered by Eileen Healey of Healey &Co. Both were well attended and thought provoking.

Planning for ICIN 2011 is already underway under theleadership of Roberto Minerva of Telecom Italia Laboratories,TPC Chair for ICIN 2011. The conference will take place from4–7 October, 2011 in Berlin. For more information, visithttp://www.icin.biz.

The conference was really international, with speakers andattendees arriving in Moscow from 49 countries. The confer-ence proceedings are available at IEEE Xplore. Some resultsof the conference are also available at the conference web site,http://www.icumt.org.

south end of the country. Boasting a pleasing climate, goldensunshine, white beaches, and lush forests, it is dubbed “the ori-ental Hawaii.” Hainan is blessed with a charming tropicallandscape, contributing to its unique folklore and culture. It isknown as a Chinese all-season garden.

For more information, please see the web site www.icait.orgor contact the authors at [email protected] [email protected].

The 2nd International Congress on Ultra Modern Commu-nications and Control Systems was held in Moscow, Russia, on18-20 October 2010. The congress was organized by SPIIRAS(an institute of the Russian Academy of Sciences), Peoples’Friendship University of Russia (Moscow), and Tampere Uni-versity of Technology (Finland). The IEEE Russia NorthwestBT/CE/COM Chapter was a technical co-sponsor of the con-ference along with the Popov Society (a professional society ofRussian radio and communication engineers). Patrons andsupporters of the congress were Nokia, Nokia Siemens Net-works, and other organizations.

The technical program of ICUMT 2010 included five keynotetalks by internationally recognized industrial and academicspeakers, seven specialized workshops, and two main congresstracks: telecommunications andcontrol/robotics. Congress pro-ceedings include about 200 papers selected frommore than 320submissions with the help of more than 250 Technical ProgramCommittee members, reviewers, and workshop chairs. Thecongress was attended by over 200 participants.

New insights in radio technologies as well as the coexis-tence of research in control systems, robotics, and telecom-munications were key topics of the congress. Keynote speakerspresented insights and trends in vehicular ad hoc networks,networking security, multicarrier solutions, society of robots,and human-robot interaction. Significant scientific contribu-tions were presented in the areas of applied problems in theo-ry of probabilities and mathematical statistics, advancedsensing and control, systems of systems, mobile computing,reliable network design, and fiber optic systems. An industrialpanel session, “The ICT Future by the Year 2015,” was orga-nized by representatives of the top management from Hewlett-Packard, Alcatel-Lucent Bell Labs, Oracle Communications,Nokia, Nokia Siemens Networks, and Telecom Italia.

ICUMT 2010 Congress in Moscow, RussiaBy Vladimir Vishnevsky, Konstantin Samouylov, Yevgeny Koucheryavy, Alexey Vinel and Dmitry Tkachenko, Russia


30

SERIES EDITORIAL

An apparently early use of spectrum sharing is the operationof wireless microphones in unused television spectrum. As moretechnologically rigorous mechanisms for spectrum sharing in thesesame bands are introduced in the content of TV white space, it isessential that the new technology be capable of operating in thesame spectrum ecosystem as these early secondary users. “DSAOperational Parameters with Wireless Microphones” by Erpek,McHenry, and Stirling discusses this important topic, and demon-strates the viability of sensing regimes to this highly heteroge-neous emission type (milliwatts from the microphone, ascompared to megawatts from a television transmitter)

For DSA to be successful, it must also provide a sustainableand effective market mechanism to ensure that spectrum uses arearbitrated to achieve their maximal utility. There is a generalinternational consensus that market-based mechanisms are instru-mental in achieving this objective. In “The Viability of SpectrumTrading Markets,” Caicedo and Weiss discuss the viability of mar-kets as applied to secondary usage, and provide concepts for howsuch markets could achieve success though sufficient spectrum liq-uidity, market participants, and extent of the spectrum resourcesthat would be accessible through such markets.

Most discussions of DSA eventually focus on the ability ofspectrum sensing regimes to detect and protect other users of thespectrum. However, to be effective, DSA must not only provideprotection to “privileged” users, but also provide effective perfor-mance for the secondary user. In “Unified Space-Time Metrics toEvaluate Spectrum Sensing,” Tandra, Sahai, and Veeravalli pre-sent a unified framework that goes beyond protection of the pri-mary user and also considers the performance of the secondaryspectrum sharing user.

Lastly, it is evident that the discussion of DSA is starting to leadto a much broader set of technologies; those that open the moregeneral question of how devices can coexist with each other indense, heterogeneous, and overlapping networks. Early experiencewith smartphones demonstrates that wireless architectures will notsatisfy the demand with just linear growth in wireless capacity: itwill be necessary to make much exponential growth in capacity asthese devices become the prevalent solution in the marketplace.

BIOGRAPHYPRESTON MARSHALL ([email protected]) is a director at the Information Sci-ence Institute at the University of Southern California’s Viterbi School ofEngineering, where he leads research programs in wireless, networking,cognitive radio, alternative computing, and related technology research. Hehas 30 years of experience in networking, communications, and relatedhardware and software research and development. For most of the lastdecade, he has been at the center of cognitive radio research, includingseven years as a program manager for DARPA, where he led key cognitiveradio and networking programs, including the neXt Generation Communi-cations (XG) program, Disruption and Delay Tolerant Networking (DTN),Sensor Networking, Analog Logic, and the Wireless Network After Next(WNaN) program. He has numerous published works, and has made manyappearances as invited or keynote speaker at major technical conferencesrelated to wireless communications. He holds a Ph.D. in electrical engineer-ing from Trinity College, Dublin, Ireland, and a B.S. in electrical engineeringand an M.S. in information sciences from Lehigh University.

n this issue, we follow up on our efforts in 2009 and 2007 tohighlight the rapidly evolving technologies of cognitive radio

and dynamic spectrum access, beginning with a guest editorial onthis topic from Preston Marshall.

We would like to continue to encourage our readers to sendus suggestions on the topics and trends in radio communicationsthey feel should be addressed in the series — we look forward toyour feedback! Joseph Evans and Zoran Zvonar

The rapid evolution of dynamic spectrum access (DSA) is evidentfrom the progression of the papers, and related events, from theIEEE International Symposium on Dynamic Spectrum Access Net-works (DySPAN). At the first DySPAN in Baltimore in 2005, therewas only one reported experiment (the Defense AdvancedResearch Projects Agency [DARPA] XG program). The bulk ofthe papers focused on what could or might happen in the DSAfield. By the second DySPAN in Dublin, the most popular area ofthe conference was the demonstration room, which crowded severaldozen operating DSA systems, or elements thereof, into a veritablesmorgasbord of sensing and adaption technology. The third DyS-PAN in Chicago coincided with the U.S. FCC announcement ofthe first official adoption of DSA principles to provide dynamicaccess to unused television channels. The last DySPAN followed byone week the release of the U.S. Broadband Task Force report,which officially recognized DSA as a core mechanism to achievethe necessary increase in spectrum access.

Following on the heals of the task force, the U.S. FCC formal-ly started inquiry on the regulatory mechanisms to adopt the prin-ciples of DSA within the broader context of regulatory regimes(ET Docket No. 10-237, Promoting More Efficient Use of Spec-trum Through Dynamic Spectrum Use Technologies). In half adecade, this technology has moved from an idealistic and poorlyunderstood technology to a potential centerpiece in a nationalstrategy to maximize the utility of the finite spectrum resource.

This progression is reflected in the articles in this month’sissue. They are not focused on farsighted vision or future techno-logical visions, but on several of the practical engineering impedi-ments to the deployment of this technology. While there is still aneed for future vision, this immediate focus is to demonstrate thistechnology in the context of some of the immediate challenges.

Perhaps a measure of the successful emergence of a technolo-gy is the necessity to plan for, and design in mitigation of, theinevitable malicious behavior. Certainly the experience with theInternet demonstrates that both malicious and greedy behaviorswill attract opportunities for inappropriate (often illegal) benefitand associated exploits if these considerations are not integratedinto the technology from the onset. In “A Bayesian Game Analy-sis of Emulation Attacks in DSA Networks,” Thomas, Komali,Borghetti, and Mähönen demonstrate that it is possible to devel-op methods to detect attempts by spectrum sharing nodes to emu-late the characteristics of “protected” spectrum users, and thusobtain advantages over other spectrum sharers. The rampant dis-tribution of malware over the Internet is certainly proof thatapparently malicious, or even vandalizing, behavior must be aconcern early in every technology, because individuals will deter-mine a mechanism to benefit from exploiting these weaknesses.


I

RADIO COMMUNICATIONS: COMPONENTS,SYSTEMS, AND NETWORKS

Joseph Evans Zoran Zvonar

THE MATURATION OF DYNAMIC SPECTRUM ACCESS FROM AFUTURE TECHNOLOGY, TO AN ESSENTIAL SOLUTION TOIMMEDIATE CHALLENGES

LYT-GUEST EDIT-Evans 2/22/11 11:38 AM Page 30

IEEE Communications Magazine • March 201132 0163-6804/11/$25.00 © 2011 IEEE

This research is funded inpart by the Air ForceResearch Lab and by theAir Force Office of Scien-tific Research. This workalso supported in part bythe European Union(ARAGORN project) andDFG through UMIC-research center facility.The views expressed inthis article are those of theauthors, and do not reflectthe official policy or posi-tion of the United StatesAir Force, Department ofDefense, or the U.S. Gov-ernment. This documenthas been approved forpublic release; distributionunlimited.

INTRODUCTION

Dynamic spectrum access (DSA) is an exciting newconcept that promises to bring flexibility to spec-trum management. Instead of relying on traditionalspectrum licenses, in which licensees have exclusiverights to a fixed, static amount of spectrum, DSAschemes allow unused spectrum to be used oppor-tunistically by other users. These opportunisticusers are called secondary users (SUs), and mustonly use spectrum not in use by primary users(PUs), radios that have priority licenses for thespectrum. To operate in either a PU or SU rolewill require either a license or operating approvalfrom a regulator such as the Federal Communica-tions Commission (FCC). Radios holding a PUlicense have access to a reserved frequency band.Radios holding an overlay SU license have theability to scavenge spectrum from whatever spec-trum is sensed to be unused.

There may be times when radios do not wantto use spectrum according to their license.Instead, the radio may pretend to hold a differ-ent license. The idea of such an emulation attack(EA) was first articulated by Chen in [1]. That

work identified a specific kind of EA called thePU emulation attack (PUEA), in which a radioemulates a PU for either selfish or maliciousreasons. When driven by selfishness, the radiosemulate to maximize their spectrum usage; whendriven by maliciousness, radios emulate todegrade the DSA opportunities of other spec-trum users. Like any kind of illegal activity,deterring either kind of EA requires a combina-tion of detection and punishment.

In this vein, the core contribution of this workis to investigate system-wide behaviors when thedetection of selfish EAs leads to punishment. Byproposing a Bayesian game framework whereradios are unsure of the legitimacy of the claimedlicense of other radios, we identify conditionsunder which a Nash equilibrium (NE) (stableoperating point) can exist. These conditions giveinsight into questions such as whether policy-abiding radios can coexist with selfish radios,with what probability selfish radios choose tolaunch EAs, and under what conditions selfishEAs are discouraged. The NE also reveals theprobability of radios challenging the license ofother radios to determine their legitimacy. Theseresults can be used to determine appropriatepolicies to keep the rate of EA arbitrarily low,something of significant interest to regulatoryagencies. Conversely, this information can beused by PU and SU licensees to determine howrampant EAs will be for their license type.

Most work on emulation attacks has beenfocused on the detection aspect of the PUEA.Chen [1] suggests a location-based authenticationscheme called LocDef, in which the location (cal-culated from the received signal strength [RSS]by a network of sensors) and waveform charac-teristics of a declared PU are compared againstthe known location and waveform characteristicsfor that transmitter. Particularly when the PUsconsist of static, well characterized users such astelevision broadcasters, this can be an effectivescheme. A clustering approach is described in [2],in which waveform features are used to catego-rize a signal as being from a PU. More generally,in [3, 4], various statistical tests are developed todetermine whether it is likely that the measured

ABSTRACT

Dynamic spectrum access has proposed tier-ing radios into two groups: primary users andsecondary users. PUs are assumed to havereserved spectrum available to them, while SUs(operating in overlay mode) must share whatev-er spectrum is available. The threat of emulationattacks, in which users pretend to be of a typethey are not (either PU or SU) in order to gainunauthorized access to spectrum, has the poten-tial to severely degrade the expected perfor-mance of the system. We analyze this problemwithin a Bayesian game framework, in whichusers are unsure of the legitimacy of the claimedtype of other users. We show that depending onradios’ beliefs about the fraction of PUs in thesystem, a policy maker can control the occur-rence of emulation attacks by adjusting the gainsand costs associated with performing or checkingfor emulation attacks.

TOPICS IN RADIO COMMUNICATIONS

Ryan W. Thomas and Brett J. Borghetti, The Air Force Institute of Technology

Ramakant S. Komali and Petri Mähönen, RWTH Aachen University

Understanding Conditions that Lead toEmulation Attacks inDynamic Spectrum Access

THOMAS LAYOUT 2/18/11 3:15 PM Page 32


RSS of the declared PU could have come froman actual PU. Unlike these works, which are pri-marily about detecting EAs, we focus on how tocope with these attacks once they are detected,and provide conditions for formulating an appro-priate response mechanism system.

BAYESIAN GAMES WITHINCOMPLETE INFORMATION

Bayesian games, or games with incomplete infor-mation, are modeled as having players of differenttypes. Knowledge of these types may or may notbe extended to all players, meaning that someplayers may not know other players’ types. Inthese games, random chance (called nature) selectstypes before the game is played according to someprobability distribution (this uncertainty is knownas incomplete knowledge). Strategy decisions in aBayesian game are made based on a combinationof players’ types, their belief in the distribution oftypes from nature, the actions available to them,and the payoffs under these combinations.

A Bayesian game contains several compo-nents; we present a simplified view of them here.First is the player set, consisting of all players inthe game. There is a set of types, where a typerepresents a particular player’s characteristics.For instance, in a fighting game, players might beof two types, strong or weak. Each player andtype has a set of strategies that represents thepossible choices in the game. In the fightinggame the strong player may have the strategies ofpunch or run, and the weak player may have thestrategies of kick or cower. When a player andtype are only allowed to select one strategy at atime (i.e. just punch or just run for the strongplayer), we say they are playing a pure strategy.Mixed strategies are a more general case. Amixed strategy is represented by some probabilitydistribution over all possible pure strategies.

Each player has a utility function ui that mapsall players’ chosen strategies and types to a realvalue, representing the preference of that partic-ular combination (higher values are consideredmore desirable). Finally, each player has a distri-bution function that represents their beliefs aboutnature’s distribution of types. In a Bayesiangame, it is assumed that everything except for theplayer’s type (including its utility, pure strategychoices, the number of players, etc.) is “commonknowledge,” meaning that all players know them,know that all players know them, and so on.

One of the more useful game theoretic-con-cepts is the NE, which describes a state fromwhich no rational player has any motivation tounilaterally change their chosen mixed or purestrategies (doing so would result in a equal orlower utility, thus giving the player no incentiveto change). The Bayesian Nash Equilibrium(BNE) is an extension of the NE definition thatincorporates the players’ types and beliefs.

The pure-strategy NE is fairly easy to conceptu-alize in a game in which both players have two purestrategies. Under pure-strategy NE, neither playerwill switch to their other pure strategy, becausedoing so alone would not increase their utility.

The same is true of the mixed strategy NE —neither player will switch to any other pure or

mixed strategy since it won’t increase their utility.To understand under what kinds of mixed strate-gies NEs arise, it helps to use a slightly differentperspective. If a mixed strategy is played suchthat the other player’s expected utilities fromeither pure strategy are equal, then that otherplayer will not prefer any particular pure ormixed strategy over any other (all will have thesame expected utility). If both players play such amixed strategy — making the other player “indif-ferent” to all mixed and pure strategies, they willbe playing the mixed strategy NE. Neither playerhas any incentive to change their mixed strategy.

SYSTEM MODELWe assume that a network consists of multipleradios attempting to access (via PU and SUlicenses) a shared block of spectrum. We assumethat SU licenses are exclusive of PU licenses,meaning that radios can have one license or theother, but not both. Furthermore, we assumethat interlopers (radios without a license) thatuse the spectrum frequently will be shut down orforced to get approval, making their presenceuncommon. Therefore, this analysis ignores unli-censed radios and dual-licensed radios. Weanticipate future work investigating the effectthese users have on the system.

Without loss of generality, we suppose that allradios are within transmission range of one anotherand suffer from interference under the sub-bandsthat are in use by other radios. (Non-interferingradios can be disregarded in our spectrum sharingmodel because they do not influence each other’sperformance and therefore can coexist.) In thismanner, the spectrum can be considered to be acommon resource of which all active radios areattempting to utilize some portion.

The bandwidth that is available for a legiti-mate PU is considered to be a fixed and reservedquantity (of course, this assumption only holdswhen the bandwidth is legitimately used; when aradio performs an EA then more than one PUmay be in a PU frequency band). In contrast, thebandwidth for SUs is dependent on the availablespectrum (we use this term to describe all spec-trum in a block not in use by a PU1)and thenumber of SUs sharing it. This forms a DSA sys-tem in which PUs have negotiated with the regu-latory body a license that allows priorityinterference-free access to a fixed subset of thespectrum block. Conversely, an SU license allowsa radio to scavenge some amount of bandwidthfrom the spectrum that is unused by the PUs.Bandwidth for SUs is distributed among all SUradios as a function of the available free band-width and number of SUs. Although the processof negotiating spectrum allocation among SUs isbeyond the scope of this work, we assume that allSUs receive an equal benefit from the process.

Although DSA proposes to relax the rules forhow spectrum is accessed, it does not propose toeliminate regulation altogether. It is expectedthat radios will utilize the spectrum block inaccordance with the license they hold. In supportof this, we assume that there is some mechanismin place that allows the verification of a radio’sPU or SU license. As discussed earlier, severalmechanisms have been suggested in the litera-ture for this, including location services, wave-

1 This term is also referredin literature as whitespace or spectrum holes.

The pure-strategy NE

is fairly easy to

conceptualize in a

game in which both

players have two

pure strategies.

Under pure-strategy

NE, neither player

will switch to their

other pure strategy,

because doing so

alone would not

increase their utility.


form identification (using either passive or activetechniques), and certificates. In this work we donot concern ourselves with the actual approachused, only assuming that there is one and it hassome non-zero cost to perform (in terms of suchresources as processing time, power consump-tion, and communication overhead).

Selfish radios (radios with the potential toperform a selfish EA) can either emulate, utiliz-ing the spectrum in the manner of a license theydo not hold, or use the spectrum legitimatelyand not perform an EA. Both types of licensees,PU and SU, if they are not policy-abiding, maychoose to perform an EA. The idea of an SUEAis less discussed than the PUEA, but may occurwhen the available spectrum for SU use is moredesirable than the spectrum reserved for PUsuse (and SU and PU licenses are exclusive ofone another). Other radios in the system canchoose to either check the credentials of thisradio (which we term challenge, since the selfishradio may or may not actually be emulating) andsuffer the verification cost discussed above, oraccept the stated role of the radio and allow theselfish radio to continue operating as it pleases.

Our model assumes the existence of a regula-tory body (e.g., the FCC) with the authority topunish violators of policy. When a violation isdetected (if a selfish radio’s EA is challenged byanother radio), the regulatory body has thecapability to employ a punishment to the viola-tor. The punishment cost can come in severalpossible forms, including financial penalties,bandwidth restrictions, timeouts, or the forfei-ture of other radio resources by the violator. Forsimplicity, we assume that punishments arefixed, meaning penalties do not change underrepeated good or bad behavior. Furthermore,once punishments have been paid, violators areallowed to resume operations in the spectrum(under the terms of their correct license).

Finally, we assume that there are enoughradios in the system that a radio knowing its ownlicense type (SU or PU) will not have an effecton its belief of the distribution of the licensetypes in the network. Furthermore, decisions toemulate and challenge are made simultaneously,

preventing the decision of one radio fromsequentially affecting the decision of the other.

The utility function for the radios is of theform

ui(s, θ)= ri(s, θ) – ci(s, θ),

where ri(s, θ) is the revenue a radio of type θiplaying against other types gets under a particularstrategy profile, s (that radio’s strategy and allother radios’ strategies), and ci(s, θ) is the costassessed under that action (leaving the total utilityas the profit). While the specific revenue valuesvary depending on the particular spectral scenariounder which the radios are operating, there arefour types of revenue that are of interest. Thefirst, rp, is the revenue a PU receives when it usesits reserved spectrum. Similarly, rs is the benefitan SU receives when it shares available spectrum(spectrum that is not occupied by a PU) withsome fixed number of other SUs. These two ben-efits are illustrated in Fig. 1a and 1b. When aradio has to share the available spectrum withone more radio than in the rs case, it gets benefitr*s. This is illustrated in Fig. 1c. Finally, when two(emulating) PUs attempt to access reserved spec-trum, they receive benefit r*p. Figure 1d illustratesthe case where the two PUs attempt to accesstwo non-overlapping spectrum bands.

Similar to the revenues, there are two types ofcosts that are of interest. The first is the regula-tory body penalty for performing an EA, ce. Theother is the cost of challenging the other radio’slicense, cc. As it is not clear that every time a vio-lation is detected the regulatory body will be ableto enact a punishment, without any loss of gener-ality ce can be considered an expected cost. Simi-larly, cc may represent the expected cost tochallenging a license if the scheme employedrequires a dynamic amount of resources.

From this system model, we define two gamesthat drive our analysis in the rest of the article,under the general assumption that the cost ofchallenging and emulating are greater than 0.The first game (the one-way game) is between apolicy-abiding SU and a selfish radio holdingeither an SU or a PU license, and the second


Figure 1. Illustrations of the four revenue variables: a), the revenue of the PU; b), the revenue of the SU; c), the revenue of an SU with anadditional SU in the available spectrum; d), the revenue of a PU when it shares spectrum with an additional PU.

(a) (b) (c) (d)

Freq

uenc

y bl

ock

PUrp

Ava

ilabl

e sp

ectr

um

Ava

ilabl

esp

ectr

um

Ava

ilabl

esp

ectr

um

SUrs

Ava

ilabl

e sp

ectr

um

SUr*s

SUr*s

Freq

uenc

y bl

ock

PUr*p

PUr*p


(the two-way game) is between two selfish SUsthat may perform an EA but are unsure if theother will challenge their license. We use two-player one-shot Bayesian games to represent theinteractions between radios. Modeling these astwo-player games is similar to the attacker/defender model used in [5, 6]. We assume thatradios determine a strategy for dealing withother spectrum users and/or using the spectrumupon the decision to use the spectrum block,based only on their knowledge of radio utilitiesand their beliefs of type distributions. Thesestrategies are applied immediately as otherradios attempt to utilize the spectrum. Theseinteractions occur quickly compared to spectrumusers arriving and leaving, so we approximatethem as one-on-one one-shot events. Futurework can relax these assumptions, providinginsight into multistage multiradio interactions.

We determine the pure and mixed strategyBNE of these two games to gain insight into thesteady-state rate of EAs in the system. We pur-sue an analytical approach here, where the val-ues of the revenues and costs are not defined,allowing these results to be used by regulatoryagencies for engineering DSA ecosystems.

THE ONE-WAYEMULATION ATTACK GAME

We now formally analyze the interaction betweena policy abiding SU and a radio of unpredictabletype (either SU or PU) that may choose to eitherlaunch a selfish EA or act true to its type. In thisgame a radio willing to perform an EA interactswith an SU that is not going perform an EA, butmay or may not choose to challenge the licenseof the other radio. This game represents a sce-nario where policy dictates that radios can onlybe challenged at particular asymmetric times(e.g., when a radio enters the spectrum for thefirst time); once radios have begun to utilize thespectrum, they are entrenched regardless of theirlegitimacy. This kind of policy might emerge if acertificate-based scheme was used for authentica-tion and certificates were only shared upon entryto a spectral band.

The player set therefore consists of two radios.Radio 1 can be one of two types, PU or SU, basedon the license granted to the radio. Radio 2 canonly be one type, an SU. While both radios knowtheir own type, radio 2’s type is known to radio 1,while radio 1’s type is unknown to radio 2. Due toradio 2’s partial information about radio 1, thereare two actions radio 1 can take: either emulateand pretend it is the opposite of its true type, or

act legitimately and present itself without decep-tion. Radio 2 has two actions available: either chal-lenge radio 1 and check its stated license or acceptits claim. The game is considered to be playedsimultaneously; the normal form matrix of utilitiesfor the two types of radio 1 is shown in Fig. 2.

For the interaction in Fig. 2, when radio 1 isof type PU, the utility outcomes are as follows. Ifradio 1 decides to emulate and radio 2 challengesit, the regulatory body ensures that radio 1 onlyuses the PU spectrum (rp) and pays the penalty(ce). Under this scenario, radio 2 gets the rev-enue of not sharing the available spectrum withan additional radio, but also must pay the cost(cc) of challenging radio 1’s license. If, instead ofchallenging, radio 2 had accepted radio 1’s emu-lation, both radios will share the available spec-trum with the other radio (r*s). If radio 1 does notemulate, it receives the benefit of the PU spec-trum. If radio 2 decides to challenge when radio1 has acted legitimately, radio 1 has committedno wrong doing and will still get the revenuefrom the PU spectrum, while radio 2 will stillhave to pay for challenging (cc) and get the samerevenue (rs). Similarly, if radio 1 acts legitimatelyand radio 2 accepts, both radios receive theirappropriate spectrum revenues without any costs.Similar logic is used to determine the payoffmatrix when radio 1 is of type SU.

We begin by investigating under what circum-stances it only makes rational sense for radio 2 toplay either challenge or accept, regardless of thestrategy of player 1 — when this occurs, the strat-egy is called dominant. We do this by investigat-ing the expected utility for radio 2 when facing allpossible strategy choices by its uncertain oppo-nent. We define radio 2’s belief that radio 1 is oftype PU to be φ and belief that it is of type SU tobe 1 – φ (radio 1’s beliefs are not interesting,since radio 2’s type is known to all). The expectedutility is calculated for every strategy-type combi-nation; for instance, the expected utility of playingchallenge against a PU playing emulate or an SUplaying legitimate is calculated by summing radio2’s utility of playing challenge against PU emulate(top left box in Fig. 2) weighted by φ and radio 2’sutility of playing challenge against SU legitimate(bottom left box in Fig. 2) weighted by 1 – φ. Thisis continued for all four combinations of strate-gies the two types can play against challenge, andall four strategies the two types can play againstaccept. We find that under some circumstancesthere is no dominant strategy, and under othersaccept is a dominant strategy, but challenge isnever a dominant strategy.

There are two general conditions where acceptis a dominant strategy for radio 2. The first


Figure 2. The normal form one-way EA game, in which the selfish radio’s (radio 1’s) type is unknown toradio 2, and radio 2’s type is known. For each strategy pair, the topmost is radio 1’s, the bottom radio 2’s.

Radio 1

Radio 2 Radio 2

Radio 1 type = PU

Emulate

Challengerp – ce,rs – cc

Acceptr*s ,r*s

Legitimaterp,

rs – cc

rp,rs

Radio 1

Radio 1 type = SU

Emulate

Challenger*s – ce,r*s – cc

Acceptrp,rs

Legitimate r*s,

r*s – cc

r*s,r*s

Our model assumes

the existence of a

regulatory body

(such as the FCC)

with the authority to

punish violators of

policy. When a

violation is detected

(if a selfish radio’s EA

is challenged by

another radio) the

regulatory body has

the capability to

employ a punish-

ment to the violator.



occurs when the revenue gained from challengingis less than cost of challenging, and the secondoccurs when radio 2 believes there are feweremulating radios. In other words, accept is adominant strategy when the expected gain fromchallenging is less than the cost of challenging.

When one of these conditions does hold (mean-ing accept is the only rational strategy to play forradio 2), either radio 1’s strategy of emulating oracting legitimately will be irrational and can beeliminated. Which one gets eliminated depends onwhether radio 1 gets more revenue from acting asa PU vs. sharing the available spectrum as an SUwith radio 2. If it gets more revenue from sharing,radio 1 will play emulate; otherwise, it will playlegitimate. Thus, there are two pure strategy NE,one in which radio 2 plays accept and radio 1 playsemulate, and one in which radio 2 plays accept andradio 1 plays legitimate.

For the mixed strategy equilibria, we find thecost of emulating affects the probability of radio2 challenging. For a fixed amount of revenuegained from emulating, higher costs of emulatinglead to lower probabilities of challenging. Inother words, as the penalties increase on emulat-ing, radio 2 has to challenge less frequently tomaintain radio 1’s indifference to either emulat-ing or acting legitimately. This makes sense, asthe cost of emulating acts to counter the incen-tive of emulating, and the greater the cost, theless often challenge has to be played to make theexpected utilities the same for radio 1 choosingbetween legitimate operation and emulate. Radio1 types that emulate do so with a probability thatmakes the expected gain for radio 2 from chal-lenging equal to the cost of challenging.

A regulatory agency may be concerned withhow likely it is that radios will perform EAsunder various revenue and cost scenarios. Weterm this probability p[EA], which is the proba-bility that a randomly encountered radio willchoose to emulate. We specifically investigatethe worst case p[EA], pmax[EA], which is definedas maximum p[EA] across all beliefs that radioshave about the fractions of types in the system.To determine this, we assume that the beliefsradios have about the distribution of types arecorrect, meaning that φ actually represents thetrue fraction of devices that are PUs. For someranges of parameter values, any belief admits apure strategy equilibrium that has player 2 emu-lating, giving a pmax[EA] of 1. For other rangesof parameters, not just any belief admits a purestrategy equilibria. In these cases, pmax[EA] isthe largest fraction (φ) of emulating types thatsupports a pure strategy equilibrium. It is worthnoting that pmax[EA] is not correlated with thecost of emulating. In other words, no matterwhat the magnitude of the penalty that the regu-latory agency places on emulating, there is nopure strategy profile that would change theworst-case probability of emulation.

THE TWO-WAYEMULATION ATTACK GAME

We now explore a different game in which eachradio has partial knowledge about the type of theother. In this game both radios are SUs that will

potentially commit an EA but do not know if theother radio will challenge them. In this gameevery claimed PU license represents an EA; how-ever, not all radios will take the effort to verifythe attack and report it to the regulatory agency.Note that the types in the two-way game are dif-ferent than in the one-way EA game: one typealways challenges the claimed license type ofother radios, and the other type always acceptsthe claimed license. Both radios know their owntype but are unaware of the other radio’s type.Each type has the same actions, emulate or actlegitimately. The game is played simultaneously;the normal form of the game for the four possi-ble combinations of types is shown in Fig. 3.

The two-way game may manifest in a sce-nario where a subset of the SUs have the capa-bility to report EAs and will therefore challengeall licenses, while others do not have this capa-bility (perhaps they are not able to communicatewith the regulatory body) and never challenge.Alternatively, it represents a scenario in whichthe challenge/accept mixed strategy is dictatedby policy rather than the radio itself.

In this game, some of the more interestingpayoffs pairs include: when both radios emulateand are both of type challenge, the radios sharethe available spectrum (r*s) and pay both thepunishment cost and the cost of checking thelicense credentials; when both radios emulateand both radios are of type accept, the radiosshare the same PU spectrum (r*p) (which, as dis-cussed earlier, may be equal to or less than rp,depending on the scenario); and when one radioperforms an EA and is of type accept while theother radio does not and is of type challenge, itleads to both radios sharing the available spec-trum (r*s) and paying either the punishment cost(ce) or the cost to challenge (cc).

Figure 3 shows at least one pure strategy equi-librium. In the case where the revenue from beingthe only SU is greater than the shared revenue asa PU, and the shared revenue as an SU is greaterthan the revenue from being the only PU, thestrategy of acting legitimately is dominant forboth types of radios — challenge and accept —regardless of their beliefs about the distributionof types. However, if these conditions do nothold, there are no easily identifiable pure strategyBNE, and we must look for a mixed strategy equi-librium. In this configuration both radios of anytype will share the spectrum as SUs because thereis no spectral incentive to perform an EA.

To determine the mixed-strategy BNE, wenote that there is no need for radios of typechallenge to mix strategies; for these radios φalone determines which strategies dominate.When a radio of type challenge faces a radio alsoof type challenge, no matter what mixed strategyeither radio selects, the utility is either r*s – ce –cc (if it emulates) or r*s – cc (if it acts legitimate-ly). In the same manner, if a radio of type chal-lenge faces a radio of type accept, no matter whatmixed strategy the accept radio selects, the utilityof the challenge radio is either rp – cc if it emu-lates or r*s – cc if it acts legitimately. In this man-ner, no mixed strategy from either type canmake radios of type challenge indifferent.

When playing a mixed strategy, the probability ofemulating for radios of type accept increases pro-

Both one-way and

two-way EA games

are games of imper-

fect information that

admit BNE in pure

and mixed strategies.

The conditions that

lead to these BNE

can serve as guide-

lines for regulatory

agencies and policy-

makers to ensure

that the rate of EA in

the system are kept

arbitrarily small.



portionally to the expected revenue from getting PUspectrum without being challenged and decreasesproportionally to the expected cost of getting caught.In other words, the radios of type accept do not likeemulating when the cost of emulating goes up orthe benefit of emulating goes down. Furthermore,because of the game symmetry, all mixed and pureBNE strategies are the same for radios 1 and 2.

In general, regulatory agencies will have aharder time creating good rules of thumb fordealing with mixed strategy BNE in this game, asthe BNE behavior of the system under mixedstrategies is strongly dictated by the relative mag-nitudes of the revenue gained and lost under dif-ferent spectral use cases. However, under allmixed strategies, increasing the cost of emulatingwill decrease the propensity for radios to emulate.

CONCLUSIONSOpening up the spectrum to allow differentlicensees to coexist in a DSA network mayencourage selfish activity. In this work we exam-ine one such manifestation of selfish behavior,emulation attacks. Selfish radios may choose toemulate (and conceal their true type) or actlegitimately from time to time, so as to increasethe expected benefit they obtain from the spec-trum, thereby giving rise to uncertainty in thesystem. We modeled the interactions between aselfish radio and policy-abiding radio (one-wayEA game) and between two selfish radios (two-way EA game) as Bayesian games.

Both one-way and two-way EA games aregames of imperfect information that admit BNEin pure and mixed strategies. The conditions thatlead to these BNE can serve as guidelines forregulatory agencies and policymakers to ensurethat the rate of EA in the system are kept arbi-trarily small. Of particular interest to regulators,we find that the penalties levied against violatorsdo not always have an effect on the BNE, andthe belief that radios have regarding type distri-bution will often dictate the BNE strategy. Thisimplies that the perceived and actual distributionof types may be as important a parameter forregulatory agencies to control as the penalties.

REFERENCES[1] R. Chen, J.-M. Park, and J. H. Reed, “Defense Against

Primary User Emulation Attacks in Cognitive Radio Net-works,” IEEE JSAC, vol. 26, Jan 2008, pp. 25–37.

[2] T. R. Newman and T. C. Clancy, “Security Threats toCognitive Radio Signal Classifiers,” Proc. Wireless@VT2009, 2009.

[3] S. Anand, Z. Jin, and K. Subbalakshmi, “An AnalyticalModel for Primary User Emulation Attacks in CognitiveRadio Networks,” Proc. IEEE DySPAN 2008, 2008.

[4] Z. Jin, S. Anand, and K. Subbalakshmi, “Detecting PrimaryUser Emulation Attacks in Dynamic Spectrum Access Net-works,” Proc. IEEE ICC ’09, June 2009, pp. 1–5.

[5] Y. Liu, C. Comaniciu, and H. Man, “A Bayesian GameApproach for Intrusion Detection in Wireless Ad HocNetworks,” GameNets ’06, 2006.

[6] F. Li and J. Wu, “Hit and Run: A Bayesian Game BetweenMalicious and Regular Nodes in MANETs,” Proc. IEEESECON, 2008, pp. 432–40.

BIOGRAPHIESRYAN W. THOMAS [M] is an assistant professor of computerengineering in the Department of Electrical and ComputerEngineering at the Air Force Institute of Technology. Hereceived his Ph.D. in computer engineering from VirginiaTech in 2007; M.S. in computer engineering from the AirForce Institute of Technology in 2001; and his B.S. fromHarvey Mudd College in 1999. He previously worked at theAir Force Research Laboratory, Sensors Directorate as a dig-ital antenna array engineer. His research focuses on thedesign, architecture, and evaluation of cognitive networks,cognitive radios, and software-defined radios.

RAMAKANT S. KOMALI [M] is currently with the Wireless Net-work Business Unit of Cisco Systems Inc. Prior to this, hewas a postdoctoral researcher in the Department of Wire-less Networks at RWTH Aachen University. He received hisPh.D. degree in electrical engineering from Virginia Poly-technic Institute and State University in August 2008. Hisresearch interests are in distributed resource allocation inwireless networks, topology control of cognitive radio net-works, analysis, design, and optimization of wireless net-work protocols, and game theory.

BRETT J. BORGHETTI is an assistant professor of computer sci-ence at the Air Force Institute of Technology. He earned aPh.D. in computer science in 2008 from the University ofMinnesota, Twin Cities; an M.S. degree in computer sys-tems in 1996 from AFIT in Dayton, Ohio; and a B.S. in elec-trical engineering in 1992 from Worcester PolytechnicInstitute, Massachusetts. His research interests focus onartificial intelligence, multi-agent systems, game theory,and machine learning.

PETRI MÄHÖNEN [SM] ([email protected]) is a full profes-sor and head of the Institute for Networked Systems atRWTH Aachen University. He joined the faculty of RWTH in2002 as Ericsson Chair of Wireless Networks. He hasworked and studied in the United Kingdom, the UnitedStates, and Finland. His scientific interests include cognitiveradio systems, networking and wireless communications,spatial statistics, and analysis of complex networks. He iswith his group active in both theoretical and experimentalresearch topics. In 2006 he received the Telenor ResearchPrize. He is serving as an Associate Editor for IEEE Transac-tion of Mobile Computing and Area Editor for the Journalof Computer Communications. He is working as scientificadvisor or consultant for different international companiesand research centers. He has been a chair or program com-mittee member for numerous conferences and workshops.He was TPC Chair for IEEE DySPAN 2010, and serves as Co-General Chair for IEEE DySPAN 2011. He is a Senior Mem-ber of ACM and a Fellow of RAS.

Figure 3. The normal form two-way EA game, in which both radios’ types are unknown to the other. For each strategy pair, the topmost isradio 1’s, the bottom radio 2’s.

Radio 1 = challengeRadio 2 = challenge

Radio 2 Radio 2 Radio 2 Radio 2Ra

dio

1 E

Emulate

r*s – ce – cc,r*s – ce – cc

Legitimate

r*s – ce – cc,r*s – cc

Lr*s – cc,

r*s – ce – cc

r*s – cc,r*s – cc

Radio 1 = challengeRadio 2 = accept

E

Emulate

rp – cc,rs – ce

Legitimate

rp – cc,rs

Lr*s – cc,r*s – ce

r*s – cc,

r*s

Radio 1 = acceptRadio 2 = challenge

E

Emulate

rs – ce,rp – cc

Legitimate

r*s – ce,r*s – cc

Lrs,

rp – cc

r*s ,

r*s – cc

Radio 1 = acceptRadio 2 = accept

E

Emulate

r*p,r*p

Legitimate

rp,rs

Lrs,rp

r*s ,r*s


INTRODUCTION

The opening of white spaces in the broadcasttelevision bands to new devices enabled bydynamic spectrum access (DSA) technologypromises a range of economic and social benefitsby enabling the use of spectrum that has laineither unused or underused.

REGULATORY PROGRESS

Across the world, regulators are becoming awareof the importance of opening up TV whitespaces for license-exempt use. Regulators in theUnited States and United Kingdom have soughtto enable these gains without impacting theoperations of existing users: mainly TV broad-casters and protected wireless microphone users.In the United States, proceedings on TV whitespaces are now concluded in rights of the FCC’sdecisions [1] in November 2008 and September2010. In the United Kingdom, Ofcom publishedproposals [2] in June 2009 for opening up TVwhite spaces to new applications.

One of the areas regulators find particularlychallenging is the determination of how to pro-tect wireless microphones, which are well estab-lished in the TV bands and integral to thebroadcast and entertainment industries. Dataconcerning real-world wireless microphone sys-tem performance is sparse, and operating prac-tice is not well documented. Difficulty inobtaining data on wireless microphone use andinaccurate methods to combine statistical factorshave led regulators toward an unnecessarily con-servative approach.

THE MOST IMPORTANT DSA ISSUEThe most important current DSA issue is to helpregulators develop spectrum access policies thatprovide a fair balance between interference toprotected legacy spectrum users and practicalimplementation. The difficulties encountered inthe FCC’s testing of prototype TV white spacedevices in the rule making process did not ini-tially lead to practical DSA rules. This is espe-cially true for sense-based rules for TV banddevices, which are currently unnecessarily con-servative in the United Kingdom (–126 dBmsensing threshold), but these rules were recentlymodified in the United States (–107 dBm).1 Thelack of reasonable spectrum access policies islikely to impede the application of sensing-basedDSA unnecessarily, while regulatory trends arefavoring DSA generally, especially geolocation-based DSA.

There are DSA interference analyses in theliterature. In [3] Dhillon et al. performed aninterference analysis at a wireless microphonereceiver with single and multiple interferers.They concluded that DSA devices have the


ABSTRACT

This article provides a comprehensive analy-sis of dynamic spectrum access operationalparameters in a typical hidden node scenariowith protected wireless microphones in the TVwhite space. We consider all relevant effectsand use an analysis framework that properlycombines probabilistic technical factors to pro-vide specific policy recommendations includingthe exclusion zone distances and the sensing-based DSA threshold detection levels. First,man-made noise measurements were taken indifferent locations and the amount of interfer-ence from man-made noise in potential wirelessmicrophone channels was analyzed. Data collec-tion results show that man-made noise levelscan be up to 30 dB above the thermal noisefloor. Furthermore, indoor-to-outdoor path lossmeasurements were conducted to determine therequired exclusion distance for DSA-enabledTV band devices to ensure reliable wirelessmicrophone operation in a typical application.The results show that the required exclusionzone can be safely and conservatively set ataround 130 m when the results from man-madenoise measurements and wireless microphonepropagation measurements are used. Addition-ally, we developed a simulation to determinethe required DSA sensing threshold levels forimpairment-free wireless microphone operation.An indoor-to-outdoor path loss model was cre-ated based on the above path loss measurementresults. This statistical path loss model was usedto determine the received signal level at TVband devices and the interference level at thewireless microphone receiver. Our results showthat the sensing threshold can be set at around–110 dBm (in a 110 kHz channel) for impair-ment-free wireless microphone operation whenman-made noise and representative propagationmodels are used.


Tugba Erpek and Mark A. McHenry, Shared Spectrum Company

Andrew Stirling, Larkhill Consultancy Limited

Dynamic Spectrum Access OperationalParameters with Wireless Microphones

1 In the September 2010decision, the FCCremoved the sensingrequirement for TV banddevices that use the geolo-cation/database methodof interference avoidance.Nevertheless, it left thedoor open for sensing-only devices andincreased the minimumrequired detection thresh-old for wireless micro-phones from –114 dBm to–107 dBm.

MCHENRY LAYOUT 2/18/11 3:12 PM Page 38


potential to cause some level of interference towireless microphones; collaborative sensing willreduce the risk significantly. However, that paperdoes not consider many of the technical issuessuch as statistical multipath propagation andprobabilistic antenna front-to-back ratios. thatregulators are concerned with. Additionally, itdoes not provide specific DSA-sense based ruleparameters such as required sensing thresholdvalues. In [4] Gurney et al. argue that the geolo-cation method (dynamically updated databases)is better than spectrum sensing. Motorola alsosupports the use of geolocation databases by TVband devices [5]. Nevertheless, geolocation meth-ods have multiple drawbacks such as:• The worst case propagation and wireless

microphone temporal use assumptions thatlead to low spectrum use

• The cost and limitations of maintaining andbeing connected to TV station locationdatabases

In [6] Buchwald et al. and in [7] Yu-chun et al.propose a disabling beacon system design, whichwill protect the wireless microphones from DSAoperation. The beacon approach providesassured protection from TV band devices, butimplementation is expensive since the systemoperator needs to purchase and deploy a beacon[8]. As a result, compared to geolocation andbeacon signal methods, the sensing-basedmethod is the most suitable method to protectwireless microphones.

CONTRIBUTION OF THIS ARTICLEIn order to help regulators make more informeddecisions in protecting wireless microphonesfrom harmful interference, the results of ouranalysis provides specific technical performanceparameter recommendations.

Geographic-based DSA rule: The minimumseparation needed to avoid harmful interferencebetween a wireless microphone system and aDSA-enabled TV band device operating in thesame UHF channel. This separation defines theexclusion zone.

Sense-based DSA rule: The minimum level towhich DSA-enabled TV band devices wouldhave to sense, to ensure that they avoid using anoccupied channel. This is related to the exclu-sion zone since, by definition, there is no inter-ference risk from white space devices, which areoutside the zone.

The rest of this article is structured as fol-lows. The next section discusses the impact ofman-made noise on wireless microphone opera-tion. We then explain the geographic exclusionDSA method and the determination of reason-able exclusion distances. The sense-based DSAmethod and the determination of reasonabledetection threshold values are then described.Finally, conclusions are given in the final section.

IMPACT OF MAN-MADE NOISE ONWIRELESS MICROPHONE OPERATIONDSA operation should impact the operation ofwireless microphones for an amount that is lessthan but comparable to the performance limi-tations due to noise. Man-made noise is often

the dominant noise source, but is rarely consid-ered in DSA analysis. This section developswireless microphone performance estimates inthe presence of man-made noise. As part ofthis study, noise and interference measure-ments were conducted in a range of locations,including private dwellings and public venues,in the Tyson’s Corner area of northern Vir-ginia in April 2009.

A BRIEF LEXICON OF NOISENoise forms a backdrop to wireless communica-tions, determining the lowest signal level thatcan be received (i.e., the receiver sensitivity).There are two key sources of noise: thermalnoise and receiver noise.

Thermal noise (also known as Johnson-Nyquist noise) is generated in electrical conduc-tors at the radio frequency input of the receiver.These conductors include the antenna and anylead connecting it to the receiver.

The receiver also generates noise, furtherlimiting its sensitivity. This latter component ofnoise, quantified in the receiver’s noise figure, isa function of the nature and configuration of thecomponents in its RF input stage. This articlebrackets thermal noise and receiver noise togeth-er, and refers to the combination as the receptionnoise floor.

In addition to noise arising in the receiver,there may be signals arising from externalsources, which the receiver can detect. Thesemay be either wanted signals, from which thereceiver can extract useful information, orunwanted signals, which impair the receiver’sability to recover the wanted signal. The unwant-ed signals are often referred to collectively asinterference or man-made noise. In the case ofwireless microphone operation in the TV bands,common examples of unwanted signals includesignals from other wireless microphones operat-ing in the vicinity and television transmissions.Harmful interference is interference that seriouslydegrades, obstructs, or repeatedly interrupts aprotected service.

TV white space devices are also a potentialsource of interference, if operating in the samechannel and sufficiently close to the microphonereceiver. The enabling regulatory framework forTV white space devices includes measures toprotect wireless microphone operations fromharmful interference, which need to be based ona solid understanding of the interference risknew DSA-enabled devices pose.

In this article the risk of serious degradationto wireless microphone operation is gauged bydetermining the carrier-to-interference-and-noise ratio (CINR). CINR is the ratio betweenthe wanted signal (referred to as the carrier) andan aggregated unwanted signal, in which man-made noise, receiver noise, and thermal noiseare all included.

The result ing value can be compareddirectly with the minimum value of CINRneeded by a wireless microphone to ensurereliable operation. The minimum value ofCINR is not published, but manufacturersindicate that 25 dB is a representative figure,appearing in the ERA report for Ofcom oncognitive access [9].

Regulatory analyses

of wireless micro-

phone protection

requirements thus

far have largely

assumed that the

noise floor at the

wireless microphone

receiver is equal to

the reception noise

floor. Little account

has been taken of

potential sources of

interference other

than white space

devices.



ESTABLISHING THE LEVEL OFMAN-MADE NOISE

Regulatory analyses of wireless microphone pro-tection requirements thus far have largelyassumed that the noise floor at the wirelessmicrophone receiver is equal to the receptionnoise floor. Little account has been taken ofpotential sources of interference other thanwhite space devices.

Sources of man-made noise include televi-sion stations, electrical equipment in homes,offices, and factories, and so on. Studio andstage environments have their own sources ofnoise, particularly other wireless microphones,as well as lighting systems, lifting machinery,and so forth.

The characteristics of man-made noise arewell understood. Indeed, the ITU has long-established guidelines on typical levels of man-made noise that are to be expected in range ofdifferent locations. However, the ITU Recom-mendation provides only mean levels, which areinsufficient to determine the risk of interferenceto wireless microphones. It is the peak levels ofsuch noise that cause problems, because thehuman ear is sensitive to even brief interruptionsor artifacts in an audio signal. Therefore, wemade noise measurements that took the noise’stemporal variation into account.

MAN-MADE NOISE MEASUREMENTSAll of the man-made noise measurements weremade at ground level. It is possible that wirelessmicrophones were sometimes used in our experi-mental area. We made measurements of wirelessmicrophone signals using the same equipment,and then we visually compared our noise mea-surements to ensure that no wireless microphonesignals were present.

It is critical to select unoccupied test frequen-

cies to make the man-made noise measurements.We used TV channels 16, 19, 21, 28, 37, 53, 56,64, 65, and 69, which are unoccupied in theTysons Corner, Virginia area. This was verifiedusing rooftop antenna measurements on top of a10-story office building.

Measurements were made in 150 potentialwireless microphone channels (200 kHz band-width), distributed over the 10 vacant UHF TVchannels, at a number of measurement points ineach location.

The first step in the measurement processwas to determine the reception noise floor ofthe measurement system considering conduct-ed and non-conducted emissions. The recep-t ion noise f loor of the measurementequipment (using a 200 kHz bandwidth) wasconfirmed as –110 dBm, given a theoreticalthermal noise floor value of –121 dBm andequipment noise figure of 11 dB). Any signalvalue above this level was interpreted as man-made noise.

An example man-made noise measurement isshown in Fig. 1. Man-made signals are not Gaus-sian-type noise and contain large temporal andspectral features that are not Gaussian noise incharacter.

The carrier-to-reception-noise-ratio (CRNR)values given next in this article refer to a recep-tion noise floor of:• –115 dBm, in the man-made noise impact

sections, corresponding to a noise figure of6 dB and a channel bandwidth of 200 kHz

• –117.5 dBm, in the exclusion zone and sens-ing threshold section, corresponding to anoise figure of 6 dB and a channel band-width of 110 kHz

These bandwidth and noise figures were chosento allow comparison with the results of ERA’sanalysis for Ofcom [9].

A sample noise power distribution plot fromthe single-family house measurements is shownin Fig. 2. It can be seen from the figure thatman-made noise can range up to 30 dB abovethe thermal noise floor. The thermal noisefloor here is –115 dBm, using a bandwidth of200 kHz and a measurement system noise fig-ure of 6 dB.

ASSESSING THE IMPACT OF MAN-MADE NOISEIn order to assess the potential impact of man-made noise, we considered each of the noiselevel samples taken, per location. For each of arange of wanted signal (carrier) levels at thewireless microphone receiver and each measure-ment sample, we calculated the CINR. If theresult was greater than 25 dB, it was deemedthat the noise level was subcritical; thus, micro-phone operation would not have been impairedin that particular channel at that time and place.The results of the calculation across the mea-surement sample base for each location are sum-marized in Table 1:• The left column of the table indicates the

signal level at the receiver in terms of itsratio to the thermal noise floor (i.e.,CRNR). The adjacent column, on theright, shows the absolute signal powerlevel received by the wireless microphonereceiver.2

Figure 1. Typical man-made noise contains discrete spectral and temporal fea-tures that is significantly different than Gaussian noise.

Frequency (MHz)Time (ms)

Church Loc3 - antenna height 61 inches; date: 16-Apr-2009; time: 14-05-50

502.5

-100

Pow

er (

dBm

)

-110

-120

-90

-80

-70

-60

-50

-40

-30

502

503503.5

504504.5

505

50

0

100

150

-30

-40

-50

-60

-70

-80

-90

-100

-110

-120

2 This was obtained byadding the thermal noisefloor level (a constantwith value –115 dBm, cal-culated by adding thereceiver noise figure of 6dB to (–174 dBm/Hz over200 kHz,)) to the CRNRvalue in the first column.



• Each of the remaining cells in each rowgives a score for each location, at thegiven CRNR value, corresponding tothe rat io of the number of samples inwhich the no i se l eve l was found to besubcritical to the total number of noiselevel samples taken in that location. Forexample, i f , in 99 of 100 samples, thenoise level was found to be subcritical,the impairment-free score would havebeen 99 percent.

• The right column of the table shows animpairment-free score calculated from sam-ples aggregated across all the locationsused.It may be observed in Table 1 that a CRNR

of around 60 dB is needed to ensure impair-ment-free scores of 100 percent in all locations.Given that wireless microphones require a mini-mum CINR of 25 dB, this implies a man-madenoise increment of around (60 dB – 25 dB =) 35dB.

It is worth remembering that the noise mea-surements described above were made in subur-ban areas. Undoubtedly the man-made noiselevels in urban and metropolitan venues areeven greater.

USERS COMPENSATE FORMAN-MADE NOISE BY ENSURINGHIGHER RECEIVED SIGNAL LEVELS

To compensate for the relatively high level ofman-made noise experienced at most majorvenues, wireless microphone users need toensure that received signal levels are muchgreater than would be needed if thermal noisewere the only consideration. These augmentedsignal levels, achieved by minimizing the dis-tance between microphone and receiver, allowwireless microphone systems to tolerate muchhigher TV band device signal levels than regula-tors have so far assumed.

GEOGRAPHIC EXCLUSIONZONE DSA METHOD

In order to estimate the required separation ofa DSA-enabled TV band device from a wirelessmicrophone receiver using the same channel, itis necessary to be able to predict the propaga-tion loss on a path between the two devices.The measurement process, described below,enabled us to compile a database of propaga-tion values over a range of distances when thetransmitter is inside a building and the receiveris outdoors.

Table 1. The potential impact of man-made noise on wireless microphone operation.

WirelessmicrophoneCRNR (dB)

Receivedpower level(dBm)

Proportion of samples where microphone operation would not havebeen impaired (%)

Location

Indoor venueparking lots

Inside singlefamily house

Insidecondo

Wolf trapparking lot* All locations

10 –104.9 0% 0% 0% 0% 0%

20 –94.9 0% 0% 0% 0% 0%

30 –84.9 0% 0% 0% 0% 0%

40 –74.9 98.3% 91.8% 84.2% 99.1% 91.5%

50 –64.9 100% 99.3% 99.5% 100% 99.6%

60 –54.9 100% 100% 100% 100% 100%

70 –44.9 100% 100% 100% 100% 100%

* Wolf Trap is a public venue, which is known as a normally quiet location.

Figure 2. Noise level distribution from measurements taken inside a single-fam-ily house.

Noise power (dBm)

Distribution of noise power inside single family house

-110-115

100

0

Num

ber

of c

ases

200

300

400

500

600

700

-105 -100 -95 -90 -85 -80



PROPAGATION LOSS MEASUREMENTS

We carried out measurements of propagationloss for 4094 possible TV band device to wirelessmicrophone receiver separation distances, rang-ing up to 2.7 km. These were at three indoorpublic venues, which were not in use at the timeand whose wireless microphone systems hadbeen switched off.

An indoor test transmitter with an emissionpower of 20 dBm at 556.36 MHz was collocatedwith each venue’s wireless microphone receiver,and coupled to an omnidirectional antenna.

A test receiver, mounted in a van, measuredthe signal strength at a large number of locationsaround the outside of the venue.

The outdoor receiver was linked to an omni-directional antenna, mounted on the roof of thevan, with its height matched to that of the testtransmitter (2 m above the ground). This elevat-ed location for the receiver antenna means thatthe measurement results understate the likelypropagation loss suffered by a signal from a realTV band device, leading to a conservative exclu-sion zone estimate.

ESTIMATING THEREQUIRED EXCLUSION ZONE SIZE

Using the propagation loss measurementsacquired through the process described above,we were able to estimate the required separationof a DSA-enabled TV band device and a wire-less microphone receiver.

For each of a range of signal (carrier) levelsat the wireless microphone receiver, it was possi-ble to calculate the propagation loss required toprevent interference from a TV band device.The calculation assumed a TV band device trans-mission power density of 4.4 dBm in a 110 kHzchannel (equivalent to 20 dBm in a 4 MHz chan-nel). The TV band device was deemed not to

cause interference in a particular position whenthe CINR remained above 25 dB at the wirelessmicrophone receiver.

Since the measurements from each of theindoor venues were similar, it was reasonable tocombine them into a single data set consisting ofa grid of 1 dB by 1 m buckets. Figure 3 showsthe scatter plot for the measured path loss data,overlaid with a red horizontal line showing thepropagation loss required between the TV banddevice and the wireless microphone receiver toensure a CINR (for the wanted signal) greaterthan 25 dB when the wireless microphone signallevel (CRNR) at the receiver is 60 dB (i.e., awanted signal carrier level of greater than –54.9dBm; see Table 1). The matching vertical redline indicates the distance beyond which all datapoints had a propagation loss equal to or greaterthan the minimum needed (i.e., all data pointsfell below the horizontal line). This provides themost conservative (largest) estimate of the sizerequired for the exclusion zone.

The reception noise floor used in this analysisis –117.5 dBm, calculated using a bandwidth of110 kHz and assuming a receiver noise figure of6 dB.

At the microphone signal level illustrated bythe horizontal red line (–57.5 dBm, correspond-ing to CRNR = 60 dB), the required propaga-tion loss to avoid impairing microphoneoperation can be seen from Fig. 3 to be around–87 dB. This minimum value of propagation losscan be seen from the figure to have beenachieved at all possible values of distance greaterthan that marked by the vertical red line, whichcan therefore safely be chosen as the boundaryof the exclusion zone.

The results of estimating exclusion zone sizefor a range of CRNR values are summarized inTable 2. The proportion of measurements madeat distances greater than or equal to the chosenexclusion zone size that meet the minimumpropagation loss requirement is referred to hereas the impairment-free score. It corresponds tothe percentage of positions outside the chosenexclusion zone at which a DSA-enabled TVband device would not have impaired micro-phone operation when operating on the samechannel. A score of 100 percent means that aTV band device operating on the same channelas the wireless microphone would not causeinterference when located anywhere outside theexclusion zone.

The estimated exclusion zone sizes (radii)corresponding to each of a range of receivedmicrophone signal levels are presented in Table2. The rightmost two columns of the table showhow the exclusion zone could be contracted iflower levels of impairment risk were tolerable.

Since man-made noise is significantly higherthan thermal noise in areas where wirelessmicrophones are used, such systems are evident-ly deployed with a much higher received signalthan would be justified from assuming only thatthermal noise applied. It is estimated that thereceived signal level used is typically in excess of60 dB above the reception noise floor (i.e.,CRNR = 60 dB). Consulting Table 2, at thisreceived signal level, a requirement for 100 per-cent impairment-free operation given these mea-

Figure 3. Required TV band device exclusion zone size for a given wirelessmicrophone signal level.

Distance (m)

Prop loss, CINRmin = 25 dB, fc = 556.36 MHz

131 m exclusion distance

CRNR = 60 dB

5000

-160

-180

Prop

agat

ion

loss

(dB

)

-140

-120

-100

-80

-60

-40

1000 1500 2000 2500



surements leads to an exclusion zone for DSA-enabled TV band devices of radius 131 m.

The reception noise floor used here is –117.5dBm, calculated assuming a microphone receiverbandwidth of 110 kHz and a receiver noise fig-ure of 6 dB, to be comparable with the valuesused in ERA’s report on cognitive access forOfcom [9].

SENSE-BASED DSA METHODThe previous section established that takingman-made noise and realistic propagation lossesinto account provides significant scope for limit-ing the exclusion zone. Regulators in the UnitedStates and United Kingdom seem to preferDSA-enabled TV band devices find vacant UHFchannels through geolocation, whereby thedevices look up which channels are vacant attheir position in a database.

However, DSA-enabled devices may also relyon spectrum sensing in order to check that achannel is free. For these devices, it is importantthat sensing thresholds are sufficiently low toprotect microphones (and TV reception), butnot so low as to make TV band devices unneces-sarily costly, difficult to produce, and liable todetect unoccupied channels as occupied. In thissection we consider how these factors impact thesensing threshold requirement.

In general, the sense-based DSA method isable to estimate the link loss between the DSA-enabled transmitter and the protected transceiv-er by measuring the received power level fromthe protected transceiver and knowing its trans-mit power level. Estimating this link loss enablesthe DSA-enabled radio to adjust its transmitpower level (or to decide to transmit or not) toavoid causing unwanted interference to the pro-tected transceiver.

In the wireless microphone situation, a pro-tected wireless microphone receiver does nottransmit a signal; hence, the receiver is a hiddennode. The DSA-enabled TV band device esti-mates the minimum likely link loss between itand the wireless microphone receiver (L3) bymeasuring the wireless microphone to TV banddevice link loss (L2).

The DSA-enabled TV band device continual-ly measures the received signal level from thewireless microphone transmitter. If the receivedsignal level is above the sensing threshold, theTV band device does not transmit on the samechannel. Because of the hidden node problem,the risk of interference to a protected wirelessmicrophone is a complex statistical function ofthe sensing threshold value.

SIMULATION DESCRIPTIONTo establish a relationship between impairment-free wireless microphone operation and thevalue chosen for the sensing threshold, aroundone million randomly chosen possible combina-tions of wireless microphone, wireless micro-phone receiver, and DSA-enabled TV banddevice positions were considered using the prop-agation loss data gathered as described above.For each position combination, a calculation wasmade of whether or not microphone operationmight have been impaired.

In order to generate the large number of pos-sible position combinations required, we used aMonte Carlo simulation. Fixing the wirelessmicrophone receiver at the center, the simula-tion generated one million different combina-tions of wireless microphone (transmitter) andTV band device positions over an area of 1 km2.

The basis for the simulation was as follows:• The wireless microphone receiver was posi-

tioned at the center of the grid.• The wireless microphone was limited to

positions within a 100 m square subset ofthe 1 km2 grid.

• The TV band device was allowed to rangeanywhere within the 1 km × 1 km grid.

• For each point in the simulation, a propaga-tion loss value was chosen at random fromthe values measured earlier for the givendistance between wireless microphone andTV band device.

• In 5 percent of the points, the propagationloss was increased by 20 dB to account forbody loss (amounting to 50,000 of the 1million simulated cases).

• The wireless microphone transmissionpower was taken as 14.8 dBm, with a sys-tem bandwidth of 110 kHz and a noise fig-ure of 6 dB used for the wirelessmicrophone receiver, yielding a receptionnoise floor of –117 dBm.

• The TV band device’s transmission powerwas taken as 20 dBm (100 mW EIRP) with-in a transmission bandwidth of 4 MHz,amounting to 4.4 dBm in a 110 kHz chan-nel.The propagation loss model used in the simu-

lation drew directly on the measurementsdescribed in the previous section. It was appliedto transmissions between wireless microphoneand TV band device as well as between TV banddevice and wireless microphone receiver, usingthe assumption that the model was applicable toall paths ending within the central 100 m2 zoneallowed for microphone roaming in the simula-tion.

No path loss assumptions were made to cal-culate the received signal level at the wireless

Table 2. Estimates of required exclusion-zone size for the data set in Fig. 1 (TVband device transmission power of 20 dBm into 4 MHz).

Impairment-free microphoneoperation score



100% 99.9% 99%

Dynamic spectrum access devicedetection threshold (dBm)

30 –87.5 732 513 280

40 –77.5 304 246 132

50 –67.5 187 131 64

60 –57.5 131 82 <50

70 –47.5 81 52 51



microphone receiver from the wireless micro-phone transmitter. The CRNR values used inTables 1 and 2 (30 dB, 40 dB, etc.) were used tocalculate the received signal level at the wirelessmicrophone receiver in the absence of TV banddevices.

For each synthesized position combinationgenerated by the simulation, the distancesbetween the DSA-enabled TV band device andwireless microphone transmitter, and TV banddevice and wireless microphone receiver werecalculated, and corresponding propagation lossvalues were retrieved from the propagation lossmeasurement base. Since the measurement baseincluded a number of possible propagation lossvalues for each value of distance, the particularvalue retrieved by the simulation was chosen atrandom from the set of applicable values for thedistance in question. In 5 percent of cases, 20 dBwas added to the propagation loss to simulatethe effect of body absorption [10].

The sensing threshold vs. failure rate plotswere created at the end of the simulation foreach different CRNR level.

ESTIMATING THE DSA SENSING THRESHOLDThe results of the simulation are presented inTable 3, with estimated sensing thresholds corre-sponding to a range of possible wanted signallevels at the microphone receiver. In the thirdcolumn, the sensing threshold value givenensures 100 percent impairment-free operation— meaning that in all the cases in the simula-tion, either the wireless microphone signal wasdetected by the DSA-enabled TV band device,or the TV band device was sufficiently separatedfrom the wireless microphone receiver for itstransmissions not to impair microphone opera-tion. For example, if the CRNR equaled 60 dB,a sensing threshold requirement of –111 dBmfor the DSA-enabled TV band device wouldhave been sufficient to protect wireless micro-phone operation from impairment when bothwere using the same microphone channel. Eitherthe TV band device would have been able todetect the wireless microphone at the specifiedthreshold, and it would have moved to an alter-

native channel, or its signal would have been suf-ficiently attenuated at the wireless microphonereceiver.

The two rightmost columns of Table 3 showhow the required sensing threshold could berelaxed if a small impairment risk to microphoneoperation were tolerable.

The reception noise floor used as the refer-ence for CRNR here is –117.5 dBm, calculatedassuming a bandwidth of 110 kHz and an equip-ment noise figure of 6 dB, comparable withERA’s analysis for Ofcom [9].

CONCLUSIONSThe conclusions of this study are as follows.

MAN-MADE NOISEThe effects of man-made noise have not proper-ly been taken into account in protection analysesto date. The reception noise floor has been con-sidered while determining the maximum allow-able interference-to-noise ratio for DSA-enabledTV band devices. On the other hand, many otherdevices impact wireless microphone operationmore than TV band devices, and this study showsthat man-made noise is one of the dominant fac-tors that interferes with wireless microphoneoperation.

Our measurements show that the peak man-made noise level can go up to 30 dB above thereception noise floor. As a result, wireless micro-phones have to have high CRNR values (>60dB) in order to operate reliably. Since wirelessmicrophones have high signal margins, interfer-ence from DSA-enabled TV band devices will benegligible. Man-made noise levels should betaken into consideration while determining therequirements for DSA operation. A reasonablerequirement for DSA-enabled devices would beto impact the noise level by no more than 10 dBless than 3 percent of the time.

DSA EXCLUSION DISTANCE METHODWe conducted propagation loss measurements insuburban areas to determine the required exclu-sion distance for DSA-enabled TV band devices.When man-made noise and representative prop-agation models are used, the required exclusionzone can be safely and conservatively set ataround 130 m.3

DSA SENSING METHODThe sensing threshold depends on statisticalparameters. DSA-enabled TV band devices canmeasure the path loss between themselves and aprotected wireless microphone transmitter, butthey cannot measure the path loss betweenthemselves and a wireless microphone receiver.As a result, there is a hidden node factor in thewireless microphone sensing threshold calcula-tions. Furthermore, multipath, blockage, andbody loss factors make the detection of wirelessmicrophone signals more difficult. All probabilis-tic parameters should be considered togetherwhile calculating the required sensing thresholdlevel instead of combining the worst case of eachof these individual factors. On the other hand,wireless microphone receivers always experiencehigh interference levels because of man-made

Table 3. Estimated sensing threshold values for DSA-enabled TV band devices(with TV band device transmission power of 20 dBm into a 4 MHz channel).

Impairment-free microphoneoperation score



100% 99.9% 99%

Dynamic spectrum access devicedetection threshold (dBm)

30 –87.5 –144 –144 –141

40 –77.5 –133 –122 –98

50 –67.5 –119 –101 –84

60 –57.5 –111 –85 –69

70 –47.5 –104 –71 >-60

3 The FCC recently modi-fied its rules for TV banddevices to reduce theexclusion distance from 1km to 400 m for devicesoperating at 100 mWEIRP or less transmitpower [1b, paragraph 61].



noise, co-channel wireless microphone signals,and broadcast TV signals. Considering the factthat wireless microphones have to have high sig-nal margins in order to operate properly, whenman-made noise and representative propagationmodels are used the required sensing thresholdcan be set at around –110 dBm (in a 110 kHzchannel).4

REFERENCES[1] FCC, “Unlicensed Operation in the TV Broadcast Bands,”

a) Second Report and Order and Memorandum Opinionand Order, 23 FCC Rcd 16807, Nov. 2008; b) SecondMemorandum Opinion and Order, FCC 10-174, Sept.23, 2010.

[2] Ofcom, “Digital Dividend: Cognitive Access, Statementn Licence-Exempting Cognitive Devices Using Inter-leaved Spectrum,” July 2009.

[3] R. S. Dhillon and T. X. Brown, “Models for AnalyzingCognitive Radio Interference to Wireless Microphonesin TV Bands,” IEEE DySPAN, 2008.

[4] D. Gurney et al., “Geo-Location Database Techniquesfor Incumbent Protection in the TV White Space,” IEEEDySPAN, 2008.

[5] “Motorola FCC Ex-Parte Filing,” Oct. 2007, pp. 2–10.[6] G. J. Buchwald et al., “The Design and Operation of the

IEEE 802.22.1 Disabling Beacon for the Protection of TVWhitespace Incumbents,” IEEE DySPAN, 2008.

[7] W. Yu-chun, W. Haiguang, and P. Zhang, “Protection ofWireless Microphones in IEEE 802.22 Cognitive RadioNetworks,” IEEE ICC, 2009.

[8] A. Hartman and E. Reihl, “Mitigating the Effects of Unli-censed Devices on Wireless Microphones,” Shure Inc.,Presentation, Nov. 2005.

[9] ERA, “Report for Ofcom on Cognitive Access,” Jan. 2009.[10] Analysis of PMSE Wireless Microphone Body Loss

Effects, COBHAM Technical Services, ERA TechnologyReport, Page 18, 2009.

ADDITIONAL READING[1] Mass Consultants Limited, “Man-Made Noise Measurement

Programme Final Report,” no. 2, Sept. 2003, p. 58.[2] Red-M, “Predicting Path Loss between Terminals of Low

Height,” Final Report, Feb. 26, 2007, Fig. 23 (a), p. 40.

BIOGRAPHIESTUGBA ERPEK ([email protected]) has beenemployed by Shared Spectrum Company as a systems engi-neer since June 2007. Her areas of interest include DSAsystem analysis, predicting interference statistics, and per-formance prediction; development and testing of DSAdetectors for the TV spectrum band; man-made noise spec-trum measurements; propagation loss model creation andvalidation measurements; and spectrum occupancy datacollection and analysis. She earned an M.S. degree in elec-trical engineering from George Mason University (GMU)with a specialization in digital signal processing and wire-less communications. She worked as a research assistant inthe Network Architecture and Performance Laboratory at

GMU. Her research topics included dynamic spectrum shar-ing in wireless networks. She graduated with a B.S. degreein electrical engineering from Osmangazi University, Eskise-hir, Turkey.

MARK A. MCHENRY ([email protected]) hasexperience in military and commercial communication sys-tems design, including research of the next generation ofadvanced wireless networks, the development of highdynamic range multi-band transceivers, as well as foundingof two high-tech companies with concentration in auto-mated spectrum management technology. In 2000 hefounded Shared Spectrum Company (SSC), which is devel-oping automated spectrum sharing technology. SharedSpectrum Company develops advanced technologies forgovernment and industry customers with challenging radiofrequency and networking needs. He was also a co-founderof San Diego Research Center (acquitted by Argon ST), awireless research and developing company supporting thetest and training community. Previously he was a programmanager at the Defense Advanced Research Projects Agen-cy (DARPA), where he managed multiple programs. He hasworked as an engineer at SRI International, NorthropAdvanced Systems, McDonnell Douglas Astronautics, Hugh-es Aircraft, and Ford Aerospace. He received the Office ofSecretary of Defense Award for Outstanding Achievementin 1997 and the Office of Secretary of Defense ExceptionalPublic Service Award in 2000. He has multiple RF technolo-gy related patents. He graduated with a B.S. in engineeringand applied science from the California Institute of Tech-nology in 1980. He received an M.S. in electrical engineer-ing from the University of Colorado and a Ph.D. in electricalengineering from Stanford University. He was named Engi-neer of the Year by the DC Council of Engineering andArchitectural Societies in February 2006. He was appointedby Secretary of Commerce Carlos Gutierrez to serve as amember of the Commerce Spectrum Advisory Committeein December 2006 and 2008.

ANDREW STIRLING ([email protected]) foundedLarkhill Consultancy Limited in 2005 to assist companiesdeveloping business around new digital distribution tech-nologies, providing regulatory and strategy input. Afterstudying physics at Imperial College, he started his careerat BBC R&D developing digital content production and dis-tribution technology. As systems group manager atPhilips/Panasonic JV he developed AV networking technolo-gy, which he helped Mercedes-Benz and its key suppliersengineer into mass production cars. He is named as inven-tor on a number of patents related to local area communi-cations. He joined consultants Cambridge Consultants/Arthur D. Little as a manager, where he assisted investorsin evaluating new businesses and helped multinationalplayers develop digital strategies. As strategy manager atITC/Ofcom he developed digital TV switchover and relatedspectrum policy, authoring a landmark report with the BBCfor the U.K. government. Larkhill's clients have included BT,the BBC, Dell, Microsoft, and the U.K. government. Onbehalf of Microsoft, Larkhill facilitates an informal coalitionof major companies interested in exploiting the TV whitespaces,- focusing on the United Kingdom and Europe. Hesits on the External Advisory Board for the EU QoSMoSproject, which is looking at how Europe can best exploitthe opportunities from secondary spectrum access.

4 The FCC recently modi-fied its rules for sensing-only TV band devices toincrease the minimumrequired detection thresh-old for wireless micro-phones from –114 dBm to–107 dBm, averaged overa 200 kHz bandwidth[1b] at 35 and AppendixB at 15.717(c).


INTRODUCTION

The radio frequency spectrum is a highly regu-lated resource whose management is usuallydeferred to a government agency in most coun-tries. The tasks related to spectrum managementencompass all the activities related to regulatingthis resource, including spectrum allocation andassignment of spectrum as well as regulationenforcement activities. For our purposes, spec-trum allocation refers to defining acceptableuses of certain bands (e.g., FM radio), whereasspectrum assignment is the process of grantingrights to particular users in a band that has beenallocated (e.g., a radio station).

Traditional spectrum allocation and assign-ment mechanisms have focused on avoidinginterference between users and on the type ofuse given to spectrum rather than on the effi-cient use of spectrum and the maximization ofeconomic benefits. Due to this, most of the spec-trum is used suboptimally most of the time withlow average occupancy values (less than 10 per-cent as reported in [1]).

Additionally, managing spectrum has become

increasingly difficult for regulatory agencies due tothe new technologies and uses for spectrum thatare continuously emerging and placing increasingdemands on this resource. Thus, more flexibleassignment mechanisms have to be put in place toadjust to this new reality while still achieving thebest usage of spectrum possible under economicor social welfare considerations [2].

Spectrum auctions have become a commontechnique for regulatory agencies to assign spec-trum to new users. Once the auction is over,however, the license holders do not get feedbackabout the current valuation of spectrum. Foreconomically driven spectrum assignment to beoptimally effective, a secondary market mustexist that allows spectrum users to optimallychoose between capital investment and spectrumuse on a continuous basis, not just at the time ofinitial assignment [2]. The interactions in themarket should take into account the geographicreusability and non-perishable characteristics ofspectrum, which make its trading different thantrading traditional market commodities.

Unlike much of the dynamic spectrum assign-ment (DSA) literature, which focuses on non-cooperative sharing, spectrum trading is a formof primary cooperative spectrum sharing thatinvolves permanent license transfers for econom-ic consideration [3]. Thus, it assigns spectrum tothose who value it most, allowing for the estab-lishment of dynamic market-driven and competi-tive wireless communication markets.

Spectrum trading markets are of growinginterest to many spectrum management agen-cies. They are motivated by their desire toincrease the use of market-based mechanismsfor spectrum management to increase spectrumefficiency. The research reported here is, inmany ways, a best case analysis to determine theviability of those markets.

PARTICIPANTS IN ASPECTRUM TRADING MARKET

To understand the organization of and interac-tions in a spectrum trading (ST) market, weneed to know what entities participate in such amarket. In [4] we elaborated a classification formarket structures that support ST. This classifi-cation considered two main types of market


ABSTRACT

Spectrum trading markets are of growinginterest to many spectrum management agen-cies. They are motivated by their desire toincrease the use of market-based mechanismsfor spectrum management and reduce theiremphasis on command and control methods.Despite the liberalization of regulations on spec-trum trading in some countries, spectrum mar-kets have not yet emerged as a key spectrumassignment component. The lack of liquidity inthese markets is sometimes cited as a primaryfactor in this outcome. This work focuses ondetermining the conditions for viability of spec-trum trading markets. We make use of agent-based computational economics to analyzedifferent market scenario and the behaviors ofits participants. Our results provide guidelinesregarding the number of market participants andthe amount of tradable spectrum that should bepresent in a spectrum trading market for it to beviable. We use the results of this analysis tomake recommendations for the design of thesemarkets.


Carlos E. Caicedo, Syracuse University

Martin B. H. Weiss, University of Pittsburgh

The Viability ofSpectrum Trading Markets

CAICEDO LAYOUT 2/18/11 3:03 PM Page 46

structures for spectrum trading: over-the-counter(OTC) markets and exchange-based marketoperation. We focus on the exchange-based casein this article.

Figure 1 illustrates an exchange-based tradingscenario. In this scenario the exchange collectsthe offers to sell and offers to buy (bids) spec-trum, determines the winning bid, and transfersthe spectrum usage right from the selling spec-trum user to the new owner of that right. Theentities that participate in exchange-based STmarkets are the following.

SPECTRUM EXCHANGEAn entity that provides and maintains a market-place or facilities for bringing together buyersand sellers of spectrum in which spectrum trad-ing transactions can take place. It also publicizesprices and anonymizes trading entities.

SPECTRUM USERAn entity that uses spectrum for a particularpurpose. A spectrum user (SU) might be actingin one of the following roles at a given momentin time:• Spectrum license holder (SLH): An entity

that owns a spectrum license and offers itfor trading in exchange for financial com-pensation. This entity can be a wireless ser-vice provider, market maker, or spectrumexchange that has been assigned a spectrumtrading band by a regulatory agency. Ingeneral, SLHs hold spectrum for specula-tion or for their own use.

• Spectrum license requestor (SLR): An enti-ty that submits bids for spectrum licenses tothe ST market with the intent to acquirethe license. SLRs obtain spectrum for spec-ulation or their own use. An entity that actsas an SLR can be a wireless serviceprovider, market maker, or company/enter-prise that acquires spectrum on behalf ofanother.

SPECTRUM REGULATORA spectrum regulator is a government entity thatoversees the ST market and defines the regula-tions for its operation. It is also responsible formaintaining a spectrum availability and assign-ment database which is updated every time aspectrum trade is completed to register the iden-tity of the new holder of spectrum.

MARKET MAKERSA market maker facilitates trading; it does notprovide services with its inventory. It acts as adealer that holds an inventory of spectrum andstands ready to trade when an SLR (buyer) orSLH (seller) desires. It gets revenue through thespread between the sell and buy prices for spec-trum, and holds a spectrum inventory for negoti-ating and speculating.

EXCHANGE-BASED ST MARKETSIn exchange-based ST markets, the spectrumexchange is the central entity for market opera-tion. In general, an exchange denotes the idea ofa central facility where buyers and sellers cantransact and which may charge fees for its ser-

vices. In the traditional sense, an exchange isusually involved in the delivery of the product.For a spectrum exchange to allow use of tradedspectrum, the required devices do not need to becollocated in the exchange, so the exchangemight not be involved in the delivery of service.We assume that spectrum exchanges make useof continuous double auctions as a mechanismto match buyers and sellers.

We consider that the spectrum exchange actsas a pooling point (POOL) if its facilities housethe communication equipment that enable thedelivery of wireless services through spectrumacquired by a buyer in the exchange. This kindof exchange also takes care of the configurationof equipment required to make the spectrumusable to the new license holder. A non-poolingpoint exchange (NOPOOL) only delivers theauthorization for use of spectrum to the buyingparty in a spectrum trade. The new SLH mustthen use this authorization to configure itsdevices to make use of the spectrum it has justacquired.

From a functional perspective a spectrumexchange can be a band manager (BM) for agiven segment of spectrum over a region or haveno band manager functionality (NOBM). BMexchanges support-leasing arrangements in addi-tion to permanent license transfers. In contrastto BM exchanges, a NOBM exchange will onlyfacilitate the trading of spectrum among entitiesin the market without holding any spectruminventory itself.

For scenarios where the exchange has BMfunctionality, SLRs send a request for spectrumto the exchange, which, if possible, will assignspectrum to the requesting entity in the form ofa timed lease within the band managed by theexchange. For a NOBM exchange the spectrumit will handle for trading will come from marketparticipants that use the exchange and makebids and offers of spectrum. It is worth mention-ing that unless the market has defined a basic


Figure 1. Spectrum trading scenario.

Regulator

Spectrum user: needsspectrum and wants

to buy it.

Trading finalization

Spectrum user: ownsspectrum and wants

to sell it.

Spectrum exchange

Posti

ngs o

f tra

dable

spec

trum

Rece

ption o

f bids

Information

gathering

Submission of bids



amount of bandwidth as a spectrum-trading unit,it will be very complicated to match bids andoffers of spectrum without incurring wastefulassignment of this resource. Although giving aparticular structure to the way a spectrum-trad-ing band should be segmented will limit its oper-ational flexibility, it also provides benefits interms of simplifying the specifications to charac-terize a particular spectrum trade and managinginterference between ST users.

From the previous discussion, the proposedclassification generates four types of spectrumexchanges that can be used to implement an STmarket. These are listed in Table 1.

MODELING OF ST MARKETSWe use agent-based computational economics(ACE) to analyze spectrum trading markets andthe behaviors of their participants which, giventheir variety, would be difficult to study withconventional statistical and analytical tools. ACEis “the computational study of economic process-es modeled as dynamic systems of interactingagents” [5]. An agent in an ACE model is a soft-ware entity with defined data and behavior.Agents can represent individuals, institutions,firms, and physical entities. ACE methods havebeen used to study cooperative secondary spec-trum sharing (in which temporary usage rightsare negotiated) [6].

A spectrum trading market modeling tool(SPECTRAD) has been developed as part ofour research work, and makes use of ACE meth-ods and concepts. With this tool, we model theparticipants in a ST market over a set of differ-ent scenarios. Our focus is to determine the con-ditions for viability of these markets. We definea viable spectrum trading market as one thatpossesses good liquidity and sustainability char-acteristics. As a first step in our analysis we havechosen to examine spectrum markets where onlyone wireless standard is used, and the unit seg-ments of tradable frequency have equal opera-tional conditions (fungibility). Future work will

examine more complicated and realistic scenar-ios. This is the best case scenario for liquidity.

When modeling markets, the agents repre-senting market participants have limited (if any)knowledge of the decisions and state of othermarket participants (bounded rationality). Agentsadapt their behavior based on their goals, andtheir interaction with the market and/or otheragents. In ACE modeling, once initial conditionshave been specified, the evolution of the model isonly dependent on the interactions among agents,and given the diversity of interactions that canarise, it is difficult to perform straightforwardcausal analysis by tracking one market partici-pant. Thus, ACE models provide a tool toobserve the aggregate behaviors that emerge ona system from the individual behaviors of itscomponents (agents). Analysis of these behaviorsover several scenarios can provide insights intocharacteristics of new markets, the effect of eco-nomic policies, and the roles of institutions.

By characterizing the trading, informationoverhead, and infrastructure costs of differentST market implementation architectures, andsince we are interested in the running behavior(sustainability) of the market once its operatinginfrastructure has been put in place, we find thatthe only differentiating factor between them iswhether the exchange is organized to work as aband manager (BM) or not (NOBM) [2]. Forthe sake of brevity we mention our model resultswith NOBM scenarios in this article.

The following subsections describe theassumptions and behaviors of the market entitiesused in our models. Further details of the imple-mentation of our models can be found in [2].

GENERAL MARKET SETUP ANDMODEL ASSUMPTIONS

The following are the assumptions used in ourmodels:• Interference conditions do not impact the

services provided over a unit of tradedspectrum.

Table 1. Types of exchanges.

Exchange type Characteristics

POOL_BM

Pooling point + band manager functionality• Use of traded spectrum is enabled and configured through equipment/infrastructure owned by the exchange.• All tradable spectrum is held by the exchange• All tradable spectrum returns to or is given by the exchange

POOL_NOBM

Pooling point only, no band manager functionality• Use of traded spectrum is enabled and configured through equipment/infrastructure owned by the exchange.• Different segments of spectrum can be activated and configured through the equipment/infrastructure of the

exchange• No spectrum inventory is held by the exchange

NOPOOL_BM

Non-pooling point + band manager functionality• All tradable spectrum is held by the exchange• All tradable spectrum returns to or is given by the exchange• Exchange grants authorizations for use of spectrum (no equipment configuration is done by the exchange)

NOPOOL_NOBMNon-pooling point, no band manager functionality• No spectrum inventory is held by the exchange• Exchange grants authorizations for use of spectrum (no equipment configuration is done by the exchange)



• Trading takes place over an exchange entityand over a single geographic service area inwhich the wireless services providers (mod-eled by SU agents) have enough radio basestations to provide service coverage.A market scenario is defined by the following

set of parameters:• Number of market participants (numSU)• Distribution of spectrum users’ valuation

level (L)• Amount of available spectrum for trading

(S)In the market scenarios considered, wireless

service requests manifest to each SU as requestsof traffic to be served for which the SU has todetermine if it has sufficient resources. The SUscan obtain resources to serve traffic by eitheracquiring spectrum in the form of basic band-width units (BBUs) or using a unit of transmis-sion of an alternate technology (AT).

Investment in AT transmission units can resem-ble investing in equipment to make better use ofspectrum already owned by the SU or in wirelinetechnology, thus avoiding the purchase of addi-tional BBUs. The choice between BBUs or ATswill be based on the economic benefit a given SUmight receive from making a selection as it tries tominimize its costs for providing wireless service.Each SU will have a fixed price for its choice ofAT unit which does not change during the life ofthe market. Thus, if an SU is acting as a spectrumlicense requestor (i.e., buyer) when the marketprice for a BBU is higher than its AT price, theSU will buy ATs; and when BBU prices are loweror equal to its AT price, the SU will buy BBUs.

BEHAVIORS FOR ST MARKET ENTITIESSpectrum User Behavior — Spectrum usersare the agents that model wireless service pro-viders (WSPs), and buy and sell spectrum inorder to meet traffic requests (buy) or obtaineconomic gain (sell). For our analysis we modelthe aggregate traffic demand for each SU withinthe ST service area with an exponential distribu-tion with a mean of μtraffic. The interval betweenchanges of traffic demand is modeled as anexponential distribution with a mean of μtchange.In this model we assume that the traffic demandsfaced by SUs are not correlated.

The SUs submit requests to buy (bids) or sell(asks) to the exchange. The exchange collectsthese requests and tries to find the best matchbetween requests to establish a trade. The SU canquery the exchange for its current market quote,which contains the minimum ask and maximumbid price posted in the market. SUs use this infor-mation in their pricing decisions. Additionally, anSU can post buy/sell orders that should executeimmediately at the current best price of the mar-ket (market order) or specify to the exchange itsdesire to buy/sell BBUs at the best price possiblebut in no event pay/sell for more/less than a speci-fied limit price (limit orders). If an SU buys ATunits, it is aware that they have a finite lifetimeand should be decommissioned in the futurebased on their mean lifetime.

NOBM Spectrum Exchange Behavior — Weassume a continuous order-driven market inwhich SUs may trade at any time they choose.

After each order is posted, the exchange updatesits order book, and if a trade can take place, ittransfers the spectrum license from the seller tothe buyer and records the details of the tradingtransaction. It also informs the regulator agentabout the trade so that it can keep track of theowner of each BBU.

After each trade or if there was no trade, theexchange announces the market quote, informingmarket participants of the current market askprice (best price at which spectrum is being soldin the market) and the current market bid price(price of the best offer to buy spectrum in themarket). This allows market participants toadapt their price behavior to make competitivebids or asks in the future.

Market Maker — The market maker (MM)provides liquidity to the market and correctsmarket imbalances. In our model the MM agentstands ready to make bids for spectrum if no SUis posting a bid, and it posts an offer to sell if noSU is on the selling side of the market. Thismakes the MM a very reactive entity that onlyintervenes in the market when there is a severeimbalance (i.e., no buyers or no sellers) in orderto keep the market alive. Using a simplified MMallows us to determine which market scenariosare viable without much economic interventionfrom entities that do not provide wireless ser-vices.

The MM has an initial inventory of BBUsassigned to it which it uses to keep a bid-ask(buy-sell) spread present at all times in the mar-ket. When market intervention by the MM is notrequired, the MM will issue a bid or ask with theobjective of getting its spectrum inventory backto its reference level, which is the same as its ini-tial spectrum inventory amount.

Regulator Agent — A regulator agent modelsa regulator entity, oversees the trades being con-ducted in the market, and updates a spectrumassignment database so that ownership of agiven BBU can be verified if needed.

FACTORS FOR ST MARKET VIABILITYOur focus is on determining the conditions forviable ST markets with respect to their liquidityand sustainability characteristics, so we selecteda set of parameters/measures that capture themain characteristics of these markets.

•Midpoint price for spectrum BBU: Thisprice gives an indication of the average price atwhich a BBU is being valued in the market. Lowvalues of this measure indicate an excess in sup-ply or low spectrum demand in the market, whilehigh values indicate low supply or high demandfor spectrum.

•Relative bid/ask spread: The bid/ask spreadis the difference between the best sell price andthe best buy price in the market. The relativebid/ask spread is the size of the bid/ask spreadrelative to the midpoint price of the quotedasset. This factor can be used as an indicator ofthe liquidity of a market [7,8]. In other words,high values of this parameter indicate low liquid-ity in the market, while low values would indi-cate high liquidity.

In our model the

market maker agent

stands ready to

make bids for

spectrum if no SU is

posting a bid, and it

posts an offer to sell

if no SU is on the

selling side of the

market. This makes

the market maker a

very reactive entity

that only intervenes

in the market when

there is a severe

imbalance.



•MM’s BBU inventory difference: Thisparameter tracks the difference between theMM’s current spectrum holdings (BBU invento-ry) and its reference level of inventory units.Substantial deviations from the reference levelwould indicate problems in the buying or sellingside of the market.

•Percentage of offered spectrum: Expressingthe amount of spectrum being offered for sale asa percentage of the total spectrum available inthe market, we find that if this percentage ishigh, the majority of the tradable spectrum isnot in use by the SUs and thus is being offeredfor sale. This would indicate low spectrum effi-ciency. In general, the lower the value of thispercentage, the more efficiency there is in termsof spectrum use.

•Number of complete market runs: For thecollection of statistics to analyze each marketscenario, 100 runs of each scenario are per-formed in SPECTRAD. Activity in a marketstarts with a series of mock auctions so that theSUs can find an initial starting price for trading.If this initial phase is not successful in finding astarting price, the market does not proceed toactual spectrum trading. This factor counts howmany of the attempted market runs were suc-cessful in finding a stable starting price and thusable to initiate spectrum trading. It is useful asan indicator of market viability since a high per-centage of complete market runs indicates thatinitiating trading is feasible and sustainable with-

out difficulty. In contrast, having a low percent-age of complete market runs would indicate thatthe market structure is not well suited to supportsustainable trading.

All of these parameters were found to beapplicable to NOBM markets. A similar analysisand determination of viability factors was con-ducted for BM markets, but it is not discussedhere. The reader can find more details on BMcases in [9].

VIABILITY CRITERIA AND RESULTSA summary of the values used for the parame-ters of the market scenarios modeled is shown inTable 2. Each run of the model was executed for5000 time ticks of which 3000 were used for thewarm-up period. Data was collected on the last2000 time ticks.

In the scenarios modeled, the variation of thetradable spectrum amount and number of SUsare related in such a way that the value of theaverage number of BBUs per spectrum user (R) isin the set {5, 10, 15, 20, 25}. For all scenarios,when R is equal to 10, on average every spec-trum user has enough spectrum to serve its aver-age traffic requirement value. Thus, R valueslower than 10 indicate an under-supply of spec-trum, while values greater than 10 indicate anover-supply of this resource to attend the aver-age traffic needs of a SU.

In order to determine the viable NOBM mar-kets based on the factors mentioned in the previ-ous section, we developed decision criteria todetermine if the behavior of a particular factorin a market is to be considered as desirable/acceptable or undesirable/unacceptable. Addi-tionally, in order to keep track of the aggregatebehavior characteristics of a market, we gave ascore to each factor with a positive value whenthe market complies positively with the desiredbehavior characteristic or a negative one when itcomplies with the undesirable behavior criteria.Based on the total scores for a market’s behav-ior, a final list of viable markets was obtained.

Most of the threshold values for each criteri-on were derived from the simulation data by tak-

Table 2. Parameters for modeled markets.

Parameter Values

Number of spectrumusers (numSU) 4, 5, 6, 10, 20, 50 (this number includes one market maker)

Distribution of spec-trum users’ valuationlevel (L) Table indicatesproportion of spectrumusers at a given valua-tion level(willingness to pay)

Dist # Low Medium High

1 1/3 1/3 1/3

2 1/2 1/4 1/4

3 1/4 1/4 1/2

Available spectrum (S)Values indicate thenumber of BBUs avail-able for trading

5*numSU, 10*numSU, 15*numSU, 20*numSU, 25*numSU.The amounts of spectrum where chosen for each value of numSU in order to have

R =⎛

⎝⎜

⎞

⎠⎟ { }

S

numSU in the set 5 10 15 20 25, , , ,

Table 3. Criteria for NOBM market scenario evaluation.

ID Factor Pass (P) Fail (F) Score P/F

C1 Percentage of completed market runs ≥70% ≤50% 2/–2

C2 Relative bid-ask spread ≤20% ≥50% 1/–1

C3 Mid-point BBU price ≥100 ≤25 1/–1

C4 Relative difference of the MM’sinventory to its reference level ≤25% ≥100% 1/–1



ing into account the values that determine break-points for different behaviors for a scenario.Table 3 summarizes the criteria to be used toevaluate and give scores to the different scenar-ios studied in this work. Factors such as the per-centage of completed market runs where givenmore weight than other factors given their rela-tive importance in the determination of viabilitycharacteristics (sustainability in this case).

VIABLE NOBM MARKETSUsing the list of criteria for scenario evaluationmentioned in Table 3, the scores for the simulat-ed scenarios are summarized in Fig. 2. We con-sider markets with scores greater than 0 asviable. Scenarios with this condition met severalof the desirable conditions for a viable market;in particular, they all have a percentage of run-ning markets greater than 50 percent. We onlyshow the scores for scenarios with user valuationdistribution #1 (as defined in Table 2) sincethere was no difference in terms of viable mar-kets among scenarios with different valuationlevel distributions.

Based on the scores, we can say that most ofthe viable market scenarios are those that haveR values that meet the condition 5 ≤ R ≤ 10 anda number of spectrum users (numSU) such that6 ≤ numSU ≤ 50. When R = 15, the viable sce-narios are those with 10 ≤ numSU ≤ 20. Figure 3shows the spectrum efficiency results from ourmodeled scenarios.

VIABILITY IMPLICATIONSNOBM spectrum trading markets are viableunder the criteria used in this work and the idealconditions mentioned earlier for markets withmore than 6 spectrum users as long as there isno oversupply of spectrum, that is, when R = 5and R = 10 (although cases with 5 SUs wereviable when R = 10).

A value of R = 5 indicates scenarios whereon average there is 50 percent less spectrum perSU to serve the SU’s average traffic require-ment. A value of R = 10 is the reference valuewhere the amount of spectrum per user is veryclose to being enough to serve a SU’s averagetraffic requirement and is where most of theviable scenarios are found. When R = 15, thereis a 50 percent oversupply of spectrum and inthis case, the viable markets are those with 10 to20 spectrum users. Thus, if there is little or nooversupply of spectrum and with a number ofspectrum users greater than or equal to 6, mostNOBM spectrum trading markets will be viable.

Some of the implications of the results fromthe models used in this work are as follows.

Number of Market Participants — Spectrumtrading is viable in markets with no excessiveoversupply or undersupply of spectrum for awide range of spectrum user values. However,when the number of SUs is less than 6, NOBMmarkets are unviable. Regulators and entitiesinterested in these markets must make sure thatenough trading participants will be in the mar-ket. The results presented here can serve as aguideline, but should be complemented by fur-ther study of market environments under differ-ent traffic (demand) patterns.

Market Makers in a ST Market — SimpleMMs as providers of liquidity, like the ones usedin the models of this work, help in the establish-ment of viable markets by holding a spectruminventory with which they can transact. Thus,regulators will not have to specify complex MMbehavior requirements or rules. Since an MMdoes not make use of its spectrum, assigning toomuch inventory to an MM would decrease spec-trum efficiency. However, the greater the inven-tory level of the MM, the better prepared itwould be to intervene in the market if there is alack of spectrum offerings. Thus, regulators willneed to define the level of spectrum holdings ofan MM to reach a desired balance of marketviability vs. spectrum efficiency.

Amount of Available Spectrum for Trading— Oversupply of spectrum negatively affectedall market scenarios considered. In particular, anoversupply of 100 percent above the level ofspectrum the SUs need to serve their average

Figure 2. Scores for NOBM scenarios.

R

Total score

25

-4

-5

Scor

e

-3

-2

-1

0

1

2

3

4

5

2015105

Number of SUs456102050

Figure 3. Percentage of assigned spectrum for NOBM scenarios.

5

20

0

Ass

igne

d sp

ectr

um (

%)

40

60

80

100

R10 15 20 25

Number of SUs456102050



traffic needs leads to unviable markets. An ade-quate amount of spectrum for trading should bemade available in the market in order to main-tain trading activity. Our scenarios showed via-bility with spectrum amounts 50 percent aboveor below the level of spectrum needed to attendaverage traffic in the market.

Spectrum Efficiency — In the viable ST mar-ket scenarios, NOBM markets provided for spec-trum efficiencies between 51 and 77 percentunder the ideal conditions of this work. Theseresults show a positive characteristic of ST mar-kets that is of great interest to regulators andSUs. Analysis of spectrum efficiency in othermarket scenarios and conditions is left for futurework.

CONCLUSIONSThe outcomes of our study can help policy mak-ers and wireless service providers (future and cur-rent) understand the required conditions forimplementing spectrum trading markets. Themarket and operational scenarios we usedassumed ideal conditions among which were theuse of a single wireless technology and the fungi-bility of spectrum in the service area where STwas operating. These restrictions were used todefine initial scenarios (almost ideal) over whichto study the dynamics of spectrum trading. Futurework will make use of our modeling tool (SPEC-TRAD) to analyze more complicated scenarios.

Our models indicate that spectrum marketscan be viable if sufficient numbers of marketparticipants exist and the amount of tradablespectrum is balanced to the demand. Given thata minimum of five to six active spectrum users(wireless service providers) are necessary in aparticular service area, it seems unlikely thatspectrum markets will be viable in mobile mar-kets unless the barriers to market entry for newservice providers are lowered. ST markets maywell create the incentive for the appearance ofnew types of wireless service providers, but with-out a truly liberalized ST market in place it isunlikely they will do so. Thus, a challenge forregulators and researchers alike will be identify-ing an appropriate band to promote spectrumtrading or to facilitate the entry of new marketparticipants, and perhaps even end users.

The matter of balancing tradable spectrum todemand will prove more challenging for regula-tors because this requires insight into servicedemand, even though this need not be too pre-cise. Thus, it will be important to develop useful(and observable) proxies that enable regulatorsto estimate how well markets are balanced. The

viability of spectrum trading in more complicat-ed trading scenarios (i.e., more than one wirelessstandard and/or non-fungible spectrum) is leftfor future work.

An important byproduct of our research isdemonstrating how ACE can be applied to thestudy of telecommunication markets. Our meth-ods and tools can be extended to the study ofother telecommunications markets or scenarioswhere adaptive behaviors of the system partici-pants are allowed and studying the emergentbehavior of the market is of interest.

REFERENCES[1] M. A. McHenry, “NSF Spectrum Occupancy Measure-

ments Project Summary,” 2005.[2] C. E. Caicedo, Technical Architectures and Economic

Conditions for the Viability of Spectrum Trading Mar-kets, Ph.D. dissertation, Univ. of Pittsburgh School ofInformation Sciences, 2009.

[3] M. Weiss and W. Lehr, “Market-Based Approaches forDynamic Spectrum Assignment,” working paper, Nov.2009; http://d-scholarship.pitt.edu/2824/.

[4] C. E. Caicedo and M. Weiss, “An Analysis of MarketStructures and Implementation Architectures for Spec-trum Trading Markets,” Telecommun. Policy ResearchConf., Fairfax, VA, 2008.

[5] L. Tesfatsion, “Agent-Based Computational Economics:A Constructive Approach to Economic Theory,” Ch. 16,Handbook of Computational Economics, vol. 2, L. Tes-fatsion and K. L. Judd, Eds., North-Holland, 2006, pp.831–80.

[6] A. Tonmukayakul and M. Weiss, “A Study of SecondarySpectrum Use Using Agent-Based Computational Eco-nomics,” Netnomics, vol. 9, no. 2, 2008, pp. 125–51.

[7] L. Harris, Trading and Exchanges: Market Microstruc-ture for Practitioners, Oxford Univ. Press, 2003.

[8] T. Chordia, R. Roll, and A. Subrahmanyam, “Liquidityand Market Efficiency,” J. Financial Economics, vol. 87,2008, pp. 249–68.

[9] C. Caicedo and M. Weiss, “The Viability of SpectrumTrading Markets,” IEEE DYSPAN ‘10, Apr. 2010, Singa-pore.

BIOGRAPHIESCARLOS E. CAICEDO BASTIDAS ([email protected]) is an assis-tant professor and director of the Center for Convergenceand Emerging Network Technologies (CCENT) of the Schoolof Information Studies at Syracuse University. He holds aPh.D. in information science from the University of Pitts-burgh, and M.Sc. degrees in electrical engineering from theUniversity of Texas at Austin, and Universidad de losAndes, Colombia. His current research interests are spec-trum trading markets, secondary use of spectrum, security,and management of data networks.

MARTIN B. H. WEISS [M‘76] holds a Ph.D. in engineering andpublic policy from Carnegie Mellon University and anM.S.E. in computer, information, and control engineeringfrom the University of Michigan. He is currently a facultymember and associate dean for academic affairs andresearch at the School of Information Sciences at the Uni-versity of Pittsburgh. He has performed techno-economicresearch in telecommunications and telecommunicationspolicy over the past 20 years, and currently works on top-ics related to cooperative secondary use of electromagneticspectrum.

Our methods and

tools can be extend-

ed to the study of

other telecommuni-

cations markets or

scenarios where

adaptive behaviors of

the system partici-

pants are allowed

and where studying

the emergent behav-

ior of the market is

of interest.



INTRODUCTION

Philosophically, frequency-agile radios’ spectrumsensing is a binary decision problem: is it safe touse a particular frequency where we are, or is itunsafe? So it seems natural to mathematicallycast the problem as a binary hypothesis test.Most researchers model the two hypotheses asprimary user present and primary user absent. Thissuggests that the key metrics should be the prob-ability of missed detection PMD and the probabilityof false alarm PFA. But is this truly the rightmodel? Does it illuminate the important under-lying trade-offs?

To understand how metrics can matter, it isuseful to step back and consider familiar capaci-ty metrics. Traditionally, the community studiedpoint-to-point links. There, Shannon capacity(measured in bits per second per Hertz) is clear-ly the important metric. However, this is notenough when we consider a wireless communica-tion network — the spatially distributed aspect iscritical, and this shows up in the right metrics.For instance, Alouini and Goldsmith in [1] pro-pose the area spectral efficiency (measured in bitsper second per square kilometer per Hertz)

when links are closely packed together, andGupta and Kumar in [2] further propose thetransport capacity (measured in bit-meters persecond per Hertz) when cooperation (e.g., multi-hop) is possible. It is these metrics that givemuch deeper insights into how wireless commu-nication systems should be designed.

Spectrum sensing is about recycling band-width that has been allocated to primary usersand thereby increasing the capacity pre-multiplierfor the secondary system. There turns out to be asignificant spatial dimension to spectrum recy-cling for a simple reason — the same frequencywill be reused by another primary transmitteronce we get far enough away. Thus, the potentialspectrum holes span both time and space.

To see why ignoring this spatial dimension ismisleading, we must first review the traditionalbinary hypothesis testing story where the centralconcept is sensitivity: the lowest received signalpower for which target probabilities of falsealarm and missed detection can be met. Moresensitive detectors are considered better and it iswell known that sensitivity can be improved byincreasing the sensing duration.

However, why does one demand very sensi-tive detectors? The strength of the primary’s sig-nal is just a proxy to ensure that we are farenough. If wireless propagation were perfectlypredictable, then there would be a single rightlevel of sensitivity. It is the reality of fading thatmakes us demand additional sensitivity. Butbecause fading can affect different detectors dif-ferently, a head-to-head comparison of the sensi-tivity of two detectors can be misleading. Instead,the possibility of fading should be incorporatedinto the signal present hypothesis itself.

The bigger conceptual challenge comes in try-ing to understand false alarms. The traditionalhypothesis-test implicitly assumes that a falsealarm can only occur when the primary users areentirely absent. But in the real world, the spec-trum sensor must also guard against saying thatit is close to the primary when it is far enoughaway. The signal absent hypothesis needs to bemodified in some reasonable way that reflectsboth these kinds of false alarms. We must takeinto account that the users doing the sensinghave some spatial distribution.

ABSTRACT

Frequency-agile radio systems need to decidewhich frequencies are safe to use. In the contextof recycling spectrum that may already be in useby primary users, both the spatial dimension tothe spectrum sensing problem and the role ofwireless fading are critical. It turns out that thetraditional hypothesis testing framework forevaluating sensing misses both of these andthereby gives misleading intuitions. A unifiedframework is presented here in which the natu-ral ROC curve correctly captures the two fea-tures desired from a spectrum sensing system:safety to primary users and performance for thesecondary users. It is the trade-off between thesetwo that is fundamental. The spectrum holesbeing sensed also span both time and space. Thesingle-radio energy detector is used to illustratethe tension between the performance in timeand the performance in space for a fixed valueof protection to the primary user.


Rahul Tandra, Qualcomm

Anant Sahai, University of California, Berkeley

Venugopal Veeravalli, University of Illinois at Urbana-Champaign

Unified Space-Time Metrics toEvaluate Spectrum Sensing

SAHAI LAYOUT 2/18/11 3:14 PM Page 54


Once both hypotheses have been appropri-ately modified, the receiver-operating-character-istic (ROC) curve appropriately reflects thefundamental trade-off in spectrum sensingbetween the safety guarantee for the primaryusers (captured by a metric we call the Fear ofHarmful Interference, FHI) and the secondaryusers’ ability to recycle the leftover spectrum forthemselves (captured by the Weighted Probabilityof Space Time Recovered metric, WPSTR). How-ever, there are two subtle, but important, issuesthat must be addressed along the way lest weend up with trivial trade-offs. Both of thesehave to do with understanding the nature of thesafety guarantee to the primary users. First, theunderlying probabilistic model regarding thespatial distribution of the secondary users shouldnot be consistent across the two hypotheses. Infact, it is better to use a worst-case spatial distri-bution under the frequency band occupiedhypothesis so that the primary users’ safetyguarantee is strong. Second, the safety guaran-tee to primary users needs to be weakened atthe start of each primary ON period. A time-domain sacrificial buffer zone needs to be intro-duced within which interference from secondaryusers is permissible; this gives the secondaryuser some time to evacuate the band and thusallows for some sensing. Without such a sacrifi-cial buffer, the trade-offs almost invariablybecome trivial [3].

Unlike the traditional sensitivity-orientedmetrics, these new metrics give a unified frame-work to compare different spectrum-sensingalgorithms and yield several new insights intothe space-time sensing problem. First, they clear-ly show that fading uncertainty forces theWPSTR performance of single-radio sensingalgorithms to be very low for desirably small val-ues of FHI. This captures the fact that a singleradio examining a single frequency cannot distin-guish whether it is close to the primary user andseverely shadowed, or if it is far away and notshadowed. Second, the metrics reveal the impor-tance of diversity and how simple non-coherentdetection can outperform matched filters in

practice. Third, an example is used to show thatthere exists a non-trivial trade-off between thespatial and temporal performance for a spec-trum sensor. In general, there exists an optimalchoice of the sensing time for which the WPSTRmetric is maximized.

SPECTRUM SENSING BYSECONDARY USERS

Spectrum holes [4], are regions in space, timeand frequency within which it is safe for a sec-ondary radio system to recycle the spectrum.The picture on the left in Fig. 1 shows there is aspatial region around every primary transmitter,called the no-talk region, within which secondaryusers are not allowed to transmit. The spectrumhole is everywhere else — shown here in green.

Intuitively, the two important dimensionsalong which a sensing algorithm should beevaluated are: the degree to which it is success-ful in identifying spectrum holes that are actu-ally there; and the amount of harmfulinterference caused to the primary system byfalsely identifying spectrum holes. An idealapproach — for example, involving a central-ized database with primary user participationand geolocation functionality for secondaryusers [5] — would, by definition, create zerounauthorized harmful interference and yetrecover all the spectrum holes.

To make the problem concrete, we now focuson a single primary user transmitting on a givenfrequency band. The picture on the right in Fig.1 illustrates that the primary transmitter (a TVtower in this example) has a protection region(gray region in the figure), and any potential pri-mary receivers within this area must be protect-ed from harmful interference. The resultingno-talk radius rn can be computed from the pro-tection radius, the transmit power of the sec-ondary radio, and the basic wireless propagationmodel [6]. The sensing problem thus boils downto deciding whether the distance from the TVtower is less or greater than rn.

Figure 1. This figure illustrates the scenario of cognitive radios acting as sensing-based secondary users recycling TV whitespaces. Thesecondary user is allowed to use the channel if it is outside both the protected region and the no-talk region (the tan-colored annulusshown in the figure) of each primary transmitter that is currently ON. The spectrum-sensing problem boils down to identifying whetherthe secondary user is within a space-time spectrum hole or not.

Protected region No-talk region

Cognitiveradio

PrimaryTx

PrimaryRx

Decodable radius

Cognitiveradio

PrimaryTx

PrimaryRx



REVIEW OF THE TRADITIONAL TIME-DOMAINFORMULATION FOR SENSING

Currently, the most popular formulation of thespectrum sensing problem casts it as a binaryhypothesis test between the following twohypotheses: primary ON and primary OFF.

The two traditional hypotheses are:

Signal absent H0 : Y[n] = W[n]Signal present H1 : Y[n] = √

—PX[n] + W[n], (1)

for n = 1, 2, …, N. Here P is the received signalpower, X[n] are the unattenuated samples (nor-malized to have unit power) of the primary sig-nal, W[n] are noise samples, and Y[n] are thereceived signal samples.

The two key metrics in this formulation are:the probability of false alarm, PFA, which is thechance that a detector falsely thinks that the sig-nal is present given that the signal is actuallyabsent; and the probability of missed detection,PMD, which is the chance that the detector incor-rectly declares the signal to be absent given thatthe signal is actually present.

The lowest signal power P at which the detec-

tor can reliably meet (PFA, PMD) targets is calledthe detector’s sensitivity. Furthermore, the mini-mum number of samples required to achieve atarget sensitivity is called the detector’s samplecomplexity. The traditional metrics triad of sensi-tivity, PFA, and PMD, are used along with thesample complexity to evaluate the performanceof detection algorithms.

DRAWBACKS WITH THETRADITIONAL FORMULATION

The key idea behind the formulation in the pre-vious section is that a detector that can senseweak signals will ensure an appropriately lowprobability of mis-declaring that we are outsidethe no-talk radius whenever we are actuallyinside. However, this formulation has some fun-damental flaws.

How Much Sensitivity Do We Really Need?— The right level of sensitivity should corre-spond to the signal power at the no-talk radius.If there were no fading, the required sensitivitywould immediately follow from the path-lossmodel. The traditional approach to deal withfading is to incorporate a fading margin into thetarget sensitivity (e.g., set the sensitivity lowenough to account for all but the 1 percent worstcase fades). However, different detectors may beaffected differently by the details of the fadingprocess. For example, a coherent detector look-ing for a single pilot tone would require a largerfading margin than a non-coherent detectoraveraging the signal power over a much widerband. Hence, thinking in terms of a single levelof sensitivity for all detectors is flawed.

How to Measure the Performance of Spec-trum Sensors — Traditionally, the frequencyunused hypothesis (H0) has been modeled asreceiving noise alone. However, it is perfectlyfine for the primary transmitter to be ON, aslong as the spectrum sensor can verify that it isoutside the primary’s no-talk radius. The realworld hypothesis H0 is actually different at dif-ferent potential locations of the secondary radio.

Building in a fading margin to the sensitivityhas the unfortunate consequence of causing thefalse-alarm probabilities to shoot up when thespectrum sensor is close, but not too close, tothe primary transmitter. This makes parts of thespectrum hole effectively unrecoverable [7]. Fig-ure 2 shows this effect in the real world.

SPECTRUM SENSING:A SPACE-TIME PERSPECTIVE

The discussion in the previous section forces usto rethink the traditional hypothesis-testing for-mulation. Fading must be explicitly included andthe reality of different potential locations mustalso be explicitly accounted for.

The received signal can be modeled as Y[n]= √

———P(R) X[n] + W[n] whenever the primary

transmitter is ON, where R is the distance of thesecondary radio from the primary transmitter.The received signal power P(R) is actually a ran-dom operator (modeled as independent of both

Figure 2. The map on the top shows the location of TV towers (red trianglesand circles) transmitting on channel 39 in the continental United States. Thelarger disks around the transmitters show the no-talk region around the TVtransmitters within which a secondary user cannot recycle the channel. Thisshows that the true spectrum hole covers about 47 percent of the total area.The effective no-talk region for a radio using the –114 dBm rule (from [5]) isshown in the bottom figure — only 10 percent of the total area remains. Thisfigure is taken from [7] where more details can be found on the availablewhitespace spectrum in TV bands.


the normalized transmitted signal and the noise)that depends on both the path loss and fadingdistributions. This gives the following compositehypothesis testing model:

(2)

where we still need to decide on the primaryuser’s ON/OFF behavior and the distributionsfor R in the two hypotheses to permit evaluationof the two kinds of error probabilities for anyspectrum sensor.

MODELING SPACEThe true position of the secondary user relativeto the primary transmitter is unknown. This iswhy we are sensing. For H1, it is natural toassume a worst-case position, generally at justwithin rn where the primary signal is presumablyweakest. A worst-case assumption makes thequality guarantee apply uniformly for all theprotected primary receivers.

Suppose we took the same approach to H0.Typically, the worst case location under H0would be just outside rn with the primary userON. After all, if we can recover this location, wecan presumably recover all the locations evenfurther away or when the primary user is OFF.Alas, this approach is fatally flawed since the sig-nal strength distributions just within rn and justoutside of rn are essentially the same. No inter-esting trade-off is possible because we are miss-ing the fundamental fact that a sensing-basedsecondary user must usually give up some areaimmediately outside of rn to be sure to avoidusing areas within rn.

Simply averaging over the distance R alsoposes a challenge. The interval (rn, ∞) is infiniteand hence there is no uniform distribution overit. This mathematical challenge corresponds tothe physical fact that if we take a single primary-user model set in the Euclidean plane, the areaoutside rn that can potentially be recovered isinfinite. With an infinite opportunity, it does notmatter how much of it we give up!

In reality, there are multiple primary trans-mitters using the same frequency. As a radiomoves away from a given primary transmitter (Rincreases), its chance increases of falling withinthe no-talk radius of an adjacent primary trans-mitter. The picture on the bottom in Fig. 3 illus-trates the Voronoi partitioning of a spatiallydistributed network of primary transmitters, andthe picture on the top shows the effective singleprimary transmitter problem with a finite area.

The key is to choose a probability measurew(r)rdr so as to weight/discount area outside rnappropriately. The rigorous way to do this is touse results from stochastic geometry and point-process theory [8]. However, the key insights canbe obtained by choosing any reasonable proba-bility measure. The numerical results here havebeen computed assuming w(r) is constant (= c)for 0 < r ≤ rn, and an exponential weightingfunction, w(r) = Aexp(–κ(r – rn)), for r > rn. Theconstant part essentially tells us the probabilityof the primary being OFF. An exponential distri-

bution is chosen for the rest because it has themaximum entropy among the set of all distribu-tions on [rn, ∞) with a given mean. In our case,this mean is related to the average minimum dis-tance between two primary towers transmittingon the same channel.

MODELING TIMEThe probability of being within the no-talk radiusin H0 seems to capture the ON/OFF behavior ina long-term average sense. But long-term aver-ages are not enough to allow us to evaluate sens-ing. Intuitively, if the primary user is coming andgoing very often, the issue of timeliness in sens-ing is more important than when the primaryuser is like a real television station and switchesstate very rarely, if at all.

Consider a secondary user that is locatedinside the no-talk radius. Let U[n] = 1 only ifthe primary transmitter is ON at time instance n.Assume that we start sensing at time instancesni, and at the end of each sensing interval, thesecondary user makes a decision of whether thefrequency is safe to use (Di = 0) or not safe touse (Di = 1). The secondary user transmits only

H 0 :Y nP R X n W n R r

W n

n[ ] =( ) [ ]+ [ ] >

[ ]

and primary ON

primary OFF

⎧⎨⎪

⎩⎪

[H 1 :Y n]] = ( ) [ ]+ [ ] ∈ [ ]P R X n W n R rn 0, ,


Figure 3. The picture on the bottom shows theVoronoi partitioning of the space between prima-ry transmitters. The multiple primary transmitterproblem is approximated as an ideal single pri-mary problem by including a spatial weightingfunction w(r) that discounts the value of areasfar away from the primary transmitter.

TX rn

TX 1rn

TX 2rn

TX 3rn

Building in a fading

margin to the

sensitivity has the

unfortunate conse-

quence of causing

the false alarm

probabilities to shoot

up when the

spectrum sensor is

close, but not too

close, to the primary

transmitter. This

makes parts of the

spectrum hole effec-

tively unrecoverable.



if the frequency is deemed to be safe, and thensenses again. This induces a random processZr[n] ∈ {0, 1} denoting the state of a secondaryuser located at a distance r from the primarytransmitter, with 1 representing an actively trans-mitting secondary user. An example scenario isshown in Fig. 4.

Intuitively, harmful interference could bequantified by measuring the fraction of the pri-mary ON time during which a secondary userlocated inside the no-talk radius is transmitting.Suppose that the primary transmitter is OFF, asecondary user senses for N samples, correctlydeclares that the primary is OFF, and hencestarts transmitting. There is still a finite non-zeroprobability that the primary comes back ONwhile the secondary is transmitting. This proba-bility depends on the duration of the secondaryuser’s transmission, but might have no connec-tion to how long N is! For example, there wouldindeed be no connection in a Poisson model (themaximum-entropy modeling choice) for primarytransmissions [3].

The secondary could thus cause interferenceeven when its spectrum sensor is as correct as itcould possibly be. If this definition of interfer-ence were to be adopted, the only way to drivethe probability of interference to zero would beto scale the secondary transmission time to zero.This would give a relatively uninteresting trade-off between the protection to the primary systemand the performance of the secondary user.

The naive definition does not recognize thatcausality implies that the initial segments of aprimary transmission are intrinsically moreexposed to interference. This is the time-domaincounterpart to the spatial status of primaryreceivers located at the edge of decodability.Just as these marginal receivers must be sacri-ficed for there to be meaningful spectrum holes,it makes sense to assume that there is a tempo-ral sacrificial buffer zone (Δ samples long) atthe beginning of every OFF to ON transition ofthe primary user (illustrated as purple regions inFig. 4). Secondary transmissions during thistime should not be considered harmful interfer-ence.

SPACE-TIME METRICS

We now define two key metrics that are similarto the traditional metrics of PFA and PMD, butare computed on the composite hypotheses inEq. 2. The trade-off between them is the funda-mental ROC curve for the problem of spectrumsensing.

Safety: Controlling the Fear of HarmfulInterference — This metric measures the worst-case safety that the sensing-based secondary usercan guarantee to the primary user under uncer-tainty. We call it the Fear of Harmful InterferenceFHI = sup0≤ r ≤ rn supFr∈FFr PFr (D = 0⎪R = r),where D = 0 is the detector’s decision declaringthat the frequency is safe to use, and FFr is the setof possible distributions for P(r) and W[n] at adistance of r from the primary transmitter. Theouter supremum is needed to issue a uniformguarantee to all protected primary receivers. Theinner supremum reflects any non-probabilisticuncertainty in the distributions of the path-loss,fading, noise, or anything else.

Performance: Success in Recovering Spec-trum Holes — By weighting the probability offinding a hole PFH(r) with the spatial densityfunction, w(r)r, we compute the weighted proba-bility of space-time recovered (WPSTR) metric:

(3)

and ∫0∞w(r)rdr = 1. The I above is shorthand for

indicator functions that take the value 1 whenev-er their subscript is true and 0 otherwise. Noticethat the integral spans locations inside and out-side the no-talk radius (0 to ∞). The nameWPSTR reminds us of the weighting of perfor-mance over space and time. 1 – WPSTR is the

WPSTR P r w r r dr

P r

M

FH

FH

M

= ( ) ( )

( ) =

∞

→∞

∫01

where,

lim IZZ nn

Mn

MZ n U

r

r

r r[ ]=( )=

→∞[ ]=

∑ >11

1

if

lim,

Innn

M

U nnM nr r

[ ]=( )=

[ ]=( )=

∑

∑>

⎧

⎨

⎪⎪⎪

⎩

⎪⎪⎪

01

01I if

,

Figure 4. The state of the primary user U[n], the sensing epochs, as well as the secondary ON/OFF processZr[n] (dashed red line) are shown in the figure. The red sensing windows indicate events when the detectordeclares the frequency to be used, and the blue sensing windows indicate when the detector declares the fre-quency to be unused. The primary sacrificial buffer zones are shown by the purple shaded regions on thefunction U[n], and the actual harmful interference events are shown by shaded tan regions on U[n].

Secondarysensing

Time

ON

Interference

Primary buffer zone (Δ)

Opportunities recovered

OFF

D=0Zr [n]SecondaryTx

U [n]PrimaryTx

D=1 D=0D=0

The naive definition

does not recognize

that causality implies

that the initial

segments of a

primary transmission

are intrinsically more

exposed to

interference. This is

the time-domain

counterpart to the

spatial status of

primary receivers

located at the edge

of decodability.



appropriate analog of the traditional PFA, exceptthat it also implicitly includes the overhead dueto the sample-complexity.

INSIGHTS FROM THENEW SPACE-TIME FRAMEWORK

ALWAYS ON PRIMARIES:PURELY SPATIAL SPECTRUM HOLES

Assume that the primary transmitter is alwaysON. This corresponds to sensing a spectrum holewhose temporal extent is infinite, so it does notmatter how long we spend sensing. To remind usof this spatial focus (and to maintain consistencywith [4]), we call the secondary performancemetric the Weighted Probability of AreaRecovered, WPAR, instead of WPSTR.

Consider a single secondary user running aperfect radiometer (i.e., one with an infinitenumber of samples). If the noise variance isperfectly known, it is straightforward to deriveexpressions for FHI and WPAR [4]. The blackcurve in Fig. 5 shows the FHI vs. WPAR trade-off. Notice that the WPAR performance at lowFHI is bad even for the perfect radiometer. Thiscaptures the physical intuition that guarantee-ing strong protection to the primary user forcesthe detector to budget for deep non-ergodicfading events. Unlike in traditional communica-tion problems where there is no harm if thefading is not bad, here there is substantial harmsince a valuable spectrum opportunity is leftunexploited.

Impact of Noise Uncertainty: SNR Walls inSpace — The real world noise power is not per-fectly known. In the traditional formulation, thisuncertainty causes the radiometer to hit an SNRwall that limits its sensitivity [9]. What happensunder these new metrics?

See [10] for the details, but the result is illus-trated by the red curve in Fig. 5. The noiseuncertainty induces a critical FHI thresholdbelow which none of the spectrum hole can berecovered (WPAR = 0). In traditional terms, thesensitivity required to budget for such rare fad-ing events is beyond the SNR wall. Just as thesample-complexity explodes to infinity as theSNR Wall is approached in terms of traditionalsensitivity, the area recovered crashes to zero asFHI approaches this critical value.

Dual Detection: How to Exploit Time-Diver-sity — The true power of these new metrics isthat they allow us to see the importance of diver-sity. This can be cooperative diversity as dis-cussed in [4], but the effect can be seen evenwith a single user. For example, one could pre-sumably exploit time diversity for multipath if webelieved that the actual coherence time is finiteNc < ∞. However, for the radiometer, all thethresholds must be set based on the primaryuser’s fear of an infinite coherence time — thespectrum sensor might be stationary. Theradiometer cannot do anything to exploit thelikelihood of finite coherence times even if thesensor is likely to be moving.

The situation is different for a sinusoidal

pilot tone, as illustrated by the blue curve inFig. 5. The best-case scenario for coherentdetection from a traditional sensitivity per-spective — infinite coherence time with nonoise uncertainty — can be worse in practicethan a simple radiometer with noise uncertain-ty. As the sinusoidal pilot is narrowband, thematched filter suffers from a lack of frequencydiversity as compared to the radiometer: fad-ing is more variable and the resulting conser-vatism costs us area.

So does the matched filter have any use inwideband settings? Yes. It gives us an opportuni-ty to deal with uncertain coherence-times. Wecan run two parallel matched filters — oneassuming an infinite coherence time and theother doing non-coherent averaging to combinematched-filtered segments of length Nc — withtheir thresholds set according to their respectiveassumptions on the coherence time. If either ofthem declares that the frequency is used, thenthe secondary user will not use this frequency.This ensures that the FHI constraint is met irre-spective of the actual coherence time. The dual-detector approach thus leads to different FHI vs.WPAR curves depending on what the mix ofunderlying coherence times is (stationary devicesor moving devices).

The dashed curve in Fig. 5 shows the perfor-mance of the matched filter running with aknown coherence time of Nc. Because it enjoystime-diversity that wipes out multipath fading, itis only limited by the same non-ergodic wide-band shadowing that limits even the widebandnon-coherent detector without noise-uncertainty.In principle however, this dual detector still hasan SNR wall due to noise uncertainty. However,to be able to illustrate this effect, Fig. 5 wasplotted using a very short coherence time Nc =100. For any realistic coherence time, the SNRwall effect would become negligible at all butextremely paranoid values for FHI.

Figure 5. Under noise uncertainty (1 dB here), there is a finite FHI thresholdbelow which the area recovered by a radiometer is zero (WPSTR = 0). Thecoherent detector (modified matched filter) has a more interesting set of plotsdiscussed in this article.

Fear of harmful interference (FHI)

Pilot detector with noise uncertaintyand finite coherence time

Radiometer withnoise uncertainty

Pilo

t det

ecto

r with

out n

oise

unc

erta

inty

and

infin

ite c

oher

ence

tim

e

Nc = 102 samples5% pilot power

x=1 dB

Radiometer without noise uncertainty

10-410-6

0.2

0

WPS

TR

0.4

0.6

0.8

1

10-2 100



ON/OFF PRIMARIES: SPACE-TIME TRADE-OFF

When the primary signal has both ON and OFFperiods, both space and time must be consideredtogether.

Memoryless Sensing Algorithms — Even thisrestrictive class of algorithms brings out someinteresting trade-offs that are absent in thespace-only scenario discussed earlier. Theassumptions we make are:• The detector senses for N contiguous sam-

ples and makes a decision based these sam-ples alone. This is clearly suboptimalbecause longer-term memory could helpsignificantly [11].

• The secondary user’s sensing and transmis-sion times are non-adaptive and fixed inadvance.

• The primary state does not change within asensing window. This approximation can leadto a lower rate of missed-detections becauseif the primary were to turn ON somewherenear the end of the sensing window, therewould be a good chance that the detectorwould not trigger. To counter this, we enforcethat the sum of the sensing-duration and thesecondary user’s transmission-duration is lessthan the buffer Δ. This ensures that the onlyway to cause unauthorized harmful interfer-ence is by having a sensing error during awindow in which the primary is ON.

Numerical Simulations — For simplicity, con-sider a radiometer facing no fading at all. See[10] for the derivation of expressions for FHI andWPSTR, but the results can also be extended toother single-user or even cooperative sensingalgorithms. The parameters used to obtain ournumeric results are chosen to match those in[12] and are described below:

The TV tower’s transmit power is assumed tobe Pt = 106 W, its protection radius rp = 134.2km, and the no-talk radius rn = 150.3 km. Thereceived power at a distance r from the TVtower is modeled as P(r) = Pt ⋅ l(r) where log l(r)is a piece-wise linear continuous function of log rchosen to approximate the InternationalTelecommunication Union (ITU) propagationcurve given in [12, Fig. 1]. Finally, the exponentin the spatial weighting function w(r) : = Aexp –κ(r – rn) is chosen to be κ = 0.02 km–1 .

The top plot in Fig. 6 shows curves depictingthe FHI vs. WPSTR performance of a radiometerfor different sensing times N. It is clear that theoptimal N is a function of the desired safety FHIThe bottom right plot takes a slice at a fixed FHI= 0.001, and considers the radiometer’s WPSTRperformance as a function of the sensing time. Itcompares this with the traditional perspective’soverall performance metric:

Notice that at very low N, essentially nothing isrecoverable since the FHI forces the detectionthreshold to be so low that noise alone usuallytriggers it. There is an optimal value for N thatbalances the time lost to sensing with the oppor-

1 1−⎛

⎝⎜

⎞

⎠⎟ −( )N

PFAΔ.Figure 6. The plot on top shows the FHI vs. WPSTR trade-off for a radiometer

as the sensing time N varies. Note that the optimal value of the sensing time isa function of the target FHI. The plots on the bottom drill down for a particu-lar FHI = 0.001 and show the trade-off between the traditional time metric Δ– N/Δ (1 – PFA) and space recovery for a radiometer. The traditional metricunderestimates the optimal sensing duration N whenever there is a spatialcomponent of the spectrum holes.

Fear of harmful interference (FHI)

Radiometer performance (Pon=0.75, Δ=2000)

10-310-4

0.5

0.4

WPS

TR

0.6

0.7

0.8

0.9

1

10-2 10-1 100

Traditional time-metric

Time vs space

0.200

Spac

e-m

etri

c

0.1

0.3

0.2

0.5

0.4

0.7

0.8

0.9

0.6

0.4 0.6 0.8 1

Number of samples (N)

WPSTR

Traditional time-metric

Space-time performance vs N (FHI=0.001)

1010

Perf

orm

ance

met

rics

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

102 103 104

N=35N=50N=100N=150

Pon=1Pon=0.75Pon=0.5Pon=0.25



tunities lost from false alarms, but the traditionalperspective is far more aggressive about settingthe sensing duration N. This is because the twotraditional hypotheses are well separated, but forpotential locations close to rn, the relevanthypotheses are much closer. As illustrated in thebottom left plot in Fig. 6, there is a tension thatmust be balanced between performance in space(which demands high-fidelity from the radiometerand hence more sample complexity) and the sole-ly time-oriented traditional performance metric.

CONCLUDING REMARKSIt is tempting to force spectrum sensors to bevery sensitive so as to guarantee protection tothe primary user (e.g., the –114 dBm rule in [5]).But the traditional metrics completely miss thatthis forces the loss of a significant portion of thespatial spectrum holes because of a presumedlack of diversity. To see the underlying trade-offs, a new joint space-time formulation is need-ed that formulates the spectrum-sensing problemas a composite hypothesis test.

Unfortunately, simple single-user strategiescannot obtain enough diversity to get a goodtrade-off. One needs to look at other sensingstrategies like dual detection, collaborative sens-ing, multiband sensing, and so on, to improveperformance. The key is to have a robust way forthe secondary user to conclude that it is indeednot deeply shadowed (not being shadowed is,after all, the typical case) and thereby avoidbeing more sensitive than is warranted.

One possibility, that deserves further investi-gation, is to exploit sensor memory. If a sec-ondary user has seen a strong primary signal inthe near past, it knows that it is probably notdeeply shadowed. This suggests that cooperativechange-detection-based algorithms can improvesensing performance in both space and time.

REFERENCES[1] M. S. Alouini and A. Goldsmith, “Area Spectral Efficien-

cy of Cellular Mobile Radio Systems,” IEEE Trans. Vehic.Tech., vol. 48, no. 4, July 1999, pp. 1047–66.

[2] P. Gupta and P. R. Kumar, “The Capacity of WirelessNetworks,” IEEE Trans. Info. Theory, vol. 46, no. 2,Mar. 2000, pp. 388–404.

[3] S. Huang, X. Liu, and Z. Ding, “On Optimal Sensing andTransmission Strategies for Dynamic Spectrum Access,”IEEE DySPAN, Oct. 2008, pp. 1–5.

[4] R. Tandra, S. M. Mishra, and A. Sahai, “What is a Spec-trum Hole and What Does it Take to Recognize One?”Proc. IEEE, May 2009, pp. 824–48.

[5] FCC, “In the Matter of Unlicensed Operation in the TVBroadcast Bands: Second Report and Order and Memoran-dum Opinion and Order,” tech. rep. 08-260, Nov. 2008.

[6] A. Sahai, N. Hoven, and R. Tandra, “Some FundamentalResults on Cognitive Radios,” Allerton Conf. Commun.,Control, Comp., Oct. 2004.

[7] S. M. Mishra, Maximizing Available Spectrum for Cogni-tive Radios, Ph.D. Dissertation, UC Berkeley, 2010.

[8] F. Baccelli and B. Blaszczyszyn, “Stochastic Geometryand Wireless Networks,” Foundations and Trends inNet., vol. 3, no. 3–4, 2009, pp. 249–449.

[9] R. Tandra and A. Sahai, “SNR Walls for Signal Detec-tion,” IEEE J. Sel. Topics Signal Process., Feb. 2008, pp.4–17.

[10] R. Tandra, A. Sahai, and V. Veeravalli, “Space-TimeMetrics for Spectrum Sensing,” IEEE DySPAN, Apr.2010, pp. 1–12.

[11] A. Parsa, A. A. Gohari, and A. Sahai, “Exploiting Inter-ference Diversity for Event-Based Spectrum Sensing,”IEEE DySPAN, Oct. 2008.

[12] S. J. Shellhammer et al., “Performance of PowerDetector Sensors of DTV Signals in IEEE 802.22WRANs,” 1st Int’l. Wksp. Tech. Policy Accessing Spec-trum, June 2006.

BIOGRAPHIESRAHUL TANDRA ([email protected]) is a senior systems engi-neer at Qualcomm Research Center, Qualcomm Inc., SanDiego, California. He currently works on the design andstandardization of next-generation WLAN systems. Hereceived his Ph.D. degree from the Department of ElectricalEngineering and Computer Sciences at the University ofCalifornia at Berkeley in 2009, where he was a member ofthe Wireless Foundations research center. In 2006 heworked as a summer intern with the Corporate R&D divi-sion of Qualcomm Inc., developing spectrum sensing algo-rithms for the IEEE 802.22 standard. Prior to that, hereceived an M.S. degree from Berkeley in 2005 and aB.Tech. degree in electrical engineering from the IndianInstitute of Technology Bombay. His research interests arein wireless communication and signal processing. He is par-ticularly interested in fundamental research questions indynamic spectrum sharing.

ANANT SAHAI [BS’94, SM’96] ([email protected]) is anassociate professor in the Department of Electrical Engi-neering and Computer Sciences at the University of Califor-nia at Berkeley, where he joined the faculty in 2002. He isa member of the Wireless Foundations center. In 2001 hespent a year at the wireless startup Enuvis developingadaptive software radio algorithms for extremely sensitiveGPS receivers. Prior to that, he was a graduate student atthe Laboratory for Information and Decision Systems at theMassachusetts Institute of Technology. His research inter-ests span wireless communication, decentralized control,and information theory. He is particularly interested inspectrum sharing, the nature of information in control sys-tems, and power consumption.

VENUGOPAL VEERAVALLI [S’86, M’92, SM’98, F’06] ([email protected]) received his B.Tech. degree in 1985 from theIndian Institute of Technology Bombay (Silver Medal Hon-ors), his M.S. degree in 1987 from Carnegie-Mellon Univer-sity, and his Ph.D. degree in 1992 from the University ofIllinois at Urbana-Champaign, all in electrical engineering.He joined Illinois in 2000, where he is currently a professorin the Department of Electrical and Computer Engineering,a research professor in the Coordinated Science Laboratory,and the director of the Illinois Center for Wireless Systems(ICWS). He served as a program director for communica-tions research at the U.S. National Science Foundation inArlington, Virginia, from 2003 to 2005. He was with Cor-nell University before he joined Illinois, and has been onsabbatical at MIT, IISc Bangalore, and Qualcomm, Inc. Hisresearch interests include distributed sensor systems andnetworks, wireless communications, detection and estima-tion theory, and information theory. He is a DistinguishedLecturer for the IEEE Signal Processing Society for2010–2011. He has been on the Board of Governors of theIEEE Information Theory Society. He has also served as anAssociate Editor for IEEE Transactions on Information Theo-ry and IEEE Transactions on Wireless Communications.Among the awards he has received for research and teach-ing are the IEEE Browder J. Thompson Best Paper Award,the National Science Foundation CAREER Award, and thePresidential Early Career Award for Scientists and Engineers(PECASE).

One possibility is to

exploit sensor

memory. If a sec-

ondary user has seen

a strong primary

signal in the near

past, it knows that it

is probably not

deeply

shadowed. This

suggests that coop-

erative change-

detection based

algorithms can

improve sensing per-

formance in both

space and time.



GUEST EDITORIAL

his feature topic is a continuation of our September2010 feature topic on standards and testbeds for cog-

nitive radio networks. In this second part of the featuretopic, two invited articles and three articles selected from apool of high-quality submissions are introduced. We hopeour readers will find these articles useful not only forunderstanding recent developments, but also for inspiringtheir own work.

Our feature topic begins with an invited article, “Wire-less Service Provision in TV White Space with CognitiveRadio Technology: A Telecom Operator’s Perspective andExperience,” contributed by M. Fitch and his colleagues atBritish Telecom. This article describes three use cases thatare of interest to telecommunications operators: futurehome networks, street coverage, and broadband access torural and underserved premises. The authors also presentresults of modeling and experimental trials, and identify anumber of technical and business challenges as well.

Wang and his colleagues at Philips Research NorthAmerica next present the article “Emerging CognitiveRadio Applications: A Survey,” which presents, primarilyfrom a dynamic spectrum access perspective, some emerg-ing applications, desirable benefits, and unsolved chal-lenges. The authors also illustrate related standardizationthat uses cognitive radio technologies to support suchemerging applications.

Following the two invited articles, the next articleoverviews the cognitive radio system (CRS) and its stan-dardization progress. Generally speaking, the CRS can becharacterized as a radio system having capabilities toobtain knowledge, adjust its operational parameters andprotocols, and learn. Many CRS usage scenarios and busi-ness cases are possible. In the article “International Stan-dardization of Cognitive Radio Systems,” Filin et al.describe the current CRS concept and describe the ongo-ing standardization progress in different international stan-dardization bodies.

The final two articles are related to testbeds, architec-tures, and prototypes of cognitive radio networks. For the

first article, “Cognitive Radio: Ten Years of Experimenta-tion and Development,” Pawelczak and his coauthors pro-vide synopses of the state-of-the-art hardware platformsand testbeds, examine what has been achieved in the lastdecade of experimentation and trials relating to cognitiveradio and dynamic spectrum access technologies, and pre-sent insights gained from these experiences in an attemptto help the community grow further and faster in the com-ing years.

The last article that appears in this issue is titled “Spi-derRadio: A Cognitive Radio Network with CommodityHardware and Open Source Software” and is contributedby Sengupta et al. In this contribution, the authors beginwith a discussion of the key research issues and challengesin the practical implementation of a dynamic spectrumaccess network. The discussion is followed by a presenta-tion of the lessons learned from the development ofdynamic spectrum access protocols, the design of manage-ment frame structures, the implementation of the dynamicspectrum access network protocol stack using software, andthe results of testbed experimental measurements.

ACKNOWLEDGMENT

We would like to thank all individuals who have contribut-ed toward this special issue. In particular, we thank all thereviewers for their high-quality professional reviews withinthe tight deadlines given to them. Special thanks are dueto the Editor-in-Chief, Dr. Steve Gorshe, for his guidance.Last but not least, we would also like to thank the publica-tions staff of IEEE Communications Society, Devika Mit-tra, Jennifer Porcello, Cathy Kemelmacher, and JoeMilizzo, whose professional advice and support throughoutthe development of this feature topic we appreciate.

BIOGRAPHIESEDWARD AU [M] ([email protected]) holds a Ph.D. degree in electronicand computer engineering from Hong Kong University of Science and Tech-nology (HKUST). As a principal engineer of Huawei Technologies, he hasworked on research and product development on 100 Gb/s and beyondoptical long-haul communications and is now leading a project on fixed

T

ADVANCES IN STANDARDS AND TESTBEDS FOR

COGNITIVE RADIO NETWORKS: PART II

Edward K. Au Dave Cavalcanti Geoffrey Ye Li Winston Caldwell Khaled Ben Letaief

LYT-GUEST EDIT-Au 2/22/11 11:39 AM Page 62


wireless transmission systems. He has actively participated in standardiza-tion organizations and industry forums. He is the primary technical repre-sentative of Huawei in the Wi-Fi Alliance and an active contributor to theOptical Interconnecting Forum (OIF), where he is a co-editor of the channelcoding project for 100Gb/s DWDM optical transmission systems, and amember of Speakers Bureau in representing OIF at industry and academicevents. He was also a working group secretary of IEEE 802.22, the firstinternational standards on cognitive radio networks. He is also stayingactive in research community. He is currently an Associate Editor of IEEETransactions on Vehicular Technology and lead guest editor for the IEEECommunications Magazine Feature Topic on Advances in IEEE Standardsand Testbeds for Cognitive Radio Networks. He is a founding member ofthe Shenzhen Chapter, IEEE Communications Society.

DAVE CAVALCANTI [M] ([email protected]) is a senior member ofresearch staff at Philips Research North America, Briarcliff Manor, New Yorksince 2005. He received his Ph.D. in computer science and engineering in2006 from the University of Cincinnati, Ohio. He received his M.S. in com-puter science in 2001 and B.Sc. in electrical engineering in 1999 (with Hon-ors) from the Federal University of Pernambuco, Brazil. His researchinterests include communication protocols for wireless networks, cognitiveradio networks, wireless sensor networks, heterogeneous networks, andcontrols and automation applications of wireless networks. He has con-tributed to standardization of several connectivity technologies in the IEEE802.15, 802.11, and 802.22 working groups. He has delivered tutorials andserved as a technical program committee member of several conferences inthe area of wireless communications and networking. He has served as aguest editor of IEEE Wireless Communications. He has served as Chair ofthe IEEE Computer Society Technical Committee on Simulation (TCSIM)since 2008.

GEOFFREY YE LI [F] ([email protected]) received his B.S.E. and M.S.E.degrees in 1983 and 1986, respectively, from the Department of WirelessEngineering, Nanjing Institute of Technology, China, and his Ph.D. degree in1994 from the Department of Electrical Engineering, Auburn University,Alabama. He was a teaching assistant and then lecturer with Southeast Uni-versity, Nanjing, China, from 1986 to 1991, a research and teaching assis-tant with Auburn University, Alabama, from 1991 to 1994, and apost-doctoral research associate with the University of Maryland at CollegePark from 1994 to 1996. He was with AT&T Labs — Research, Red Bank,New Jersey, as a senior and then principal technical staff member from1996 to 2000. Since 2000 he has been with the School of Electrical andComputer Engineering at Georgia Institute of Technology as an associateand then a full professor. He has held the Cheung Kong Scholar title at theUniversity of Electronic Science and Technology of China since March 2006.His general research interests include statistical signal processing andtelecommunications, with emphasis on OFDM and MIMO techniques, cross-layer optimization, and signal processing issues in cognitive radios. In theseareas he has published about 200 papers in refereed journals or conferencesand filed about 20 patents. He has also written two books. He has served oris currently serving as an editor, a member of editorial board, and guest edi-tor for about 10 technical journals. He organized and chaired many interna-tional conferences, including technical program vice-chair of the IEEE ICC’03. He was selected as a Distinguished Lecturer for 2009–2010 by IEEECommunications Society, and won the 2010 IEEE Communications SocietyStephen O. Rice Prize Paper Award in the field of communications theory.

WINSTON CALDWELL ([email protected]) received his B.Eng. degree inelectrical engineering from Vanderbilt University and his M.S. degree inelectrical engineering from the University of Southern California. He is alicensed Professional Engineer in the state of California with over 16 yearsof electrical engineering experience, specializing in RF propagation andantenna design. He is vice president of spectrum engineering for News Cor-poration’s Fox Technology Group. In the past he has served as a systemsengineer in the computer industry with EMC Corporation and as a seniordata systems and telemetry engineer in the aerospace industry with theBoeing Company. To facilitate his up-to-date knowledge and understandingin the changing world of frequency spectrum technologies and policies, heis an active member and presenter of engineering analyses at the ITU, IEEE,MSTV, NABA, NAB, and SMPTE. He is a founding member of IEEE P802.22,Chairman of the IEEE 802.22.2 Recommended Practice Task Group, andLiaison to the IEEE 802.18 Radio Regulations — Technical Advisory Group.

KHALED BEN LETAIEF [F] ([email protected]) received a B.S. degree with distinc-tion in electrical engineering from Purdue University, West Lafayette, Indi-ana, in December 1984. He received M.S. and Ph.D. degrees in electricalengineering from Purdue University in August 1986 and May 1990, respec-tively. From January 1985 and as a graduate instructor in the School ofElectrical Engineering at Purdue University, he taught courses in communi-cations and electronics. From 1990 to 1993 he was a faculty member atthe University of Melbourne, Australia. Since 1993 he has been at HongKong University of Science & Technology, where he is currently dean ofengineering. He is also Chair Professor of Electronic and Computer Engi-neering as well as director of the Hong Kong Telecom Institute of Informa-tion Technology and the Wireless IC System Design Center. His currentresearch interests include wireless and mobile networks, broadband wire-less access, OFDM, cooperative networks, cognitive radio, MIMO, andbeyond 3G systems. In these areas he has over 400 journal and conferencepapers, and has given invited keynote talks as well as courses all over theworld. He has three granted and 10 pending U.S. patents. He has served asa consultant for different organizations and is the founding Editor-in-Chiefof IEEE Transactions on Wireless Communications. He served on the editori-al boards of other prestigious journals including IEEE Journal on SelectedAreas in Communications — Wireless Series (as Editor-in-Chief). He hasbeen involved in organizing a number of major international conferencesand events. These include serving as Co-Technical Program Chair of theIEEE ICC — Circuits and Systems, ICCCS ’04; General Co-Chair of the 2007IEEE Wireless Communications and Networking Conference; Technical Pro-gram Co-Chair of IEEE ICC ’08, and Vice General Chair of IEEE ICC ’10. Hehas served as an elected member of the IEEE Communications SocietyBoard of Governors and IEEE Distinguished Lecturer. He also served asChair of the IEEE Communications Society Technical Committee on WirelessCommunications, Chair of the Steering Committee of IEEE Transactions onWireless Communications, and Chair of the 2008 IEEE TechnicalActivities/Member and Geographic Activities Visits Program. He is a memberof the IEEE Communications and Vehicular Technology Societies’ FellowEvaluation Committees as well as of the IEEE Technical Activities Board/PSPBProducts & Services Committee. He is the recipient of many distinguishedawards including the Michael G. Gale Medal for Distinguished Teaching(highest university-wide teaching award at HKUST); 2007 IEEE Communica-tions Society Publications Exemplary Award; eight Best Paper Awards; andthe prestigious 2009 IEEE Marconi Prize Paper Award in Wireless Communi-cations. He is Vice-President for Conferences of IEEE ComSoc.

GUEST EDITORIAL

LYT-GUEST EDIT-Au 2/22/11 11:39 AM Page 63


INTRODUCTION

Cognitive radio (CR) [1, 2] is being intensivelyresearched as the enabling technology for sec-ondary access to TV white space (TVWS). TheTVWS spectrum comprises large portions of theUHF spectrum (and VHF in the United States)that is becoming available on a geographicalbasis for sharing as a result of the switchover

from analog to digital TV. This is where sec-ondary users can, using unlicensed equipment,share the spectrum with the digital TV transmit-ters and other primary (licensed) users such aswireless microphones. Such secondary access isfree, but is conditional upon not causing harmfulinterference to the primary users. The avoidanceof interference through the use of cognitivetechniques, databases and sensing is describedlater in this article.

The total capacity associated with TVWSturns out to be quite significant. For example,modeling commissioned by Ofcom revealed thatover 50 percent of locations in the United King-dom have more than 150 MHz of TVWS avail-able for cognitive access. This offers scope for aconsiderable amount of capacity, and when thewide-area coverage ability of the UHF frequencyrange is also taken into account plus the factthat its use is free, TVWS becomes an attractiveproposition for the use cases described in thisarticle. Signals in the TV bands travel much fur-ther in cluttered environments than WiFi or 3Gsignals, and they penetrate into buildings withmuch less loss.

Both in the United States [3, 4] and morerecently in the United Kingdom [5, 6], the regu-lators have given conditional endorsement tothis new sharing mode of access, and there isalso significant industry effort underway towardstandardization, trials, and testbeds. Theseinclude geolocation databases and sensing forprimary user protection, agile transmission tech-niques, and the so-called etiquette protocols forintra- and intersystem coexistence in TVWS.

So far the majority of research on cognitiveaccess to TVWS has been focused on a singlecognitive device. However, the provision of com-mercial services based on the technology (e.g.,unlicensed mobile broadband or wireless homenetworks) will involve situations with systems ofmultiple cognitive equipment types that maybelong to either the same or different serviceproviders. Feasibility studies of CR for suchcommercial applications therefore require sys-

ABSTRACT

Currently there is a very fundamental changehappening in spectrum regulation, possibly themost fundamental ever in its history. This is theenabling of spectrum sharing, where primary(licensed) users of the spectrum, are forced toallow sharing with secondary users, who uselicense-exempt equipment. Such sharing is freefor the secondary users, subject to the conditionthat they do not cause harmful interference tothe primary users. The first instance of suchsharing is occurring with the UHF digital TVspectrum, in what is commonly called TV whitespace. Regulators such as the FCC in the UnitedStates and Ofcom in the United Kingdom haveindicated that other spectrum will follow suit.Cognitive radio is an enabling technology thatallows such sharing. Following recent rulings byFCC and Ofcom and the emergence of a seriesof related industry standards, CR operation inTVWS is moving from the research domaintowards implementation and commercialization,with use-cases that are of interest to telecomoperators. In this article we describe three suchuse cases: future home networks, coverage of thestreet from inside buildings, and broadbandaccess to rural and underserved premises. Wepresent results of modeling and trials of techni-cal feasibility, undertaken by the Innovate andDesign team at BT. Based on our experience wedraw conclusions regarding the feasibility andcommercial importance of these use cases, andidentify some of the remaining technical andcommercial challenges.

COGNITIVE RADIO NETWORKS (INVITED ARTICLE)

Michael Fitch, Maziar Nekovee, Santosh Kawade, Keith Briggs, and Richard MacKenzie, BT

Wireless Service Provision in TV White Space with Cognitive RadioTechnology: A Telecom Operator’sPerspective and Experience

FITCH LAYOUT 2/18/11 3:09 PM Page 64


tem-level studies performed in the context ofreal-life service scenarios.

Implementations of TVWS services are likelyto start with point-to-multipoint deployments(i.e., with zero mobility), such as rural broad-band access and backhaul to small 3/4G cells,and later progress to more mobile and quality ofservice (QoS)-aware systems. Access to TVWSwill enable more powerful public Internet con-nections with extended coverage and improveddownload speeds.

Time-division duplex (TDD) systems arepreferable to frequency-division duplex (FDD)when using TVWS, since FDD requires a fixedseparation of base station transmit and terminaltransmit frequencies, and this condition restrictsthe number of TVWS channels available. TDDis free from this restriction and is also bettersuited to asymmetrical links; this points towards802.11x (WiFi), 802.16 (WiMAX), and ThirdGeneration Partnership Project (3GPP) TimeDivision Long Term Evolution (TD-LTE) beingsuitable candidates that have mature standards.The requirement to avoid adjacent channels maybe imposed if the TVWS transmitters have out-of-band emissions that are too high. The combi-nation of FDD and adjacent-free would severelyreduce the amount of spectrum available and isbest avoided.

We all know that there is a growing demandfor high-data-rate wireless services such as data,video, and multimedia to users in homes, inoffices, and on the move. At the same time gov-ernments have pledged to close the digital dividebetween urban and rural communities by provid-ing universal broadband (2 Mb/s in the UnitedKingdom) to citizens regardless of their geo-graphical location, and there is sometimes fund-ing available to assist the process. Consequently,telecom operators are under pressure to cost-effectively provide such universal broadband ser-vices to rural communities, and are investigatingthe wireless option as an alternative to DSL orfiber. Network operating costs and site acquisi-tion costs will cause traditional macro networkdesigns to be uneconomical in providing equiva-lent coverage and capacity using licensed 3G and(in the near future) LTE bands because of theneed for smaller cells, this need arising bothfrom demand for higher bit rates and to supportmore users in the system. Therefore, alternativesolutions based on CR technology operating ona exempt-exempt basis in TVWS spectrum arebecoming commercially interesting, in particularfor fixed line operators that have a significantfiber and copper infrastructure, as well as poten-tial new entrants, such as Google and Microsoft.TVWS may provide a viable and highly scalablealternative to conventional solutions based oncellular and/or WiFi technologies.

TVWS SPECTRUM AVAILABILITYAND DATABASE STRUCTURE

The commercial case for using TVWS will dependupon the amount of spectrum that becomes avail-able for sharing, upon how the availability of thisspectrum varies with location and upon transmitpower allowed by cognitive devices [7].

Figure 1 shows the allocation of the UHFspectrum in the United Kingdom after the com-pletion of the digital switchover (DSO) fromanalog to digital TV [5]. The 128 MHz of spec-trum marked in green (16 channels) is thecleared spectrum which Ofcom plans to licensethrough auctions. The 256 MHz of spectrum (32channels) marked in purple is the so-calledinterleaved spectrum which can be used on ageographical basis for exempt-exempt use byCRs. Finally, the channel marked in pink (chan-nel 38) is licensed by Ofcom for exclusive accessby wireless microphones and other programmaking and special events (PMSE) equipment.This is a safe haven for such devices, but it is notsufficient for large events that commonly useover 100 wireless microphones. Hence, these pri-mary users will use other channels as well, result-ing in uncertainty of where in the spectrum thesedevices will be. There are two ways of dealingwith this uncertainty. First, the TV transmittersand radio microphones can be registered in adatabase, and then the CR device can interro-gate the database periodically to find out whichchannels are free. Such periods will typically be2 h. Second, the CR device can sense the spec-trum to detect when channels are free. In thefuture we believe that both methods will be usedtogether in order to have flexibility and achievemaximum efficiency in secondary use of TVWS.In the short term, however, geolocation databas-es seem to provide a technically feasible andcommercially viable solution.

These are important processes that must befollowed by TVWS devices which distinguishthem from both licensed and currently usedexempt-exempt devices (e.g., as used in theindustrial, scientific, and medical [ISM] bands).

Figure 2 shows an idea for a TVWS databasestructure, where the national regulator eitherowns or contracts out the supply of a centraldatabase, and there are several secondarydatabases that are typically owned by networkoperators. It is likely that the regulator will wantto certify the algorithms used in all the databas-es. Such algorithms will be used to determinewhich channels can be used by the secondarydevices and the transmit power they can employwithout causing interference. The regulator isconcerned only with protection of primary ser-vices and does not recommend or mandate anymethod of negotiation between secondary users.Obviously such negotiation is required for rea-sons of fairness, and this will be built into eti-quette between the secondary databases shownin Fig. 2.

Ofcom in the United Kingdom is currentlyclarifying whether they have the necessary legalpowers to regulate databases, and they will gainsuch powers from the government if needed.

The central database contains the boundariesof the primary users, and algorithms to calculatethe TVWS channels and the powers that cansafely be used without causing interference. Thelocation certainty input to this database is used tocalculate the required protection margin. Thegreater the uncertainty in location of the cogni-tive device, the higher the margin and the lowerthe allowed transmit power. Therefore, if thetopology of a cognitive network is master-slave,

The commercial case

for using TVWS will

depend upon the

amount of spectrum

that becomes

available for sharing,

upon how the

availability of this

spectrum varies with

location and upon

transmit power

allowed by

cognitive devices.



the locations of the slaves will be different fromthat of the master; one way of handling this situ-ation is to enter a low value for the location cer-tainty. In this way, the location of the group iscatered for by the master unit, giving a largeuncertainty about its own position.

This method will result in an overly conserva-tive allocation of channels and transmit powers.A better idea might be to have a structure asshown in Fig. 2, where the secondary databasecontains more detailed information about thesystem topology such as antenna directivities,and the QoS and mobility needs of the individu-al radio links so that the spectrum portfolio fromthe central database can be more efficiently bro-kered. The secondary database would optimizethe location information that is given to the cen-tral database, to obtain a potentially wider choiceof channels and higher transmit powers. The sec-ondary databases ca broker fairness and tradingbetween secondary users, and would modify con-tent and algorithm parameters using input fromusers and sensing. Methods for preventing spec-trum hogging are under study, as are securemethods for interrogating and validating thedatabases. This two-step database approach is

being developed in European collaborativeFramework 7 project QoSMOS [8] of which BTis co-coordinating partner. Another Europeanproject where BT plays a leading role isQUASAR [9], which is developing techno-eco-nomic decisions to support methodologies andtools that can support operators in quantifyingthe value of TVWS to their business, based onspecific service provision requirements andfuture customer’s demand.

Internally, BT has developed a set of model-ing tools to quantify the availability and variabili-ty of TVWS spectrum across the UnitedKingdom. These make use of publicly availablecoverage maps of digital terrestrial TV (DTT),which were generated via computer simulationsfrom Ofcom’s database of location, transmitpower, antenna height, and transmit frequenciesof the United Kingdom’s DTT transmitters andrepeaters. It combines these coverage maps withterrain data and simplified propagation modelingcalculations to obtain estimates for the vacantTVWS frequencies at any given location in theUnited Kingdom. The amount of spectrum avail-able is adversely affected if we impose the condi-tion that the adjacent channel is kept free.

Combining this field strength data with prop-agation modeling of cognitive devices, the avail-able TVWS spectrum for cognitive access at anygiven location has been computed with a spatialresolution of 100 m. Herein also lies a problem,because the amount of available TVWS appearsto increase as the spatial resolution reducesbecause smaller shadowed regions begin to becounted, a kind of fractal effect. What is missingis a standard method of computing the amountof TVWS spectrum available. Identification ofsuitable metrics and time/spatial resolutions isidentified as a future research topic, and the dif-ferent national regulators should agree on andspecify these parameters.

The terrain data used in our modeling tool isbased on the STRM2 data elevation data set,which at present is the most complete high-reso-lution digital topographic database of Earth.Finally, using statistics of housing distribution inthe United Kingdom, we have also extracted thepopulation weighted TVWS availability, which isimportant in investigating the feasibility of

Figure 1. The U.K. UHF TV band after the completion of digital switchover [5].

Channelfrequency

(MHz)

21

470-478

22

478-488

23

488-494

24

494-502

25

502-510

26

510-518

27

518-528

28

528-534

29

534-542

30

542-550

31

550-558

32

558-568

33

568-574

34

574-582

35

582-600

36

500-508

37

508-606

38

606-614

39

614-622

40

622-630

41

630-638

42

638-648

43

648-654

44

654-662

45

662-670

46

670-678

47

678-680

48

680-694

49

694-702

50

702-710

51

710-718

52

718-726

53

726-734

54

734-742

55

742-750

56

750-758

57

758-766

58

766-774

59

774-782

60

782-790

61

790-798

62

798-806

63

806-814

64

814-822

65

822-830

66

830-838

67

838-846

68

846-854

69Retained/interleaved spectrum Cleared spectrum PMSE

854-862

Figure 2. A possible database structure.

Inputs:link quality and mobility requirements, topology, antenna directivity, etc.Outputs:channels, powers, time to live, etc.

Regulatorycontrol Inputs:

location, location certaintyOutputs:channels, powers, time to live

Nationalregulator

Centraldatabase

Interface to MAC

Secondarydatabase

Secondarydatabase


broadband provision to BT’s customers in bothurban and rural areas.

Figure 3 shows, as an example, the availabili-ty and frequency decomposition of TVWS spec-trum for low power CRs, such as those used forhome networks, in Central London, after DSO.In this figure, vacant channels are shown aswhite bars while occupied channels are leftblack. As can be seen the available TVWS chan-nels are highly non-contiguous. This feature cangreatly restrict the use of TVWS for high-throughput applications, such as HDTV, byexisting WiFi and 4G access technologies, asmodulation techniques implemented in thesetechnologies can only operate using contiguousportions of spectrum. QoSMOS is developingflexible radio technology to bond channels evenif they are not contiguous.

In the case of London, if we are constrainedto contiguous spectrum for pooling purposed,there is only 16 MHz available out of a total of96 MHz. We shall discuss some of the conse-quences of this limitation in the next section.

Figure 4 is a map of the United Kingdom,with 1km square pixels that are colored accord-ing to the amount of TVWS that will be avail-able after DSO. It can be seen that largeamounts of TVWS are available in remote areaslike Wales and Scotland.

Figure 5 shows our computed population-weighted complementary cumulative distributionfunction of TVWS availability for the entireUnited Kingdom, once again for low-powerdevices (about +10 dBm EIRP). In some futurehigh-power use-cases, such as rural broadband(up to about +3 dBm EIRP), CRs maybe con-strained not to use vacant TV channels adjacentto those used for TV broadcasting. The bluecurve in Fig. 5 shows the CDF of TVWS avail-ability without this constraint while the red curveis the CDF with this constraint. As can be seenfrom this figure, just over 70 percent of the U.K.population has potential access to at least 100MHz of TVWS when no adjacent channel con-straint is imposed. Although the spectrum avail-ability is somewhat reduced when this constraintis imposed, there is still a considerable band-width available (at least 50 MHz for just over 50percent of population).

USE CASESWe classify the use cases into the following threescenarios:• Indoor services, which generally require

small coverage, and hence power levels thatare either significantly lower, due to betterpropagation characteristics in theVHF/UHF bands, or comparable to thatused in current ISM bands

• Outdoor coverage from indoor equipment,which requires penetration through wallswith medium range coverage ( a few hun-dred meters), and hence power levels thatare generally higher than or comparable tothat in the ISM bands

• Outdoor services, which may require signifi-cantly higher transmit power levels than arecurrently permitted in the ISM bands (com-parable to those used by cellular systems)

Most indoor applications of TVWS canalready be realized using WiFi and Zigbee tech-nology operating in the 2.4 GHz and 5 GHz ISMbands. The main advantage of using TVWS isthat the additional capacity offered will help torelieve congestion, in particular in the 2.4 GHzband. This use can also result in better indoorpropagation of signals through the home. Fur-thermore, the lower frequencies of TVWS bandscan result in lower energy consumption (roughlyan order of magnitude) compared to WiFi/Zig-bee. This is a particularly interesting advantagefor use case scenarios that involve battery-pow-ered devices (laptops, smartphones, sensors, etc).

In the following sections we describe threeuse cases in more detail, with a discussion of themodeling and trial studies performed at BT inorder to examine their technical feasibility andpotential business benefits.

WIRELESS MULTIMEDIA STREAMING FORCONNECTED HOMES

Currently operators are rolling out next-genera-tion access networks. This means that opticalfibers are being installed either all the way tohomes or to street cabinets with very high ratedigital subscriber line (VDSL) providing the lastlinks from the cabinets to homes. This willenable fixed broadband speeds of between40–100 Mb/s on uplink and downlink to homes.Such connection speeds are required in order tosupport streaming of high-definition multimediacontent to homes via the Internet, includinghigh-definition TV (HDTV) and on-demand AVcontent, such as BBC’s iPlayer and BT You-View. With this kind of high-data-rate contentbrought to homes, a new challenge for wirelessis to provide distribution within the home envi-ronment. Furthermore, some home users maywish to view content on mobile and portable ter-minals, which typically have low-resolutionscreens, while others maybe using HDTV termi-nals and set-top boxes, which in many cases needto be connected wirelessly to the Internet.


Figure 3. Free channels in London area are fragmented over the band (blackbars).

Channel number2520 30 35

Central London

45

Occupied

Free channels (8 MHZ)

40 50 55 60 65



Currently wireless multimedia streaminginside homes is mainly supported using WiFitechnology operating in the 2.4 GHz band. Inurban areas this band is already suffering fromcapacity limitations due to a combination ofinterference caused by the high density ofdevices that use the band and the well-knowninefficiencies of WiFi’s distributed coordinationmechanisms. One potential solution is to migrateto the currently non-congested 5 GHz exempt-exempt band, which has 160 MHz of spectrum inband A and 220 MHz in band B. One problemwith using the 5 GHz for whole-home contentdistribution is that it does not deliver high band-widths into all rooms in even a medium-sizedhouse, in particular into rooms that are on a dif-ferent floor from the access point. Simulationsand tests performed by BT indicate that WiFi at5 GHz using 802.11n can deliver less than 10Mb/s into rooms directly above or below theaccess point, and less than 1 Mb/s if there is acombination of floors and walls to penetrate,where the access point is near the floor in a cor-

ner of a ground floor room, which is where mostpeople locate it. The opening up of TVWS spec-trum for exempt-exempt use could provide analternative delivery mechanism that resolvesboth the interference and range issue of the cur-rent WiFi solutions and this alternative is beingintensely considered by some of the industryplayer in the CogNea alliance [10], which hasalready developed a TVWS standard primarilyfocused on whole-home multimedia distribution[11]. Also the IEEE currently has a workinggroup developing the 802.11af amendment whichwill allow for 802.11 deployment in TVWS.

We have performed system simulation stud-ies to assess the feasibility of using TVWS spec-trum for multimedia distribution. We considerdeployment scenarios in urban environmentsthat involve thousands of home access pointsdensely packed in a small area of a city likeLondon. In order to realistically model serviceprovision in TVWS spectrum in such scenarios,the impact of interference from neighboringhomes, the spatial distribution of houses andthe presence of walls and other obstacles withineach house needs to be considered. Finally, theavailability and frequency decomposition ofTVWS can vary strongly with location and thisfactor also needs to be incorporated into anymeaningful feasibility study.

The deployment scenario used in the studyreported here is 1 km2 of a typical urban areain London (Bayswater) that contains about5000 houses and office buildings. The houseand street layout data for the study was highlydetailed, and was imported from a geographicinformation system (GIS) database. The maxi-mum access point density was derived fromavailable surveys and was fixed at 20 percent;that is, 1000 randomly selected houses wereassumed to have a home access point. Eachaccess point is then associated with a clientwithin a maximum distance of 12 m, which isrepresentative of U.K. homes. Propagation ofWiFi and TVWS signals within each house andbetween houses are derived as a function ofdistance, geometry of houses, and number ofwalls and floors, using appropriate propagationmodels.1 The availability of TVWS channelsfor the scenario studies is derived from themodel described earlier. It is then assumedthat a home access point queries a centrallymanaged database for TVWS channel avail-ability via its fixed line connection. The homebase station and the associated clients are con-figured in master-slave architecture, with thehome base station advertising its presence andchannel availabil ity via regular beacons,enabling slaves to establish a communicationlink with the master.

The effect of interference from access pointsand clients in neighboring homes is modeled interms of the achievable signal-to-interference-and-noise ratio (SINR) at the client, and isderived from the aggregate traffic load of all sur-rounding client-access point pairs which operateco-channel with the pair under consideration.We considered three different traffic loadings inthe study: interference traffic profile of a persis-tent 2 Mb/s video stream, interference trafficprofile of a mix of voice, video, and data at 2Figure 4. TVWS distribution across the United Kingdom: red: little; white: much.



Mb/s, and interference traffic profile of mixvoice, video and data at 6 Mb/s. These trafficprofiles are later referred to as Video, 2 Mb/s,and 6 Mb/s, respectively.

Simulations were performed of the achievabledata-rates at the client under the above deploy-ment and interference scenarios assuming opera-tion in the 2.4 GHz band, the 5 GHz band andin the available TVWS frequencies, using thesame random algorithm to assign the availablechannels to customer’s home access points. Inthe case of 2.4 and 5 GHz bands, the use of a 2× 1 antenna diversity technique (IEEE 802.11n)was assumed while, for TVWS spectral band,single antennas were assumed at both accesspoint and client due to the relatively large anten-na size required at UHF. Finally in order tomatch the peak data rate of 2.4 and 5 GHzbands, which have 20 MHz wide channels, three8 MHz TVWS bands were aggregated.

The stochastic nature of offered traffic, clientdistance from the home access point and distri-bution of home networks means that the resultsof our study are best represented in terms ofprobability distributions. Tables 1 and 2 showthe percentage of clients that can be served fromthe access point with a data rate of 2 and 6 Mb/s,respectively, for different interference trafficprofiles.

It can be seen from the above tables for lowtraffic loading, the TVWS spectral band outper-forms the 2.4 GHz and 5 GHz spectral bands.However, for high traffic loadings, which corre-spond to applications like wireless HDTVstreaming, better performance is achieved usingthe 5 GHz band. The main reason underlyingthe above results is the subtle interplay betweencoverage and interference effects. By operatingin TVWS spectrum a larger coverage withineach home can be achieved than is possiblewhen using the 2.4 GHz and 5 GHz bands. How-ever, at the same time the interference to neigh-boring homes is more severe due to better

propagation and penetration characteristics. Asa consequence there is an optimal transmittingpower of around +3 dBm, in contrast to the sig-nificantly higher operating points for the 2.4 and5 GHz WiFi bands which is +10 to 20 dBm.

Although transmit power is only one of thefactors that determines energy consumption ofdevices, our finding does point out an importantpotential advantage of operation in TVWS interms of energy/battery power saving which, toour knowledge, had been overlooked previously.Overall our results show that TVWS spectrum onits own is perhaps unable to support future high-data rate multimedia streaming applications butis best used in conjunction with either the 2.4 or5 GHz band, as a way for congestion relief orcoverage extension. Furthermore, triplet channelbonding in TVWS is required in order to achievethe same base performance rate as in 2.4 and 5GHz bands, which may not be always feasiblewith current access technologies due to the non-contiguous nature of TVWS spectrum (Fig. 1).

HIGH-SPEED BROADBAND WIRELESSACCESS FROM THE INSIDE-OUT

Wireless LAN (WLAN) operating in the ISMbands and using the IEEE 802.11x (i.e., standardWiFi technology) is one of the fastest and cheap-est broadband wireless access (BWA) systems,and has seen a high growth rate. Examples ofdeployment include WiFi hotspots such as BTOpenzone, municipal WiFi networks and publicWiFi networks, such as BT-FON. With FON,residential broadband customers share a portionof their home WiFi broadband bandwidth foroutdoor public use. The two traffic streams are

Figure 5. Population-weighted complementary cumulative distribution functionof TVWS availability. Red is when adjacent channels cannot be used.

MHz free (red: adjacent channels also free)

TVWS estimated from database

500

0.2

0.0

Frac

tion

of

popu

lati

on

0.4

0.6

0.8

100 150 200 250

Table 1. Performance comparison for 2 Mb/s ser-vice requirement.

Table 2. Performance comparison for 6 Mb/s ser-vice requirement.

Interferencetraffic profile 5 GHz 2.4 GHz TVWS

2 Mb/s 97% 97% 99%

6 Mb/s 97% 90% 97%

Video 97% 75% 97%

Interferencetraffic profile 5 GHz 2.4 GHz TVWS

2 Mb/s 98% 83% 98%

6 Mb/s 95% 70% 70%

Video 93% 55% 50%

1 At fine-grained simula-tion level we used anapproach based on divid-ing the square kilometerarea into 1 m2 pixels andpropagating signals fromone pixel to anotheraccording to the underly-ing topology of houses,and location of walls andfloors.



kept separate for security reasons and the resi-dential users’ traffic receives higher priority onuplink and downlink. The outdoor use of suchWiFi home networks is currently receiving ahigh level of interest from telecom operators andcustomers alike. Smartphones equipped withWiFi can roam onto the WiFi networks fromcellular networks, obtaining a higher data-rate inmost cases. In the United Kingdom, for exam-ple, there are currently over 1 million BT-FONsubscribers/providers, amounting to 6.3 percentof the residential broadband customers so thereis already significant coverage and there is poten-tial for further growth. Furthermore, mobileoperators like T-mobile have begun to seriouslylook at the use of WiFi indoor and outdoor net-works as a means for offloading the exponential-ly growing mobile data traffic from their highlyovercommitted cellular networks.

Unfortunately, despite the high density ofhome access points available for communitysharing, the coverage provided by such commu-nity WiFi networks is currently rather patchyhence limiting the usefulness and commercialsuccess of such networks. This is due to a combi-nation of the relatively stringent regulatory capson transmit power levels of WiFi in the UnitedKingdom and the rest of Europe, e.g., 20 dBm(100 mW) in the 2.4 GHz band and also due tothe high wall and floor penetration loss sufferedby signals in the ISM bands.

We performed system studies for severalurban and sub-urban service areas in Londonand elsewhere in order to investigate whether,by switching operation of the network from theISM bands to TVWS spectrum, the above cover-age limitations could be overcome. The architec-ture of the resulting network is shown in Fig. 6.Once again our proposed architecture is basedon the use of geo-location technology for loca-tion determination combined with a databaselook-up. Each access point is assumed to be con-nected to one, or multiple, databases which pro-vide information on the unused TVWS channelsthat are available at the location of the accesspoint — and information on maximum transmitpower levels that could be used in each channel.Furthermore, the use of master-slave technologymeans that the necessary functionalities for

database lookup and channel selection need tobe implemented only in the access points, sokeeping the complexity and cost of end-userdevices to a minimum. Users with a TVWSmodem, or dongle, can connect from outside tohome access points via a TVWS channel that isperiodically advertised via beacons by eachaccess point.

For the chosen deployment environment,dense urban, urban and sub-urban, a range ofdeployment densities of BT-FON access pointswere investigated in the study. Interferenceeffects from neighboring home hubs were mod-eled for a worst-case traffic profile comprising of2 Mb/s persistent video streaming traffic. Statis-tics on achievable data rates as a function oflocation of the outdoor client were then comput-ed on a 1m2 grid resolution level. Figure 7 sum-marizes the results of system studies which wereperformed for the dense urban deployment sce-nario. It shows the effect of switching operationfrequency of access points from 5 GHz to 2.4GHz and then to TVWS UHF bands on theachievable broadband (2 Mb/s) coverage of theBT-FON network. We found that, due to inter-ference effects for each band considered, thereis an optimal deployment density of access pointsbeyond which the coverage does not furtherimprove, and can even start degrading.

From Fig. 7 it can be seen that coverage isvery patchy when the system operates at 5 GHzand some improvement is gained by switching to2.4 GHz. The most striking result is achievedwhen home access points switch operation toTVWS where, with only a 20 percent deploy-ment density, a blanket 2 Mb/s indoor-outdoorcoverage is achieved. Note that this broadbandcoverage level at 2 Mb/s data rate is 25 timeshigher than that achievable with 3G technologiessuch as HSPA. It is economically much moreviable than the broadband coverage that couldbe offered by 4G cellular technologies due to therelatively low infrastructure and site acquisitioncosts.

RURAL BROADBANDIn the United Kingdom right now, about 15 per-cent of households (2.75 million) cannot accessbroadband (2 Mb/s). These are located mainly,but not exclusively, in rural areas. The problemis apparently not limited to the United Kingdomas latest EU statistics show that 30 percent ofthe EU’s rural population has no access to highspeed Internet.

The problem is caused mainly by long metallines, either between the exchange and the houseor in the backhaul to the exchange itself. Forsome situations, the cost of upgrading the linesor backhaul is very high when considering therelatively few users who would benefit, or theterrain through which the lines pass. The thresh-old capital cost, beyond which the commercialcase disappears for the large operator, is around$2000 per house. Consequently, operators in theUnited Kingdom and elsewhere are looking seri-ously into wireless alternatives and TVWS is atempting proposition. The topology here isessentially point-to-point or point-to-multipoint,which indicates that WiFi, fixed WiMAX, orTD-LTE are potential options for the air inter-

Figure 6. Indoor-to-outdoor broadband wireless access network using homeaccess points operating in TVWS.

Nomadic userWiFi TVWS or3G/4G TVWS

Mobile userTVWS+3G/4G

TVWS database

Home hub

Network

Home hub

Network



face since CR technology is agnostic to the actu-al air interface used.

Rural broadband using LTE is being pursuedby Verizon in the United States and T-mobile inGermany and both have secured cleared DTTspectrum in the 700 MHz band which they planto use for its delivery. Another option would beto use unlicensed wireless broadband provisionin the ISM bands using a combination of WiFi in5 GHz band (for point-to-point wireless back-haul) and WiFi in 2.4 GHz band (for point-to-multipoint) distribution. This option is currentlybeing exploited in the United States, mainly onsmall scales for local communities, where thereare currently around 3000 Wireless Service Pro-viders (WISP) that use this model.

The WISP community model is incompatiblewith large operators such as BT for two reasons.First, large operators are required by regulationto unbundle, or offer network transport to multi-ple communications providers (CPs). Thereforea large operator will build out infrastructurewhich has the necessary connection points andnetwork management processes for this to bepossible, rather than build a community networkwhich is essentially just one CP. Secondly, it isnot economically viable for a community net-work to use a large operator for backhaul to theInternet. This is because the CP has to pay forboth backhaul (a wholesale function) and Inter-net provision (a retail function) — and a largeoperator is forbidden by regulation to subsidiesone with the other. It may cost a total of £20kper year to connect a community network to theInternet at 8 Mb/s, which can connect perhaps50 households — and this is expensive comparedwith, say, a satellite network operator who is notbound by such regulation.

TVWS is attractive for use with rural broad-band, first because it is free and secondlybecause it is fairly stable; the primary users willonly rarely change their use of the band in suchareas. Since it is fixed point-to-multipoint, thereis also a low probability that other secondaryusers will cause interference, although of coursemechanisms must be in place to make this prob-ability approach zero.

The large operator must, though, make provi-sion for when too little TVWS is available. Theworst scenario is where an operator begins toprovide broadband services using TVWS andthen changes occur either with primary users orother secondary users appear which means thatsuch services must be reduced or suspended.Such provision could be the reservation of a fewchannels in each region, which can be used inthis situation as a fallback — and the operatorwould need to pay the regulator for such reser-vation. This topic is only just starting to be dis-cussed between operators and regulators, andlittle progress has so far been made. The opti-mum amount of spectrum to reserve in eachregion is identified as a further research chal-lenge.

In the US the provision of cost-effective ruralbroadband in TVWS has been under evaluationfor a number of years, and this has been one ofthe initial motivations for FCC to open thesebands for cognitive access. Furthermore, theIEEE 802.22 family of standards has been devel-

Figure 7. Achievable indoor-outdoor broadband coverage using BT-FON accesspoints operating at 5 GHz (top) 2.4 GHz (center), and TVWS (bottom).



oped specifically with rural broadband in mind[12]. The situation in the United Kingdom is dif-ferent because the TVWS is exclusively in theUHF band, the channels are wider at 8 MHzand the population density distribution is differ-ent. We have investigated the feasibility of ruralbroadband provision in TVWS in the UnitedKingdom and first indications are that it may betechnically and commercially feasible. Conse-quently BT is setting up TVWS testbeds in Scot-land and England to make measurements ofcoverage, capacity and efficiency in practical sit-uations.

The aim of one BT study was to investigatehow many houses could and should be connect-ed using TVWS. Houses which should not beconnected using TVWS are those which alreadyhave adequate DSL provision and these are gen-erally where the total line length is less thanabout 3 km. There are very few houses in theUnited Kingdom that are further than 8 kmfrom an exchange, so that the window of oppor-tunity for TVWS in rural broadband is for hous-es that are between 3 and 8 km. We call suchhouses not-spots. Each isolated community is,therefore characterized by two parameters; thenumber of not-spots in the cluster within thisdistance window and the distance between thefurthermost not-spot and the nearest exchange.The first parameter determines the requiredwireless bandwidth that needs to be available inorder to provide 2 Mb/s broadband to customerswhile the maximum distance determines thetransmit power level required to establish therequired link.

We have used a database of U.K. housingdensity together with an internal database of the2.75 million U.K. not-spots and 5500 BTexchanges to count the number that can be cov-ered with TVWS. We cannot divulge the actualresults, but the number is high enough to bepotentially commercially feasible, provided thata satisfactory resolution is obtained to the fall-back problem discussed above.

Not-spots in rural areas tend to be clusteredinto groups, where the largest group is around60 but more typically they are 10 to 30. Assum-ing that customer’s traffic is bursty, and using anoverbooking factor of 10, it follows that therequired data rate on the downlink for the

largest groups should be around 2 * 60/10 = 12Mb/s. In the worst case, this would need to bedelivered over an 8 km distance via the TVWSlink. Our initial estimate indicates that it shouldbe possible to provide such a data rate via aTVWS link in the UHF band, provided thetransmit power allowed by the regulator is notless than 4W. This aligns, coincidentally perhaps,with the 4W transmit power limit imposed byFCC on fixed TVWS devices in the UnitedStates. However, unlike the FCC, Ofcom has notimposed any maximum transmit power limits ondevices that rely on a geo-location database forincumbent detection but has left the choice ofthis parameter to be determined by the database.We have constructed a laboratory testbed toassess the air interface performance in respect oftransmit powers and bandwidth — and also theMAC layer functions; a basic diagram of thesetup is shown in Fig. 8.

In the testbed, the TVWS link is formedbetween two Ubiquity software-defined radiomodules and Yagi antennas. The TVWS used is762 MHz, transmitted under the authority of atest and development license. We have achieved6 Mb/s throughput in the 8 MHz channel, butare currently working to improve this throughthe use of higher-level modulation schemes,multiple-input multiple-output (MIMO), andimproved MAC functions, toward the 12 Mb/swe believe is necessary.

CONCLUSIONA regulatory framework for secondary utilizationof TV white spaces is well underway in both theUnited States and the United Kingdom, whileimportant steps in this direction are being takenwithin the European Union and elsewhere.Using results from our recent research at BT, wehave shown in this article that cognitive access toTVWS is a significant new opportunity for oper-ators to provide a range of improved and newwireless services. In addition to the three usecases described here, another important applica-tion we are investigating includes high-capacitywireless connectivity for future data-intensivesmart utility grids and vehicular communicationnetworks.

The first use cases that will employ TVWSwill be fixed point-to-point links such as ruralbroadband access. This is forecast to occur inthe next one to two years. Looking ahead, theuse-cases will become more mobile and withvariable and managed QoS, such as indoor tooutdoor coverage via super WiFi communitynetworks and possibly femtocells. Mobile net-work operators are interested in the use ofTVWS for cellular extension and rural access.The challenges to be overcome include:• Quantifying the amount of TVWS spectrum

that is available, which is being addressedin some EC part-funded projects likeCogEU and QUASAR [9].

• Provision of reliable service and managedmobility and QoS, which is also beingaddressed in some EC projects such asQoSMOS.

• Agreement across Europe and the UnitedStates on regulatory aspects.

Figure 8. Block diagram of the testbed.

TVWS

WiFi

Router/gateway

Internet



• There are currently many overlapping stan-dards emerging from the EuropeanTelecommunications Standards Institute(ETSI), IEEE, and International Telecom-munication Union (ITU), which may leadto fragmentation of the market.

• The growth of the necessary ecosystem sothat terminals and equipment are availableat reasonable cost.

• CR equipment certification procedures.• Development of a new value chain including

databases and over-the-top services basedon location. The FCC is setting up a newadvisory group to look into this.

• Optimization of fallback spectrum methods.• Database structure and etiquette.• Flexibility of radio devices.• MAC layer development and its interaction

with the databases.

ACKNOWLEDGMENTThe authors thank their colleagues at BT andour partners in the EU FP7 projects QoSMOSand QUASAR for useful discussions.

REFERENCES[1] J. Mitola III and G. Q. Maguire, “Cognitive Radio: Mak-

ing Software Defined Radio More Personal,” IEEE Pers.Commun., vol. 6, no. 4, Aug. 1999.

[2] A. Wyglynski, M. Nekovee, and T. Hou, Eds., CognitiveRadio Communications and Networks: Principles andPractice, Academic Press, 2009.

[3] FCC, “Second Report and Order in the Matter of Unli-censed Operation in the TV Broadcast Bands (ET Docketno. 04-186), Additional Spectrum for UnlicensedDevices Below 900 MHz and in 3 GHz Band (EC DocketNo 02-380),” Nov. 14, 2008.

[4] FCC, “Second Memorandum Opinion and Order in theMatter of Unlicensed Operation in the TV BroadcastBands (ET Docket no. 04-186), Additional Spectrum forUnlicensed Devices Below 900 MHz and in 3 GHz Band(EC Docket No 02-380),” Sept. 23, 2010.

[5] Ofcom, “Statement on Cognitive Access to InterleavedSpectrum,” July 1, 2009; http://www.ofcom.org.uk/con-sult/condocs/cognitive/statement.

[6] Ofcom, “Digital Dividend: Geolocation for CognitiveAccess,” http://www.ofcom.org.uk/consult/condocs/cogaccess/cogaccess.pdf.

[7] M. Nekovee, “Cognitive Radio Access to TV WhiteSpaces: Spectrum Opportunities, Commercial Applica-tions and Remaining Technology Challenges,” Proc.IEEE DyPAN, Singapore, Apr. 2010.

[10] Cognitive Networking Alliance; http://www.cognea.org/.[11] ECMA Std. 392, “MAC and PHY for Operation in TV

White Space,” Dec. 2009.

[12] IEEE 802.22, “WG on WRANs (Wireless Regional AreaNetworks)”; http://www.ieee802.org/22/.

[8] QoSMOS; http://www.ict-qosmos.eu.[9] QUASAR; http://www.quasarspectrum.eu.

BIOGRAPHIESMICHAEL FITCH ([email protected]) works in the Researchand Technology part of BT Innovation and Design, leadinga small research team specializing in physical and systemsaspects of wireless communications. He has been with BTsince 1989 working in various research and developmentroles, currently working on a number of collaborative pro-jects on emerging wireless technologies such as LTE, smallcells, and cognitive radio. In addition he provides consul-tancy to other parts of BT on wireless matters. Previousexperience is with satellite systems and mobile radio sys-tems. He holds a first degree in math and physics, a Ph.D.in satellite communications, and is a member of the IET.

MAZIAR NEKOVEE leads research on cognitive radio and newparadigms for spectrum access at BT, and provides adviceto BT’s Spectrum Strategy Group. His research focuses onanalysis, performance modeling, and algorithm develop-ment for complex networked systems. He obtained hisM.Sc. in electrical engineering (cum laude) from Delft Uni-versity of Technology in the Netherlands and his Ph.D. intheoretical physics from the University of Nijmegen, also inthe Netherlands. He is a recipient of the prestigious Indus-try Fellowship from the Royal Society. He is the author ofover 60 papers in peer-reviewed journals and conferences,and holds several patents. He has been involved in themanagement of a number of national, European, andinternational collaborative projects.

KEITH BRIGGS has a Ph.D. in mathematics from MelbourneUniversity, and has published in the areas of dynamical sys-tems, computational number theory, biomathematics, andstatistical mechanics. He has worked in the last 10 years onmathematical and statistical modeling of communicationssystems. He also publishes in the field of historical linguis-tics, especially as applied to toponyms.

SANTOSH KAWADE received his M.Sc. in telecommunicationsengineering from University College London and is current-ly studying part-time toward a doctorate degree at Univer-sity College London. Since joining BT, he has contributedto research toward understanding the fundamental perfor-mance limits of wireless networks, radio wave propagation,and interference modeling. He has published a number ofacademic papers in this area.

RICHARD MACKENZIE received his M.Eng. in electronic engi-neering from the University of York, United Kingdom, in2005 and his Ph.D. in electronic and electrical engineer-ing from the University of Leeds, United Kingdom, in2010. His Ph.D. focused on improving QoS over wirelesshome networks, with a focus on real-time video. Hejoined BT Innovate & Design in 2009, where he works asa wireless researcher. His work mainly involves MAC layeranalysis and wireless testbed implementations for cogni-tive radio.

Looking ahead, the

use cases will

become more

mobile and with

variable and man-

aged QoS, such as

indoor to outdoor

coverage via super

WiFi community net-

works and possibly

femtocells. Mobile

network operators

are interested in the

use of TVWS for cel-

lular extension and

rural access.



INTRODUCTION

Current spectrum allocations are based on acommand-and-control philosophy; that is, spec-trum is allocated for a particular application(e.g., TV broadcasting), and such allocations donot change over space and time. There havebeen several important developments in the pastfew years in the spectrum policy and regulatorydomains to accelerate opportunistic uses of spec-trum. The most recent of these are the publica-tion of the National Broadband Plan in March2010 [1], the publication of the final rules forunlicensed devices in the TV bands in Septem-ber 2010 [2], and the ongoing proceeding forsecondary use of the 2360–2400 MHz band formedical body area networks (MBANS) [3]. Cog-nitive radio (CR) technology plays a significantrole in making the best use of scarce spectrum tosupport the increasing demand for emergingwireless applications, such as TV bands forsmart grid, public safety, broadband cellular, andthe MBAN band for medical applications. Inorder to take advantage of these new opportuni-ties, a number of standards (e.g. IEEE 802.22[4], IEEE 802.11af, ECMA 392 [5], IEEESCC41, and ETSI RRS [6]) are either in devel-opment or have already been completed.

REGULATION

NATIONAL BROADBAND PLAN

The National Broadband Plan (NBP) is a policydocument that was the culmination of almost ayear’s worth of work by the Federal Communi-cations Commission (FCC) with input fromindustry and government agencies on how toformulate spectrum policy in order to facilitatebroadband usage for the coming years. One ofthe main recommendations of the NBP is tofree up 500 MHz of spectrum for broadbanduse in the next 10 years with 300 MHz beingmade available for mobile use in the next fiveyears. The Plan proposes to achieve this goal ina number of ways: incentive auctions, repackingspectrum, and enabling innovative spectrumaccess models that take advantage of oppor-tunistic spectrum access and cognitive tech-niques to better utilize spectrum. The Planurges the FCC to initiate further proceedings onopportunistic spectrum access beyond thealready completed TV white spaces (TVWS)proceedings.

TV WHITE SPACES REGULATIONThe major worldwide regulatory agenciesinvolved in developing rules for the unlicenseduse of TV white spaces are the FCC in the Unit-ed States, the Office of Communications(Ofcom) in the United Kingdom, and the Elec-tronic Communications Committee (ECC) ofthe Conference of European Post and Telecom-munications (CEPT) in Europe.

The FCC released the final rules for “Unli-censed Operation in the TV Broadcast Bands”in September 2010 [2]. This was the culminationof many years of deliberations on the subject,starting with the first NPRM in May 2004 andfollowed by laboratory and field testing of sens-ing devices through 2007 and 2008 and the sec-ond report and order in 2008[7]. A recent studyshows the opportunity provided by TV whitespaces is shown to be potentially of the sameorder (~62MHz) as the recent release of “beach-front” 700MHz spectrum for wireless data ser-vice [8], while New America Foundation hasanother estimate of 15–40 channels available inmajor cities [9]. The main features of the rulesas set forth in this order are as follows:

ABSTRACT

Recent developments in spectrum policy andregulatory domains, notably the release of theNational Broadband Plan, the publication offinal rules for TV white spaces, and the ongoingproceeding for secondary use of the 2360–2400MHz band for medical body area networks, willallow more flexible and efficient use of spectrumin the future. These important changes open upexciting opportunities for cognitive radio toenable and support a variety of emerging appli-cations, ranging from smart grid, public safetyand broadband cellular, to medical applications.This article presents a high-level view on howcognitive radio (primarily from a dynamic spec-trum access perspective) would support suchapplications, the benefits that cognitive radiowould bring, and also some challenges that areyet to be resolved. We also illustrate relatedstandardization that uses cognitive radio tech-nologies to support such emerging applications.

COGNITIVE RADIO NETWORKS (INVITED ARTICLE)

Jianfeng Wang, Monisha Ghosh, and Kiran Challapali

Philips Research North America

Emerging Cognitive Radio Applications: A Survey

WANG LAYOUT 2/18/11 3:20 PM Page 74


•TV band devices (TVBDs) are divided intotwo categories: fixed and personal/portable.Fixed TVBDs operate from a known, fixed loca-tion and can use a total transmit power of up to4 W effective isotropic radiated power (EIRP),with a power spectral density (PSD) of 16.7mW/100 kHz. They are required to either have ageolocation capability or be professionallyinstalled in a specified fixed location and havethe capability to retrieve a list of available chan-nels from an authorized database. Fixed TVBDscan only operate on channels that are not adja-cent to an incumbent TV signal in any channelbetween 2 and 51 except channels 3, 4, and 37.Personal/portable devices are restricted to chan-nels 21–51 (except channel 37) and are allowed amaximum EIRP of 100 mW with a PSD of 1.67mW/100 kHz on non-adjacent channels and 40mW with a PSD of 0.7 mW/100 kHz on adjacentchannels, and are further divided into two types:mode I and mode II. Mode I devices do notneed geolocation capability or access to adatabase. Mode II devices must have geoloca-tion capability and the means to access adatabase for list of available channels.

•Sensing was a mandatory feature to protectincumbents in the previous ruling [7] but is nowan optional feature in fixed mode I and mode IIdevices. Incumbent protection will be throughthe use of authorized databases that have toguarantee security and accuracy of all communi-cations between it and fixed or mode II devices.Geolocation means in mode II devices have tobe accurate within ±50 m. Since sensing isoptional, in order to maintain up-to-data chan-nel availability information, Mode II devicesneed to check their location every 60 s and, ifthe location changes by more than 100 m, haveto access the database for an updated channellist. In order to facilitate mobility, mode IIdevices are allowed to download channels for anumber of locations within an area and use achannel that is available within that area withoutthe need to access the database as long as itdoes not move outside the area. In addition, anew mechanism is defined in the rules to ensurethat mode I devices that do not have geolocationare within the receiving range of the fixed ormode II device from which it obtained the list ofchannels on which it could operate. This is the“contact verification” signal, which needs to bereceived by the mode 1 device every 60 s, or else

it will have to cease operation and reinitiate con-tact with a fixed or mode II device.

•A sensing-only device is a personal/portableTVBD that uses spectrum sensing only to deter-mine a list of available channels. Sensing onlydevices may transmit on any available channelsin the frequency bands 512-608 MHz (TV chan-nels 21–36) and 614–698 MHz (TV channels 38-51), and are allowed a maximum transmit powerof 50 mW with a PSD of 0.83 mW/100 kHz onnon-adjacent channels and 40 mW with a PSDof 0.7 mW/100 kHz on adjacent channels. Inaddition, sensing only device must demonstratewith an extremely high degree of confidence thatthey will not cause harmful interference toincumbent radio services. The required detectionthresholds are: ATSC digital TV signals: –114dBm, averaged over a 6 MHz bandwidth; NTSCanalog TV signals: –114 dBm, averaged over a100 kHz bandwidth; and Low power auxiliary,including wireless microphone, signals: –107dBm, averaged over a 200 kHz bandwidth. ATVBD may start operating on a TV channel ifno TV, wireless microphone or other low powerauxiliary device signals above the detectionthreshold are detected within a minimum timeinterval of 30 secs. A TVBD must perform in-service monitoring of an operating channel atleast once every 60 secs. After a TV, wirelessmicrophone or other low power auxiliary devicesignal is detected on a TVBD operating channel,all transmissions by the TVBD must cease withintwo seconds.

•Safe harbor channels for wireless micro-phone usage are defined in all markets to be thefirst available channel on either side of Channel37. TVBDs cannot operate on these channels. Inaddition, licensed and unlicensed wireless micro-phone users can register in the database if theycan demonstrate that they require adequate pro-tection from interference.

Table 1 summarizes the various parametersand potential applications of TVBDs enabled bythe US TVWS rules.

Meanwhile, Ofcom, the regulatory body inthe UK, has also made significant progress indeveloping regulations for the TV white spaceswith a first consultation released on February 16,2009, and a further statement in July 2009 [10].The detailed rules have yet to be released butfirst indications are that TVBDs will requireeither sensing or gelocation/database access. The

Table 1. TVBD parameters and applications.

Fixed device Mode II, mode I Sensing only

Channels 2~51 except 3, 4, 37. Non-adjacentchannels only 21–51 except 37 21–51 except 37

Power limit 4 W 100 mW 50 mW

Incumbent protectionmechanisms Database Database, contact

verification (mode I) Sensing

Potential applicationsSmart grid (network gateway, smartmeters), cellular backhaul (BS, relaystation), MBMS (BS)

Public safety, fem-tocell, MBMS (CE)

Public safety,femtocell

Ofcom has made

progress in develop-

ing regulations for

the TV white spaces

with a first consulta-

tion released on

February 16, 2009,

and a further state-

ment in July 2009.

The detailed rules

have yet to be

released but first

indications are that

TVBDs will require

either sensing or

gelocation/database

access.



sensing levels being proposed for sensing-onlydevices are –120 dBm for digital TV and –126dBm for wireless microphones.

The ECC has just begun working on cognitiveradio in the TV bands within its newly createdgroup SE 43 [11] which is tasked with definingthe technical and operational requirements ofoperating in the TV white spaces. Draft ECCreport 159 [12] was released on Sept 30, 2010 forpublic consultation. This report will be used asthe starting point for regulatory activities withinthe ECC.

MBANS REGULATORY ACTIVITIES IN THE USThe proposal to allocate the frequency band2360–2400MHz for MBANS on a secondarybasis was initially made in the US by GE in 2007[13], followed by a NPRM issued by the FCC in2009 [3]. The principal incumbent in this band inthe US is Aeronautical Mobile Telemetry(AMT), which uses 2360–2390 MHz, and thereare a number of proposals under considerationby the FCC that would allow MBAN devices assecondary users to coexist with the primary ATSuser, without either creating interference to orbeing subject to interference from AMT services.These include exclusion and coordination zonesas well as additional interference mitigationmechanisms, such as Listen-Before-Transmit andAdaptive Frequency Selection (LBT/AFS). Sincethe proposed transmit power for MBANS isquite low (1 mW in 2360–2390 MHz and 20 mWin 2390–2400 MHz), simulations have shown thatthese techniques would work well to protectAMT from interference while also maintainingthe quality of service required for the MBANSapplication [14, 15]. Thus, spectrum utilization ismaximized by allowing opportunistic use of theunderused frequency band 2360-2400 MHz forMBANS applications, instead of allocating newspectrum exclusively for this purpose. This pro-ceeding is still under consideration by the FCCand final rules are yet to be published.

In Europe, activities have been restricted toLow Power-Active Medical Implants (LP-AMI).Draft ECC Report 149 [16] considers the feasibil-ity of frequency bands 2360–2400 MHz,2400–2483.5 MHz, 2483.5–2500 MHz and2700–3400 MHz for LP-AMI, and concludes withthe recommendation that the frequency band2483.5–2500 MHz is the most promising candi-date band for this purpose, based on analysis ofinterference between LP-AMI and existing andproposed future incumbents in these bands.External medical telemetry or MBANs, is notconsidered in this report; however, proposals havebeen made to initiate an activity to explore theuse of the 2360–2400 MHz frequency band forMBANs in Europe in order to harmonize withthe anticipated regulation in the United States.

SMART GRID NETWORKSTransformation of the 20th-centrury power gridinto a smart grid is being promoted by manygovernments as a way of addressing energy inde-pendence and sustainability, global warming andemergency resilience issues [17, 18]. The smartgrid comprises three high-level layers, from anarchitectural perspective: the physical power

layer (generation and distribution), the commu-nication networking layer, and the applicationslayer (applications and services, e.g., advancedmetering, demand response, and grid manage-ment). A smart grid transforms the way power isgenerated, delivered, consumed and billed.Adding intelligence throughout the newly net-worked grid increases grid reliability, improvesdemand handling and responsiveness, increasesefficiency, better harnesses and integrates renew-able/distributed energy sources, and potentiallyreduces costs for the provider and consumers.

Sufficient access to communication facilitiesis critically important to the success of smartgrids. A smart grid network would typically con-sist of three segments [17]:• The home/building area networks (HANs)

that connect smart meters with on-premiseappliances, plug-in electrical vehicles, anddistributed renewable sources (e.g., solarpanels)

• The advanced metering infrastructure(AMI) or field area networks (FANs) thatcarry information between premises (viasmart meters) and a network gateway (oraggregation point), which will often be apower substation, a utility pole-mounteddevice, or a communications tower

• The wide area networks (WANs) that serveas the backbone for communicationbetween network gateways (or aggregationpoints) and the utility data centerWhile HANs can use WiFi, Zigbee, and

HomePlug, and WANs can leverage the fiber-based IP backbone or even the broadband cellu-lar network infrastructure, appropriatetechnologies for AMI/FANs are still under con-sideration. The dimension of an AMI/FAN couldrange from a few hundred meters to a few kilo-meters or more (e.g., in rural areas). Bandwidthrequirements are estimated in the 10–100 kb/srange per device in the home or office building[17]. This may scale up quickly with the numberof devices on a premise if appliance-level datapoints as opposed to whole-home/building dataare transmitted to the network gateway. Powerline communication (PLC) is used in some AMIbut has bandwidth and scalability problems.Moreover, the safety issues associated withground fault currents are of concern as well.Some wireless meter readers currently use the900 MHz unlicensed band. This is not withoutcomplications, however, since this band will soonbecome crowded due to the growth of unli-censed devices including smart meters. IEEE802.15.4g, the Smart Utility Networks (SUN)Task Group, is currently working to create aphysical layer (PHY) amendment for AMI/FANby using license-exempt frequency bands such as700 MHz–1 GHz and the 2.4 GHz band. Itremains to be seen how 802.15.4g handles inter-ference, which is common to unlicensed devicesoperating in these bands. The cellular network isan alternative for AMI/FAN as well. However,the investment and operation costs could behigh. Moreover, cellular networks themselvesface bandwidth challenges as cellular data trafficgrows dramatically year by year. Cellular net-works also have coverage issues in certain places(e.g., rural areas).

Adding intelligence

throughout the

newly networked

grid increases grid

reliability, improves

demand handling

and responsiveness,

increases efficiency,

better harnesses and

integrates renew-

able/distributed ener-

gy sources, and

potentially reduces

costs for the provider

and consumers.



Cognitive-radio-based AMI/FANs may offermany advantages such as bandwidth, distanceand cost, as compared with other wireline/wire-less technologies in certain markets. Figure 1illustrates a CR-based wide area AMI/FAN. Inthis case, the network gateway and smart metersare equipped with CR and dynamically utilizeunused/underutilized spectrum to communicatewith each other directly or via mesh networkingover a wide area with minimal or no infra-structure. The network gateway connects with aspectrum database over a WAN and serves as thecontroller to determine which channel(s) to usefor the AMI/FAN based on the location andtransmission power needed for smart meters.Taking TVWS as an example, since networkgateways and smart meters are both fixed, theycan operate in the fixed mode and use transmis-sion power up to 4 W EIRP. With the high trans-mission power and superior TV band propagationcharacteristics, the network gateway may reachall the smart meters with one or two hops (e.g.,covering an entire town). In rural areas availableTVWS channels could be abundant, so channelavailability would not be an issue.

There are several other standardizationgroups currently working on the incorporation ofcognitive radio technologies to utilize TVWS tosupport applications such as smart grid networks,particularly AMI/FANs. Within the IEEE, thefollowing groups are developing standards forTVWS: The IEEE 802.22 Working Group isnearing completion of the standard for TVWS-based wireless regional area networks for rangesup to 10–100 km, which could be used for large-scale smart grid networks; an IEEE 802.15 studygroup (SG) has been created recently to investi-gate the use of TVWS; and IEEE 802.11af isspearheading the development of an IEEE 802.11amendment for TVWS operation for WLANs.

Like other unlicensed devices, CR-enabledAMI/FAN devices are not immune from interfer-ence or congestion, especially if they are hetero-geneous and not coordinated with each other.This may introduce issues such as reliability anddelay, and limit the applicability of unlicenseddevices for more critical grid control or real-time

smart grid applications. CR-enabled AMI/FANsshould go beyond just dynamic spectrum accessand develop self-coexistence mechanisms to coor-dinate spectrum usage, and may even prioritizespectrum use according to the class of smart gridtraffic (e.g., real-time vs. non-real-time, emergen-cy report vs. demand response). The IEEE802.19.1 Working Group is currently working ondeveloping a standard for wireless coexistence inthe TVWS and may help mitigate interferenceissues among CR-based AMI/FANs. Further-more, CR-enabled AMI/FANs should also con-sider how to interoperate with other wirelesstechnologies such as wireless cellular networks inorder to make the smart grid more resilient, scal-able, accessible, and of better quality.

PUBLIC SAFETY NETWORKSWireless communications are extensively used byemergency responders (e.g., police, fire, andemergency medical services) to prevent orrespond to incidents, and by citizens to quicklyaccess emergency services. Public safety workersare increasingly being equipped with wirelesslaptops, handheld computers, and mobile videocameras to improve their efficiency, visibility,and ability to instantly collaborate with centralcommand, coworkers, and other agencies. Thedesired wireless services for public safety extendfrom voice to messaging, email, web browsing,database access, picture transfer, video stream-ing, and other wideband services. Video surveil-lance cameras and sensors are becomingimportant tools to extend the eyes and ears ofpublic safety agencies. Correspondingly, datarates, reliability, and delay requirements varyfrom service to service.

On the other hand, the radio frequenciesallocated for public safety use [19] have becomehighly congested in many, especially urban, areas[20]. Moreover, first responders from differentjurisdictions and agencies often cannot commu-nicate during emergencies. Interoperability ishampered by the use of multiple frequencybands, incompatible radio equipment, and a lackof standardization.

Figure 1. Smart grid networks.

CR basedwide areaAMI/FAN

HAN

GreenHome

GreenHome

Spectrumdatabase

Networkgateway

(e.g., fixed TVBD)

Smart meter(e.g., fixed TVBD)

WAN

GreenHome

GreenHome

With the high trans-

mission power and

superior TV band

propagation charac-

teristics, the network

gateway may reach

all the smart meters

with one or two

hops, e.g., covering

an entire town. In

rural areas, available

TVWS channels

could be abundant

so channel availabili-

ty would not be

an issue.


In coping with the above challenges, the U.S.Department of Homeland Security (DHS)released its first National Emergency Communi-cations Plan (NECP) in July 2008. The morerecently released National Broadband Plan [1]clearly reflects the effort to promote public safe-ty wireless broadband communications. The rec-ommendations include creating a public safetybroadband network, creating an administrativesystem that ensures access to sufficient capacityon a day-to-day and emergency basis, and ensur-ing there is a mechanism in place to promoteinteroperability.

Cognitive radio was identified as an emergingtechnology to increase efficiency and effective-ness of spectrum usage in both the NECP reportand the National Broadband Plan. With CR,public safety users can use additional spectrumsuch as license-exempt TVWS for daily opera-tion from location to location and time to time.With appropriate spectrum sharing partnershipswith commercial operators, public safety workerscan also access licensed spectrum and/or com-mercial networks. For example, the public safetycommunity could roam on commercial networksin 700 MHz and potentially other bands both inareas where public safety broadband wirelessnetworks are unavailable and where there is cur-rently an operating public safety network butmore capacity is required to respond effectivelyto an emergency.

Figure 2 illustrates public safety communica-tions with incorporation of CR networking tech-nologies. In this case, location-aware and/orsensing-capable CR devices together with thespectrum coordinator in the back office respondto the emergency and coordinate with users(including primary and secondary users)

in/around the incident area to ensure the emer-gency responders have sufficient capacity andmeans for communications on the field andto/from infrastructure. In addition, CR canimprove device interoperability through spec-trum agility and interface adaptability, or a net-work of multiple networks. CR devices cancommunicate directly with each other by switch-ing to common interface and frequency. Further-more, with help of multi-interface orsoftware-defined radio (SDR), CR can serve asthe facilitator of communications for otherdevices which may operate in different bandsand/or have incompatible wireless interfaces. Asillustrated in Fig. 2, such CR devices (communi-cation facilitators) can be located in a few pow-erful emergency responders’ vehicles andwireless access points. This lifts the burden offthe handheld devices for each to have CR capa-bility to mitigate the issue that different emer-gency responders may use different radios todayand very likely in the future as well.

It remains to be seen how CR technologieswill support priority delivery and routing of con-tent through its own network as well as publicnetworks, thus protecting time-sensitive life-sav-ing information from loss or delay due to net-work congestion. This goes beyond spectrumawareness to content awareness, from the PHYto the application layer.

Standardization remains key to the success ofCR. ECMA 392 standard is the first internation-al standard that specifies PHY and mediumaccess control (MAC) layers to enable person-al/portable devices to operate in TVWS. WhileECMA 392 is not designed specifically for publicsafety, it may be suitable for the following rea-sons. ECMA 392 supports dynamic channel useby using both geolocation-based databases aswell as sensing, and can be adapted to complywith local spectrum regulations. Compared toother existing standards, ECMA 392 not onlysupports flexible ad hoc networking but alsoquality of service (QoS), which is required foron-field emergency communications.

CELLULAR NETWORKSThe use of cellular networks is undergoing dra-matic changes in recent years, with consumers’expectations of being always connected, any-where and anytime. The introduction of smart-phones, the popularity of social networks,growing media sites such as Youtube, Hulu, andflickr, introduction of new devices such as e-readers, have all added to the already high andgrowing use of cellular networks for convention-al data services such as email and web-browsing.This trend is also identified in the FCC’s vision-ary National Broadband Plan [1].

This presents both an opportunity and a chal-lenge for cellular operators. The opportunity isdue to the increased average revenue per userdue to added data services. At the same time,the challenge is that in certain geographicalareas, cellular networks are overloaded, duepartly to limited spectrum resources owned bythe cellular operator. Recent analysis [21] sug-gests that the broadband spectrum deficit is like-ly to approach 300 MHz by 2014, and that


Figure 2. Public safety networks.

WAN

Spectrumcoordinator

Accesspoint

(e.g., fixed TVBD)Communicationvehicle

(e.g., mode IITVBD)

CR basedMANET

Emergency 911EMS Helicopter Operations

(e.g., mode ITVBD)



making available additional spectrum for mobilebroadband would create value in excess of $100billion in the next five years through avoidanceof unnecessary costs.

With the FCC’s TVWS ruling, new spectrumbecomes available to cellular operators. In thelong term, television band spectrum that is cur-rently not described as white spaces may alsobecome available to cellular operators, as dis-cussed in the National Broadband Plan. Specifi-cally, the plan discusses the possibility for currentlicense holders of television spectrum to volun-tarily auction their licenses, in return for part ofthe proceeds from the auction. The plan envi-sions that this newly freed spectrum could beused for cellular broadband applications (hencethe name of the plan).

Many papers have investigated the applica-tion of spectrum sensing or spectrum sharing incellular networks [6, 22, 23]. Figure 3 illustrateshow cognitive radio technologies can augmentnext generation cellular networks like LTE andWiMAX to dynamically use these newly avail-able spectrums either in the access or backhaulparts of their networks. A spectrum coordinatorcan be added in the non-access stratum (NAS)to allow cellular networks to dynamically leasespectrum from spectrum markets and/or identifysecondary license exempt spectrum opportunitiesto meet the cellular traffic demand given a loca-tion and time period. The base stations (includ-ing relay stations) configure channels to operateaccording to the instructions of the spectrumcoordinator and aggregate the spectrum for use.

For access network applications, two usecases can be envisioned. The first is hotspots,such as game stadiums and airports, where alarge number of users congregate at the sametime. Take the example of a stadium: usersincreasingly have phones equipped with camerasthat can capture pictures or videos of events atthe game and upload them to media sites orsend them to their friends. Such picture andvideo data puts enormous strain on the cellularnetwork. In Cisco’s study 60 percent of growth isexpected from such picture and video data.Today, some of this data can be offloaded toISM band WiFi networks. However, due to thelarge amount of data generated in a small area(“hotspot”), both cellular networks and ISMband WiFi networks, are likely to be overloaded.If this data can be offloaded to additional spec-trum, such as TVWS, the cellular network canthen be used for voice applications in a morereliable fashion, thus benefiting both the userand cellular operator.

The second access network application is sim-ilar to a femtocell. Today several cellular opera-tors sell a mini-cell tower (looks like a WiFiaccess point) that consumers may buy and installin their homes. Typical users of femtocell arethose that have bad coverage in certain parts oftheir homes, such as basements. These femtocelldevices operate on the same frequencies asthose of cellular operators. However, these fem-tocell devices have several issues. First, due tothe fact that femtocell devices and cellular net-works operate on the same frequency, the quali-ty of the network suffers when these twonetworks interfere with each other. Second, the

coverage of these devices is limited. TV whitespace radio coverage is significantly improveddue to the better propagation characteristics andin addition, there is no interference between thefemtocell and main cell.

A somewhat different issue than the dataoverload or spotty coverage discussed above alsocan be noted with cellular networks. Rural areas(to be more precise, areas with low populationdensity distribution) are known to have poorcoverage. Cellular operators have rights to usetheir spectrum nationwide, however, choose notto deploy their networks in rural areas. The rea-son for this is that a significant part of the costsof a cellular operator is infrastructure costs.These costs cannot be recovered in rural areasdue to lack of sufficient number of subscribers ina given area. With white space spectrum, forexample, being made available for unlicenseduse, cellular operators can use them for back-haul, to connect their cell towers to their back-bone networks, thus reducing labor intensivebackhaul cables installation, and thus providecoverage to more customers in unserved andunderserved areas.

Some design considerations need to be keptin mind in using additional spectrum given thatthe transmission requirements associated withthe additional spectrum could vary significantlyfrom that of the primary cellular spectrum. TakeTVWS as an example. The FCC rules as dis-cussed above put certain restrictions on different

Figure 3. Cellular networks.

Base station(e.g., fixed TVBD)

(e.g., Mode II ormode I TVBD)

Relay stationUnlicensedspectrum

Femtocell

(e.g., Fixed ormode IITVBD)

Hot spots (e.g., Mode I TVBD)

Licensedspectrum

Leasedspectrum

Spectrumaggregation

Primary userdatabase

Spectrummarket

External IP network

CellularIP networkbackbone

Spectrumcoordinator

$



device types. For data offloading between basestations and CPE, base stations would operate infixed mode, and CPE can only operate in modeI. The PSD and strict emission mask require-ment may restrict mode I personal/portabledevices for uplink transmission. Therefore, formode I devices, a class of receiver-only whitespace devices might easily be possible in thenear term, enabling broadcast type or mainlydownlink applications with minimal return chan-nel interactivity over cellular or another returnchannel. However, the economic viability of suchan application remains to be seen. On the otherhand, the backhaul scenario as discussed abovewill have fewer issues.

WIRELESS MEDICAL NETWORKS

In recent years there has been increasing interestin implementing ubiquitous monitoring ofpatients in hospitals for vital signs such as tem-perature, pressure, blood oxygen, and electrocar-diogram (ECG). Normally these vitals aremonitored by on-body sensors that are then con-nected by wires to a bedside monitor. TheMBAN is a promising solution for eliminatingthese wires, thus allowing sensors to reliably andinexpensively collect multiple parameters simul-taneously and relay the monitoring informationwirelessly so that clinicians can respond rapidly[24]. Introduction of MBANs for wireless patientmonitoring is an essential component to improv-ing patient outcomes and lowering healthcarecosts. Through low-cost wireless devices, univer-sal patient monitoring can be extended to mostif not all patients in many hospitals. With suchubiquitous monitoring, changes in a patient’scondition can be recognized at an early stageand appropriate action taken. By getting rid ofwires and their management, the associated risksof infection are reduced using MBANs. Addi-

tionally, MBANs would increase patient comfortand mobility, improve effectiveness of caregivers,and improve quality of medical decision making.Patient mobility is an important factor in speed-ing up patient recovery.

Quality of service is a key requirement forMBANs, and hence the importance of having arelatively clean and less crowded spectrum band.Today, MedRadio and WMTS band are used inmany medical applications, but the bandwidth islimited and cannot meet the growing need [24,25]. The 2.4 GHz industrial, scientific, and medi-cal (ISM) band is not suitable for life-criticalmedical applications due to the interference andcongestion from IT wireless networks in hospi-tals. By having the 2360–2400 band allocated forMBANs on a secondary basis, QoS for theselife-critical monitoring applications can be betterensured. Moreover, the 2360–2400 MHz band isimmediately adjacent to the 2400 MHz band forwhich many devices exist today that could easilybe reused for MBANS, such as IEEE 802.15.4radios. This would lead to low-cost implementa-tions due to economies of scale, and ultimatelylead to wider deployment of MBANs and henceimprovement in patient care.

MBAN communication will be limited totransmission of data (voice is excluded) used formonitoring, diagnosing, or treating patients.MBAN operation is permitted by either health-care professionals or authorized personnel underlicense by rule. It is proposed that the 2360–2400MHz frequency band be classified into two bands:2360–2390 MHz (band I) and 2390–2400 MHz(band II). In the 2360–2390 MHz band, MBANoperation is limited for indoor use only to thosehealthcare facilities that are outside exclusionzones of AMT services. In the 2390–2400 MHzband, MBAN operation is permitted everywhere:all hospitals, in homes, and in mobile ambu-lances. There are a number of mechanisms forMBAN devices to access spectrum on a sec-ondary basis while protecting incumbents andproviding a safe medical implementation. Anunrestricted contention-based protocol such asLBT is proposed for channel access. The maxi-mum emission bandwidth of MBAN devices isproposed to be 5 MHz. The maximum transmitpower is not to exceed the lower of 1 mW and10logB dBm (where B is the 20 dB bandwidth inmegahertz) in the 2360–2390 MHz band and 20mW in the 2390–2400 MHz band. The maximumaggregated duty cycle of an MBAN is not toexceed 25 percent. A geographical protectionzone along with an electronic key (e-key) MBANdevice control mechanism is further used to limitMBAN transmissions. E-key device control isused to ensure that MBAN devices can accessthe 2360–2390 MHz frequency band only whenthey are within the confines of a hospital facilitythat is outside the protection zone of AMT sites.

Figure 4 illustrates both in-hospital and out-of-hospital solutions for using 2360–2390 MHz.Any hospital that plans to use the AMT spec-trum for an MBAN has to register with anMBAN coordinator. The MBAN coordinatordetermines if a registered hospital is within pro-tection zones of AMT sites (with possible coor-dination with primary users). If a hospital isoutside protection zones, the MBAN coordina-

Figure 4. Medical body area networks.

In-hospitalsolution

Out-hospitalsolution

Hospital

Hospital ITnetwork

Bodysensors Controller Controller

Band II

MBAN MBAN

Transit

Accesspoint

E-key toenable band I

Band I+

Band II

MBANcoordinator

Primaryuser

database

WAN

Primary user (ATS)protection zone

Bodysensors



tor will issue an e-key specifically for that hospi-tal to enable MBAN devices within that hospitalto access AMT spectrum. Without a valid e-key,by default MBAN devices can only use the2390–2400 MHz band. The distribution of e-keysto MBAN devices that are connected to the hos-pital IT network can be automatically doneeither through wired or wireless links. MBANdevices must have a means to automatically pre-vent transmissions in the 2360–2390 MHz AMTband when devices go outdoors. Once a sensorin an MBAN loses its connection to its hubdevice, it stops transmission within the2360–2390 MHz AMT spectrum or transitions tothe 2390–2400 MHz band. The 2390–2400 MHzband can be used anywhere without restrictionand hence without an e-key. Simulations haveshown that these technologies would work wellto protect AMT from interference while alsomaintaining the QoS required for the MBANapplications [13, 14].

The IEEE has been working on MBAN stan-dardization. In addition to ongoing activities inIEEE 802.15.6 on BANs, 802.15 Task Group 4jwas started in December 2010 to specificallydevelop standards for MBANs in the 2360–2400MHz band by leveraging the existing IEEE802.15.4 standard.

CONCLUSIONMany milestones, both regulatory and technical,have been reached in opening spectrum formore flexible and efficient use, and this trendwill continue. Cognitive radio technology plays asignificant role in making the best use of scarcespectrum to support fast growing demand forwireless applications, ranging from smart grid,public safety, broadband cellular, to medicalapplications. Standard development organiza-tions (SDOs) have begun to develop standardsto take advantage of the opportunities. However,challenges still remain since CR-enabled net-works have to coexist with primary as well assecondary users and need to mitigate interfer-ence in such a way that they can better supportsuch applications from end to end.

REFERENCES[1] Connecting America: The National Broadband Plan,

http://download.broadband.gov/plan/national-broad-band-plan.pdf.

[2] Unlicensed Operations in the TV Broadcast Bands, Sec-ond Memorandum Opinion and Order, FCC 10-174,Sept. 23, 2010.

[3] Amendment of the Commission’s Rules to Provide Spec-trum for the Operation of Medical Body Area Networks,Notice of Proposed Rulemaking, FCC, ET Docket no. 08-59.

[4] IEEE 802.22 WG on WRANs (Wireless Regional AreaNetworks), http://www.ieee802.org/22/.

[5] ECMA 392: MAC and PHY for Operation in TV WhiteSpace, 1st ed., Dec. 2009.

[6] M. Mueck et al., “ETSI Reconfigurable Radio Systems:Status and Future Directions on Software Defined Radioand Cognitive Radio Standards,” IEEE Commun. Mag.,vol. 48, Sept. 2010, pp. 78–86.

[7] Unlicensed Operation in the TV Broadcast Bands, Sec-ond Report and Order, FCC 08-260, Nov. 14, 2008.

[8] K. Harrison, S. M. Mishra, and A. Sahai, “How Much White-Space Capacity is There?” 2010 IEEE Symp. New Frontiersin Dynamic Spectrum, Singapore, 6–9 Apr. 2010.

[9] B. Scott and M. Calabrese, “Measuring the TV ‘WhiteSpace’ Available for Unlicensed Wireless Broadband,”New America Foundation, Tech. Rep., Jan. 2006.

[10] Digital Dividend: Cognitive Access, http://www.ofcom.org.uk/consult/condocs/cognitive.

[11] SE 43: Cognitive Radio Systems in White Spaces,http://www.cept.org/0B322E6B-375D-4B8F-868B-3F9E5153CF72.W5Doc?frames=no&.

[12] Dract ECC Report 159, “Technical and OperationalRequirements for the Possible Operation of CognitiveRadio Systems in the “White Spaces” of the FrequencyBand 470–790 MHz,” http://www.ero.dk/D9634A59-1F13-40D1-91E9-DAE6468ED66C?frames=no&.

[13] Ex-parte comments of GE Healthcare in Docket 06-135, http://fjallfoss.fcc.gov/ecfs/document/view?id=6519820996.

[14] Reply Comments of Philips Healthcare Systems inDocket 08-59, http://fjallfoss.fcc.gov/ecfs/document/view?id=7020244837.

[15] Reply Comments of GE Healthcare in Docket 08-59,http://fjallfoss.fcc.gov/ecfs/document/view?id=7020244842.

[16] Analysis on Compatibility of Low-Power Active MedicalImplant (LP-AMI) Applications Within the FrequencyRange 2360–3400 MHz, in Particular for the Band2483.5-2500 MHz, with Incumbent Services,http:/ /www.ero.dk/0FFA3C12-E787-4868-9D02-CC9CA6D5F335?frames=no&.

[17] DOE, Communications Requirements of Smart GridTechnologies, report, Oct. 5, 2010

[18] D. J. Leeds, The Smart Grid In 2010: Market Seg-ments, Applications and Industry Players, Gtm Research,July 2009.

[19] T. L. Doumi, “Spectrum Considerations for Public Safe-ty in the United States,” IEEE Commun. Mag., vol. 44,no. 1, Jan. 2006, pp. 30–37.

[20] L. E. Miller, “Wireless Technologies and the SAFECOM SoRfor Public Safety Communications,” NIST report, 2005.

[21] Mobile Broadband: The Benefits Of Additional Spec-trum, FCC, OBI tech. paper no. 6, Oct. 2010

[22] T. Kamakaris, M. M. Buddhikot, and R. Iyer, “A Casefor Coordinated Dynamic Spectrum Access in CellularNetworks,” 1st IEEE Int’l. Symp. New Frontiers inDynamic Spectrum Access Networks, Baltimore, MD,pp. 289–98.

[23] I. F. Akyildiz et al., “A Survey on Spectrum Manage-ment in Cognitive Radio Networks,” IEEE Commun.Mag., vol. 46, no. 4, Apr. 2008, pp. 40–48.

[24] M. Patel and J. Wang, “Applications, Challenges, andProspective in Emerging Body Area Networking Tech-nologies,” IEEE Wireless Commun., vol. 17, no. 1, Feb.2010, pp. 80–88.

[25] B. Zhen et al., “Frequency Band Consideration of SG-MBAN,” IEEE 802.15-MBAN-07-0640-00, Mar. 2007.

BIOGRAPHIESJIANFENG WANG ([email protected]) is a seniormember of research staff at Philips Research North Ameri-ca. He received his Ph.D. in electrical and computer engi-neering from the University of Florida in 2006. His researchinterests include cognitive radio, medical body area net-works, wireless tele-health, M2M, and smart grid commu-nications. His work has led to over 30 publications injournals and conferences. He has been serving as technicaleditor of ECMA 392 and is coexistence group leader ofIEEE 802.22.

MONISHA GHOSH [SM] ([email protected]) is aprincipal member of research staff at Philips Research cur-rently working on cognitive and cooperative radio net-works. She received her Ph.D. in electrical engineering in1991 from the University of Southern California. From1991 to 1998 she was a senior member of research staff inthe Video Communications Department at Philips Research.From 1998 to 1999 she was at Lucent Technologies, BellLaboratories, working on wireless cellular systems. Herresearch interests include estimation and information theo-ry, error correction, and digital signal processing for com-munication systems.

KIRAN CHALLAPALI [M] ([email protected]) is aprincipal member of research staff at Philips ResearchNorth America. He has been with Philips since 1990. Hegraduated from Rutgers University with an M.S. degree inelectrical engineering in 1992. He currently leads CogNeA,an industry alliance to bring cognitive radio solutions tomarket. He has published over 25 technical papers in IEEEjournals and conferences, and has about 25 patents, issuedor pending.

Challenges still

remain since

CR-enabled networks

have to coexist with

primary as well as

secondary users and

need to mitigate

interference in such

a way that they can

better support such

applications from

end to end.



The views presented inthis article are those of theauthors and do not neces-sarily reflect the views ofITU, IEEE, ETSI, orECMA. This research wasconducted under a con-tract of R&D for radioresource enhancement,organized by the Ministryof Internal Affairs andCommunications, Japan.

INTRODUCTION

The current radio environment is character-ized by its heterogeneity. Different aspects ofthis heterogeneity include multiple operatorsand services, various radio access technolo-gies, different network topologies, a broadrange of radio equipment, and multiple fre-quency bands.

Such an environment has a lot of technicaland business opportunities. Examples are jointmanagement of several radio access networks(RANs) within one operator to balance the loadof these networks, detecting and using unusedspectrum in the allocated frequency bands with-out interrupting the operation of the primary

users of such frequency bands, and spectrumtrading between several operators.

To exploit such opportunities, the concept ofthe cognitive radio system (CRS) has been devel-oped. In general, the CRS can be characterizedas a radio system having capabilities to obtainknowledge, adjust its operational parametersand protocols, and learn. Given such broadunderstanding of the CRS, many usage scenariosand business cases are possible.

Various CRS deployment scenarios and usecases can be classified using two types of CRS:heterogeneous and spectrum sharing. Examplesof heterogeneous CRSs are management of sev-eral RANs within one operator, and reconfigura-tion of base stations and terminals. An exampleof a spectrum sharing CRS is detecting andusing unused spectrum.

Currently international standardization ofCRS is performed at all levels, including theInternational Telecommunication Union (ITU),IEEE, European Telecommunications StandardsInstitute (ETSI), and European Association forStandardizing Information and CommunicationSystems (ECMA), where each of these organiza-tions is considering multiple CRS deploymentscenarios and business directions. This articledescribes the current concept of the CRS andshows a big picture of CRS standardization toassist both academia and industry in selectingimportant research topics and promising busi-ness directions.

CRS CONCEPT

DEFINITION OF CRSThe first definition of cognitive radio was givenby J. Mitola [1]. He defined cognitive radio asfollows: “The term cognitive radio identifies thepoint at which wireless personal digital assistantsand the related networks are sufficiently compu-tationally intelligent about radio resources andrelated computer-to-computer communicationsto: (a) detect user communications needs as afunction of use context, and (b) to provide radioresources and wireless services most appropriateto those needs.”

Recently the understanding of the technologyhas evolved into the CRS concept. Academiaand industry from different countries and pro-jects have come up with mostly the same under-

ABSTRACT

The current radio environment is character-ized by its heterogeneity. Different aspects ofthis heterogeneity include multiple operatorsand services, various radio access technologies,different network topologies, a broad range ofradio equipment, and multiple frequency bands.Such an environment has a lot of technical andbusiness opportunities. Examples are joint man-agement of several radio access networks withinone operator to balance load of these networks,detecting and using unused spectrum in the allo-cated frequency bands without interrupting theoperation of the primary users of such frequencybands, and spectrum trading between severaloperators. To exploit such opportunities, theconcept of cognitive radio system has been devel-oped. Many CRS usage scenarios and businesscases are possible. This has triggered a lot ofstandardization activity at all levels, including inthe International Telecommunication Union,IEEE, European Telecommunications StandardsInstitute, and European Association for Stan-dardizing Information and Communication Sys-tems; each of these organizations is consideringmultiple CRS deployment scenarios and businessdirections. This article describes the current con-cept of the CRS and shows the big picture ofinternational standardization of the CRS. Under-standing of these standardization activities isvery important for both academia and industryin order to select important research topics andpromising business directions.

COGNITIVE RADIO NETWORKS

Stanislav Filin, Hiroshi Harada, Homare Murakami, and Kentaro Ishizu, National Institute of Information

and Communications Technology

International Standardization ofCognitive Radio Systems

FILIN LAYOUT 2/18/11 3:08 PM Page 82


standing of the CRS concept. The definition ofCRS developed by Working Party (WP) 1B ofthe ITU — Radiocommunication Sector (ITU-R) represents this common understanding.

ITU-R WP 1B defines the CRS as “a radiosystem employing technology that allows the sys-tem: to obtain knowledge of its operational andgeographical environment, established policiesand its internal state; to dynamically andautonomously adjust its operational parametersand protocols according to its obtained knowl-edge in order to achieve predefined objectives;and to learn from the results obtained” [2].

The following key capabilities of the CRS areunderlined in the ITU-R WP 1B definition: thecapability to obtain knowledge, the capability toadjust operational parameters and protocols,and the capability to learn.

CRS CAPABILITIESThe capability to obtain knowledge is one of thethree key characteristics of CRS. Such knowl-edge includes the following key components:• CRS operational radio environment• CRS geographical environment• Internal state of CRS• Established policies• Usage patterns• Users’ needs

The CRS operational radio environment ischaracterized by, for example, the current statusof spectrum usage, indication of the availableradio systems and their assigned frequencybands, coverage areas of these radio systems,and interference levels.

The CRS geographical environment is char-acterized by, for example, positions of radiosthat are components of the CRS and other radiosystems, orientation of antennas of radios of theCRS and other radio systems, and distribution ofusers in the geographic area of the CRS.

The internal state of the CRS can be charac-terized by configuration of the CRS (e.g., fre-quency bands and protocols used by its radios),traffic load distribution, and transmission powervalues.

The established policies may describe fre-quency bands allowed to be used by the CRSunder certain conditions, where such conditionsmay include maximum level of transmissionpower in operating and adjacent frequencybands, rules that the CRS shall follow to avoidcausing harmful interference to other radio sys-tems.

The usage patterns may collect behavior ofthe CRS, other radio systems, and users. Theusers’ needs may be described by user prefer-ences or policies. Examples of such user prefer-ences are requests for high bandwidth, low delay,fast download time, and low cost.

In order to obtain knowledge, CRS can usevarious approaches, including:• Collecting information from component

radio systems• Geolocation• Spectrum sensing• White space database access• Access to a cognitive pilot channel (CPC)

Component radio systems of the CRS typical-ly perform a lot of measurements, such as

received signal power, signal-to-interference-plus-noise ratio, and load. Also, they are awareof their current state, for example, frequencybands and radio access technologies (RATs)used by base stations and terminals, and trans-mission power values. All this information con-tributes a lot to the knowledge of the CRS.

Positions of radios (e.g., base stations andterminals) that are components of the CRS andother radio systems can be obtained using geolo-cation. Geolocation can be performed duringprofessional installation or using a localizationsystem (e.g., Global Positioning System andwireless positioning system).

White space database access and spectrumsensing is very important in some deploymentscenarios of the CRS. These two approaches areused to identify white spaces, detect primaryusers, and identify white spaces, while they mayalso be used to detect secondary users.

Access to a CPC is also very important insome CRS deployment scenarios. The CPCserves as a means to exchange informationbetween components of the CRS, and in suchcases the CPC is typically considered part of theCRS.

The second key characteristic of the CRS isits capability to dynamically and autonomouslyadjust its operational parameters and protocolsaccording to the obtained knowledge in order toachieve some predefined objectives. Such adjust-ment consists of two stages: decision making andreconfiguration.

Typically, the CRS includes an intelligentmanagement system responsible for makingdecisions regarding parameters and protocolsthat need to be adjusted. Following the decisionsmade, the CRS performs reconfiguration of itsradios. Such reconfiguration may include changeof the following parameters: output power, fre-quency band, and RAT.

These three examples are regulated parame-ters;that is, they are typically specified by radioregulations. This is the new capability of theCRS, compared to commonly used adaptationmethods like power control or adaptive modula-tion and coding.

The third key characteristic of the CRS is itscapability to learn from the results of its actionsin order to further improve its performance.

Figure 1 summarizes the CRS concept. Themain components of the CRS are the intelligentmanagement system and reconfigurable radios.The four main actions of the CRS are obtainingknowledge, making decisions, reconfiguration,and learning.

The knowledge used by the CRS includesknowledge about operational radio and geo-graphical environment, internal state, establishedpolicies, usage patterns, and users’ needs. Themethods to obtain this knowledge include get-ting information from component radio systemsof the CRS, geolocation, spectrum sensing,access to the CPC, and database access.

Using the obtained knowledge, the CRSdynamically and autonomously makes reconfigu-ration decisions according to some predefinedobjectives (e.g., in order to improve efficiency ofspectrum usage). Based on the decisions made,the CRS adjusts operational parameters and

White space

database access and

spectrum sensing is

very important in

some deployment

scenarios of the CRS.

These two approach-

es are used to identi-

fy white spaces and

to detect primary

users and to identify

white spaces, while

they may also be

used to detect sec-

ondary users.



protocols of its reconfigurable radios. Suchparameters include output power, frequencyrange, modulation type, and RAT.

Also, the CRS can learn from its decisions toimprove its future decisions. The results oflearning contribute to both obtaining knowledgeand decision making.

TYPES OF CRS

Many scenarios of CRS deployment, CRS usecases, and CRS business cases are possible. Weidentify two key types of the CRS: heteroge-neous type CRS and spectrum sharing type CRS.

In the heterogeneous type CRS, one or sever-al operators operate several radio access net-works (RANs) using the same or differentRATs. Frequency bands allocated to theseRANs are fixed.

These RANs may have different types of basestations. One type of base station is legacy,designed to use a particular RAT to providewireless connection to terminals. Another typeof base station is reconfigurable, having capabili-ty to reconfigure itself to use different frequencybands allocated to the operator and use differentRATs as specified by the radio regulations forthese frequency allocations.

Operators provide services to users havingdifferent terminals. One type of terminal is lega-cy, designed to use a particular RAT. Such a ter-minal can connect to one particular operator orother operators having roaming agreements withthe home operator. Another type of terminal isreconfigurable. Such a terminal has capability toreconfigure itself to use different RATs. Corre-spondingly, such a terminal can hand overbetween different RANs using different RATsoperated by different operators. Optionally, areconfigurable terminal can support multiplesimultaneous links with RANs.

Within the heterogeneous CRS, severaldeployment scenarios are possible. In one sce-nario the CRS has only legacy base stations,while some of the terminals are reconfigurable.Such terminals can make decisions to reconfig-ure themselves to connect to different compo-nent RANs inside the CRS. Also, terminalreconfiguration can be managed or supportedfrom the network side. For this purpose, somemanagement entities are deployed on the net-work side.

Figure 1. CRS concept summary.

- Information from components of CRS- Geo-location- Spectrum sensing- Access to Cognitive Pilot channel

- Radio environment- Geographical environment- Internal state- Established policies- Usage patterns- User’s needs

Distributeddecision making

e.g. SDR

Database access

Learning

Intelligent management system

Reconfigurable radios

Obtainingknowledge

- Output power- Frequency range- Modulation type- Radio access technology- Protocols

Decision andadjustment

Figure 2. Heterogeneous type CRS: cross-device and cross-network handover.

Cross-device handover Switch of terminal during communication Automatic handover according to user’s context

Cross-network handover Switch of network in use during communication Handover according to context (content, speed, price, etc.)

High-speed mobile ring network Scalable mobile core network supporting fast handover

Basic access signaling

Ubiquitous networking server

Corresponding node

Cellular network

Cross-device handover

Wireless LAN

Cross-network handover Wireless LAN

Cross- device handover

Cellular network

Cross-network handover



The heterogeneous CRS is considered, forexample, in the following standards: IEEE1900.4, IEEE draft standard P1900.4.1, andIEEE 802.21.

One example of this deployment scenario isshown in Fig. 2 [3]. In this example a reconfig-urable terminal performs cross-network hand-over and cross-device handover. Anotherexample is shown in Fig. 3 [4], where a reconfig-urable terminal performs cross-operator hand-over. Also, it can support multiple simultaneouslinks with different RANs.

One more deployment scenario of the hetero-geneous CRS is when a mobile wireless routerserves as a bridge between multiple radio sys-tems and terminals as shown in Fig. 4 [5]. Such amobile wireless router has the capability to com-municate with different radio systems using dif-ferent RATs while providing connection toterminals using one RAT. In this case terminalsdo not need to have reconfiguration capability tocommunicate with different radio systems.

All these deployment scenarios of the hetero-geneous CRS are possible within the currentradio regulations.

In the spectrum sharing CRS several RANsusing the same or different RATs can share thesame frequency band. One deployment scenarioof this type of CRS is when several RANs oper-ate in unlicensed or lightly licensed spectrum,where the CRS capabilities can enable coexis-tence of such systems.

Another deployment scenario is when a sec-ondary system operates in the white space of atelevision broadcast operator frequency band. Insuch a scenario, the CRS capabilities should pro-

vide protection of primary service (televisionbroadcast) and coexistence between secondarysystems.

Spectrum sharing CRS is considered, forexample, in the following standards: IEEE1900.4, IEEE draft standard P1900.4a, IEEE

Figure 3. Heterogeneous type CRS: cross-operator multi-link handover.

WLAN private

network 2

WLAN private

network 1

WLAN access point

Cognitive terminal

W-CDMA commercial

network

PHS commercial

network

WLAN private

network 3

Aggregation proxy server

Application server

Supporting NAT/HTTP proxy IP core network

Figure 4. Heterogeneous type CRS: mobile wireless router.

Connected via WLAN

PCs and PDAs supporting WLAN access

Mobile wireless router

Internet connections via WLANs, commercial HSDPA and PHS network

Communication devices (CF type, USB type)

Internet



draft standard P1900.6, IEEE 802.11y, IEEEdraft standard P802.11af, IEEE draft standardP802.19.1, IEEE draft standard P802.22, IEEEdraft standard P802.22.1, and standard ECMA-392.

One example of a spectrum sharing CRS isshown in Fig. 5 [5]. In this example a cognitiveradio base station (CRBS) and cognitive radioterminals (CRTs) sense the spectrum to detecttemporarily vacant frequency bands. Based onthe sensing results , the CRBS and CRTsreconfigure themselves to use these vacant fre-quency bands. Such reconfiguration can besupported by a management entity on the net-work side called the network reconfigurationmanager.

These deployment scenarios of the spectrumsharing CRS are possible within the nationalradio regulations of some countries.

INTERNATIONALSTANDARDIZATION OF CRS

Due to very large interest in the CRS, its stan-dardization is currently performed on all levels,including the ITU, IEEE, ETSI, and ECMA.

In the ITU, ITU-R WPs 1B and 5A arecurrently preparing reports describing theCRS concept and the regulatory measuresrequired to introduce the CRS. In the IEEEseveral Working Groups (WG) in StandardsCoordination Committee (SCC) 41 on Dynam-

ic Spectrum Access Networks and the 802LAN/MAN Standards Committee are stan-dardizing CRSs and their components. InETSI, Technical Committee (TC) on Recon-f igurable Radio Systems (RRS) has beendeveloping reports describing different com-ponents of the CRS, as well as reports on theCRS concept and the regulatory aspects of theCRS. In the ECMA, Task Group 1 of Techni-cal Committee 48 has standardized a CRS forTV white space.

CRS STANDARDIZATION IN THE ITUITU-R WP 1B is currently developing a workingdocument toward draft text on World RadioConference 2012 (WRC 12) agenda item 1.19.Agenda item 1.19 is “to consider regulatorymeasures and their relevance, in order to enablethe introduction of software-defined radio andcognitive radio systems, based on the results ofITU-R studies, in accordance with Resolution956 (WRC 07)” [6].

To prepare the working document, WP 1Bhas developed definitions of the software definedradio (SDR) and CRS [2]. Also, WP 1B hassummarized the technical and operational stud-ies and relevant ITU-R Recommendations relat-ed to the SDR and CRS. WP 1B has consideredthe SDR and CRS usage scenarios in differentradio services. Also, WP 1B has considered therelationship between SDR and CRS. Currently,WP 1B is considering the international radioregulation implications of the SDR and CRS, aswell as, methods to satisfy WRC 12 agenda item1.19.

The ITU-R WP 5A is currently developingthe working document toward a preliminary newdraft report, “Cognitive Radio Systems in theLand Mobile Service” [7]. This report willaddress the definition, description, and applica-tion of cognitive radio systems in the land mobileservice.

The following topics are currently consideredin the working document:• Technical characteristics and capabilities• Potential benefits• Deployment scenarios• Potential applications• Operational techniques• Coexistence• Operational and technical implications

CRS STANDARDIZATION IN IEEEIEEE SCC 41 — IEEE SCC 41 is developingstandards related to dynamic spectrum accessnetworks. The focus is on improved use of spec-trum, including new techniques and methods ofdynamic spectrum access, which requires manag-ing interference and coordination of wirelesstechnologies, and includes network managementand information sharing [8].

The 1900.1 WG developed IEEE 1900.1,“Standard Definitions and Concepts for Dynam-ic Spectrum Access: Terminology Relating toEmerging Wireless Networks, System Function-ality, and Spectrum Management.” This stan-dard creates framework for developing otherstandards within the IEEE SCC 41.

The 1900.4 WG developed IEEE 1900.4,r“Architectural Building Blocks Enabling Net-

Figure 5. Spectrum sharing type CRS.

Cognitive radio (CR) base station

Cognitive radio (CR) terminal

Network reconfiguration manager (NRM)

1. Collection and Analysis

2. Instruction of reconfiguration

3. Notification of operational frequencies



work-Device Distributed Decision Making forOptimized Radio Resource Usage in Heteroge-neous Wireless Access Networks.” IEEE 1900.4defines the architecture of the intelligent man-agement system of a CRS. Both the heteroge-neous and spectrum sharing CRS are supportedby the IEEE standard 1900.4.

Currently, the 1900.4 WG is developing twonew draft standards: P1900.4.1 and P1900.4a.Development of draft standard P1900.4.1, “Inter-faces and Protocols Enabling Distributed Deci-sion Making for Optimized Radio ResourceUsage in Heterogeneous Wireless Networks,”started in March 2009. P1900.4.1 uses IEEE1900.4 as a baseline standard. It provides adetailed description of interfaces and serviceaccess points defined in IEEE 1900.4. Develop-ment of draft standard P1900.4a, “Architectureand Interfaces for Dynamic Spectrum AccessNetworks in White Space Frequency Bands,”started in March 2009 together with P1900.4.1.P1900.4a amends IEEE 1900.4 to enable mobilewireless access service in white space frequencybands without any limitation on the radio inter-face to be used.

The 1900.5 WG is developing draft standardP1900.5, “Policy Language Requirements andSystem Architectures for Dynamic SpectrumAccess Systems.” P1900.5 defines a vendor-inde-pendent set of policy-based control architecturesand corresponding policy language requirementsfor managing the functionality and behavior ofdynamic spectrum access networks.

The 1900.6 WG is developing draft stan-dard P1900.6, “Spectrum Sensing Interfacesand Data Structures for Dynamic SpectrumAccess and other Advanced Radio Communi-cation Systems.” P1900.6 defines the logicalinterface and data structures used for theinformation exchange between spectrum sen-sors and their clients in radio communicationsystems.

On March 8, 2010 the ad hoc on white spaceradio was created within IEEE SCC41. The pur-pose is to consider interest in, feasibility of, andnecessity of developing a standard defining radiointerface (media access control and physical lay-ers) for a white space communication system.

IEEE 802 — IEEE 802 WGs are defining CRSsand components of the CRS [9]. The activity todefine CRSs is currently performed in the 802.22and 802.11 WGs, while the activity to specifycomponents of a CRS is currently performed in802.21, 802.22, and 802.19 WGs.

The draft standard P802.22 is entit led“Draft Standard for Wireless Regional AreaNetworks Part 22: Cognitive Wireless RANMedium Access Control (MAC) and PhysicalLayer (PHY) Specifications: Policies and Pro-cedures for Operation in the TV Bands.” Itspecifies the air interface, including the cogni-tive MAC and PHY, of point-to-multipointwireless regional area networks, comprised ofa professionally installed fixed base stationwith fixed and portable user terminals operat-ing in the unlicensed VHF/UHF TV broadcastbands between 54 MHz and 862 MHz (TVwhite space).

The IEEE standard 802.11y is entitled “IEEE

Standard for Information Technology —Telecommunications and Information Exchangebetween Systems — Local and MetropolitanArea Networks — Specific Requirements —Part 11: Wireless LAN Medium Access Control(MAC) and Physical Layer (PHY) Specifica-tions — Amendment 3: 3650–3700 MHz Opera-tion in USA.” This standard defines themechanisms (e.g., new regulatory classes, trans-mit power control, and dynamic frequency selec-tion) for 802.11 to share frequency bands withother users.

Draft standard P802.11af is entitled “IEEEStandard for Information Technology —Telecommunications and Information ExchangeBetween Systems — Local and MetropolitanArea Networks — Specific Requirements —Part 11: Wireless LAN Medium Access Control(MAC) and Physical Layer (PHY) Specifications— Amendment: TV White Spaces Operation.”It is an amendment that defines standardizedmodifications to both the 802.11 physical layersand MAC layer to meet the legal requirementsfor channel access and coexistence in the TVWhite Space.

IEEE 802.21 is entitled “IEEE Standard forLocal and Metropolitan Area Networks —Part 21: Media Independent Handover Ser-vices.” It defines extensible media-access-inde-pendent mechanisms that enable theoptimization of handover between heteroge-neous IEEE 802 networks, and facilitate hand-over between IEEE 802 networks and cellularnetworks.

Draft standard P802.22.1 is entitled “Stan-dard to Enhance Harmful Interference Protec-tion for Low Power Licensed Devices Operatingin TV Broadcast Bands.” It specifies methodsfor license-exempt devices to provide enhancedprotection to low-powered licensed devices fromharmful interference when they share the samespectrum.

Draft standard P802.19.1 is entitled “IEEEStandard for Information Technology —Telecommunications and Information ExchangeBetween Systems — Local and MetropolitanArea Networks — Specific Requirements —Part 19: TV White Space Coexistence Methods.”It specifies radio-technology-independent meth-ods for coexistence among dissimilar or indepen-dently operated TV band device networks anddissimilar TV band devices.

CRS STANDARDIZATION IN ETSI TC RRSIn ETSI standardization of the CRS is per-formed in the TC RRS [10].

ETSI Technical Report (TR) 102 682,“Functional Architecture for the Managementand Control of Reconfigurable Radio Systems,”was published in July 2009. It provides a feasi-bility study on defining a functional architec-ture for reconfigurable radio systems, in termsof collecting and putting together all manage-ment and control mechanisms targeted atimproving the utilization of spectrum and theavailable radio resources. This denotes thespecification of the major functional entitiesthat manage and direct the operation of areconfigurable radio system, as well as theiroperation and interactions.

Using the obtained

knowledge, the CRS

dynamically and

autonomously makes

reconfiguration

decisions according

to some predefined

objectives (e.g., in

order to improve

efficiency of

spectrum usage).

Based on the

decisions made, the

CRS adjusts

operational

parameters and

protocols of its

reconfigurable

radios.



ETSI TR 102 683, “Cognitive Pilot Channel,”was published in September 2009. It provides afeasibility study on defining and developing theconcept of the CPC for reconfigurable radio sys-tems to support and facilitate end-to-end con-nectivity in a heterogeneous radio accessenvironment where the available technologiesare used in a flexible and dynamic manner intheir spectrum allocation context.

ETSI TR 102 802, “Cognitive Radio SystemConcept,” was published in February 2010. Itformulates the harmonized technical concept forCRSs. Both infrastructure as well as infrastruc-tureless radio networks are covered. Based onthe system concept, the identification of candi-date topics for standardization is the key targetof this study, also including a survey of relatedactivities in other standard development organi-zations.

ETSI TR 102 803, “Potential RegulatoryAspects of Cognitive Radio and SoftwareDefined Radio Systems,” was published inMarch 2010. This report summarizes the studiescarried out by ETSI TC RRS related to the CRSand SDR. In particular, the study results havebeen considered for items of potential relevanceto regulation authorities.

ETSI TC RRS is currently developing adraft TR, “Operation in White Space Frequen-cy Bands.” This draft report will describe howradio networks can operate on a secondarybasis in frequency bands assigned to primaryusers. The following topics are currently con-sidered: operation of the CRS in UHF whitespace frequency bands, methods for protectingprimary users, system requirements, and usecases.

Also, ETSI TC RRS is currently developingdraft technical specification, “Coexistence Archi-tecture for Cognitive Radio Networks on UHFWhite Space Frequency Bands.” This draft spec-ification will define system architecture for spec-trum sharing and coexistence between multiplecognitive radio networks. The coexistence archi-tecture is targeted to support secondary users inUHF white space frequency bands.

CRS STANDARDIZATION IN ECMAIn ECMA, standardization of the CRS is per-formed in Task Group 1 of Technical Commit-tee 48.

Standard ECMA-392, “MAC and PHY forOperation in TV White Space,” was published inDecember 2009 [11]. It specifies MAC and phys-ical layers for personal/portable cognitive wire-less networks operating in TV bands. Also,ECMA-392 specifies a number of incumbentprotection mechanisms that may be used to meetregulatory requirements.

CONCLUSIONSIn general, the CRS can be characterized as aradio system having capabilities to obtain knowl-edge, adjust its operational parameters, and pro-tocols, and learn. Many CRS usage scenariosand business cases are possible.

Currently, international standardization ofCRS is being performed at all levels, includingthe ITU, IEEE, ETSI, and ECMA, where each

of these organizations is considering multipleCRS deployment scenarios and business direc-tions. This article has described the current con-cept of the CRS and has shown the big pictureof international standardization of the CRS. Fig-ure 6 summarizes the international standardiza-tion of the CRS.

REFERENCES[1] J. Mitola and G.Q. Maguire, “Cognitive Radio: Making

Software Radios More Personal,” IEEE Personal Com-mun., vol. 6, no. 4, Aug. 1999, pp. 13–18.

[2] ITU-R SM.2152, “Definitions of Software Defined Radio(SDR) and Cognitive Radio System (CRS),” Sept. 2009.

[3] M. Inoue et al., “Context-Based Network and Applica-tion Management on Seamless Networking Platform,”Wireless Personal Commun., vol. 35, no. 1–2, Oct.2005, pp. 53–70.

[4] H. Harada et al., “A Software Defined Cognitive RadioSystem: Cognitive Wireless Cloud,” IEEE GLOBECOM,Nov. 2007, pp. 249–99.

[5] H. Harada et al., “Research and Development on Het-erogeneous Type and Spectrum Sharing Type CognitiveRadio Systems,” 4th CrownCom, June 2009.

[6] “Working Document towards Draft CPM Text on WRC-12 Agenda Item 1.19, Annex 5 to Document 1B/158,”Feb. 2010.

[7] “Cognitive Radio Systems in the Land Mobile Service,Working Document towards a Preliminary Draft NewReport ITU-R [LMS.CRS], Annex 12 to Document5A/601-E,” Nov. 2010.

[8] IEEE DYSPAN Standards Committee; http://grouper.ieee.org/groups/scc41/.

[9] IEEE 802 LAN/MAN Standards Committee; http://www.ieee802.org/.

[10] M. Mueck et al., “ETSI Reconfigurable Radio Systems:Status and Future Directions on Software Defined Radioand Cognitive Radio Standards,” IEEE Commun. Mag.,vol. 48, no. 9, Sept. 2010, pp. 78–86.

[11] ECMA-392 Std., “MAC and PHY for Operation in TVWhite Space,” Dec. 2009.

BIOGRAPHIESSTANISLAV FILIN [SM] (sfi l [email protected]) is an expertresearcher with the National Institute of Information andCommunications Technology (NICT), Japan. He is currentlyserving as NICT representative to ITU-R WP 5A and WP 1B,ETSI TC RRS, and the IEEE 1900.4 WG. He has been a vot-ing member of IEEE SCC 41 on dynamic spectrum accessnetworks. In IEEE 1900.4 he has been serving as technicaleditor and chair of several subgroups. He was a votingmember of the IEEE 1900.6 WG. He participated in theIEEE 802 EC SG on TV white space. He is a voting memberof IEEE 802.11 and 802.19. He was chair of the IEEE SCC41group on WS radio. In 2009 he received the IEEE SA SBaward for contribution to the development of IEEE 1900.4-2009.

HIROSHI HARADA ([email protected]) is director of the Ubiq-uitous Mobile Communication Group at NICT and is alsodirector of NICT’s Singapore Wireless Communication Labo-ratory. He joined the Communications Research Laboratory,Ministry of Posts and Communications, in 1995 (currentlyNICT). Since 1995 he has researched SDR, cognitive radio,dynamic spectrum access networks, and broadband wire-less access systems on the microwave and millimeter-waveband. He has also joined many standardization committeesand forums in the United States as well as Japan, and hasfulfilled important roles for them, especially IEEE 802.15.3c,IEEE 1900.4, and IEEE1900.6. He currently serves on theBoard of Directors of the SDR Forum and as Chair of IEEESCC41 (IEEE P1900) since 2009 and Vice Chair of IEEEP1900.4 since 2008. He was also Chair of the IEICE Techni-cal Committee on Software Radio (TCSR), 2005–2007, andVice Chair of IEEE SCC41 in 2008. He is involved in manyother activities related to telecommunications. He is a visit-ing professor of the University of Electro-Communications,Tokyo, Japan, and is the author of Simulation and Soft-ware Radio for Mobile Communications (Artech House,2002).

HOMARE MURAKAMI ([email protected]) received his B.E. andM.E. in electronic engineering from Hokkaido University in

ETSI TC RRS is cur-

rently developing

draft technical speci-

fication on “Coexis-

tence Architecture

for Cognitive Radio

Networks on UHF

White Space Fre-

quency Bands.” This

draft specification

will define system

architecture for spec-

trum sharing and

coexistence between

multiple cognitive

radio networks.



1997 and 1999. He has worked in the CommunicationsResearch Laboratory, Ministry of Post and Telecommunica-tions, since 1999, which is now reorganized as NICT. He iscurrently a senior researcher in the Ubiquitous MobileCommunications Group of NICT. He worked at AalborgUniversity from 2003 to 2005 as a visiting researcher. Hisinterest areas are cognitive radio networking, IP mobility,new transport protocols supporting wireless communica-tions, and naming schemes.

KENTARO ISHIZU ([email protected]) received M.E. and Ph.D.degrees from Kyushu University, Japan, in 2003 and 2005,respectively, with a major in computer science. He hasbeen working for NICT since 2002. He has been engaged inR&D projects on heterogeneous wireless networks, dis-tributed content delivery networks, and cognitive wirelessnetworks.

Figure 6. Summary of international standardization on CRS.

ITU

ITU-R

WP 5A

WP IB Definitions of SDR and CRS

Draft CPM text on WRC-12 agenda item 1.19

CRS in the land mobile service

IEEE

SCC41

1900.4

1900.1 1900.1 Definitions and concepts for dynamic spectrum access

1900.4 Architecture for optimized radio resource usagein heterogeneous wireless access networks

1900.5 P1900.5 Policy language requirements and architectures

1900.6 P1900.6 Spectrum sensing interfaces and data structures

Ad-hoc WS radio

P1900.4.1 Interfaces and protocols for optimized radioresource usage in heterogeneous wireless access networks

P1900.4a Architecture and interfaces for dynamic spectrumaccess networks in white space frequency bands

ETSI

TC RRS

ECMA

TC 48

WG1 Potential regulatory aspects of CRS and SDR

CRS concept

WG3 Functional architecture for RRS

Cognitive pilot channel

TGI

- Finished

ECMA-392 MAC and PHY for operation in TV white space

Operation in white space frequency bands

Coexistence architecture for cognitive radio networks onUHF white space frequency bands

802

802.11 802.11y Wireless LAN: 3650-3700 MHz operation in USA

P802.11af Wireless LAN: TV white spaces operation

802.19 P802.19.1 TV white space coexistence methods

802.21 P802.21 Media independent handover services

802.22 P802.22 Wireless RAN: Policies and procedure for operationin the TV bands

P802.22.1 Enhancing harmful interference protection

- Ongoing

Currently internation-

al standardization of

CRS is performed at

all levels, including

ITU, IEEE, ETSI, and

ECMA, where

each of these

organizations

considers multiple

CRS deployment

scenarios and

business directions.


INTRODUCTION

Cognitive radio (CR), in its original meaning, isa wireless communication paradigm utilizing allavailable resources more efficiently with the abil-ity to self-organize, self-plan, and self regulate[1]. In its narrow, however far more popularizeddefinition, CR-based technology aims to combatscarcity in radio spectrum using dynamic spec-trum access (DSA) [2]. DSA technologies arebased on the principle of opportunistically usingavailable spectrum segments in a somewhatintelligent manner.

Implementation and experimentation workhas ramped up in the latter half of the decade.Because of the complexities involved in design-ing and developing CR systems [3, 4], moreemphasis has been placed on the development ofhardware platforms for full experimentation andtesting of CR features. Since 1999, the first timethe term cognitive radio was used in a scientificarticle [1], numerous different platforms andexperimental deployments have been presented.These CR testbeds differ significantly in theirdesign and scope. It is now appropriate to askhow mature these platforms are, what has been

learned from them, and if any trends from theanalysis of functionalities provided by these plat-forms can be identified. This article answersthese questions.

This article has three main sections and con-tributions. First, we present a primer on thecommon systems being used for CR researchand development. The following section focuseson overviews of the key events in recent yearsthat have helped progress the field of CR andDSA technologies. We then present insightsgained from these experiences and look ahead athow the community can grow in the comingyears. We conclude in the final section.

CR IMPLEMENTATION:PLATFORMS AND SYSTEMS

We briefly review the most popular existinghardware and software radio systems, dividingthese platforms into two headings. First, we dealwith reconfigurable software/hardware systems,where the majority of the radio functionality,like modulation/coding/medium access control(MAC) and other layer processing, is performedin software. The burden in terms of processingand functionality on the radio frequency (RF)front-end is intended to be minimal in thesecases. Second, we take a look at composite sys-tems comprising a combination of purely soft-ware and hardware-based signal processingelements (e.g., field-programmable gate arrays[FPGAs]).

RECONFIGURABLESOFTWARE/HARDWARE PLATFORMS

We begin by focusing on three research-orientedsystems: OSSIE, GNU Radio, and Iris.

OSSIE — The Open Source SCA Implementa-tion::Embedded (OSSIE) project is an opensource software package for SDR development[5]. OSSIE was developed at Virginia Tech, andhas become a major Linux-based open sourceSDR software kit, sponsored by the U.S. Nation-


This material is based onwork supported by ScienceFoundation Ireland underGrant no. 03/CE3/I405 aspart of CTVR at the Uni-versity of Dublin, TrinityCollege, Ireland.

ABSTRACT

The year 2009 marked the 10th anniversaryof Mitola and Maguire Jr. introducing the con-cept of cognitive radio. This prompted an out-pouring of research work related to CR,including the publication of more than 30 specialissue scientific journals and more than 60 dedi-cated conferences and workshops. Although thetheoretical research is blooming, with manyinteresting results presented, hardware and sys-tem development for CR is progressing at aslower pace. We provide synopses of the com-monly used platforms and testbeds, examinewhat has been achieved in the last decade ofexperimentation and trials relating to CR, anddraw several perhaps surprising conclusions.This analysis will enable the research communityto focus on the key technologies to enable CR inthe future.


Przemyslaw Pawelczak, University of California, Los Angeles

Keith Nolan and Linda Doyle, University of Dublin, Trinity College

Ser Wah Oh, Institute for Infocomm Research

Danijela Cabric, University of California, Los Angeles

Cognitive Radio: Ten Years ofExperimentation and Development

PAWELCZAK LAYOUT 2/22/11 1:56 PM Page 90


al Science Foundation (NSF) and the Joint Tac-tical Radio System (JTRS), among others.OSSIE implements an open source version ofthe Software Communication Architecture(SCA) development framework supporting SDRdevelopment initiated by the U.S. Departmentof Defense, and it supports multiple hardwareplatforms. Further information is available athttp://ossie.wireless.vt.edu. OSSIE is mostly usedat Virginia Tech.

GNU Radio — Arguably, the software definedradio (SDR) system with the most widespreadusage is the open source GNU Radio project(http://www.gnuradio.org). It supports hard-ware-independent signal processing functionali-ties. Beginning in 2001 as a spin-off of theMassachusetts Institute of Technology’s (MIT’s)PSpectra code originating from the Spec-trumWare project, the GNU Radio softwarewas completely rewritten in 2004. Signal pro-cessing blocks are written in C and C++, whilethe signal flow graphs and visualization toolsare mainly constructed using Python. GNURadio is currently one of the official GNU pro-jects having strong support from the interna-tional development community. A wide rangeof SDR building blocks are available, includingones commonly used to build simple CR-likeapplications (e.g., energy detection). The GNURadio project prompted the development ofthe Universal Software Radio Peripheral(USRP) hardware by Ettus Research LLC,described later.

Iris — Iris is a dynamically reconfigurable soft-ware radio framework developed by the Uni-versity of Dublin, Trinity College. This is ageneral-purpose processor-based rapid proto-typing and deployment system. The basic build-ing block of Iris is a radio component written inC++, which implements one or more stages ofa transceiver chain. Extensible Markup Lan-guage (XML) is used to specify the signal chainconstruction and characteristics. These charac-teristics can be dynamically reconfigured tomeet communications criteria. Iris works inconjunction with virtually any RF hardwarefront-end and on a wide variety of operatingsystems.

A wide range of components have beendesigned for Iris that are focused on CR-like sys-tems. Multiple sensing components ranging fromsimple energy detection to more sophisticatedfilter bank and feature-based detection compo-nents are available. A suite of components fordynamically shaping and sculpting waveforms tomake best use of available white space, or com-ponents that enable frequency rendezvousbetween two systems on frequencies that are notknown a priori, have also been developed. Fordevelopment purposes Iris can also interfacewith Matlab. Iris is predominantly used by thedevelopment group at the University of Dublin,Trinity College.

RF Front-Ends — GNU Radio and Iris aredesigned to carry out the majority of signal pro-cessing in software. However, each systemrequires a minimal hardware RF front-end.

USRP — The most commonly used RF front-end, especially in the research world, is the Uni-versal Software Radio Peripheral (USRP). TheUSRP is an inexpensive RF front-end and acqui-sition board with open design and freely avail-able documentation and schematics. The USRPis highly modular; a range of different RF daugh-terboards for selected frequency ranges may beconnected.

Two types of USRP are available. USRP 1.0contains four high-speed analog-digital convert-ers (ADCs) supporting a maximum of 128Msamples/s at a resolution of 14 bits with 83 dBspurious-free dynamic range, an Altera CycloneFPGA for interpolation, decimation, and signalpath routing, and USB 2.0 for the connectioninterface. USRP 2.0 replaces the Altera FPGAwith a Xilinx Spartan 3-2000 FPGA, gigabitEthernet, and an ADC capable of 400 Msam-ples/s with 16-bit resolution. The reader isdirected to http://wwww.ettus.com for furtherinformation.

Other RF Front-Ends — A limited number ofother RF front-ends are also available for usewith these systems. These include the Scaldioflexible transceiver from IMEC, Belgium(http://www2.imec.be/ be en/research/green-radios/cognitive-radio.html), and the MaynoothAdaptable Radio System from the National Uni-versity of Ireland, Maynooth [6].

COMPOSITE SYSTEMSThe boundary between hardware and softwareframeworks (or platforms) is not as straightfor-ward as might be assumed. The emphasis inreality is on reconfigurability. A number of com-posite platforms exist which have both softwareand hardware components that can be used tofacilitate CR systems. Composite systems differfrom reconfigurable software/hardware plat-forms in that composite systems contain all therequired components (dedicated hardware andsoftware, documentation, ready-made softwarepackages and modules, etc.) that allow for imme-diate CR development.

Iris began life on a general-purpose processorbut has also migrated to an FPGA platform. Onthe FPGA platform, components can be run insoftware on the PowerPC and/or in hardware onthe FPGA logic. The main Iris framework runson the PowerPC with many of the componentsmentioned above in the FPGA logic.

BEE — The Berkeley Emulation Engine (BEE)and its successor BEE2 are two hardware plat-forms developed by the University of Californiaat Berkeley Wireless Research Center. BEE2consists of five Xilinx Vertex-II Pro VP70FPGAs in a single compute module with 500giga-operations/s. These FPGAs can parallelizecomputationally intensive signal processing algo-rithms even for multiple radios. In addition todedicated logic resources, each FPGA embeds aPowerPC 405 core for minimized latency andmaximized data throughput between micropro-cessor and reconfigurable logic. To support pro-tocol development and interfaces between othernetworked devices, the PowerPC on one of theFPGAs runs a modified version of Linux and a

Composite systems

differ from

reconfigurable

software/hardware

platforms in that

composite systems

contain all the

required components

(dedicated hardware

and software, docu-

mentation, ready

made software pack-

ages and modules,

etc.) that allow for

immediate CR

development.



full IP protocol stack. Since FPGAs run at clockrates similar to those of the processor cores, sys-tem memory, and communication subsystems, alldata transfers within the system have tightlybounded latency and are well suited for real-time applications. In order to interface this real-time processing engine with radios and otherhigh-throughput devices, multigigabittransceivers (MGTs) on each FPGA are used toform 10 Gb/s full-duplex links. Eighteen suchinterfaces per BEE2 board are available, allow-ing 18 independent radio connections in an arbi-trary network configuration. The BEE2 withnetwork and Simulink capabilities can be usedfor experimenting with CRs implemented onreconfigurable radio modems and in the pres-ence of legacy users or emulated primary users.Further information is available at http://bee2.eecs.berkeley.edu.

WARP — The Wireless Open-Access ResearchPlatform (WARP) (http://warp.rice.edu) fromRice University, Houston, Texas, is a completehardware and software SDR design. WARPhardware is very similar in approach to theUSRP. A motherboard serves as an acquisitionboard, while daughterboards serve as data col-lection boards. As of December 2009, two ver-sions of motherboards were available. Theversion 2.2 motherboard is connected to a PCvia a gigabit Ethernet interface. Motherboardprocessing is performed by a Xilinx Virtex-IIPro FPGA. Four independent motherboardscan be connected at the same time. ADCs oper-ating at 65 Msamples/s with 14-bit resolutionare available. Software development for WARPis multilayered. It ranges from low-level very-high-speed integrated circuit hardware descrip-tion language (VHDL) coding to Matlabmodeling. Xilinx Matlab extensions for VHDLare available, and the code for WARP is widelyopen. As of December 2009, 21 demo imple-mentations of different wireless functionalitiesusing WARP originated from Rice Universityitself, while 17 are from other institutionsaround the world.

KUAR — The Kansas University Agile Radio(KUAR) was an experimental hardware plat-form intended for the 5.25–5.85 GHz unli-

censed national information infrastructure(UNII) frequency band with a tunable rangeof 30 MHz [7]. It featured a Xilinx Virtex IIPro P30 FPGA with embedded PC for signalprocessing, four independent interfacesbetween the FPGA and embedded PC, andused an ADC with 105 Msamples/s and 14-bitresolution. The KUAR approach allows splitprocessing between the embedded PC plat-form and FPGA. KUAR uses modified GNURadio software to implement its signal pro-cessing features.

Other Platforms — Many other custom SDRplatforms are available that are unique inboth hardware and software design. However,we need to emphasize that these platformssimply provide appropriate hardware and soft-ware for the digital processing required, inte-grated with an RF front-end. Hence, the userof these products does not need to look for astandalone RF front. Some commercial plat-forms such as the Lyrtech solut ions(http://www.lyrtech.com) among others alsoexist but are not considered in this article.The summary of described components, alongwith additional parameters, is presented inTable 1.

OTHER SYSTEMSIn addition to the software-centric and compos-ite systems described in this article, it is impor-tant to note that several standalone componentshave also been developed. The need for spec-trum sensing, an important aspect of CR func-tionality, has been a driver for this developmentwork. Examples include Rockwell Collins,IMEC, and sensing devices from the Institute forInfocomm Research (I2R), Singapore, which isaddressed later in this article.

Finally, there are some well known DSA-focused SDR platforms that are not used direct-ly in CR experimentation at the moment. Themost prominent ones include the JapaneseNational Institute of Information and Commu-nications Technology SDR Platform [6, Sec.3.3], FlexRadio and PowerSDR used mainly foramateur radio work (http://www.flex-radio.com), and SoftRock kits (http://www.dspradio.org).

Table 1. Summary of popular development solutions for CR, see also [6, Table 2].

USRP2 KUAR WARP BEE2

RF bandwidth (MHz) 100 30 40 64

Frequency range (GHz) DC-5 (non continuous) 5.25–5.85 2.4–2.5 (4.9–5.87) 2.39–2.49

Processing architecture FPGA FPGA FPGA FPGA

Connectivity gigabit Ethernet USB/Ethernet gigabit Ethernet Ethernet

No. of antennas 2 2 4 18

ADC performance 400 MS/s, 16 bit 105 MS/s, 14 bit 125 MS/s, 16 bit 64 MS/s, 12 bit

Community support yes no (defunct) yes no

In addition to the

software-centric and

composite systems

described in this

article, it is important

to note that several

stand-alone compo-

nents have also been

developed. The need

for spectrum sens-

ing, an important

aspect of CR func-

tionality, has been a

driver for this devel-

opment work.



BUILDING CR AND DSA SYSTEMS:EXPERIMENTATION AND TRIALS

Following the brief synopses of the key systemsenabling SDR and CR development, we proceedto the second main part of this article. We startwith describing the experimental results of multi-ple platform interactions during recent SDR,CR, and DSA-focused conferences.

OVERVIEW OF IMPORTANT CR EXPERIMENTSConference Demonstrations — In the latterpart of the last decade, some independent con-ference venues featured demonstration sessions.The information relating to these events formsour starting point. We focus mostly on thedemonstrations presented at IEEE DySPANand SDR Forum (now Wireless InnovationForum) conferences, which are the most recog-nized and largest directly related events in thecommunity.

A demonstration track was first establishedin the IEEE DySPAN conference series in 2007.Since that year there have been a total of 22demonstrations. The SDR Forum annual techni-cal symposium, run by the SDR Forum since1996, organized their first demonstration trackin 2007. The demonstrations presented that yearcomprised only SDR platforms and develop-ment kits for engineers. In 2008 real demonstra-tions were presented. In total, 12 demoplatforms were shown, among them three thatwere related to DSA. During the 2009 SDRForum conference event, 10 demonstrationswere presented, among them three related toDSA systems. Important demos presented out-side of these two venues are also included inthis survey. The Association for ComputingMachinery (ACM) MobiCom ’09 included onlyone CR-like demo from RWTH, Aachen Uni-versity, Germany. In 2008 ACM MobiCom fea-tured one CR demo from Microsoft Research,China. ACM SIGCOMM ’09 included onedemonstration from the University of Dublin,Trinity College.

The survey data for this article were collect-ed as follows. From the publicly available dataon each demonstration, we have extracted infor-mation related to the waveforms used, frequen-cy ranges, form of spectrum sensing, transmit orreceive capabilities, control channel usage, typeof application used, sponsoring body, and num-ber of developers. We focused only on actualdemonstrations, ignoring demos that were eitherpresenting development frameworks only, orbased on SDR and reconfigurable platformsthat were not related to CR or DSA systems. Intotal, we have identified 41 relevant demonstra-tions. For detailed information on each demon-stration platform the reader is referred to therespective conference proceedings. The data areas follows:• IEEE DySPAN ’10:

–Wright State University, Army Research,United States: “Spectrally Modulated Spec-trally Encoded Platform”; sponsored byinternal funds–University of Dublin, Trinity College, Ire-land, European Union (EU): “OFDM

Pulse-Shaping for DSA; Multi-CarrierCDMA for DSA”; both sponsored by Sci-ence Foundation Ireland–Institute for Infocomm Research, Singa-pore: “Communication in TV WhiteSpaces”; sponsored by Singapore Agencyfor Science, Technology and Research–IMEC, Belgium, EU: “Wideband Spec-trum Sensor”; sponsored by internal funds–RWTH, Germany, EU: “Policy Engine forHome Networks”; sponsored by GermanResearch Foundation and EU ARAGORNProject; “OFDM Adaptation Based onSpectrum Sensing”; sponsored by GermanResearch Foundation; “DecomposableMAC Framework”; sponsored by GermanResearch Foundation and EU 2PARMAproject–Communications Research Center, Cana-da: “WiFi Network with Spectrum Sens-ing”; sponsored by internal funds–University of Notre Dame, United States:“Primary User Traffic Pattern Detection”;sponsored by U.S. National Science Foun-dation and National Institute of Justice

• SDR Forum ’09:–University of Oulu, Finland, EU: “MobileAd Hoc Network with Opportunistic CRMAC”; sponsored by internal funds–IMEC, Belgium, EU: “Wideband Spec-trum Sensor”; (also IEEE DySPAN ’10),sponsored by internal funds–University of Piraeus, Greece, Alcatel-Lucent, Germany, EU: “Dynamic RadioAccess Technique Re-Configuration”; spon-sored by the EU E2R Project

• ACM MobiCom ’09:–RWTH, Germany, EU: “CR Capacity Esti-mation”; sponsored by German ResearchFoundation and EU ARAGORN project

• ACM SIGCOMM ’09:–University of Dublin, Trinity College, Ire-land, EU: “An FPGA-Based AutonomousAdaptive Radio”; sponsored by ScienceFoundation Ireland

• SDR Forum ‘08:–University of Dublin, Trinity College, Ire-land, EU: “Cyclostationary SignatureEmbedding and Detection” (see IEEE DyS-PAN ‘07); sponsored by Science Founda-tion Ireland–Shared Spectrum Company, United States:“XG Radio”; sponsored by DARPA XGProgram–Virginia Tech, United States: “MultinodeCR Testbed”; sponsorship informationunknown

• ACM MobiCom ’08:–Microsoft Research, China: “WiFi Net-work on TV Bands”; sponsored by internalfunds

• IEEE DySPAN ’08:–TU Delft, University of Twente, Nether-lands: “Non-Continuous OFDM with Spec-trum Sensing”; sponsored by Dutch AAFFreeband Program–Philips Research, United States: “IEEE802.11a with Frequency Adaptation”; spon-sored by internal funds–Adaptrum, United States: “Wireless Micro-

In the latter part of

the last decade,

some independent

conference venues

featured demonstra-

tion sessions. The

information relating

to these events

forms our starting

point. We focus

mostly on the

demonstrations pre-

sented at the IEEE

DySPAN and SDR

Forum (now Wireless

Innovation Forum)

conferences.



phone Detection”; sponsored by internalfunds–University of Dublin, Trinity College, Ire-land, EU: “Cyclostationary SignatureEmbedding and Detection” (also SDRForum ‘08); “Point to Point DSA Link withSpectrum Sensing”; both sponsored by Sci-ence Foundation Ireland–Virginia Tech, United States: “Heteroge-neous Cooperative Multinode DSA net-work”; sponsored by U.S. National Instituteof Justice, National Science Foundation,and DARPA–Institute for Infocomm Research, Singa-pore: “Transmission over TV WhiteSpaces”; sponsored by the Singapore Agen-cy for Science, Technology and Research–Motorola, United States: “WiFi-LikeOperation in TV Bands”; sponsored byinternal funds–Omesh Networks, United States: “ZigBee-Based Self-Configured Network”; spon-sored by internal funds–Rockwell Collins, United States: “Spec-trum Sensor and Signal Classifier”; spon-sored by the DARPA XG Program–Shared Spectrum Company, United States:“XG Radio”; sponsored by the DARPAXG Program–University of South Florida, United States:“Spectrum Sensing with Feature Detec-tion”; sponsored by internal funds–University of Utah, United States: “HighResolution Spectrum Sensing”; sponsoredby internal funds

• IEEE DySPAN ’07:–Shared Spectrum Company, United States:“XG Radio”; sponsored by the DARPAXG Program

–Motorola, United States: “WiFi-Like Net-work in Licensed Bands”; sponsored byinternal funds–Virginia Tech, United States, University ofDublin, Trinity College, Ireland, EU: “Cog-nitive Engine-Based Radio Reconfigura-tion”; sponsored by Science FoundationIreland–University of Dublin, Trinity College, Ire-land, EU: “Cyclostationary SignatureEmbedding and Detection” (also SDRForum ’08); sponsored by Science Founda-tion Ireland–University of Kansas, United States:“KUAR Presentation”; sponsored by theU.S. National Science Foundation,DARPA, and the Department of the Interi-or National Business Center–QinetiQ, United Kingdom: “SpectrumMonitoring Framework”; sponsored byinternal funds–SRI International, United States: “PolicyReasoner Combined with SSC XG Radios”;sponsored by the DARPA XG Program–University of Dublin, Trinity College, Ire-land, EU: “Extensions to XG Policy Lan-guage”; sponsored by Science FoundationIreland–University of Twente, Netherlands, EU:“Spectrum Monitoring Device”; sponsoredby the Dutch Adaptive Ad Hoc Free BandWireless Communications (AAF) Program

IEEE DySPAN ‘07 — During the first ever trialof its kind during IEEE DySPAN ’07, QinetiQ(a U.K. Ministry of Defense contractor) andShared Spectrum Company carried out a simul-taneous transceiver operation test in the UHFband. Data from the evaluation are not publical-ly available as it was considered proprietaryinformation. However, it was found that theShared Spectrum Company’s detect-and-avoidsystem could coexist with a very fast hopping sin-gle-carrier system in the same frequency band.Further information regarding the demonstra-tions is available at http://www.ieee-dyspan.org/2007. A wireless trial licence was issued by theCommission for Communications Regulation(Comreg) in Ireland for multiparty trials in thiscase. Further information is available at http://www.testandtrial.ie.

IEEE DySPAN ‘08 — IEEE DySPAN ’08 fea-tured 13 live demonstrations comprising Tx/Rxand Rx-only systems. A special temporaryauthority license was issued by the FCC forthe 482–500 MHz frequency range, allowingmultiple companies and academic institutionsto occupy and interfere with each other forthe duration of the event. The University ofDublin — Trinity College, Shared SpectrumCompany (using XG nodes), I2R, Universityof Utah, Stevens Institute of Technology,OMESH Networks, Virginia Tech, andMotorola demonstrated DSA transceiver sys-tems. Adaptrum, Philips, the University ofSouth Florida, Anritsu, Rockwell Collins, andTU Delft carried out signal detection andanalysis work using these transmissionsources. This location features several high-

Figure 1. Measurement results example from IEEE DySPAN '08. In the water-fall plot the narrowband signal is a FM transmission and the broadband signalis an XG radio. The waterfall spans approximately 60 seconds of measure-ment.



power analog TV transmitters in the immedi-ate vicinity. The trials demonstrated that DSAsystems and networks could be establishedand maintained even in close proximity tothese high-power TV services and even inChicago’s extremely crowded RF environ-ment. Further information is avai lable athttp://www.ieee-dyspan.org/2008.

Figure 1 is an example waterfall plotobtained using an Anritsu MS2721B handheldanalyzer inside the conference demo room span-ning approximately 1 min. The wideband signalis Shared Spectrum Company’s orthogonal fre-quency-division multiplexing (OFDM) signalfrom the XG nodes. This was operating on a dono harm basis and simply vacated any channelwhere the received signal level from a non-XGsignal exceeded –90 dBm. In one scenario, anarrowband FM signal modulated with a 1 kHzsine wave was swept up and down in the fre-quency band to serve as a potential interferer toXG. It is clearly seen that the XG signal didmove to a vacant channel. This proved thatDSA is possible even in the shadow of extreme-ly powerful adjacent channel TV transmissions.However, this also demonstrated the weaknessof an energy detection do no harm approach.As an example of a simple denial of serviceattack demonstration, it was possible to triggerthe XG signal to change channels as the detec-tion system was energy-threshold-based. Insome cases the XG and narrowband sourceappear on the same frequency. This is becausethe transmitted power of the narrowband inter-ferer was reduced, and did not exceed the XGsystem detection threshold.

IEEE DySPAN ‘10 — IEEE DySPAN ’10 fea-tured 10 demonstrations of DSA systems. Whilesome of the demonstrations possessed the capa-bility to transmit, as was the case with the Uni-versity of Notre Dame and CommunicationsResearch Centre Canada devices, all of themused license-exempt bands only. Two demos, onefrom RWTH, Aachen University, and one fromUniversity of Dublin, Trinity College, demon-strated the capability of non-contiguous OFDMtransmission end effective subcarrier suppressiontechniques, again showing the demonstrationusing the license-exempt channels only.

Key Commercial Experimentation and Trials— This section presents brief overviews of keycommercial trials and experimentation work car-ried out in recent years that have broken newground and helped influence the direction of CRand DSA research.

DARPA XG Experimentation — DARPA XGradio was manufactured by Shared SpectrumCompany in the early 2000s [8]. It is an imple-mentation of a DSA system using interferencedetection and avoidance techniques. A policyengine is used for frequency selection and access.The XG radio uses the IEEE 802.16 physicallayer, with a 1.75 MHz bandwidth OFDM signaland 20 dBm transmit power. All nodes in thenetwork use a common frequency, despite theavailability of more channels at a certain pointof time.

One of the most interesting field trial resultswere presented in [8] by the Defense AdvancedResearch Projects Agency (DARPA) XG pro-gram. The DARPA XG trial was presumablythe first private CR system trial ever. OnAugust 15–17, 2006 the U.S. Department ofDefense’s DARPA demonstrated the capabili-ties of XG radios to work on a CR-like basis.Tests were performed at different locations inVirginia. Six mobile nodes were involved in thedemonstrations, and as the authors claim, ademonstration was successful, proving that theidea of l isten before talk communicationequipped with policy-based reasoning in radioaccess is fully realizable. The system demon-strated very short channel abandon times ofless than 500 ms (i.e., the time during whichthe device ceased communication at a certainchannel and vacated it) and short reestablish-ment times (i.e., less than 200 ms) given thelack of pre-assigned frequencies. The reestab-lishment time is the time taken for the deviceto select a new channel and resume communi-cations.

The channel abandon goal of 500 ms wasmostly met, and problems were mostly due tosoftware and IEEE 802.16 modem glitches.During the experiment U.S. Department ofDefense radios were operating in the 225–600MHz range, and XG radios were selectingunused frequency channels in this range (i.e.,one out of six possible), where the number of allpossible channels to select was an implementa-tion choice.

Experiences from Spectrum Sensing in theTV Bands — The most prominent hardwaretrial for spectrum sensing thus far has been theFCC field trial conducted in 2008 by the Officeof Engineering and Technology (OET). Fivehardware prototypes from Adaptrum, I2R Singa-pore, Microsoft Corporation, Motorola Inc., andPhilips Electronics North America were submit-ted for examination. The tests covered TV sig-nals and Part 74 wireless microphone signals, ina laboratory controlled environment as well asthe actual field. All devices supported sensing ofTV signals, while the I2R, Microsoft, and Philipsdevices also supported wireless microphonesensing.

TV Sensing Laboratory Test: In general, alldevices exhibited good sensitivities (better thanthe –114 dBm threshold established by the FCC[9]) in the laboratory single channel test. ThePhilips device in particular achieved the bestsensitivity in a clean signal environment whilethe Microsoft device had the best performancein captured signal tests. Most devices were ableto maintain good sensitivities when the adjacentchannel power was within manageable levels forthe devices [10, Table 3-1] for adjacent channeltest results. However, the sensitivities were notdetermined in some cases due to insufficientselectivity, receiver desensitization, or devicemalfunction. From the measurable detectionthresholds, the I2R device threshold was betterthan –114 dBm for all cases except for one whenthe N + 1 adjacent signal level is at –28 dBm.The Philips device exhibited the best perfor-mance at low adjacent signal level of –68 dBm.

In some cases, the

XG and narrowband

source appear on

the same frequency.

This is because the

transmitted power of

the narrowband

interferer was

reduced, and did not

exceed the XG

system detection

threshold.


Nevertheless, the future spectrum sensing hard-ware development should tackle the issues oflack of receiver selectivity and receiver desensiti-zation, especially when the adjacent channelshave high powers.

TV Sensing Field Test: Four test conditions(Table 2) were considered by the FCC [10]. Twoof these test conditions involved the white spacedevice (WSD) operating within the service con-tour of a station assigned to the channel. Forcondition I, the broadcast signal was viewable ona representative consumer TV, and for conditionII, the broadcast signal was not viewable on arepresentative consumer TV. For condition II,we note that there is no mechanism to deter-mine whether a TV signal actually exists in themeasurement locations.

All devices, under condition I tests, met theintended probability of detection of over 90 per-cent for ATSC channels. The geolocationdatabase approach from Motorola was able toidentify occupied channels with 100 percentaccuracy. For identification of unoccupied chan-nels, the I2R device exhibited the best perfor-mance, but not with complete reliability.Ironically, the geolocation-database-basedapproach did not exhibit the best performance inthis aspect, presumably due to incomplete infor-mation in the database. This shows that spec-trum sensing alone works to some degree, butthe performance could be further enhancedespecially in the identification of unoccupiedchannels. Combining a geolocation databasewith spectrum sensing may be a better optiondepending on the specific deployment scenarioin mind.

Wireless Microphone Test: The field tests forwireless microphone sensing were performedwith the I2R and Philips devices at two loca-tions. The Philips device reported all of thechannels on which the microphones were des-ignated to transmit as occupied whether themicrophone was transmitting or not. The I2Rdevice indicated several channels as availableeven when the microphones were on. The wire-less microphone field tests at first glance didnot seem to give convincing results in the capa-bility of the submitted WSDs to detect wirelessmicrophone signals reliably. Nevertheless, the

White Space Coalition (WSC) later found outthat the wireless microphone operators wereimproperly transmitting signals on many chan-nels occupied by TV broadcast signals withinthe protected TV service contours during thefield trials [11]. Even so, there is so far nocomprehensive trial that proves the acceptableperformance of wireless microphone signaldetection. As an alternative, the WSC pro-posed to use beacons for protecting wirelessmicrophone signals.

OBSERVATIONS FROMCR PLATFORMS’ INTERACTIONS

We now proceed to the third and final part ofthis article. We focus on the many interestingconclusions that may be drawn from the obser-vation of the development progress of demon-stration platforms for CR-like systems andnetworks presented earlier. Some of these mayseem to be surprising and contradict the com-mon feeling about the way these networks areevolving. We also suggest recommendations tohelp the community evolve faster and advancethe field of research. These are summarizedbelow.

THERE ARE PRACTICALLY NO COMPREHENSIVECR DEMONSTRATION PLATFORMS

Almost all testbeds presented publicly aremore or less focused on DSA functionality.From the surveyed demos, there is not a singleone that presents at least a feature of CR thathas been proposed in [1], like artificial intelli-gence (AI) usage in spectrum selection. Wepresume that the field is not mature enough toprovide meaningful demonstrations with AIfeatures. The more exciting AI functionalitytends to lend itself better to scenarios involvingnetworks, distributed resources, and higher-plane functionality featuring teamwork andcollaboration [12].

We encourage open collaboration betweenresearch groups to help progress toward com-prehensive demonstrations better linked toreal-world scenarios. The IEEE DySPAN


Table 2. Probabilities of proper channel classification.

PrototypeATSC channels NTSC channels

UnoccupiedCondition I Condition II Condition I Condition II

Adaptrum 91% 51% 89% 30% 75%

I2R 94% 30% 25%1 10%1 81%

Motorola (geolocation) 100% 100% 100% 100% 71%

Motorola (sensing) 90% 48% — — 64%

Philips 100% 92% 100% 100% 15%

Note: 1 I2R’s white space device did not support NTSC but was tested by the FCC for NTSC anyway.

We encourage open

collaboration

between research

groups to help

progress toward

comprehensive

demonstrations bet-

ter linked to real-

world scenarios. The

IEEE DySPAN

demonstrations

series provided a

glimpse of what

value could be

generated from

these collaborative

activities.



demonstrations series provided a glimpse ofwhat value could be generated from these col-laborative activities. Further public dissemina-tion of outcomes from these activities in theform of website content and publicly availablevideos would significantly increase the visibili-ty and impact of this work. This in turn wouldincrease the prospects of collaboration andjoint project opportunities with external groupsaround the world.

OPEN SDR PLATFORMS DOMINATE THERESEARCH MARKET

As seen in Fig. 2a, the majority of demonstra-tions use GNU Radio and either the USRP ordedicated RF front-ends. This demonstratesthat open source SDR development kits andopen hardware platforms are proving to be themost accessible university research platform forDSA-related research. On the other hand,other open source software components sup-porting development of CR-like systems, suchas WARP, Iris, and OSSIE, described earlier,are mostly used by the universities that devel-oped them.

Open sourcing is a valuable means of entic-ing new users, supporting a wide range ofdevelopment ecosystems, and increasing theimpact of a research platform. Research insti-tutions are encouraged to explore this option.Additional opportunities in the form ofbespoke development work, greater employ-ment opportunities for the researchersinvolved, and the prospects of a developmentlifetime not restricted by the duration of theproject are potential indirect outcomes fromthis approach.

MANY TESTBEDS ARE NOT DSA IN THESTRICT MEANING OF THE TERM

Surpris ingly , the majority of platformsenabling real-world communication and pre-sented in the past couple of years are designedto work in license-exempt bands, where norequirements on primary user protection arepresent. However, certain issues (e.g., theinterference impact of secondary opportunis-tic usage on primary users, and adjacent chan-nel and dynamic range issues) simply cannotbe analyzed properly unless deployed in a fre-quency band with active real-world incum-bents . In addit ion to these technicalconstraints, market mechanisms and economicdrivers including light licensing and incentiveauction schemes cannot be properly trialed inlicense-exempt bands.

Spectrum regulators can provide wireless testand trial licensing options to help facilitateexperiments in non-license-exempt spectrum thatmore closely meet real-world incumbent scenar-ios. The Commission for Communications Regu-lation (Comreg) in Ireland, the Office ofCommunications (Ofcom) in the United King-dom, and the FCC (through their special tempo-rary authority license mechanism) are examplesof regulators that offer these options. Weencourage research groups to avail of theseopportunities where possible.

OFDM IS TYPICALLY THEDESIGN CHOICE FOR WAVEFORMS

Referring to Fig. 2b, the majority of waveformsused have been OFDM-based (includingDARPA XG). In addition, some prototypes arebased on IEEE 802.11 standards where OFDMis a standard spectrum access scheme. USRP-based testbeds use OFDM to implement non-contiguous forms of this spectrum access scheme,which allows for the dynamic notching and shap-ing of subcarriers to accommodate detectedincumbent frequency user activity. Some otherdemonstrations not using OFDM are available,like recent University of Dublin, Trinity College

Figure 2. Current status of CR demonstration platforms presented in this arti-cle: a) hardware platforms used; b) waveforms used; c) types of signal detec-tion used (OTS: off the shelf, SC: single carrier, SMSE: spectrally modulatedspectrally encoded).

(a)

Dedicated

6

0

Num

ber

8

10

4

2

12

USRP IRIS 802.11 OTS SDR KUAR WARP

(c)

Energy0

Num

ber

5

20

10

15

Other Cyclostationarity Feature

(b)

OFDM

6

0

Num

ber

8

10

4

2

12

802.11 xPSK MC-CDMA SMSE SC 802.15.4 802.16



demonstrations with multicarrier code-divisionmultiple access (MC-CDMA).

Single-carrier (SC) waveform-based researchshould continue. SC schemes can alleviate theneed for highly linear power amplifiers andbackoff as is the case for OFDM, thus helpingreduce the cost of user terminals. Single-carri-er frequency-division multiple access (SC-FDMA) is a variant of OFDM being used forLong Term Evolution (LTE) and LTE-Advanced terminals, the successor to HighSpeed Download Packet Access (HSPDA).The research community can therefore stand topotentially benefit from extending their exist-ing OFDM-based work to target SC-FDMA,carrier aggregation, and other related LTE-based technologies.

ENERGY DETECTION IS THEMOST POPULAR SIGNAL DETECTION METHOD

Energy detection is used by the majority ofthe s y s tems addres sed in th i s a r t i c l e todetect the presence of other users in a bandof in te res t . Energy de tec t ion o f fe r s agreater detection speed and less computa-tional complexity than cyclostationary fea-ture analysis , for example. However, thiscomes a t a cos t . Energy detect ion i s nothighly regarded for accuracy in low signal-to -no i se ra t io ca ses , a s no ted ear l i e r .Among those demos enabling energy detec-tion only, a few enable cooperation in spec-trum sensing. However, it was found duringthe DySPAN demons t ra t ions tha t th i smethod i s subopt imal and easy to abuse.There are many other interesting and morereliable sensing approaches in existence inthe literature, including cyclostationary fea-ture analysis [13, 14] and filter bank tech-n iques [15 ] , wh ich l end themse lves toimplementat ion on a var iety of the p lat -forms mentioned in this article (Fig. 2c).

GEOLOCATION AND SENSING ARE NEEDED FORMAXIMUM RELIABILITY BUT AT A COST

The FCC WSD tests demonstrated that a combi-nation of geolocation and sensing yielded thebest results in condition I and II tests. However,the ability to sense signals down to the estab-lished thresholds may have implications in termsof significantly higher terminal costs than if ageolocation database approach was used on itsown.

Cost i s a major fac tor inf luenc ing themarket adoption of WSD-based technolo-gies. Further real-world trials are requiredto determine whether sensing and the asso-ciated costs can be significantly reduced ifgeo loca t ion -based approaches can beemployed to meet the regulatory guidelines.The outcomes of this work would also helpshape regulatory policy in terms of a stancetha t ba lances the need for pr imary userpro tec t ion , and he lp ing new marke t s toemerge and evolve. These factors would inturn help to increase the market adoptionprospects of new white space-based tech-nologies.

LACK OF APPROPRIATE RF FRONT ENDS

A key bottleneck in CR experimentation hasalways been (and we believe continues to be) theavailability of appropriate frequency-agile RFfront-ends that can easily be coupled with theparts of the CR that carry out the digital pro-cessing — be they pure software systems likeGNU Radio or a mix of hardware and softwarelike the BEE. The USRP has been the most suc-cessful product to do just this, especially in termsof accessibility for researchers (Fig. 2a).

We have approached the stage where out-of-the-laboratory tests are now required to signifi-cantly progress the field of research. The RFfront-end requirements must therefore evolve tosupport this work. Increased transmit power, fre-quency range coverage, smaller form factors,increased support for add-on modules, anincreased range of interfaces, weatherproofhousings, and more adaptable power sourcefacilities are key to facilitating this shift in focus.The research community needs to engage withlarge equipment vendors to demonstrate ideasand prototype solutions to promote developmentof new RF front-ends in sufficient quantities toprovide for larger-scale research and commercialactivities.

SMALL AND CENTRALIZED SYSTEMS ARE THEDESIGN CHOICE FOR MOST OF THE PLATFORMS

Designers have full control over their platformswith a centralized approach. This avoids theneed for a control channel (19 out of 28 sur-veyed platforms focusing on networking had nocontrol channel enabled); however, it means sac-rificing the flexibility of the design. Most demoshave two nodes, some have three, and there area few that might have a few more. Thus, testbedsare small and not of a substantial enough size toreally explore or uncover networking issues.There is much less focus on cognitive networks,and when a network focus is present the scenar-ios typically target single-digit numbers of nodesand centralized scenarios.

The time to increase the scope of the researchvision has now arrived. The research communityis urged to expand their testbed plans to exam-ine larger-scale and distributed multinode sce-narios over wider geographical areas.Collaborative efforts are now beginning to focuson this more, however. Key activities in Europe,for example, include the European ScienceFoundation’s European Cooperation in Scienceand Technology (COST) IC0902 and COST-IC0905 (COST-TERRA) projects, which focuson applying CR across layers, devices, and net-works, and developing a harmonized techno-eco-nomic framework for CR and DSA acrossEurope. Further information on these is avail-able at http://cost-terra.org and http://newyork.ing.uniroma1.it/IC0902.

NO DRAMATIC INCREASE IN THE NUMBER OFAVAILABLE CR AND NETWORK PROTOTYPES

The number of papers presented including cog-nitive radio as a keyword increases exponentiallyevery year. However, every year IEEE DySPAN

Cost is a major factor

in influencing the

market adoption of

WSD-based tech-

nologies. Further

real-world trials are

required to deter-

mine whether sens-

ing and the

associated costs can

be significantly

reduced if geoloca-

tion-based approach-

es can be employed

to meet the regula-

tory guidelines.



has received a similar number of demonstrationsubmissions. IEEE DySPAN ’07, ’08, and ’10received 13, 15, and 12 submissions, respectively.

More industry-led research is now required toincrease the number of prototype systems fromthe small set of systems focused on long-termresearch-only concept ideas.

ONLY ONE THIRD OF THEPRESENTED DEMOS ARE FROM THE UNITED

STATES

Although the United States still dominates inresearch and development of CR-like systems,due to worldwide interest, almost 60 percent ofthe demos are from Canada, the EU, and Asia.

UNIVERSITIES DOMINATE THEDEMONSTRATION MARKET

As an emerging technology, DSA-based systemsare the basis for patent generation and otherintellectual property protection endeavors. Thisis one of the reasons why publicly viewable com-mercial offerings appear to be slow to emerge.On the other hand, university-created proto-types and research publications concerningthese tend to emerge more quickly and involvepublic dissemination of the work through aca-demic publications to help build the researchprofile and status of the research group andacademic institution.

MORE EMPHASIS IS NEEDED ONREPORTING FAILURES

The development path of an emerging tech-nology includes failures as well as successes.In many cases, the reasons why a particularDSA or CR approach was not successful canbe perhaps even more important than thesmall number of scenarios where the systemdoes live up to its claims. While some techni-cal reports focus on problems associated withDSA-related systems like [10], research publi-cations tend not to focus on this valuableinformation. By report ing the reasonsapproaches may not work, the research com-munity can avoid repeating the same mistakesand evolve faster.

EACH DEMONSTRATION WAS DEVELOPED BY ASMALL NUMBER OF PEOPLE

Thanks in part to ready-made SDR systems,available documentation, and, in the case of theUSRP, an active community of developers, thenumber of people involved in demonstrationscan be limited. For the case of surveyed demosfrom the previous section, the average numberof developers is approximately three.

ABSENCE OF IEEE 802.22 DEMONSTRATIONSInterestingly among all presented demonstra-tions, not a single one implemented the IEEE802.22 protocol stack. Although some compo-nents for IEEE 802.22 have already been devel-oped (e.g., the spectrum sensing module of[16]), none of the universities and companieshave focused on these networks. Not only are

demos and testbeds for IEEE 802.22 missing;there is also a lack of literature on WRAN net-works that directly take into account specifica-tions of the standard to evaluate itsperformance [3, 4].

CONCLUSIONSIn this article we have presented a survey ofstate-of-the-art hardware platforms and testbedsrelated to CR concepts. We broke this workdown into three sections. First, we present aprimer on the common systems being used forCR research and development. Synopses of thekey events in recent years that have helpedprogress the field of CR and DSA technologiesfollow this. Finally, we present insights gainedfrom these experiences in an attempt to help thecommunity grow further and faster in the com-ing years.

ACKNOWLEDGMENTSThe authors would like to thank Rahman DoostMohammady and Jörg Lotze for providing initialdata for the demonstration survey.

REFERENCES[1] J. Mitola III and G. Q. Maguire, Jr., “Cognitive Radio:

Making Software Radios More Personal,” IEEE PersonalCommun., vol. 6, no. 4, Aug. 1999, pp. 13–18.

[2] J. Hoffmeyer et al., “Definitions and Concepts forDynamic Spectrum Access: Terminology Relating toEmerging Wireless Networks, System Functionality, andSpectrum Management,” IEEE 1900.1-2008, Oct. 2,2008.

[3] F. Granelli et al., “Standardization and Research in Cog-nitive and Dynamic Spectrum Access Networks: IEEESCC41 Efforts and Other Activities,” IEEE Commun.Mag., vol. 48, no. 1, Jan. 2010, pp. 71–79.

[4] Q. Zhao and B. M. Sadler, “A Survey of Dynamic Spec-trum Access: Signal Processing, Networking, and Regu-latory Policy,” IEEE Signal Process. Mag., vol. 24, no. 3,May 2007, pp. 79–89.

[5] C. R. Aguayo González et al., “Open-Source SCA-BasedCore Framework and Rapid Development Tools EnableSoftware-Defined Radio Education and Research,” IEEECommun. Mag., vol. 47, no. 10, Oct. 2009, pp. 48–55.

[6] R. Farrell, M. Sanchez, and G. Corley, “Software-DefinedRadio Demonstrators: An Example and Future Trends,”Hindawi Int’l. J. Digital Multimedia Broadcasting, 2009.

[7] G. J. Minden et al., “KUAR: A Flexible Software-DefinedRadio Development Platform,” Proc. IEEE DySPAN,Dublin, Ireland, Apr. 17–20, 2007.

[8] M. McHenry et al., “XG Dynamic Spectrum Access FieldTest Results,” IEEE Commun. Mag., vol. 45, no. 6, June2007, pp. 51–57.

[9] FCC, “Second Report and Order and MemorandumOpinion and Order, in the Matter of ET Docket no. 08-260 and ET Docket no. 02-380,” tech. rep., Nov. 14,2008; http://hraunfoss.fcc.gov/edocspublic/attach-match/FCC-08-260A1.pdf.

[10] S. K. Jones et al., “Evaluation of the Performance ofPrototype TV-Band White Space Devices Phase II,” FCCTech. Rep., Oct. 15, 2008; http://hraunfoss.fcc.gov/edocs public/attachmatch/DA-08-2243A3.pdf.

[11] Harris Wiltshire & Grannis LLP, “Ex Parte Filing inResponse to FCC ET Docket Nos. 04-186, 02-380,” Aug.19, 2008, Philips, in the name of the White SpaceCoalition.

[12] K. E. Nolan and L. E. Doyle, “Teamwork and Collabora-tion in Cognitive Wireless Networks,” IEEE WirelessCommun., vol. 14, no. 4, Aug. 2007, pp. 22–27.

[13] A. Tkachenko, D. Cabric, and R. W. Brodersen, “Cyclo-stationary Feature Detector Experiments using Recon-figurable BEE2,” Proc. IEEE DySPAN ‘07, Dublin, Ireland,Apr. 17–20, 2007.

[14] P. D. Sutton, K. E. Nolan, and L. E. Doyle, “Cyclosta-tionary Signatures in Practical Cognitive Radio Applica-tions,” IEEE JSAC, vol. 26, no. 1, Jan. 2008, pp. 13–24.

Interestingly, among

all presented

demonstrations, not

a single one

implemented the

IEEE 802.22 protocol

stack. Although

some components

for IEEE 802.22 have

already been

developed (see the

spectrum sensing

module of [16]),

none of the

universities and

companies have

focused on these

networks.



[15] B. Farhang-Boroujeny and R. Kempter, “Multicarrier CommunicationTechniques for Spectrum Sensing and Communication in CognitiveRadios,” IEEE Commun. Mag., vol. 46, no. 4, Apr. 2008, pp. 80–85.

[16] J. Park et al., “A Fully Integrated UHF-Band CMOS Receiver with Multi-Resolution Spectrum Sensing (MRSS) Functionality for IEEE 802.22 Cog-nitive Radio Applications,” IEEE J. Solid-State Circuits, vol. 44, no. 1,Jan. 2009, pp. 258–68.

BIOGRAPHIESPRZEMYSLAW PAWELCZAK [S’03, M‘10] ([email protected]) received hisM.Sc. degree from Wroclaw University of Technology, Poland, in 2004and his Ph.D. degree from Delft University of Technology, The Nether-lands. From 2004 to 2005 he was a staff member of Siemens COMSoftware Development Center, Wroclaw, Poland. During fall 2007 hewas a visiting scholar at the Connectivity Laboratory, University of Cali-fornia, Berkeley. Since 2009 he has been a postdoctoral researcher atthe Cognitive Reconfigurable Embedded Systems Laboratory, Universityof California, Los Angeles. His research interests include cross-layeranalysis of opportunistic spectrum access networks. He is a Vice-Chairof the IEEE SCC41 Standardization Committee. He was a coordinatorand an organizing committee member of cognitive radio workshops col-located with IEEE ICC in 2007, 2008, and 2009. Since 2010 he has beena co-chair of the demonstration track of IEEE DySPAN. He was therecipient of the annual Telecom Prize for Best Ph.D. Student in Telecom-munications in The Netherlands in 2008 awarded by the Dutch RoyalInstitute of Engineers.

KEITH NOLAN ([email protected]) received his Ph.D. degree in electronicengineering from the University of Dublin, Trinity College, Ireland, in 2005.He is a research fellow with the Telecommunications Research Centre(CTVR) at the University of Dublin, Trinity College. He has served as organiz-er, chair, and co-chair of demonstrations for IEEE DySPAN symposia, andon numerous TPCs for conferences concerning cognitive radio and dynamicspectrum access technologies. He currently serves on the managementcommittee for COST Actions IC0902 and IC0905 (COST-TERRA), and is alsoa technical co-author of the IEEE P1900.1 standard.

LINDA DOYLE ([email protected]) is a member of faculty in the School ofEngineering, University of Dublin, Trinity College. She is currently direc-tor of CTVR, a national research center that is headquartered in TrinityCollege and based in five other universities in Ireland. CTVR carries outindustry-informed research in the area of telecommunications, andfocuses on both wireless and optical communication systems. She isresponsible for the direction of CTVR as well as running a large researchgroup that is part of the center. Her research group focuses on cognitiveradio, reconfigurable networks, spectrum management and telecommu-nications, and digital art.

SER WAH OH [SM] ([email protected]) obtained his B.Eng. from theUniversity of Malaya, Malaysia, in 1996, and Ph.D. and M.B.A. degreesfrom Nanyang Technological University (NTU), Singapore, in 1999 and2010, respectively. He is currently a research scientist and project managerat the Institute for Infocomm Research (I2R), Singapore. He oversees TVwhite space activities in I2R, and is currently looking into application of TVwhite space on the smart grid. In 2008 he successfully led a team ofresearchers to contribute TV white space technologies to the field trial con-ducted by the U.S. FCC, which resulted in subsequent approval of TV whitespace in the United States. He was previously in charge of algorithm devel-opment for 3G WCDMA over a software-defined radio platform. At thesame time, he also serves as technical adviser for Rohde & Schwarz andComSOC Technologies. From 2005 to 2008 he concurrently held the posi-tion of adjunct assistant professor in NTU. Prior to I2R, he was a technicalmanager at STMicroelectronics in charge of teams in the Singapore andBeijing R&D Centers. He was responsible for 3G WCDMA and TD-SCDMAphysical layer development. He is also a recipient of the 2010 Ernst &Young Cash Prize Award as the Top MBA Graduate, the 2009 Institution ofEngineers Singapore Prestigious Engineering Achievement Award, and theIEEE ICT 2001 Paper Award. He has served as Demo Chair, Publicity Chair,and Track Chair, and on the TPCs for various conferences and seminars. Hehas published over 30 papers and several invited papers, and holds fourU.S. patents with several pending.

DANIJELA CABRIC ([email protected]) received a Dipl. Ing. degree from theUniversity of Belgrade, Serbia, in 1998 and an M.Sc. degree in electricalengineering from the University of California, Los Angeles, in 2001. Shereceived her Ph.D. degree in electrical engineering from the University ofCalifornia, Berkeley, in 2007, where she was a member of the BerkeleyWireless Research Center. In 2008 she joined the Faculty of Electrical Engi-neering at the University of California, Los Angeles as an assistant profes-sor. Her key contributions involve the novel radio architecture, signalprocessing, and networking techniques to implement spectrum-sensingfunctionality in cognitive radios. She has written three book chapters andover 25 major journal and conference papers in the fields of wireless com-munications and circuits and embedded systems. She was awarded aSamueli Fellowship in 2008 and an Okawa Foundation research grant in2009.

CALL FOR PARTICIPATION

IEEE TTM is a unique event for industry leaders, academics and decision making government officials who direct R&D activities, plan research programs or manage portfolios of researchactivities.

This is the first ever organized Symposium of future technologies and will cover in a tutorial way a selected set of potentially high impact emerging technologies, their current state of maturity and scenarios for the future.

The Symposium brings world renowned experts to discuss theevolutionary and revolutionary advances in technology landscapes as we look forward to 2020. All the presentations in this Symposium are given by invited World leading experts with excellent opportunity for informal interaction between the attendees and senior business leaders and world-renowned innovators. On-site visits of local companies will be organized. Plenary Topics and Sessions:

� Impact of Technology on Environment� Smart Grid� China’s Path to Technology leadership� Future Directions in Wireless and Future Mobile Services� Future of silicon based microelectronics� Carbon nanostructures and conducting polymers� Energy Harvesting and Storage� E Health and Advances in Biomedical Engineering� Cloud Computing� Internet of Things� Digital Content at Home

Keynote and Invited Speakers:� Nobuhiro Endo, CEO NEC Corporation, Japan� George Arnold, National Coordinator for Smart Grid

Interoperability, NIST, USA� Philipp Zhang, Chief Scientist, Huawei, China� Ghavam Shahidi, IBM Fellow, Director of Silicon

Technology, IBM Watson Research Center, USA� Joe Weinman,Worldwide Lead, Communications, Media,

and Entertainment Ind. Solutions, HP, USA� Peter Hartwell, Senior Scientist, HP Laboratories, USA� Hugh Bradlow, Chief Tech. Officer, Telstra, Australia� Michael Austin, VP, BYD America, USA� Raj Jammy, Dir., Silicon Front End Proc., Sematec, USA� Jia Ma, Chief Scientist, Wuxi SensiNet Institute, China� Roberto Saracco, Dir., Telecom Italia Future Ctr., Italy� Tero Ojanperä, Executive VP, Head of Services and Dev.

Experience, Nokia, Finland� Minoru Etoh,Managing Director of Multimedia

Laboratories, NTT DoCoMo� Pierre Mars, VP Appl. Engineering, CAP XX, Australia

Further Information: http://www.techbeyond2020.ust.hk

YCM2869.indd 1 2/22/11 10:15:49



INTRODUCTIONDynamic spectrum access (DSA) networkingallows unlicensed users/devices (“secondaryusers”) to opportunistically access a licensedspectrum band owned by “primary users” subjectto certain spectrum etiquettes and regulations. Itis expected that DSA will alleviate some of theradio spectrum scarcity problem. Cognitiveradios (CRs) enable spectrum sensing, DSA, anddynamic spectrum management. A CR sensesprimary licensed bands and detects the presenceor absence of primary users in these bands. Thensecondary users either release or occupy theseprimary bands depending on whether or not theprimary users are present in these bands, respec-tively. Details on definitions and regulatoryaspects of CRs can be found in [1].

Practical implementation of a CR faces sever-al challenges in terms of hardware design, soft-ware stack implementation, interfacing the CR

device with policy servers, and so on. Examplesof some these issues are the following:

•Synchronization: When two communicatingCRs decide to move to a new band or channel,they must successfully synchronize with eachother to resume communication. Therefore,implementing protocols for accurate synchro-nization message (e.g., available channel list)exchange, resynchronizing in the new band asquickly as possible to prevent loss of data fromupper layers, data buffering strategies during thesynchronization process, planning for the situa-tion when a new band is not available immedi-ately, and so forth must be considered.

•Hardware delays: Using open source soft-ware (as explained in later sections) may resultin the radio hardware being reset and restartedduring channel switching. This hardware resetprocess configures the medium access control(MAC) layer, and adapts the radio to the poten-tially modified transmission and receptionparameters in the new band. This hardwarereset/restart process causes significant delaysduring channel switching.

In this article we describe SpiderRadio: thesetup and implementation of a software-drivenCR using off-the-shelf IEEE 802.11a/b/g hard-ware supported by the Atheros chipset. The soft-ware abstraction hides the physical (PHY) layerdetails from the upper layers in the modifiednetwork protocol stack, as discussed later.

The software abstraction layer is pro-grammable and allows SpiderRadio to configurethe transmission/reception parameters automati-cally to operate in any unused frequency band inthe allowable spectrum bands. The implicationof this feature is that SpiderRadio can be on sev-eral wireless networks at the same time operat-ing on different frequency bands. It can also beconnected to an infrastructure based wirelessnetwork and an ad hoc network simultaneously.This is in contrast to current radios which canonly be configured to operate statically in anyone frequency band connecting to one network.Some general guiding principles are derivedbased on our experience in SpiderRadio-basedDSA testbed experiments.

ABSTRACT

In this article we present SpiderRadio, a cog-nitive radio prototype for dynamic spectrumaccess networking. SpiderRadio is built usingcommodity IEEE 802.11a/b/g hardware and theopen source MadWiFi driver. This helps us indeveloping and testing our prototype withouthaving to buy and manage several licensed spec-trum bands. We begin with a discussion of thekey research issues and challenges in the practi-cal implementation of a dynamic spectrum accessnetwork. Then the lessons learned from thedevelopment of dynamic spectrum access proto-cols, designing management frame structures,software implementation of the dynamic spec-trum access network protocol stack, and testbedexperimental measurement results are present-ed. Several trade-offs in prototype implementa-tion complexity vs. network performance arealso discussed. We also identify potential securi-ty vulnerabilities in cognitive radio networks,specifically as applied to SpiderRadio, and pointout some defense mechanisms against these vul-nerabilities.


S. Sengupta, John Jay College of Criminal Justice

K. Hong, R. Chandramouli ,and K. P. Subbalakshmi, Stevens Institute of Technology

SpiderRadio: A Cognitive Radio Networkwith Commodity Hardware andOpen Source Software

SENGUPTA LAYOUT 2/18/11 3:13 PM Page 101


RELATED WORK

The majority of current research in CR-enabledDSA focuses on the theoretical aspects [2, 3, ref-erences therein] with relatively fewer attempts tobuild working prototypes. In [4] a software-defined CR prototype is developed that is ableto sense spectrum in the UHF band based onwaveform analysis, but no dynamic channelswitching upon detection of primary devices isexplored. A feature detector design for TVbands with emphasis on the PHY layer is pre-sented in [5]. In [6] a CR network prototype isbuilt based on field programmable gate array(FPGA), and a virtual sensing mechanism isdeveloped. In [7] a CR prototype is built withoff- the-shelf IEEE 802.11 devices for spectrumsensing. Primary incumbent detection based oncounting PHY/cyclic redundancy check (CRC)errors is proposed. A primary device is emulatedby a Rohde & Schwarz sine-wave signal genera-tor with IEEE 802.11 access cards operating assecondary devices. However, dynamic frequencyswitching upon detection of primary devices isnot considered here.

The research in [8, 9] investigates adaptivechannel width in wireless networks by focusingon spectrum assignment algorithms to handlespectrum variation and fragmentation. Limita-tions in most of the above mentioned works arethat they do not comprehensively address themajor DSA requirements of fast physical switch-ing, data loss issues at the time of physicalswitching, synchronization failure and overheadissues, and hidden incumbents challenges. Syn-chronization failure between two secondarydevices upon switching will prove fatal for sec-ondary network data communication as effectivethroughput will drop drastically or, even worse,there may be loss of communication. Thus, inthis work, we discuss the SpiderRadio prototypethat addresses algorithmic and implementationissues for sensing-based dynamic frequencyswitching and communication.

CR PROTOTYPEIMPLEMENTATION CHALLENGES

A CR prototype MAC will have many featuressimilar to any existing standard MAC (e.g.,IEEE 802.11 or IEEE 802.16). However, somedistinguishing requirements for DSA make theimplementation highly challenging.

In DSA, when a CR node is switched on, itmay use an etiquette such as listen before talk byscanning all the channels to find out whether anyincumbent in the interfering zone is using anyparticular channel, and builds a spectrum usagereport of vacant and used channels. Unlike theexisting single-frequency radio devices (whichoperate using only one static frequency), CRnodes need to discover their communicationpeers through extensive channel scanning andbeacon broadcasting [10]. Once a CR nodelocates broadcasts from communicating peers, itthen tunes to that frequency and transmits backin the uplink direction with the radio node iden-tifier. Authentication and connection registra-tions are then done gradually. Due to such an

extensive connection establishment procedure atthe beginning, when the number of candidatefrequency channels is large, the initial neighbordiscovery process is likely to become highly timeconsuming. The available channel list may alsochange randomly due to the random arrival/departure of primary users in these bands.Unless the MAC layers of the communicationCR nodes synchronize in a different band pro-actively within a certain delay threshold, networkconnectivity may be lost. If network connectivityis lost, the nodes must go through the highlytime consuming neighbor discovery processrepeatedly. An efficient and robust synchroniza-tion mechanism is thus crucial.

Another challenge for the nodes is themethod and implementation to exchange the listof currently available channels and the channelto which they will resume communication upondetection of a primary in the current operatingchannel.

Upon a successful channel switch and syn-chronization, the wireless card must reconfigureitself to the new frequency channel; thus, itneeds to stop the data flow from the upper lay-ers. This operation will adversely affect the per-formance at the higher layers, degrading thedata throughput performance unless some reme-dial actions are taken to enhance the DSA MAC.

Note that, despite these challenges, dynamicchannel switching must still be simple with agoal toward fast switching, reduced synchroniza-tion failure, reduced synchronization overhead,and increased effective throughput.

SPIDERRADIO SYSTEM DESIGNAs discussed before, the SpiderRadio prototypeis based on IEEE 802.11a/b/g wireless accesscards built with Atheros chipsets. The buildingblock for the software stack (Fig. 1) is the Mad-wifi driver (http://madwifi-project.org/). Madwificontains three sublayers: the IEEE 802.11 MAClayer, wrapper (an interface to lower layer) ofthe Atheros Hardware Abstraction Layer (HAL),and Atheros Hardware Abstraction Layer (theonly closed source component). For the Spider-Radio , the IEEE 802.11 MAC layer is modifiedto speed up and increase the reliability of chan-nel switching, while the wrapper of AtherosHAL is modified to build special hardwarequeues for the prototype.

MODIFICATIONS TO ATHEROS HAL WRAPPERWe propose and implement two special hard-ware queues that become active whenever anydynamic channel switching action needs to betriggered. The first hardware queue is synchro-nization queue (sync queue), which is used fortransmitting synchronization management framesonly. The synchronization management framesare special-purpose frames and are used for syn-chronization between the communicating nodesat the time of switching. Whenever any of thetwo communicating CR nodes sense the necessi-ty for channel switching (initiator node), it thenenables the sync queue and triggers the channelswitching request management frame from thesync queue with the ongoing data communica-tion. Inside the synchronization management

We propose and

implement two

special hardware

queues that become

active whenever any

dynamic channel

switching action

needs to be

triggered. The first

hardware queues are

the synchronization

queue and data

buffer queue.



frame, we pack the destination channel informa-tion (called the candidate frequency channel(s),i.e., to which the CR nodes desire to switch toupon vacating the current channel).

The second hardware queue, data bufferqueue, is enabled when the communicating CRnodes are physically switching channels and theMAC for both nodes is being configured withthe transmission and reception parameters in thenew frequency band. With data buffer queueenabled, we allocate a local memory for buffer-ing the data temporarily from the upper layer sothat no data from the upper layer will be lostand the switching scheme will not have anyadverse effect. These modifications are imple-mented within the Madwifi driver in such a waythat dynamic channel switching in the PHY/MAC layer is hidden from the upper layers, notaffecting the upper layer functionalities at all,thus creating smooth, seamless switching.

EXTENDED MANAGEMENT FRAME STRUCTUREWe use an extended management frame fordynamic channel switching and synchronizationpurposes between two SpiderRadio nodes. Toexplain the new management frame, we beginwith a discussion of the standard IEEE 802.11MAC frame structure and MAC header [11].

In the IEEE 802.11 MAC header, a 2-bit typefield indicates whether the frame is a control,management, or data frame, while a 4-bit sub-type field indicates different subtypes of framesunder one particular type of a frame. For exam-ple, with type field set for a management frame,there can be 16 different subtypes of manage-ment frames. Ten different subtypes of manage-ment frames are already defined in IEEE 802.11:beacon, probe request, probe response, associationrequest, association response, re-associationrequest, re-association response, disassociation,authentication, and de-authentication frame. Sixmore subtypes for 802.11 MAC managementframes could be defined out of which we use onesubtype of a management frame for channelswitching and synchronization. Under this sub-type, four more extended subtypes (signifyingswitching request, switching response, confirma-tion request, and confirmation response frames)are defined. Identification for these extendedsubtypes is built in the first two bytes of theframe body. The necessity and detailed usage ofthese four different extended managementframes are explained later.

In Fig. 2, the structure of the switchingrequest/response frame is shown. A 2-byte Sub-Type Identification field indicates this as a chan-nel switch request/response frame. In the UnitedStates, there are three non-overlapping channelsin IEEE 802.11g and 13 non-overlapping chan-nels in IEEE 802.11a. A 2-byte destination chan-nel bitmap is enough to create a bitmap for allthese 16 channels. For bitmapping more chan-nels, the 2-byte destination channel bitmap canbe extended. The 8-byte timestamp indicates thetime when this request frame is prepared fortransmission.

Similar to the switching request and responseframe, confirm management frames are auxiliarysynchronization management frames. A nodereceiving a confirmation response packet will

compare the current channel information fromthe confirmation request packet with the chan-nel it is operating on currently. If they are thesame, this node will copy the current channeland confirmation count field to the confirmationresponse frame bit by bit and send it back to thetransmitter.

BITMAP CHANNEL VECTORIn order to address the hidden incumbent prob-lem [12], we embed the candidate channelsinformation inside the channel switching requestmanagement frame, instead of the initiator CRnode attempting to convey a channel switchingrequest using only one frequency channel’sinformation. The number of candidate channelsis updated dynamically by the initiator nodedepending on the feedback received from thereceiving CR node. The reason behind transmit-ting a synchronization message with multiplecandidate frequencies is that even if the receiv-ing CR node encounters a licensed incumbenttransmission (hidden to the initiator CR node),it still has ways to choose other candidate chan-nel(s) and report this incumbent transmission tothe initiator using a similar management framecalled a channel switch response managementframe. With this mechanism, even with the pres-ence of a hidden node incumbent, risk of syn-chronization failure is reduced significantly.Figure 2 shows the proposed channel switchingrequest management frame structure in detail.Recall that there are three non-overlap channelsin 802.11g and 13 non-overlap channels in802.11a which are emulated as primary bands inour testbed experiments. Clearly, with primarydevices dynamically accessing the bands, theavailability of the spectrum bands for SpiderRa-dio nodes changes dynamically. Since we usemultiple candidate frequency channels sent by

Figure 1. Proposed protocol stack for SpiderRadio.

Modified Madwifi

Upper layer (TCP/IP)

Atheros IEEE 802.11 a/b/g wireless interface card

802.11 media access control layer(modified for processing sync frames)

Atheros hardware abstraction layer

Atheros HAL wrapper(modified for sync frame queue

and data buffer queue)



initiator nodes, embedding absolute information(spectrum band frequency) of candidate fre-quency channels would again invoke the chal-lenge of a variable length management frame.This may consume more time to decode theheader information for variable length frames.

In order to solve this issue, we use a bitmapchannel vector for sending candidate channel(s)information. Since we have 16 non-overlappingchannels, we implement the length of thisbitmap channel vector as 2 bytes in the MACpayload, as shown in Fig. 2, thus mapping theavailability of each channel to a single bit. Whena channel is available (candidate), the corre-sponding bit will be set to 1; otherwise, it will beset to 0.

Note that the advantage of using a bitmapchannel vector for transmitting candidate chan-nel information is that a fixed length manage-ment frame can be used even though thechannel availability information is variablelength. The fixed length bitmap channel vectoris sufficiently easy and quick to decode. More-over, with the usage of a bitmap channel vector,the management frame becomes easily scalable.If there are more than 16 non-overlapping chan-nels in any system, we only need to expand theprogrammable bitmap vector field for that sys-tem. Following the destination channel bitmapvector field, the next field signifies the finaltimeout switching time from the initiator’s per-

spective. This field is designed to indicate whenthe initiator node will timeout from the currentsynchronization mechanism (if no synchroniza-tion could be established; i.e., even after multi-ple switching requests, no response frame isreceived from the other communicating CRnode), vacate the current channel, and start theresynchronization attempt through quick prob-ing following the destination (candidate) chan-nel bitmap vector.

DYNAMIC FREQUENCYSWITCHING IMPLEMENTATION

Two SpiderRadio secondary devices communi-cating with each other on a frequency channelmust vacate the channel upon detecting thearrival of a primary device (or for coexistence)on that particular channel and must switch to anew channel to resume communication. Toenable efficient spectrum switching, each nodemaintains the spectrum usage report in a localspectrum usage report database (SURD), whichkeeps track of the bands occupied by the prima-ry user or available open spectrum bands. Whena channel switching event is triggered, the sec-ondary devices have three requirements:• Switch as fast as possible to reduce waste of

time and resume data communicationquickly.

Figure 2. Detailed structure of the switching request/response frame.

Protocolversion

Bits: 2Frame control field:

Type

2

ToDS

1

FromDS

1

Pwrmgt

1

WEP

1

Rsvd

1

Moredata

1

Retry

1

Morefrag

1

SubType

4

Framecontrol

Subtypeidentification

15 14Position of bits:

Correspond channel

Correspond Freq (MHz):

13 12 ... ... ... 1 0

165 161 157 153 149 64 60 56 52 48 44 40 36 11 6 1

5825 5805 5785 5765 5745 5320 5300 5280 5260 5240 5220 5200 5180 2457 2437 1412

Frame body ofswitching

request/responseframe

Bytes: 2

Finaltimeout

swtichingtime

8

Timestamp

8

Destination channel bitmap (for request)Destination channel (for response)

Destination channel bitmap vector (for request frame):

2

Bytes: 2

DurationID

2

Address2

802.11 MAC header

6

Address3

6

Framebody

0-2312

CRC

4

Address4

6

Sequencecontrol

2

Address1

6

The advantage of

using bitmap chan-

nel vector for trans-

mitting candidate

channel information

is that fixed length

management frame

can be used even

though the channel

availability informa-

tion is variable

length. The fixed

length bitmap chan-

nel vector is suffi-

ciently easy and

quick to decode.



• Switch successfully to reduce synchroniza-tion failure so that the nodes do not end upin different channels and lose communica-tion.

• Keep the overhead for synchronization assmall as possible to maximize effective datathroughput.With the above goals in mind, we next discuss

the implementation of three gradually improvingversions of channel switching protocols for Spi-derRadio in increasing order of complexity.Note that each version is more robust and com-plex, and requires more overhead than its prede-cessor.

DYNAMIC FREQUENCY SWITCHING: VERSION 1In version 1 two SpiderRadio nodes communi-cating on a frequency channel, upon detectingprimary user activity on this particular channel,trigger a frequency switching procedure andmove to a new channel. The dynamic frequencyswitching procedure is initiated by one of theSpiderRadio nodes, which transmits a channelswitching request management frame to the othernode for synchronization. It then moves to thenew channel. (The channel switching requestmanagement frame is explained in detailabove.) The node initiating channel switchingrequest management frame is called the Initia-tor SpiderRadio, while the other node is calledthe Receiver SpiderRadio. Receiver SpiderRa-dio, upon receiving the channel switchingrequest management frame, switches to the newchannel indicated in the payload of the man-agement frame and resynchronizes with the Ini-tiator. In Fig. 3a we present the synchronizationprocedure for the channel switching requestmanagement frame.

The advantage of this method is its simplicityand reduced overhead. Moreover, dynamic syn-chronization is possible with initiation of a chan-nel switch request management frame carrying

the new channel information, thereby makingchannel switching quite fast.

However, this method also has its drawbacksin terms of robustness, as follows:

•As synchronization is heavily dependent onthe channel switch request management frame,if this frame is lost, there is a high probability ofsynchronization failure as the initiator would endup being in the new channel, whereas the receiv-er will still be in the old channel, resulting in theloss of communication.

•As initiator initiates the channel switchrequest management frame, the new channelinformation (to which the nodes would move to)is inserted in this frame by the initiator from itslocal SURD. The problem with such a protocolis that the receiver may have a primary deviceoperating in the new frequency channel in itsvicinity, however, the initiator does not have anyinformation about this in its local SURD. As aresult, the initiator would again end up being inthe new channel, whereas the receiver will stillremain in the old channel, thereby resulting inthe loss of communication.

•Another key issue of this method is thedetermination of the initiator node. If both com-municating SpiderRadio nodes detect the arrivalof a primary device and simultaneously initiate achannel switch request management frame withinformation about the new channel from theirlocal SURDs, there is a probability of synchro-nization failure. As both the SpiderRadio nodesnow act as initiators without knowing the statusof the other node, both nodes will move to theirspecified new channels. Unless new channelsselected by both the initiators are the same, syn-chronization failure is bound to happen.

In Fig. 3b we present the experimental syn-chronization failure probability of version 1. Wefind that with the network traffic decreasing, thesynchronization failure probability decreases aswell. However, as observed from Fig. 3b, syn-

Figure 3. a) Dynamic channel switching with switching request management frame; b) synchronization failure probability results for ver-sion 1.

Network traffic congestion (kbytes/s)3000 2000 1000 500 10

2

0

Sync

hron

izat

ion

failu

re p

roba

bilit

y (in

%)

4

6

8

10

12

14

16

18Dynamic switching: version 1

(a) (b)

2. Switching

Initiator

New Channel

Data communication onnew channel

1. Channelswitch request

frameOld channel

Data communication onold channel

Receiver

Initiator Receiver

2. Switching

New channel

Old channel



chronization failure probability in version 1 isvery high, thus making this version a poor choicefor SpiderRadio.

DYNAMIC FREQUENCY SWITCHING: VERSION 2In this method synchronization is no longer

dependent on the channel switching requestmanagement frame only. We introduce anothermanagement frame called the channel switchresponse management frame (discussed in detailpreviously).

The synchronization protocol in version 2 issummarized as follows.Step 1: The initiator sends a channel switching

request management frame carrying infor-mation on new channel(s).

Step 2: Upon successful reception of therequest frame, the responder transmits achannel switching response managementframe carrying information about theagreed new channel back to the initiator asacknowledgment; then the responderswitches to the new channel.

Step 3: The initiator, after receiving theresponse frame, switches to the new channel.As the initiator switches only after receiving

the response management frame, the chance ofsynchronization failure reduces significantly (Fig.4a). Using version 2, we can also solve the prob-lem of both nodes being initiators. If both theSpiderRadio nodes initiate channel switchrequest management frame, the one with earliertimestamp will win. The eight byte timestampfield from the switching request managementframe will let both the nodes decide on the win-ner and other node will automatically follow therole of responder.Even with the enhancement, version 2 faces adrawback in terms of higher channel switchingrequest frame loss probability as shown in Fig.4b. This is because only one channel switchingrequest frame is transmitted for channel switch-ing initiation thus making the switching requestframe highly loss-prone.

DYNAMIC FREQUENCY SWITCHING: VERSION 3

To avoid synchronization failure due to the lossof the channel switching response managementframe, we introduce two more types of manage-ment frames, confirm request management frameand confirm response management frame. In Fig.5a, we present the version 3 implementation ofthe dynamic channel switching protocol.

The synchronization protocol in version 3 issummarized as follows.Step 1: The initiator sends channel switching

request management frame(s) carryinginformation of new channel(s).

Step 2: Upon successful reception of therequest frame, the responder initiates achannel switching response managementframe carrying information about the newchannel back to the initiator as acknowledg-ment; then the responder switches to thenew channel.

Step 3: The initiator, after receiving theresponse frame, will switch to the new chan-nel.

Step 4: The responder will monitor the datacommunication on the new channel.

Step 5: If no data communication is receivedfrom the initiator, the responder will send aconfirm request management frame to theinitiator.

Step 6: If the initiator is on the new channel, itwill send a confirm response frame, andcommunication on the new channel willresume.

Step 7: If the responder could not receive aconfirm response, it will consider that theinitiator is in the old channel, and go backto the old channel and try to repeat theprotocol if it is within the time thresholdpermitted by the primary device standard.In order to make version 3 more robust, we

also configure SpiderRadio so that the initiatorsends multiple channel switching request man-agement frames to reduce the loss probability of

Figure 4. a) Synchronization failure probability with version 2; b) channel switching request frame loss probability with version 2.

Network traffic congestion (kbytes/s)

(a)

3000

0.05

0

Sync

hron

izat

ion

failu

re p

roba

bilit

y (%

)

0.1

0.15

0.2

0.25

0.3

0.35

2000 1000 500 10

Dynamic switching: version 2


(b)

30000

Man

agem

ent

fram

e lo

ss p

roba

bly

(%)

5

10

15

2000 1000 500 10

Dynamic switching: version 2



switching request frames. In our experiment weprogram the initiator to transmit three switchingrequest management frames. The result is pre-sented in Fig. 5b, which shows very low lossprobability for switching request frames, thusmaking version 3 highly robust. From experi-ments, we also compare the synchronization fail-ure probabilities of all three versions. Thesynchronization failure probability (in percent-age) for version 3, even under very high networktraffic congestion (approximately 3000 kbytes/s),turns out to be almost negligible (0.0050 per-cent) as compared to versions 1 and 2.

TESTBED SETUP ANDEXPERIMENTAL RESULTS

For conducting extensive experiments with Spi-derRadio enabled nodes, we built two groups ofSpiderRadio prototypes, one for indoor testingand the other for outdoor testing.

Each node of the indoor group is a standarddesktop PC running the Linux 2.6 operating sys-tem. They were all equipped with Orinoco 802.11a/b/g PCMCIA wireless cards. There is no PCM-CIA slot for the desktop PC, so we use an ENE-CB1410 PCMCIA-to-PCI adapter card forconverting the PCMCIA devices to operate onthe desktop PC.

In the outdoor group SpiderRadio isdeployed on two laptops running the Linux 2.6operating system: a Compaq NC4010 and a DellInspiron 700m. Both were equipped withOrinoco 802.11 a/b/g PCMCIA wireless cards.The TX powers of these wireless devices wereset to 100 mW. Another laptop running Win-dows Vista and equipped with Wi-Spy 2.4xactedas a monitor in the testbed. These Orinocodevices are equipped with Atheros 5212 (802.11a/b/g) chipsets. For our testbed setup, the prima-ry user bands were emulated using the 900 MHz,2.4 GHz, and 5.1 GHz Wi-Fi spectrum bands.The primary user communication was emulated

in two ways: two cordless phones communicatingwith each other using the intercom feature andan Agilent signal generator (e4437b) operatingin the Wi-fi bands. The SpiderRadio node wasconfigured to be the secondary user device forthe experiments. For the purpose of sensing anddetecting the arrival of the primary user, weimplemented the spectrum sensing methodologybased on observed PHY errors, received signalstrengths, and n-moving window strategy as pro-posed in our earlier work [13]. We placed Spi-derRadio nodes at a distance of 5–20 m fromeach other, communicating with TCP datastreams. We carried out experiments under fivedifferent network traffic congestion scenarios: 3Mbytes/s, 2 Mbytes/s, 1 Mbytes/s, 0.5 Mbytes/s,and 10 kbytes/s during day and night times. Notethat since the testbed is located in Hoboken,New Jersey (in close proximity to Manhattan,New York City), the radio interference is signifi-cantly different during day and night. Interfer-ence due to students using the Stevens campuswireless network also varies significantly betweennight and day. In Fig. 6a we present the averagetime to synchronize under all five network trafficcongestion scenarios. For showing the effective-ness of version 3 with three switch request man-agement frames, we compare this with a simplerversion 3 where only one switch request manage-ment frame is transmitted. The comparisonresult is shown in Fig. 6a. The first observationfrom this plot is that when the network trafficcongestion decreases, the average time to syn-chronize also decreases for both mechanisms.However, the more interesting observation isthat with higher network traffic, version 3 withthree management frames performs much betterthan the older version 3 (i.e., with one manage-ment frame). The difference in the performancedecreases gradually with a decrease in networktraffic congestion. At the lowest network trafficcongestion (i.e., 10 kbytes/s), version 3 with onemanagement frame results in better performancethan that with three management frames. This is

Figure 5. a) Dynamic frequency switching: version 3; b) channel switching request frame loss probability with version 3.


(b)

3000

0.5

0

Man

agem

ent

fram

e lo

ss p

roba

bilit

y (%

)

1

1.5

2

2.5

3

2000 1000 500 10

Dynamic switching with 3 mgmt frames: version 3

(a)

3. ww

itching

Initiator

New channel

5. If no normal datareceive, send confirm

request frame

6. Confirmresponse frame

4. Directly normal data confirm

Old channel

Data communication onold channel

Responder

Initiator Responder

3. ww

itching

7. If Confirm

fail: Back to oldchannel and repeat protocol

New channel

Old channel1. Channelswitch request

frame

2. Channelswitch

response frame



because with very low network traffic, the lossprobability of a management frame is very low,so the need for redundant management framesto reduce loss probability is no longer needed.Thus, it can be concluded that at night time (orwhen network traffic congestion is very low oralmost zero), version 3 with one managementframe might be a better choice compared tothree management frames to reduce overheadfor the same performance.

The effective throughput is shown in Fig. 6b.The results are shown for different switchingintervals (1, 2, 3, 5, and 10). For benchmarkingpurposes, we calculate the ideal maximumthroughout achievable under the same operatingenvironment and conditions without any fre-quency switching. The dotted line in the figuredepicts the maximum possible throughput (3.353Mbytes/s — benchmark). As evident from thefigure, the proposed CR system demonstrateshigh throughput even with very high frequentswitching; and, as is obvious, with less frequentswitching (switching every 5 or 10 s), thethroughput achieved is almost the same as thebenchmark throughput, proving the effectivenessof the proposed CR prototype.

FUTURE DIRECTIONS:SECURING THE SPIDERRADIO

Recent research in the area of CR security [14]has underlined the need to consider security inthe design stages of the CR network (CRN).Several flavors of denial-of-service (DoS) attackscan be launched on the CRN if the architectureand protocols are not designed specifically toavoid these problems. In keeping with that spirit,we discuss some potential security issues Spider-Radio will need to address and some potentialsolutions to these problems.

Since the primary differentiating factorbetween wireless networks and CRNs is theneed to sense and switch between spectrum

bands, most of the unique security threats toCRNs come from these two functionalities. Wefocus on the vulnerabilities existing in switching/synchronization functionality of CRNs, ratherthan those that occur in sensing [15].

Earlier, several protocols for resynchroniza-tion have been proposed for SpiderRadio. Allthe protocols assume that the channel to whichthe initiator and Receiver SpiderRadios mustswitch (rendezvous), has already been deter-mined. A malicious node with an intent to jamor desynchronize communication between thesenodes can easily do so with very minimalresource expenditure on its part, by tracking thetwo communicating nodes and successively jam-ming these channels. In order to prevent thistype of DoS, the security of the rendezvoussequence must be guaranteed at least to someextent. Several solutions will have to be co-optedto achieve this goal. These include a securepseudo-random rendezvous sequence with mean-ingful convergence guarantees (to ensure fasterchannel synchronization), efficient cryptographicauthentication of the switching request frame aswell as the response management frame (in ver-sion 2 of the protocol), and confirmation frames(version 3). All solutions will have to be opti-mized for time to convergence.

CONCLUSIONSSpiderRadio’s software abstraction-based imple-mentation platform at the MAC layer hides thePHY layer details from the higher layers in anetwork protocol stack efficiently. The special-purpose queues built into the stack help alleviatehigher layer packet losses during dynamic chan-nel switching. The three versions of the pro-posed dynamic channel switching protocols seemto gracefully trade off complexity for achievablethroughput. These protocols also achieve fastchannel switching with negligible synchronizationfailure rate between the transmitter and thereceiver. The empirical throughput observed in

Figure 6. a) Average time to synchronize; b) average effective throughput with various frequency intervals.


(a) (b)

30000

5

0

Ave

rage

tim

e to

syn

chro

nize

(m

s)

10

15

20

25

30

35

2000 1000 500 10

Version 3 with 3 management framesVersion 3 with 1 management frame

Switching intervals (s)

21

2.2

2

Ave

rage

eff

ecti

ve t

hrou

ghpu

t (M

byte

s/s)

2.4

2.6

2.8

3

3.2

3.4

3.6

3.8

4

3 5 10

Effective throughput without switchingEffective throughput with switching



testbed experiments for different channel switch-ing intervals is close to the ideal throughputwithout fixed channel access. This implies thatthe channel switching protocol and implementa-tion in SpiderRadio is fast enough for practicaldynamic spectrum access networking applica-tions.

ACKNOWLEDGMENTThis work is supported by grants from theNational Institute of Justice #2009-92667-NJ-IJ,the National Science Foundation #0916180, andPSC-CUNY Award #60079-40 41.

REFERENCES[1] F. Akyildiz et al., “Next Generation/Dynamic Spectrum

Access/Cognitive Radio Wireless Networks: A Survey,”Comp. Net., vol. 50, no. 13, 2006, pp. 2127–59.

[2] H. Salameh, M. Krunz, and O. Younis, “Mac Protocolfor Opportunistic Cognitive Radio Networks with SoftGuarantees,” IEEE Trans. Mobile Computing, vol. 8, no.10, Oct. 2009, pp. 1339–52.

[3] S.-Y. Tu, K.-C. Chen, and R. Prasad, “Spectrum Sensingof OFDMA Systems for Cognitive Radio Networks,” IEEETrans. Vehic. Tech., vol. 58, no. 7, Sept. 2009, pp.3410–25.

[4] H. Harada, “A Software Defined Cognitive Radio Proto-type,” IEEE 18th Int’l. Symp. Pers., Indoor and MobileRadio Communications (PIMRC), pp. 1-5, 2007.

[5] R. DeGroot et al., “A Cognitive-Enabled ExperimentalSystem,” IEEE DySPAN, 2005, pp. 556–61.

[6] Y. Yuan et al., “KNOWS: Cognitive Radio Networks overWhite Spaces,” IEEE DySPAN, Apr. 2007, pp. 416–27.

[7] K. Shin et al., “An Experimental Approach to SpectrumSensing in Cognitive Radio Networks with off-the ShelfIEEE 802.11 Devices,” 4th IEEE CCNC, Jan. 2007, pp.1154–58.

[8] P. Bahl et al., “White Space Networking with Wi-Fi LikeConnectivity,” Proc. ACM SIGCOMM 2009 Conf. DataCommun., 2009, pp. 27–38.

[9] R. Chandra et al., “A Case for Adapting Channel Widthin Wireless Networks,” Proc. ACM SIGCOMM 2008Conf. Data Commun., 2008, pp. 135–46.

[10] C. Cordeiro et al., “IEEE 802.22: the First WorldwideWireless Standard based on Cognitive Radios,” IEEEDySPAN, 2005, pp. 328–37.

[11] “IEEE Standard for Information Technology —Telecommunications and Information Exchangebetween Systems — Local and Metropolitan area Net-works — Specific Requirements — Part 11: Wireless LanMedium Access Control (MAC) and Physical Layer (PHY)Specifications,” Mar. 2007.

[12] S. Sengupta et al., “Enhancements to Cognitive RadioBased IEEE 802.22 Air-Interface,” IEEE ICC, 2007, pp.5155–60.

[13] K. Hong, S. Sengupta, and R. Chandramouli, “Spider-Radio: An Incumbent Sensing Implementation for Cog-nitive Radio Networking using IEEE 802.11 Devices,”IEEE ICC, 2010.

[14] F. Granelli et al., “Standardization and Research inCognitive and Dynamic Spectrum Access networks: IEEESCC41 Efforts and Open Issues,” IEEE Commun. Mag.,Jan. 2010.

[15] G. Jakimoski and K. Subbalakshmi, “Towards SecureSpectrum Decision,” IEEE Int’l. Conf. Commun., Symp.Sel. Areas Commun., June 2009.

BIOGRAPHIESSHAMIK SENGUPTA [M] ([email protected]) is an assis-tant professor in the Department of Mathematics andComputer Science, John Jay College of Criminal Justice ofthe City University of New York. He received his B.E. degree(first class honors) in computer science from Jadavpur Uni-versity, India, in 2002 and his Ph.D. degree from theSchool of Electrical Engineering and Computer Science,University of Central Florida, Orlando, in 2007. His researchinterests include cognitive radio, dynamic spectrum access,game theory, security in wireless networking, and wirelesssensor networking. Shamik Sengupta serves on the orga-nizing and technical program committee of several IEEEconferences. He is the recipient of an IEEE GLOBECOM2008 best paper award.

KAI HONG ([email protected]) received his B.S. degree inautomatic control from Beijing Institute of Technology in2004 and his M.S. degree in automatic control from Bei-jing Institute of Technology in 2007. He is currently a Ph.D.candidate in computer engineering at Stevens Institute ofTechnology, where he is a member of the Media Security,Networking, and Communications (MSyNC) Laboratory. Hisresearch focus is in the area of Cognitive Radio, dynamicspectrum access networks and wireless security.

R. CHANDRAMOULI [M] ([email protected]) is a professor inthe Electrical and Computer Engineering (ECE) Departmentat Stevens Institute of Technology. His research in wirelessnetworking, cognitive radio networks, wireless security,steganography/ steganalysis, and applied probability isfunded by the NSF, U.S. AFRL, U.S. Army, ONR, and indus-try. He served as an Associated Editor for IEEE Transac-tions on Circuits and Systems for Video Technology(2000–2005). Currently, he is the Founding Chair of theIEEE ComSoc Technical Sub-Committee on Cognitive Net-works, Technical Program Vice Chair of the IEEE ConsumerCommunications and Networking Conference (2007), andChair of the Mobile Multimedia Networking Special Inter-est Group of the IEEE Multimedia Communications Techni-cal Committee.

K. P. (SUBA) SUBBALAKSHMI [M] ([email protected]) is anassociate professor in the ECE Department at Stevens Insti-tute of Technology. Her research interests lie in cognitiveradio networks, wireless security, and information forensicsand security. Her research is supported by grants from theU.S. National Science Foundation, National Institute of Jus-tice, and other Department of Defense agencies. She servesas Chair of the Security Special Interest group of IEEE Com-Soc’s Multimedia Technical Committee. She has given tuto-rials on cognitive radio security at several IEEE conferencesand has served as Guest Editior for several IEEE specialissues in her area of interest.

The three versions of

the proposed

dynamic channel

switching protocols

seem to gracefully

trade-off complexity

for achievable

throughput. These

protocols also

achieve fast channel

switching with negli-

gible synchronization

failure rate between

the transmitter and

the receiver.


110

SERIES EDITORIAL

environments, personalized services/media, virtual sport groups,online gaming, and edutainment.

Scientists and engineers worldwide from industry, researchcenters, and universities are working toward the future mediaInternet architecture. In this special issue, we have striven to givean overview of the industrial and research community viewpointby balancing the article selection to include the highest qualitypapers from both arenas.

We start the special issue with the article entitled “CURLING:Content-Ubiquitous Resolution and Delivery Infrastructure forNext-Generation Services” by Ning Wang et al. Apart from anoverview of the most prominent content-centric networkapproaches, the article highlights some of the most importantchallenges in the future media Internet, such as security, qualityof service, scalability, reliability, and network management, con-sidered from a telecom operator perspective.

Solutions based on content-oriented networks (CONs) areamong the most prominent approaches toward the future mediaInternet. In comparison to IP networking, within a CON, hostsidentification is replaced by content identification and content filename is independent of the content file/segment location. In IPnetworking, a user should know which source server holds the con-tent file of interest (spatial coupling) and communicate with thatserver throughout the content delivery (temporal coupling). Inorder to support this delivery method, search engines return asresults to queries pointers to locations (URL) rather than pointersto the content itself.2 In CONs, the content generation and con-tent consumption are decoupled in time and space, so that contentis delivered based purely on its name (routing by name). More-over, in IP networking a host address is irrelevant to its contentname, which results in phishing and pharming attacks, while inCONs the authenticity of the contents can be easily verified. Thenext article, entitled “A Survey on Content-Oriented Networkingfor Efficient Content Delivery,” by Jaeyoung Choi et al. presents acomprehensive survey on content naming and name-based routing,quantitatively compares CON routing proposals, and evaluates theimpact of the publish/subscribe paradigm and in-network caching.

Most Internet traffic is due to video content; thus, efficientvideo coding and streaming is significant. Contrasting the conven-tional client-server model, in peer-to-peer (P2P) distribution mod-els, video is delivered to the end users not directly from the serverbut in a fully distributed fashion by converting users to contentredistributors. This may result in a major economy in scale, espe-

he Internet has become the most important medium forinformation exchange and the core communication environ-

ment for business relations as well as for social interactions. Everyday millions of people all over the world use the Internet for aplethora of daily activities including searching, information accessand exchange, multimedia communications enjoyment, buying andselling goods, and keeping in touch with family and friends, just toname a few. Statistics show (Fig. 1) that Internet usage hasachieved a penetration of 77.4 percent in North America, 61.3 per-cent in Australia, 58.4 percent in Europe, and an average of 28.7percent of the total worldwide population as of 2010 [1]. This cor-responds to more than a fourfold increase (444 percent actually)over a period of 10 years. If we consider that Asia and Africacount together for more than 70 percent of the world’s populationand have currently the lowest penetration rates with much room togrow as their economies also develop, there is no doubt that manymore people will acquire Internet access over the next 10 years.

Moreover, it is a common belief that besides growing, the Inter-net is evolving toward even richer and more immersive experiences.Advances in video capturing and creation will lead to massive cre-ation of new (user generated) multimedia content and Internetapplications, including 3D videos, immersive environments, net-work gaming, and virtual worlds. Overall Internet traffic is expectedto reach an average of 767 exabytes1 per year in the period2010–2014, four times the amount of data currently circulating inthe Internet [2]. To have an idea of the amount of data this volumerepresents, this is the equivalent of 12 billion DVDs transferredover the Internet every month at the end of the forecast period.

In this respect, future media Internet will not simply be afaster way to go online. The increasing flood of traffic and thenew communication needs will pose many challenges to the net-work infrastructure. Overdimensioning (adding more powerfulrouters, more fiber, etc.) is only a temporary solution. At somepoint in time, structural changes will become necessary. While theextent of the architectural changes can be debated [3], it is notcontested that essential elements of the current Internet infra-structure will need to be, to an extent, redesigned. New methodsof content finding and streaming, diffusion of heterogeneousnodes and devices, new forms of (3D) user-centric/user generatedcontent provisioning, the emergence of software as a service, andinteraction with improved security, trustworthiness, and privacy.In this evolving environment, rich 3D content as well as communi-ty networks (peer-to-peer, overlays, and clouds) are expected togenerate new models of interaction and cooperation, and be ableto support new innovative applications, like virtual collaboration


T

FUTURE MEDIA INTERNET

Theodore Zahariadis Giovanni Pau Gonzalo Camarilo

1 1 exabyte = 1018 bytes = 1000 petabytes = 1 million terabytes

2 One approach to decouple the search engines results from the contentlocation is provided by the EC FP7 project COAST, “Content AwareSearching Retrieval and Streaming” (http://www.coast-fp7.eu)

LYT-GUEST EDIT-Zahariadis 2/22/11 5:21 PM Page 110


cially in cases of highly popular videos and proper selection ofpeer nodes. Furthermore, content needs to be displayed on a vari-ety of devices featuring different sizes, resolutions, computationalcapabilities, and Internet access. If video is encoded in a scalableway, it can be adapted to any required spatio-temporal resolutionand quality in the compressed domain, according to peers’ band-width and end-user context requirements. The next article, “Peer-to-Peer Streaming of Scalable Video in Future InternetApplications” by Toni Zgaljic et al., presents a fully scalable exten-sion (scalable video coding, SVC) of the latest H.264/MPEG-4AVC video coding standard, and describes successful experimentsof streaming SVC encoded videos over P2P networks.

Many researchers envision that the future networked mediaapplications will be multisensory, multi-viewpoint, and multi-streamed, relying on (ultra) high definition and 3D video. Theseapplications will place unprecedented demands on networks forhigh-capacity, low-latency, and low-loss communication paths. Thenext article, “Improving End-to-End QoE via Close Cooperationbetween Applications and ISPs” by Bertrand Mathieu et al., advo-cates the development of intelligent cross-layer techniques that, onone hand, will mobilize network and user resources to provide net-work capacity where it is needed, and, on the other hand, willensure that applications adapt themselves and the content they areconveying to available network resources. Aiming to improve thequality of experience (QoE) and optimize network traffic, the arti-cle presents an architecture based on cooperation between theapplication providers, users, and communications networks.

The high volumes of content create specific trends for efficientinformation mining and content retrieval. In the near future,search engines should not respond to a user query by just findingthe most popular content, but what the user is actually seeking.As such, personalization and contextual issues should be takeninto account. The next article, “System Architecture for EnrichedSemantic Personalized Media Search and Retrieval in the FutureMedia Internet” by María Alduán et al., describes a system archi-tecture that handles, processes, delivers, and finds digital mediaby providing the methods to semantically describe contents with amultilingual-multimedia-multidomain ontology, annotate contentagainst this ontology, process the content, and adapt it to the net-work and network status. The article presents the architecture,and the modules’ functionalities and procedures, including thesystem application model, to the future media Internet concepts.

Finally, in the future media Internet users will request newmethods of communication and interaction, with much better QoE,well beyond today’s communication forms. It is expected that voiceover IP, videoconferencing, IPTV, email, instant messaging, and so

on will be completed by virtual environments and 3D virtualworlds, where friends and colleagues will meet, chat, and interact inmore natural ways. The last (but not least) article of this specialissue, “Automatic Creation of 3D Environments from a SingleSketch Using Content-Centric Networks” by Theodoros Semertzidiset al., describes an innovative core application that provides aninterface where the user sketches in 2D the scene of a virtual net-worked world, and the system exploits dynamically similarity searchand retrieval capabilities of the search-enabled content-centric net-work to fetch 3D models that are similar to the drawn 2D objects.The retrieved 3D models act as the building components for anautomatically constructed 3D scene.

Before we leave you to enjoy this special issue, as guest editorswe would like to thank all authors, who invested a lot of work intheir really valuable contributions, and also all reviewers, whodedicated their precious time in providing numerous commentsand suggestions. Last but not least, we would also like to acknowl-edge the enlightening support of the Editor-in-Chief, Dr. SteveGorshe and the publication staff.

REFERENCES[1] Internet World usage statistics, www.Internetworldstats.com/stats.html[2] CISCO Visual Networking Index Forecast 2010.[3] T. Zahariadis, Ed., “Fundamental Limitations of Current Internet and the

Path to Future Internet,” European Commission Future Internet Archi-tecture (FIArch) Experts Group, 2nd draft, Dec. 2010.

BIOGRAPHIESTHEODORE ZAHARIADIS [M] ([email protected]) received his Ph.D. degreein electrical and computer engineering from the National Technical Univer-sity of Athens, Greece, and his Dipl.-Ing. degree in computer engineeringfrom the University of Patras, Greece. Currently he is associate professor atthe Technological Education Institution of Chalkida, Greece, and chief tech-nical officer at Synelixis Solutions Ltd. From 1997 to 2003, he was withLucent Technologies, first as technical consultant to ACT, Bell Labs, NewJersey, and subsequently as technical manager of Ellemedia Technologies,Athens, Greece, while from 2001 to 2006 he was also chief engineer atHellenic Aerospace Industry. Since 1996 he is involved in various EC fundedprojects and currently chairs the EC Future Media Internet ArchitectureThink Tank (FMIA-TT) and the EC Future Internet Architecture (FIArch)Group. He is a member of the Technical Chamber of Greece and ACM. Hiscurrent research interests are in the fields of broadband wireline/wireless/mobile communications, content-aware networks, and sensor networking.Since 2001 he has been a Technical Editor of IEEE Wireless Communicationsand has served as principal guest editor of many special issues of maga-zines and journals.

GIOVANNI PAU ([email protected]) is a research scientist at the ComputerScience Department of the University of California, Los Angeles. Heobtained a Laurea degree in computer science and a Ph.D. in computerengineering from the University of Bologna. He served as Vice Chair andsecretary of the ComSoc Multimedia Technical Committee and as Vice Chairfor North America. He served as Technical Program Committee Vice-Chairfor IEEE ICC ’06 General Symposium, Technical Program Committee Co-Chair for IEEE International Workshop on Networking Issues in MultimediaEntertainment (NIME ’06), and Technical Program Committee Co-Chair forNIME ’04) in conjunction with IEEE GLOBECOM, He is Associate Editor-in-Chief and co-founder of IEEE Communications Society Multimedia and aSteering Committee member for IFIP MedHocNet Mediterranean Ad HocNetworking Workshop since 2002. He serves as Associate Editor of the Else-vier International Journal of Ad Hoc Networks and Springer InternationalJournal on Peer-to-Peer Systems. His current research interests includemobility, wireless multimedia, peer-to-peer and multimedia entertainment,vehicular networks, and future Internet archtectures.

GONZALO CAMARILLO ([email protected]) works for EricssonResearch in Finland. He received M.Sc. degrees in electrical engineeringfrom the Stockholm Royal Institute of Technology, Sweden, and Universi-dad Politecnica de Madrid, Spain. His research interests include signaling,multimedia applications, transport protocols, and networking architectures.He has authored a number of RFCs, books, patents, and scientific paperson these areas. He has co-authored, among other standards, the SessionInitiation Protocol specification (RFC 3261). He has served on the InternetArchitecture Board and has chaired a number of Internet Engineering TaskForce (IETF) working groups. Currently, he is director of the Real-TimeApplications and Infrastructures area at the IETF. He is also IETF liaisonmanager to the Third Generation Partnership Projects.

SERIES EDITORIAL

Figure 1. World Internet penetration rate, 2010 (source: InternetWorld Stats [1])

0 10% 20% 30% 40% 50% 60% 70% 80% 90%

Africa

World Avg.

Asia

Middle East

Latin Ameria/Caribbean

Oceania/Australia

North America

Penetration Rate

Europe

77.4%

61.3%

10.9%

21.5%

29.8%

58.4%

34.5%

28.7%

LYT-GUEST EDIT-Zahariadis 2/22/11 5:21 PM Page 111

INTRODUCTION

The original Internet model focused mainly onconnecting machines, whereby addresses point tophysical end hosts, and routing protocols com-pute routes to specific destination endpoints.Nowadays the Internet is primarily used fortransporting content/media, where a high volumeof both user-generated and professional digital

content (webpages, movies/songs, live videostreams, etc.) is delivered to users who are usu-ally only interested in the content itself ratherthan the location of the content sources. Humanneeds along with the nature of communicationtechnologies have transformed the Internet intoa new content marketplace, generating revenuefor various stakeholders. In fact, the Internet israpidly becoming a superhighway for massivedigital content dissemination.

In this context, many researchers have advo-cated a transition of the Internet model fromhost-centric to content-centric, with various archi-tectural approaches proposed [1–7]. Many ofthese proposals support the key feature of loca-tion independence, where content consumers donot obtain explicit location information (e.g.,the IP address) of the targeted content source apriori, before issuing the consumption request[1–3, 5, 7]. Nevertheless, location requirementsare sometimes still demanded by both contentconsumers and providers. On one hand, contentproviders may want their content accessed onlyby content consumers from a specific region(known as geo-blocking); for example, BBCiPlayer, Amazon Video-on-Demand, AppleiTunes Store, and Sina video services. On theother hand, content consumers may prefer con-tent originated from specific regions in theInternet; for instance, a U.S.-based shoppermight only like to check the price of an itemsold in Amazon online stores in North Americarather than anywhere else in the world. Today,this is typically achieved through the user’sexplicit input in the URL (e.g., Amazon.com


ABSTRACT

CURLING, a Content-Ubiquitous Resolutionand Delivery Infrastructure for Next GenerationServices, aims to enable a future content-centricInternet that will overcome the current intrinsicconstraints by efficiently diffusing media contentof massive scale. It entails a holistic approach,supporting content manipulation capabilities thatencompass the entire content life cycle, fromcontent publication to content resolution and,finally, to content delivery. CURLING providesto both content providers and customers highflexibility in expressing their location preferenceswhen publishing and requesting content, respec-tively, thanks to the proposed scoping and filter-ing functions. Content manipulation operationscan be driven by a variety of factors, includingbusiness relationships between ISPs, local ISPpolicies, and specific content provider and cus-tomer preferences. Content resolution is alsonatively coupled with optimized content routingtechniques that enable efficient unicast and mul-ticast-based content delivery across the globalInternet.


Wei Koong Chai, University College London

Ning Wang, University of Surrey

Ioannis Psaras and George Pavlou, University College London

Chaojiong Wang, University of Surrey

Gerardo García de Blas and Francisco Javier Ramon Salguero, Telefónica Investigación y Desarrollo

S.A.U.

Lei Liang, University of Surrey

Spiros Spirou, Intracom SA Telecom Solutions

Andrzej Beben, Warsaw University of Technology

Eleftheria Hadjioannou, PrimeTel PLC

CURLING: Content-UbiquitousResolution and Delivery Infrastructurefor Next-Generation Services

WANG2 LAYOUT 2/18/11 3:18 PM Page 112


and Amazon.ca) and supported by name resolu-tion through the standard Domain Name Sys-tem (DNS) [8], with the relevant requestsdirected toward the specific regional web server.Similar practice can be observed in multimedia-based content access (e.g., video-on-demandservices), where consumers have specificrequirements regarding the location/area of con-tent sources.

In this article, we introduce a new Internet-based content manipulation infrastructure —CURLING: Content-Ubiquitous Resolution andDelivery Infrastructure for Next GenerationServices. The objective is to both accurately andefficiently hit (or not hit) content objects in spe-cific regions/areas of the Internet, based on userrequirements and preferences. Such anapproach, deployed by Internet service pro-viders (ISPs), allows both content providers andconsumers to express their location require-ments when publishing/requesting content,thanks to the supported content scoping/filteringfunctions. In particular, instead of following theconventional DNS-like approach, where a con-tent URL is translated into an explicit IPaddress pointing to the targeted content server,the proposed content resolution scheme is basedon hop-by-hop gossip-like communicationbetween dedicated content resolution server(CRS) entities residing in individual ISP net-works. Content resolution operations can bedriven by a variety of factors, including the busi-ness relationships among ISPs (provider/cus-tomer/peer), content consumer preferences andlocal ISP policies. This resolution approach isnatively coupled with content delivery processes(e.g., path setup), supporting both unicast andmulticast functions. Specifically, a content con-sumer simply issues a single content consump-tion request message (capable of carrying his/herlocation preferences on the content source can-didate(s)), and then individual CRS entities col-laboratively resolve the content identifier in therequest, in a hop-by-hop manner, toward thedesired source. Upon receiving the request, theselected content source starts transmitting therequested content to the consumer. During thiscontent resolution operation, multicast-like con-tent states are installed along the resolutionpath so that the content flows back immediatelyupon completion of the resolution process. Byexploiting multicast delivery techniques, weincrease the sustainability of the system in viewof the expected explosion of content in theInternet.

BUSINESS MODEL

We first present a basic business model thatinvolves relevant stakeholders and their businessinteractions. The following top-level roles can beenvisaged:• Content providers (CPs): the entities that

offer content to be accessed and consumedacross the Internet. These include bothcommercial CPs and end users who publishtheir content in the Internet.

• Content consumers (CCs): the entities thatconsume content as receivers.

• ISPs: Equipped with the CURLING con-tent-aware infrastructure, ISPs are respon-sible for dealing with the contentpublication requests from CPs, and contentconsumption requests from consumers, andfor the actual delivery of the content, possi-bly with quality of service (QoS) awareness.Figure 1 shows the business interactions

between individual roles. Since CPs rely on theunderlying content-aware infrastructure ownedby ISPs , they are expected to establish a servicelevel agreement (SLA) involving relevant pay-ment to the ISP for content publication services(CP-ISP SLA). In addition, since ISPs offer con-tent searching/location and delivery services toCCs, a CC-ISP SLA can be established. Some-times, CCs may need to pay CPs for consumingcharged content (e.g., pay-per-view). This can becovered by the CC-CP SLA between the two.Finally, business contracts are also establishedbetween ISPs (ISP-ISP SLA), given a provider-customer or peering relationship between them.A low-tier ISP needs to pay its provider ISP notonly for content traffic delivery, but also for del-egated content publication/resolution services onbehalf of its own customers, including directlyattached CPs and consumers.

THE CURLING ARCHITECTUREOur solution requires a form of aggregatablelabels capable of being sequentially ordered towhich we refer to as content identifiers (IDs). Acontent to be published and accessed is allocat-ed a globally unique content ID. Multiple copiesof the same content that are physically stored atdifferent sites in the Internet share one exclu-sive ID.

Content manipulation operations rely on twodistinct entities in the CURLING architecture:• The content resolution server (CRS), which

handles content publication requests, dis-covers the requested content, and supportscontent delivery

• The content-aware router (CaR), which col-laborates with its local CRS(s) to enforcereceiver-driven content delivery pathsAt least one CRS entity is present in every

domain for handling local publication requestsand content consumption requests, and interact-

Figure 1. Business model.

Internet service provider

CC-CP SLA

CC-ISP SLA CP-ISP SLA

ISP-ISP SLA

Contentconsumer

Contentprovider



ing with other neighboring CRS entities for con-tent publication/resolution across domains. BothCPs and consumers are configured to know theirlocal CRS. The number of CRSs in each domaindepends on performance and resilience consid-erations. Figure 2 depicts the functional view ofthe CURLING architecture. The internal struc-ture of the CRS entity consists of three logicalcomponents. The content management block isresponsible for dealing with requests from bothCPs and CCs (via CRS-CP and CRS-CC inter-faces, respectively), including content ID alloca-tion and entry creation upon new contentregistrations, and also content ID lookup uponeach content consumption request from a CC. Adedicated content record repository is also main-tained, including not only content ID lookupinformation, but also ingress and egress(es)CaR(s) within the local domain for each activecontent session being delivered in the network.The inter-CRS protocol component enables thecommunication between neighboring CRSs forhandling interdomain content publication/con-sumption requests. Finally, the monitoring mod-ule gathers necessary near-real-time informationon content server and underlying network condi-tions for supporting optimized content resolu-tion and delivery configuration operations.

CRSs communicate with other entities viaspecialized interfaces:• Inter-CRS interface enables interaction

among CRSs in neighboring domains, espe-cially when they cooperate in content publi-cation and searching for a requestedcontent across domains.

• CRS-CP interface connects content serversowned by CPs with CRSs, and allows CPsto publish content, optionally with scopingrequirements on potential CCs. This inter-face is also responsible for passing informa-

tion on server load conditions to a CRS forenabling optimized content resolution oper-ations.

• CRS-CC interface connects CC devices withthe CRSs and allows consumers requestingand receiving content with scoping/filteringpreferences on candidate content sources.

• CRS-CaR interface allows a CRS to activelyconfigure relevant CaRs for each contentsession (e.g., content state maintenance). Italso gathers necessary information from theunderlying network that will be used foroptimized content resolution processes.A CaR is the network element that natively

processes content packets according to their IDs.Generally, it is not necessary for every router inthe network to be a CaR, and typically CaRs areplanted at the network boundary as ingress andegress points for content delivery across ISP net-works. The function of CaRs will be specifiedlater with the description of the content deliveryprocess.

HOP-BY-HOP HIERARCHICALCONTENT-BASED OPERATIONS

We envisage the following three-stage contentoperation life cycle: publication, resolution, anddelivery. The task of content resolution is to:• Identify the desired content source in the

Internet according to the requested contentID and optionally CC preferences

• Trigger the content transmission by theselected content server

Once the content server starts the transmissionof the content upon receiving the content con-sumption request, the content delivery functionis responsible for enforcing the actual deliverypath back to the consumer.

Figure 2. High-level architecture of the hop-by-hop hierarchical content resolution approach.

Contentserver

Content provider

Scopingpreference

Consume

Consume

Register

Serverawareness

CC

-CRS interface

CP-C

RS interface

CRS-C

C interface

CRS-C

P interface

Content consumer

QoSrequirements Inter-CRS

protocol

Inter-CRS interface

CRS-CaR interface

Monitoringmodule

Publication

Contentmanagement

CRS Consume Publish

Other CRS

Resolution

Delivery

CaR configuration

CaR CaR

Network

Network awareness

Scoping/filtering

preferences



CONTENT PUBLICATION

Content publication is the process of makingcontent available across the Internet. It consistsof two stages.

Stage 1: Content Registration — It beginswith the CP notifying the local CRS that a newcontent is now available via a Register mes-sage. In the case where multiple copies of thesame content are available at different locations,the CP is responsible for informing the localCRS(s) of each content server hosting that spe-cific content copy. Upon reception of the Reg-ister message, the CRS registers this contentby creating a new record entry in its local con-tent management repository containing a global-ly unique content ID assigned to that content,and the explicit location of the content (i.e., IPaddress of the content server).

Stage 2: Content Publication Dissemina-tion — Once the content is registered to a CRS,this CRS is responsible for publishing it globallyto ensure successful discovery by potential con-sumers. This is achieved through the dissemina-tion of the Publish message across CRSs inindividual domains according to their businessrelationships. A Publish message is created bythe CRS where the content is actually registeredby the CP. By default, each CRS disseminates anew Publish message towards its counterpartin the provider domain(s) until it reaches a tier 1ISP network. Each CRS receiving a new Pub-lish message updates its content managementrepository with a new record entry containingthe content ID and the implicit location of thecontent (i.e., the IP prefix associated with theneighboring domain from where the Publishmessage has been forwarded). Following thisrule, each CRS effectively knows the locations ofall the content within its own domain (explicitly)and those under it (its customer domains, implic-itly). Peer domains, however, will not know thecontent records of each other.

We introduce the concept of scoped publica-tion to allow publication of content only to spe-cific areas in the Internet as designated by theCP. This feature is able to natively supportregionally-restricted content broadcasting ser-vices such as BBC iPlayer and Amazon VoDthat are only available within the United King-dom and United States, respectively. We achievethis through the INCLUDE option embedded inthe Register/Publish messages where theCP specifies a scoped area in the Internet (e.g.,only the IP prefix associated with the local ISPnetwork where the content is registered). A spe-cial case of scoped publication is the wildcardmode (denoted by an asterisk symbolizing alldomains) for which the CP has no restrictions onthe geographical location of potential consumersin the Internet.

Figure 3 illustrates different scenarios in thepublication process. It depicts the domain-levelnetwork topology with each circle logically rep-resenting a domain containing a CRS entity. Wefirst assume that CP S1 registers a content item(assigned with ID X1 by the local CRS in thestub domain A.A.A) to the entire Internet by

issuing a Register message with a wildcard.Each intermediate CRS along the publicationpath creates a content entry for X1 associatedwith the IP prefix of its customer domain fromwhere the Publish message has been forward-ed. For clarity, the Publish messages are omit-ted in the figure for other scenarios. Ourapproach also allows local domain policies toinfluence the publication process (e.g., domainB.A has the policy of NOT propagating contentX2 originated from the multi-homed domainA.B.B to its own provider). S3 illustrates thescoped registration by only registering contentX3 to tier-2 domain A.A from this CP. This effec-tively limits the access of content X3 to domainA.A and its customer domain A.A.A. Finally,records for different copies of the same contentcan also be aggregated. For instance, both S4and S5 host one copy of content X4 respectively,but the two Publish messages from B.B.A andB.B.B are merged at B.B, in which case domainB only records aggregated location information(X4 → B.B). A content consumption request forX4 received at B.B can be forwarded to eitherB.B.A or B.B.B based on performance conditionssuch as content delivery path quality or serverload.

CONTENT RESOLUTIONIn the content resolution process, a content con-sumption request issued by a CC is resolved bydiscovering the location of the requested contentand is finally delivered to the actual contentsource to trigger the content transmission. A CCinitiates the resolution process via a Consumemessage containing the ID of the desired con-tent. The primary resolution procedure followsthe same provider route forwarding rule in thepublication process (i.e., the Consume messagewill be further forwarded to its provider(s) if theCRS cannot find the content entry in its localrepository). If a tier-1 domain is not aware ofthe content location, the request is forwarded toall its neighboring tier-1 domains until the con-tent consumption request is delivered to theidentified content source. If the content is notfound after the entire resolution process, anError message is returned to the requesting CCindicating a resolution failure.

The scoping functions can also be applied inthe resolution process, either embedded in therequest from a CC or actively issued by a CRSfor route optimization purposes during the con-tent delivery phase. The function allows a CC toindicate preferred ISP network(s) as the sourcedomain of the requested content. Specifically, aCC may use the INCLUDE option in Consumemessages, which carry one or multiple IP prefix-es to indicate from where he/she would like toreceive the content.1 Since a set of explicit IPprefixes for candidate content source is carriedin the Consume message, the corresponding res-olution process becomes straightforward: eachintermediate CRS only needs to forward therequest (splitting required in the presence ofmultiple non-adjacent IP prefixes) towards thetargeted IP prefix(es) directly according to theunderlying BGP routes. In case multiple inter-domain routes are available towards a specificprefix, the most explicit one will be followed, as

In the content

resolution process, a

content consumption

request issued by a

CC is resolved by

discovering the

location of the

requested content

and is finally

delivered to the

actual content

source to trigger the

content transmission.

1 It is not always requiredthat CCs know the actualIP prefix of the domainsthey prefer but their localCRSs may be responsiblefor translating the regioninformation (e.g., domainnames) into IP prefixesthrough standard DNSservices.



is consistent with today’s interdomain routingpolicy. In Fig. 4, CC C1 issued a Consume mes-sage for content X1 indicating its preference forcontent source in domain A or its customerdomains. This Consume message is then explic-itly forwarded towards A from B following theunderlying BGP routing, but without splitting itto C despite that a copy of X1 is also accessiblefrom C’s customer domain C.A. This scoping-based content resolution path is illustrated withthe solid line in the figure.

The filtering function in content resolutionoperations has complementary effect to scoping.Instead of specifying the preferred networks, theCC has the opportunity to indicate unwanteddomains as possible sources of the desired con-tent. The filtering function is enabled via theEXCLUDE option in Consume messages. It isimportant to note the fundamental difference inresolving content consumption requests withscoping and filtering functions. In contrast to thescoping scenario in which a content consumptionrequest is explicitly routed towards the desiredIP prefix(es) according to the BGP route, in thefiltering case, each request is routed based onthe business relationship between domains (simi-lar to content publication operations). Consideragain Fig. 4 with CC C2 requesting X1 with theexclusion of domain C and its customer domains.Since it is multihomed, the request is sent toboth its providers A.B and B.A (see the dashedline in the figure). However, at the tier 1 level,domain C is excluded when resolving this request

even though a copy of content X1 can be foundin the customer domain of C.

A wildcard in a content consumption requestcan be regarded as a special case whereby theCC does not have preferences on the geographi-cal location of the content source. The wildcard-based resolution is illustrated in Fig. 4 via therequest from consumer C3 for content X2 (dot-ted lines). We see that B splits the request toboth A and C at the tier-1 level. Since only A hasthe record entry for X2, the request is resolvedto S2.

Through these illustrations, we show thatbidirectional location independence can beachieved in the sense that neither CCs nor CPsneed to know a priori the explicit location ofeach other for content consumption. In particu-lar, CCs may include implicit content scoping/fil-tering information when requesting content. Thecontent resolution system then automaticallyidentifies the server in the desired area thathosts the content. On the CP side, when contentis published, scoping can be applied such thatthe content can only be accessed by consumersin the designated area in the Internet. We showin the following section that, thanks to the multi-cast-oriented content delivery mechanism, thecontent server is not aware of the explicit loca-tion of the active consumers of that content.

We conducted simulation experiments basedon a real domain-level Internet sub-topologyrooted at a tier-1 ISP network (AS7018). Thisfour-tier network topology is extracted (with

Figure 3. Content publication process.

Tier-1

Tier-2

Tier-3

S5::Register(*,X4)Content aggregation example

S4::Register(*,X4)S2::Register(*, X2)Multi-homed

S1::Register(*, X1)Wildcard mode

S3::Register(INCLUDE(A.A), X3)Scoped publication

Publish(A.A, X1)

Publish(A.A.A, X1)

A B

A.A

A.A.A

Peering link

A.B.B B.B.A B.B.B

A.B B.A B.B

Provider-customer link

X1->A.AX2->A.B

X1->A.A.AX3->S3

X4->B.B

X1->S1 X2->S2 X4->S4 X4->S5

X4->B.B.A, B.B.B

X2->A.B.B

X2->A.B.B



aggregation) from the CAIDA dataset [9], withexplicit business relationships between neighbor-ing nodes. Content sources and consumers wererandomly distributed in the domains of thistopology. According to our results, the averagelength of the content resolution paths betweenindividual CCs and resolved content sources is4.4 domain-level hops (i.e., the content is onaverage 4.4 autonomous systems [ASs] awayaccording to the resolution paths). This is a verygood result and also consistent with the generalobservation that Internet interdomain sessionsare of similar length based on Border GatewayProtocol (BGP) routing and the power-law inter-domain Internet topology.

CONTENT DELIVERYIn CURLING, content delivery paths areenforced in a receiver-driven multicast mannerthat needs state maintenance based on contentIDs. As described, content consumption requestsare resolved through a sequence of CRSs accord-ing to either the business relationships betweenISPs (in wildcard and filtering modes) or theBGP reachability information on the scopedsource prefix (in scoping mode). In both cases,once a CRS has forwarded the content con-sumption request to its next hop counterpart, itneeds to configure the local CaRs that will be

involved in the delivery of content back from thepotential server. Specifically, once a CRSreceives a content consumption request from itscounterpart in the previous hop domain and for-wards it towards the next hop CRS, it needs tocorrespondingly install the content ID state atthe local egress and ingress border CaRs con-necting to the two neighboring domains.2 Thedetermination of ingress/egress CaRs for eachcontent consumption request is purely based onthe BGP reachability information across net-works. Within each domain, the communicationbetween the non-physically connected ingressand egress CaRs can be achieved either by estab-lishing intradomain tunnels that traverse non-content-aware core IP routers, or nativelythrough the content-centric network routing pro-tocols [1]. Therefore, the actual domain-levelcontent delivery path is effectively the reversepath followed by the delivery of the original con-tent consumption request. It is worth mentioningthat CRSs do not directly constitute the contentdelivery paths, in which case the configurationinteraction between the CRS and localingress/egress CaRs is necessary.

Let us take Fig. 5 for illustration. We assumethat currently CC C1 (attached to domain 2.1/16)is consuming live streaming content X from serv-er S (attached to domain 1.2.1/24). The content

Figure 4. Content resolution in scoping, filtering, and wildcard modes.

C

BA

A.A A.B B.A B.B

A.A.A A.B.B B.B.A B.B.B

S2S1

C1::Consume(INCLUDE(A), X1)With scoping function

C3::Consume(*, X2)Wildcard mode

C2::Consume(EXCLUDE(C), X1)With filtering function

C.A

X1->A.AX2->A.B

X1->A.A.AX3->S3

Peering link


X4->B.B.A,B.B.B

X4->S5X4->S4

X4->B.B

X1->S3

X1->C.A

X2->A.B.B

X2->A.B.B

X2->S2X1->S1

2 In case of a failed con-tent resolution, contentstates temporally main-tained at CaRs can beeither timed out or explic-itly torn down by the localCRS.



delivery path traverses a sequence of intermedi-ate domains, and each of the correspondingingress/egress CaRs is associated with a star thatindicates the content state maintained for con-tent delivery. As mentioned, these states areconfigured by the local CRSs during the contentresolution phase. Now CC C2 (attached todomain 1.1/16) issues a consumption request forthe same content. Upon receiving the request,the local CRS forwards it to its provider coun-terpart in domain 1/8, as it is unaware of thecontent source location. Since the CRS in 1/8knows that content flow for X is being injectedinto the local network via the originally config-ured ingress CaR 1.0.0.2, it then updates its out-going next-hop CaR list by adding a new egress1.0.0.3 leading towards CC C2. Thus, a newbranch is established from CaR 1.0.0.2 which isresponsible for delivering the content back tothe new consumer C2 (the dashed line), butwithout any further content resolution process.

The proposed content delivery operation isalso supported by a routing optimization tech-nique for path switching from provider routesto peering routes. In the figure, once the CRSin domain 1.1/16 noticed that the content flowwith source address belonging to prefix 1.2.1/24has been injected into the local domain viaingress CaR 1.1.0.1 via the provider route, andit also knows from the local BGP routing infor-mation that there exists a peering routetowards the content source, it then issues anew scoping-based content consumptionrequest: Consume(INCLUDE{1.2.1/24},X)to the CRS in domain 1.2/16 in the peeringroute towards the source. Upon receiving therequest, the CRS in 1.2/16 updates the localCaR 1.2.0.1 by adding a new outgoing next-hopCaR 1.1.0.1. As a result, a new branch via thepeering route is established towards C2. Oncethe ingress CaR 1.1.0.1 has received the con-

tent via the interface connecting to 1.2.0.1, itprunes the old branch via the provider route(the dashed line). This content delivery pathoptimization effectively reduces content trafficwithin top-tier ISP networks and also possiblyreduces the content delivery cost for customerdomains. Of course, this operation is not nec-essary if a CRS is allowed to send content con-sumption requests to its peering counterparts(in addition to the provider direction) duringthe resolution phase. However, such an optionwill incur unnecessarily higher communicationoverhead in disseminating content consump-tion requests, especially when the peeringroute does not lead to any source that holdsthe requested content.

We are also interested in the actual benefitfrom such inter-domain routing optimizationtechniques for cost-efficient content deliveryacross the Internet, especially from the viewpoint of tier-1 ISPs that constitute the Internetcore. We used the same domain-level topologyas previously described for evaluating the corre-sponding performance. According to our results,the content traffic (in terms of the number ofmedia sessions) traversing higher tier 1 and 2ISPs can be reduced by 8.7 percent throughpeering route switching, and in particular by asubstantial 28.1 percent in tier-1 ISPs. This isbeneficial given that less traffic traverses tier-1domains through relatively long paths.

DISCUSSION ON SCALABILITYThe domain-level hop-by-hop content resolutionstrategy presented follows a similar style to thatproposed in [3]. However, through the new scop-ing and filtering functions, our architecture pro-vides the necessary flexibility for both CPs andCCs to publish/request content at/from theirdesired area(s). The scalability of the system,

Figure 5. Multicast-based content delivery process.

S: X

Content server S

Contentconsumer C1

Contentconsumer C2

Consume(INCLUDE{1.2.1/24}, X)

1.0.0.11.0.0.3

1.0.0.21.1.0.1

1.1/161.2/16

1.2.1/24

2.1/16

2/8

1.2.0.1

CRS

1/8

CRS

CRS

CRS

CRS CRS

Peering link


According to our

results, the content

traffic traversing

higher tier 1 and 2

ISPs can be reduced

by 8.7 percent

through peering

route switching, and

in particular by a

substantial 28.1 per-

cent in tier-1 ISPs.

This is beneficial

given that less traffic

traverses tier-1

domains through

relatively long paths.



thus, is dependent on the amount and popularityof content in each CRS, with the most vulnerableCRSs being those that maintain the highestnumber of popular content entries. This is incontrast with intuition that the most strainedCRSs will be the tier-1 ones, since content publi-cations and requests may often not reach thetier-1 level based on our approach. Again, wetake BBC iPlayer as an example where both thecontent publication and consumption requestsare restricted to IP prefixes only from the Unit-ed Kingdom. In addition to that, local domainpolicies may also override the default publica-tion route (see S2 in Fig. 3).

Business incentives also present a naturalload distribution mechanism for our system. Weforesee ISPs charging higher publication tariffsfor popular content published at higher tierdomains (with tier-1 domains being the mostexpensive) which can be potentially accessed bya higher number of consumers. This mechanismforms a business tussle from the CPs’ point ofview when provision of wider access is coupledwith higher monetary cost. Instead, a CP maystrategically replicate content to multiple lower-tier regional ISPs (by applying scoping functionsthere) in which they believe their content will belocally popular.

Finally, our system allows aggregation in twoways. First, as illustrated in Fig. 3 for S4 and S5,the record for the same content can be mergedduring the publication process among CRSs.Second, a block of sequential content IDs shouldbe allocated to interrelated content so that theycan be published in one single process. This ruleexploits the fact that a specific CP usually offerscontent with some relationship with each other(e.g., all episodes of a television series). Thisallows for coarser granularity in the publicationprocess whereby the CP can send only one Pub-lish message to publish all the related content.The local CRS still assigns a unique content IDfor each content, but the IDs are sequentiallyconnected. The onwards publication process willonly involve the entire block of IDs rather thanthe individual content records, especially towardshigh-tier ISPs.

CONCLUSIONSIn this article, we present CURLING, a newcontent-based Internet architecture that sup-ports content publication, resolution and deliv-ery. Content providers can cost-efficientlypublish content based on its expected popularityin different regions by scoping its publicationwhile content consumers can express their loca-tion preferences by scoping/filtering their con-tent consumption requests. The processes aredevised so that both sides are oblivious of theircounterpart’s location, resulting in a bi-direction-al location independence paradigm, but withoutsacrificing content providers’ and consumers’location preferences. The proposed route opti-mization mechanism enhances the efficiency ofcontent delivery by using content states estab-lished during the resolution process and initiat-ing content delivery path switching; it mimics assuch interdomain multicast delivery, which hasseen very slow deployment until now.

ACKNOWLEDGMENTS

This work was undertaken under the Informa-tion Society Technologies (IST) COMET pro-ject, which is partially funded by the Commissionof the European Union. We would also like tothank our project partners who have implicitlycontributed to the ideas presented here.

REFERENCES[1] V. Jacobson et al., “Networking Named Content,” ACM

CoNEXT ’09, 2009, pp. 1–12.[2] P. Jokela et al., “LIPSIN: Line Speed Publish/Subscribe

Inter-networking,” Proc. ACM SIGCOMM ‘09, Barcelona,Spain, Aug. 2009.

[3] T. Koponen et al., “A Data-Oriented (and Beyond) Net-work Architecture,” Proc. ACM SIGCOMM ‘07, Kyoto,Japan, Aug. 2007.

[4] P. Francis and R. Gummadi, “IPNL: A NAT-ExtendedInternet Architecture,” Proc. ACM SIGCOMM ‘01, SanDiego, CA, Aug. 2001, pp. 69–80.

[5] I. Stoica et al., “Internet Indirection Infrastructure,”Proc. ACM SIGCOMM ‘02, Pittsburgh, PA, Aug. 2002,pp. 73–86.

[6] D. Clark et al., “FARA: Reorganizing the Addressing Archi-tecture,” Proc. ACM SIGCOMM, FDNA Wksp., Aug. 2003.

[7] M. Caesar et al., “Routing on Flat Labels,” Proc. ACMSIGCOMM ‘06, Pisa, Italy, Sept. 2006, pp. 363–74.

[8] P. Mockapetris, “Domain Names — Concepts and Facili-ties,” IETF RFC 1034, Nov. 1987.

[9] CAIDA dataset; http://www.caida.org/research/topolo-gy/#Datasets.

BIOGRAPHIESWEI KOONG CHAI ([email protected]) is a research fellow inthe Department of Electronic and Electrical Engineering,University College London (UCL), United Kingdom. Hereceived a B.Eng. degree in electrical engineering from Uni-versiti Teknologi Malaysia in 2000 and an M.Sc. (Distinc-tion) in communications, networks, and software, and aPh.D. in electronic engineering, both from the University ofSurrey, United Kingdom, in 2002 and 2008, respectively.His research interests include content-centric networks,QoS and service differentiation, resource optimization, andtraffic engineering.

NING WANG ([email protected]) is a lecturer at the Uni-versity of Surrey. He received his B.Eng. (Honors) degreefrom Changchun University of Science and Technology, P.R.China, in 1996, his M.Eng. degree from Nanyang Universi-ty, Singapore, in 2000, and his Ph.D. degree from the Uni-versity of Surrey in 2004. His research interests includeInternet content delivery techniques, end-to-end QoS provi-sioning, and traffic engineering.

IOANNIS PSARAS ([email protected]) is a research fellowin the Department of Electronic and Electrical Engineering,UCL. He received a diploma in electrical and computerengineering, and a Ph.D. degree from Democritus Universi-ty of Thrace, Greece, in 2004 and 2008, respectively. Hewon the Ericsson Award of Excellence in Telecommunica-tions in 2004. He has worked at DoCoMo Eurolabs andEricsson Eurolab. His research interests include congestioncontrol, delay-tolerant networks, and user- and content-centric networks.

GEORGE PAVLOU ([email protected]) holds the Chair ofCommunication Networks at the Department of Electronicand Electrical Engineering, UCL. Over the last 20 years hehas undertaken and directed research in networking, net-work management, and service engineering, having exten-sively published in those areas. He has contributed to ISO,ITU-T, IETF, and TMF standardization activities, and hasbeen instrumental in a number of key European and U.K.projects that produced significant results. He is currentlythe technical manager of COMET.

CHAOJIONG WANG ([email protected]) holds a B.Sc. incomputing and information technology from the Universityof Surrey, and an M.Sc. in computer science from OxfordUniversity. He is a Ph.D. student in electrical engineering atthe Centre for Communication Systems Research (CCSR),University of Surrey.

The proposed route

optimization mecha-

nism enhances the

efficiency of content

delivery by using

content states estab-

lished during the res-

olution process and

initiating content

delivery path switch-

ing; it mimics as

such inter-domain

multicast delivery

which that has seen

very slow deploy-

ment until now.



GERARDO GARCIA DE BLAS ([email protected]) is a research engi-neer in the IP Network Technologies group in TelefónicaI+D. He holds a Master’s in telecommunications engineer-ing from the Technical University of Madrid, Spain. Since2002 he has worked in Telefónica I+D, with his main focuson network evolution, network planning, and traffic analy-sis. He is currently coordinating the architecture definitionin the EU FP7 project COMET.

FRANCISCO JAVIER RAMON-SALGUERO ([email protected]) received hisMaster’s in telecommunications engineering from MálagaUniversity, Spain, in 2000, and his Master’s in economics atUNED, Spain, in 2006. With Telefónica I+D since 2000, cur-rently he heads the IP Network Technologies group, coordi-nating research on long-term evolution of Internettechnologies. Since 2010 he has led the FP7-COMET con-sortium, focused on providing to the future Internet a uni-fied approach to content location, access, and deliverywith the appropriate network resources.

LEI LIANG ([email protected]) is a research fellow at CCSRat the University of Surrey, from which he received his B.E.degree in 1998, M.S. degree in 2001, and Ph.D. degree in2005. With strong knowledge in multiparty communica-tions, network performance measurement and analysis, IPnetworking QoS, IP over satellite networks, IP multicastover satellite network, and IP network security, he hasbeen heavily involved in 14 EU and U.K. projects since2001.

SPIROS SPIROU ([email protected]) is a senior engineer atIntracom Telecom, Greece, working on IPTV, content-awarenetworking, and network management. Previously, he wasa research associate on data acquisition and grid comput-ing at NCSR Demokritos, Greece, and CERN, Switzerland.He holds a B.Sc. (1997) in computer engineering, a post-graduate diploma (1998) in mathematics, and an M.Sc.(2000) in neural networks. He chaired the European FutureInternet Socio-Economics Task Force.

ANDRZEJ BEBEN ([email protected]) received M.Sc. andPh.D. degrees in telecommunications from Warsaw Univer-sity of Technology (WUT), Poland, in 1998 and 2001,respectively. Since 2001 he has been an assistant professorwith the Institute of Telecommunications at WUT, wherehe is a member of the Telecommunication Network Tech-nologies research group. His research areas include IP net-works (fixed and wireless), content-aware networks, trafficengineering, simulation techniques, measurement meth-ods, and testbeds.

ELEFTHERIA HADJIOANNOU ([email protected])received a Dipl.-Ing. degree from the Electrical Engineeringand Computer Science Department, Aristotle University ofThessaloniki, and an M.B.A. from the University of Macedo-nia, Greece. Previously, she was a research associate at theInformation System Laboratory (ISlab), University of Mace-donia, where she was involved in a number of projects rel-evant to eGovernment and eParticipation. Currently, she isa member of the R&D Department at Primetel PLC, analternative telecommunication provider in Cyprus.



INTRODUCTION

Video traffic has been and will be increasinglyprevalent in the Internet. Some video contentproviders (CPs, e.g., YouTube and Hulu) haveeven begun to provide high-definition videostreaming services. As the bit rate of multimediatraffic increases, the TCP/IP architecture mayreveal its inefficiency in delivering time-sensitivemultimedia traffic.

Another important multimedia application ismulticasting/broadcasting over IP networks (e.g.,IPTV). However, the endpoint-based Internet isnot suitable for multicast/broadcast due to issuesincluding multicast address assignment and com-plex group management. Such ineptness leads tolimited deployment and complicated multicastframeworks.

At present, many voluminous contents (mostof which are multimedia [1]) are delivered tonumerous users by peer-to-peer (P2P) systemssuch as BitTorrent. In BitTorrent, for each con-tent file there is a tracker, which informs a newpeer of other peers. A peer exchanges missingparts (called chunks) of the content file withother peers. However, from a networking per-spective, the delivery performance of BitTorrentis inefficient since a peer can download chunksonly from a small subset of peers who may bedistantly located. In general, P2P systems havelimited information on peers downloading the

same content and the network topology amongthem (e.g. proximity).

In these “content-oriented” applications/ser-vices, an end user cares not about hosts, butabout contents. However, the current Internetrelies on the host-to-host communication model.This mismatch leads to application/service-spe-cific solutions, which may be costly and/or ineffi-cient. Two representative examples are:• Web caches and content delivery networks

(CDNs) transparently redirect web clientsto a nearby copy of the content file.

• P2P systems enable users to search andretrieve the content file.To address the above mismatch, there have

been studies on content-oriented networking(CON)1 (e.g. [2–4]). They strive to redesign thecurrent Internet architecture to accommodatecontent-oriented applications and services effi-ciently and scalably. The essence of CON lies indecoupling contents from hosts (or their loca-tions) not at the application level, but at the net-work level. Note that these proposals also solveor mitigate other Internet problems such asmobility and security.

We argue that the new CON paradigm will:• Free application/service developers from re-

inventing application-specific deliverymechanisms

• Provide scalable and efficient delivery ofrequested contents (e.g., by supporting mul-ticast/broadcast/anycast naturally)

In this article, we classify the prior studies onCON, discuss their technical issues, and identifyfurther research topics. After demonstrating theperformance of CON proposals, we concludethis article.

CONTENT-ORIENTED NETWORKING

A CON architecture can be characterized byfour main building blocks:• How to name the contents• How to locate the contents (routing)• How to deliver/disseminate the contents• How to cache the contents “in” the network

ABSTRACT

As multimedia contents become increasinglydominant and voluminous, the current Internetarchitecture will reveal its inefficiency in deliver-ing time-sensitive multimedia traffic. To addressthis issue, there have been studies on content-oriented networking (CON) by decoupling con-tents from hosts at the networking level. In thisarticle, we present a comprehensive survey oncontent naming and name-based routing, anddiscuss further research issues in CON. We alsoquantitatively compare CON routing proposals,and evaluate the impact of the publish/subscribeparadigm and in-network caching.


Jaeyoung Choi, Jinyoung Han, Eunsang Cho, Ted “Taekyoung” Kwon, and Yanghee Choi,

Seoul National University

A Survey on Content-OrientedNetworking for Efficient Content Delivery

1 Many studies use theirown terminology such asdata-oriented, content-centric, and content-based. In this article, weuse content-oriented as ageneric term.

CHOI LAYOUT 2/18/11 3:05 PM Page 121


There are relatively many studies on the firsttwo components, to be classified in this section.The last two topics need more investigationunder CON environments, which will be dis-cussed later. Before presenting the taxonomy, letus discuss common characteristics in CON pro-posals [2–4].

A CON has three characteristics distinct fromIP networking. First, a CON node2 performsrouting by content names, not by (host) locators.This means two radical changes:• Identifying hosts is replaced by identifying

contents.• The location of a content file is indepen-

dent of its name.An IP address has both the identifier and

locator roles; hence, IP networking has problemslike mobility. By splitting these roles, CON haslocation independence in content naming androuting, and is free from mobility and multihom-ing problems.

Second, the publish/subscribe paradigm is themain communication model in CON: a contentsource announces (or publishes) a content file,while a user requests (or subscribes to) the con-tent file. In IP networking, a user should knowwhich source holds the content file of interest(spatial coupling), and the two hosts should beassociated throughout the delivery (temporalcoupling) [5]. However, with the publish/sub-scribe paradigm, we can decouple the contentgeneration and consumption in time and space,so contents are delivered efficiently and scalably(e.g., multicast/anycast).

Third, the authenticity of contents can easilybe verified by leveraging public key cryptogra-phy. In IP networking, a host address seen by auser is irrelevant to its content name, whichresults in phishing and pharming attacks. Forcontent authentication in CON, either a self-cer-tifying content name [2, 4] or a signature in apacket [3] is used. We skip the security-relatedexplanations here; see [2, 3] for details.

CONTENT NAMINGWe classify naming schemes in CON into threecategories: hierarchical, flat, and attribute-based.

Hierarchical Naming — CCN [3] and TRIAD[6] introduce a hierarchical structure to name acontent file. Even though it is not mandatory, acontent file is often named by an identifier3 likea web URL (e.g. /www.acme.com/main/logo.jpg),where / is the delimiter between components ofa name. Thus the naming mechanism in the pro-posals can be compatible with the current URL-based applications/services, which may imply alower deployment hurdle. Its hierarchical naturecan help mitigate the routing scalability issuesince routing entries for contents might be aggre-gated. For instance, if all the contents whosenames start with www.acme.com are stored in asingle host, we need a single routing entry (tothe host) for these contents. However, as con-tent files are replicated at multiple places, thedegree of aggregation decreases. For instance, ifpopular contents are increasingly cached by in-network caching, the corresponding routingentries that have been aggregated should be splitaccordingly. Note that components in a hierar-

chical name (e.g., www.acme.com and logo.jpg)have semantics, which prohibits persistent nam-ing. Persistence refers to a property that once acontent name is given, people would like toaccess the content file with the name as long aspossible. For example, if the ownership of a con-tent file is changed, its name becomes mislead-ing with the above naming.

Flat Naming — To avoid the above shortcom-ings, DONA [2] and PSIRP [4] employ flat andself-certifying names by defining a content iden-tifier4 as a cryptographic hash of a public key.Due to its flatness (i.e., a name is a randomlooking series of bits with no semantics), persis-tence and uniqueness are achieved. However,flat naming aggravates the routing scalabilityproblem due to no possibility of aggregation. Asflat names are not human-readable, an addition-al “resolution” between (application-level)human-readable names and content names maybe needed.

Attribute-Based Naming — CBCB [7] identi-fies contents with a set of attribute-value pairs(AVPs). Since a user specifies her interests witha conjunction and disjunction of AVPs, a CONnode can locate eligible contents by comparingthe interest with advertised AVPs from contentsources. It can facilitate in-network searching(and routing), which is performed by externalsearching engines in the current Internet. How-ever, it has drawbacks such as:• An AVP may not be unique or well defined.• The semantics of AVPs may be ambiguous.• The number of possible AVPs can be huge.

NAME-BASED ROUTINGCON should be able to locate a content filebased on its name, which is called name-basedrouting. The prior studies can be classifieddepending on whether there is a systematicstructure in maintaining routing tables of CONnodes.

Unstructured Routing — Like IP routing, thisapproach assumes no structure to maintain rout-ing tables; hence, the routing advertisement (forcontents) is mainly performed based on flooding.CCN suggests inheriting IP routing, and thus hasIP compatibility to a certain degree. Therefore,CCN might be deployed incrementally with cur-rent IP networking. CCN just replaces networkprefixes (in IP routing) with content identifiers,so the modification of IP routing protocols andsystems may not be significant. Just as networkprefixes are aggregatable in IP routing, so arehierarchical content identifiers in CCN routing.However, as a content file is increasingly repli-cated or moved, the level of aggregation dimin-ishes. Moreover, the control traffic overhead(i.e., the volume of announcement messageswhenever a content file is created, replicated, ordeleted) would be huge.

Structured Routing — Two structures havebeen proposed: a tree and a distributed hashtable (DHT). DONA is the most representativetree-based routing scheme. Routers in DONAform a hierarchical tree, and each router main-

2 A CON node refers to anode that performs CONfunctionalities like contentrouting and caching, whilea node may indicate anIP router as well as aCON node.

3 In this article, a nameand an identifier are usedinterchangeably.

4 They also add a label(which has semantics)into a content identifier;however, the label can beinterpreted only by end-points (i.e. publishers andsubscribers), not by in-network nodes.

Persistence refers to

a property that once

a content name is

given, people would

like to access the

content file with the

name as long as

possible. For exam-

ple, if the ownership

of a content file is

changed, its name

becomes misleading

with the above

naming.



tains the routing information of all the contentspublished in its descendant routers. Thus, when-ever a content file is newly published, replicated,or removed, the announcement will be propagat-ed up along the tree until it encounters a routerwith the corresponding routing entry. Thisapproach imposes an increasing routing burdenas the level of a router becomes higher. The rootrouter should have the routing information of allthe contents in the network. Since DONAemploys non-aggregatable content names, thisscalability problem is severe.

On the other hand, PSIRP [4] adopts hierar-chical DHTs [8]. The flatness of a DHT imposesan equal and scalable routing burden amongrouters. If the number of contents is C, eachrouter should have log(C) routing entries. How-ever, the DHT is constructed by random anduniform placement of routers, and thus typicallyexhibits a few times longer paths than a tree thatcan exploit the information of network topology.Also, the flatness of a DHT often requires for-warding traffic in a direction that violates theprovider-customer relation among ISPs; forinstance, a customer ISP does not want toreceive a packet from its provider ISP if the des-tination is not located inside.

Table 1 compares the CON studies with afocus on naming and routing characteristics. AsCCN and TRIAD adopt hierarchical naming,content names are aggregatable and IP compati-ble. However, their flooding-based (i.e., unstruc-tured) routing incurs significant control traffic. Ifall the contents of the same publisher are storedin a single host, their routing entries are aggre-gated into a single one; thus, the routing scala-bility is proportional to the number of publishernodes. TRIAD considers only this case. In CCN,however, the contents can be replicated, whichmay split the aggregated routing entries; in theworst case, the routing burden becomes on theorder of the number of contents.

DONA and PSIRP employ flat names forpersistence. As they have systematic routingstructures, the control traffic will not be substan-tial. In DONA, the root node of the tree shouldhave the routing information of all the contents.

Meanwhile, the DHT structure of PSIRP leviesthe routing burden of a logarithm of the numberof contents on every node.

With AVP-based content names, CBCBenables in-network searching and establishessource-based multicast trees to deliver contentsbetween publishers and subscribers. As eachattribute can be selected or not in a searchquery, the number of routing entries of a routermay be proportional to 2A where A is the totalnumber of attributes. Furthermore, its controloverhead will be high since each new query mayhave to be flooded across the network.

FURTHER ISSUES IN CONMULTISOURCE DISSEMINATION

The current Internet architecture is designedwith point-to-point connectivity since early stageapplications rely on packet exchanges betweentwo hosts. As the Internet becomes increasinglypopular, however, new applications requiring dif-ferent connectivities have emerged: one-to-many(1:N) and many-to-many (M:N).

1:N connectivity represents content dissemi-nation from a single source to multiple recipi-ents; representative applications are onlinestreaming and IPTV services. To support suchapplications with the point-to-point TCP/IParchitecture, the Internet Engineering TaskForce (IETF) standardized the IP multicastingframework, which is deployed in limited situa-tions like a separate network for intradomainIPTV services. Compared to IP multicasting,CON accommodates 1:N connectivity naturallyby the publish-subscribe paradigm in terms ofcontent naming and group management [2, 4].However, its link efficiency is not different fromIP multicasting. Thus, let us focus on M:N con-nectivity.

M:N connectivity takes place among multiplesources and multiple recipients. There are twokinds of M:N connectivity applications:• M instances of 1:N connectivity (e.g., video-

conference).• M sources disseminate different parts of a

content file to N recipients.

Table 1. Taxonomy of CON proposals in terms of naming and routing criteria is summarized. Note that therouting scalability of each proposal is in proportion to either N, C, A, or its logarithm/exponential. Here, N,C, and A are the numbers of publisher nodes, contents, and attributes in the entire network, respectively.

Naming Naming advantages Routing structure Routingscalability

Controloverheads

CCN Hierarchical Aggregatable, IPcompatible Unstructured N (best)

C (worst) High

DONA Flat Persistent Structured (tree) C Low

PSIRP Flat Persistent Structured(hierarchical DHT) logC Low

TRIAD Hierarchical Aggregatable, IPcompatible Unstructured N High

CBCB (attribute, value)pairs In-network searching Source-based

multicast tree 2A High

The current Internet

architecture is

designed with point-

to-point connectivity

since early stage

applications rely on

packet exchanges

between two hosts.

As the Internet

becomes increasingly

popular, however,

new applications

requiring different

connectivities have

emerged: one-to-

many and

many-to-many.



We focus on the latter case, whose representa-tive applications are P2P systems like BitTor-rent. Another application is multi-user onlinegaming in which different but partially overlap-ping game data are transmitted to players.

Substantiating M:N connectivity requiresapplication/service-specific overlays or relaymechanisms in the current Internet. However,CON can disseminate5 contents more efficientlyat the network level by spatial decoupling of thepublish/subscribe paradigm and content aware-ness at network nodes.

Figure 1 illustrates how inefficiently a contentfile is distributed by current P2P operations.Suppose two P2P overlays are formed, and thepeers in both overlays wish to download thesame file. Unfortunately, peers in each overlayare distant; hence, the throughput is poor, whichhappens frequently in reality. This is becauseP2P systems are application-level solutions thatcannot exploit network topology information. Incontrast, CON can efficiently disseminate a con-tent file among subscribers since CON nodes(R2 and R4) will help them download the con-tent file also from the other overlay.

Disseminating a content file from multiplesources is tightly coupled with name-based rout-ing. That is, in order to exploit multiple sourcesin disseminating the same content in CON, eachCON node may have to keep track of individualsources of the same content (e.g., CCN andDONA). In this case, a CON node can seek toretrieve different parts of the requested contentin parallel from multiple sources to expedite dis-semination. To the best of our knowledge, thereis no prior study on this multisource dissemina-tion at a networking level. Depending on round-trip times and traffic dynamics of the path toeach source, the CON node should dynamicallydecide/adjust which part of the content file is tobe received from each source.

Another relevant issue is what routing infor-mation should be stored and advertised by eachCON node for multiple sources of the same con-tent. For instance, suppose a CON node receivesrouting advertisements from two sources of thesame content, and learns that one source is closeand the other source is very far. It may not beuseful to announce both of the sources sinceretrieving data from the far source would be

inefficient. The more sources of the same con-tent the CON node learns of, the more selective-ly it may have to propagate the routinginformation of the sources. How to design arouting protocol for multiple sources, bothintradomain and interdomain, is a crucial issue.

IN-NETWORK CACHINGThe advantages of in-network caching in an ISPmay be twofold:• To reduce the incoming traffic from neigh-

bor ISPs to lower the traffic load on itscross-ISP links6 (and hence its expense fortransport link capacity)

• To improve the delay/throughput perfor-mance by placing the contents closer totheir users

The latter has the same rationale as CDNs. Theusefulness of caching is already proven by thecommercial success of CDNs. In-network cachingis also attractive to content providers (CPs) sinceit can mitigate the capital expense on their con-tent servers. We believe an ISP with in-networkcaching capability can also offer CDN-like busi-nesses to CPs if the majority of potential sub-scribers to the CPs are connected to the ISP(e.g., [9]).

Considering the above incentives, it is viableto cache popular content files in CON nodes (ortheir corresponding storage servers); other stud-ies (e.g., [2–4]) also suggest introducing in-net-work caching. While there are well studiedcaching policies such as least recently used andleast frequently used replacement policies atindividual nodes, the performance of in-networkcaching can be further improved by coordinatingmultiple CON nodes in a distributed fashion.There have been recent studies on distributedcaching (e.g., how to locate caching points [9]and how to cache contents [10]). However, asthey assume IP networking, their work is limitedin that• Only a single source (or cache) delivers the

content file to a subscriber.• Limited topologies (e.g. tree) or places (e.g.

point of presence) are taken into account.Thus, we need to reformulate the distributedcaching problem in CON environments; forinstance, multisource dissemination and generalnetwork settings may have to be considered.

Another (maybe smaller) topic is how todesign a signaling protocol among CON nodesto support distributed caching. For instance, arouting protocol may have to be extended tofacilitate coordinated caching among CON nodeswithout significant signaling traffic overhead.The major design issue would be that the morefrequently the content files are replaced in acache, the more routing information may have tobe advertised to enhance the network-widecaching performance.

PERFORMANCE EVALUATIONUsing ns-2, we evaluate:• The effect of routing structures on the reso-

lution time (or delay) to locate a contentfile

• How much traffic load can be reduced byin-network caching network-wide

Figure 1. Dissemination of the same file in the two separate overlays will beinefficient since peers are distantly located.

P1

P5R1

R2

R3

R4R5

R6

P2

P3P4

P2P overlay 1

P2P overlay 2

5 Even though CON doesnot care whether a singleor multiple recipients sub-scribe to a particular con-tent file, we implicitly use“deliver” for 1:1 connec-tivity and “disseminate”for 1:N and M:N connec-tivities.

6 By reducing incomingtraffic from its providerISP, its connection feemay also be lowered.



In addition to tree and DHT structures, weintroduce a new routing structure: two-tier. Wefirst explain how the network topology is con-structed and content requests are generated forexperiments. Then we describe the two-tier rout-ing structure, followed by the simulation results.

End hosts, which publish or subscribe to con-tent files, are collocated with CON nodes forsimplicity. Using GT-ITM models, we generate aphysical transit-stub topology, where a singletransit domain connects 10 stub domains. Thereare 310 nodes total whose links have 10 Gb/sbandwidth capacity, among which 100 nodes inthe stub domains are selected as CON nodes.There are 10 nodes in the transit domain, whichdo not serve as CON nodes since they normallyhave higher traffic loads (typically with higherlink capacities) and hence may not be suitablefor CON operations (e.g., in-network caching).

Reflecting the Internet traffic statistics [1],four types of content files are published in theend hosts: video, audio, software, and web con-tents, with 68, 9, 9, and 14 percent in terms ofthe content volume, respectively. For each con-tent type, 1000 content files are published andevenly distributed among the end hosts. Thepopularity (or request probability) of a particularcontent file is determined by the Zipf distribu-tion whose parameter is set to 1.0. The arrivalrate of subscriptions (or request rate) is set to0.5 s–1. For details like the subtypes of each con-tent type and the file size of each content sub-type, see [1].

TWO-TIERThrough our earlier experiments, we made thefollowing observations:• As a tree can be formulated with network

topology information (e.g., hop countbetween nodes), tree routing achieves high-er throughput than DHT routing.7

• DHT routing is more scalable in terms ofrouting burden and more resilient tonode/link failures due to multiple pathsthan tree routing.

Thus, we introduce a hybrid approach whoserouting structure consists of two tiers: a DHT isthe high tier, and a tree is the low tier.

At the low tier, CON nodes form a tree struc-ture. As the tree covers only a part of the wholenetwork, the routing scalability issue is not sig-nificant. At the high tier, CON nodes form aDHT. That is, only the root node of each treewill participate in the DHT. Therefore, a queryfor a content file published in the same tree willbe serviced within the tree. If a query is for acontent file outside the tree structure, the DHTstructure is exploited to forward the query to thecorresponding tree where the requested contentis published. Figure 2 illustrates how a query isforwarded across the two trees via the DHT intwo-tier.

COMPARISON OF ROUTING STRUCTURESWe compare tree, two-tier, and DHT in terms ofa resolution delay, which refers to how long ittakes for a content request to arrive at the con-tent publisher. Also, we measure how the rout-ing structure resiliently routes content requestsin presence of node failures.

Figure 3 shows the resolution delays of thethree routing structures. The tree structure out-performs the DHT since the DHT topology isconstructed without any information on thephysical topology. Thus, a content request goesback and forth among CON nodes of the DHTto reach its corresponding publisher node. Theperformance of two-tier falls between because itis a hybrid approach.

Figure 4 shows the successful resolution ratioas the node failure rate increases. Here R meansthat each of 100 CON nodes will fail for onehour once during a simulation run on average.The performance gain of the DHT over the treeis noticeable since there are multiple paths

Figure 2. Two-tier name-based routing architecture.

X: Subscriber of AY: Publisher of A

: End host

: CON node

Manager ofcontent A

ContentID of A

Root of the tree-network thatcontains content A

Hash (A) = 2

X

Y

32

1

Figure 3. Resolution delay comparison between three structured approaches.

Simulation time (s)200000

3

2.5

Reso

luti

on t

ime

(s)

3.5

4

4.5

5

5.5

6

6.5

7

40000 60000

TreeDHT2-tier

7 Usually, a node’s posi-tion in a DHT is deter-mined randomly, e.g. by ahash function.



among nodes in the DHT, and a failure of ahigher-level node in the tree results in morerouting failures. In the case of two-tier, there are10 trees; each tree is formulated by the CONnodes in the same stub domain. The root nodesof the 10 trees participate in the DHT as well.From Figs. 3 and 4, the two-tier structure com-promises on the trade-off between the tree (per-formance) and the DHT (resilience). Recall thatthe nodes of the two-tier will have less routingburdens than the tree.

NETWORK TRAFFIC LOAD

In the second experiment, we demonstrate howmuch traffic load is reduced by CON. We com-pare DONA (tree structure), two-tier, and thecurrent Internet. The cache replacement policyfor the CON nodes is Least

The cache size of each CON node is 5 Gbytes.Note that there is no publish/subscribe paradigmand in-network caching in the current Internet.

Figure 5 shows how much CON proposals(DONA and two-tier) can reduce the network-wide traff ic load by the publish/subscribeparadigm and in-network caching. The perfor-mance metric is the product of hop count andlink bandwidth (consumed to deliver con-tents). As the simulation time goes on, theproduct of hop count and bandwidth diminish-es in CON proposals due to the cache effect.Each plot is the average of 1000 s. Thus, thecache effect appears almost from the begin-ning. Sometimes, two-tier exhibits slightlypoorer performance than DONA due to DHToverlay inefficiency.

CONCLUSIONSTo fundamentally solve the mismatch betweencontent-oriented Internet usage and host-basedInternet architecture, content-oriented network-ing studies have been proliferated with a focuson naming and routing. In this article, we classifyand compare the prior proposals in terms ofnaming and routing criteria. Also, we identifytwo important research topics in CON environ-ments:• How to disseminate contents from multiple

sources• How to decide which contents to cache in

distributed environmentsWe then compare CON routing structures

and demonstrate the performance gain of CONover IP networking. For future work, an interest-ing topic would be to compare the in-networkcaching part of CON proposals and CDN solu-tions in terms of network traffic load mitigation.

ACKNOWLEDGMENTThis publication is partially based on work per-formed in the framework of the Project COAST-ICT-248036, which is supported by the EuropeanCommunity. This work is also supported by NAPof Korea Research Council of Fundamental Sci-ence and Technology, and the IT R&D programof MKE/KEIT (10035245: Study on Architectureof Future Internet to Support Mobile Environ-ments and Network Diversity). The ICT at SeoulNational University provides research facilities.Professor Ted “Taekyoung” Kwon was on sab-batical leave at Rutgers University, where hehad in-depth discussions with ProfessorDipankar Raychaudhuri.

REFERENCES[1] H. Schulze and K. Mochalski, ipoque -internet study

2008/2009, http://portal.ipoque.com/downloads/index/study.

[2] T. Koponen et al., “A Data-Oriented (and Beyond) Net-work Architecture,” SIGCOMM ’07, 2007, pp. 181–92.

[3] V. Jacobson et al., “Networking Named Content,”CoNEXT ’09, New York, NY, 2009, pp. 1–12.

Figure 4. Robustness comparison between three structured approaches.

1R

0.2

0

Succ

ess

rati

o

0.4

0.6

0.8

1

1.2

1.5R 2R 2.5R 3R

TreeDHT2-tier

Figure 5. The impact of publish/subscribe paradigm and in-network caching onthe content delivery.

Simulation time (s)

400000

5

0

Band

wid

th (

MB/

s) *

hop

coun

t

10

15

20

25

30

35

40

45

80000 120000 160000

InternetDONA2-tier



[4] K. Visala et al., “An Inter-Domain Data-Oriented Rout-ing Architecture,” ReArch ’09: Proc. 2009 Wksp. Re-architecting the internet, New York, NY, 2009, pp.55–60.

[5] P. Eugster et al., “The Many Faces of Publish/Sub-scribe,” ACM Computing Surveys 35, 2003, pp. 114–31.

[6] M. Gritter and D. R. Cheriton, “An Architecture for Con-tent Routing Support in the Internet,” 3rd UsenixSymp. Internet Technologies and Sys., 2001, pp. 37–48.

[7] A. Carzaniga et al., “A Routing Scheme for Content-Based Networking,” IEEE INFOCOM ’04, Hong Kong,China, Mar. 2004.

[8] P. Ganesan, “Canon in g major: Designing Dhts withHierarchical Structure,” ICDCS, 2004, pp. 263–72.

[9] J. Erman et al., “Network-Aware Forward Caching,”WWW 2009, 2009.

[10] S. BORST et al., “Distributed Caching Algorithms forContent Distribution Networks,” INFOCOM 2010, 2010.

BIOGRAPHIESJAEYOUNG CHOI ([email protected]) received his B.S.degree in computer science and engineering from SeoulNational University, Korea, in 2004. He is currently workingtoward his Ph.D. degree at the School of Computer Scienceand Engineering, Seoul National University. His researchinterests include content-oriented networking, peer-to-peernetworking/application, and global Internet infrastructureincluding inter/intradomain routing protocols.

JINYOUNG HAN ([email protected]) received his B.S.degree in computer science from Korea Advanced Instituteof Science and Technology (KAIST), Daejeon, in 2007. He iscurrently working toward his Ph.D. degree at the School ofComputer Science and Engineering, Seoul National Univer-sity. His research interests include peer-to-peer networks,content-centric networks, and ubiquitous computing.

EUNSANG CHO ([email protected]) received his B.S.degree in computer science and engineering from SeoulNational University in 2008. He is currently working towardhis Ph.D. degree at the School of Computer Science andEngineering, Seoul National University. His research inter-ests include peer-to-peer, content-centric, and delay toler-ant networking.

TED “TAEKYOUNG” KWON ([email protected]) is with theSchool of Computer Science and Engineering, SeoulNational University. He was a visiting professor at Rutgersuniversity in 2010. Before joining Seoul National University,he was a postdoctoral research associate at the Universityof California Los Angeles and City University of New York.He obtained his B.S., M.S., and Ph.D. at Seoul NationalUuniversity. He was a visiting student at IBM T. J. WatsonResearch Center and the University of North Texas. Hisresearch interest lies in future Internet, content-orientednetworking, and wireless networks.

YANGHEE CHOI ([email protected]) received his B.S. in elec-tronics engineering at Seoul National University, M.S. inelectrical engineering at KAIST, and D.Eng. in computerscience at Ecole Nationale Superieure des Telecommuni-cations Paris, France, in 1975, 1977, and 198,4 respec-t ively. He worked at the Electronics andTelecommunications Research Institute (Korea), CentreNational D’Etude des Telecommunications (France), andIBM Thomas J. Watson Research Center (United States)before joining Seoul National University in 1991. He waspresident of the Korea Institute of Information Scientistsand Engineers. He is dean of the Graduate School ofConvergence Science and Technology, president of theAdvanced Institutes of Convergence Technologies, andchair of the Future Internet Forum of Korea. He has pub-lished over 600 papers in network protocols and archi-tectures.



INTRODUCTION

Multimedia applications over the Internet arebecoming popular due to the widespread deploy-ment of broadband access. In conventionalstreaming architectures the client-server modeland the usage of content distribution networks(CDNs) along with IP multicast were the mostdesirable approaches for many years. Theclient/server architecture, however, severely lim-its the number of simultaneous users in videostreaming. The reason is the bandwidth bottle-neck at the server side, since usually many clientsrequest the content from the server. A CDNovercomes the same bottleneck problem byintroducing dedicated servers at geographically

different locations, resulting in expensive deploy-ment and maintenance.

Compared to conventional approaches, amajor advantage of peer-to-peer (P2P) stream-ing protocols is that each peer involved in con-tent delivery contributes its own resources to thestreaming session. Administration, maintenance,and responsibility for operations are thereforedistributed among the users instead of handledby a single entity. As a consequence there is anincrease in the amount of overall resources inthe network, and the usual bottleneck problemof the client-server model can be overcome.Therefore, a P2P architecture extends exception-ally well with large user bases, and provides ascalable and cost-effective alternative to conven-tional media delivery services. The main advan-tage of P2P systems is bandwidth scalability,network path redundancy, and the ability to self-organize. These are indeed attractive featuresfor effective delivery of media streams. Never-theless, several problems are still open and needto be addressed in order to achieve high qualityof service and user experience. In particular, thebandwidth capacity of a P2P system is extremelyvarying, as it relies on heterogeneous peer con-nection speeds, and directly depends on thenumber of connected peers.

To cope with varying bandwidth capacitiesinherent to P2P systems, the underlying videocoding/transmission technology needs to supportbit-rate adaptation according to available band-width. Moreover, displaying devices at the userside may range from small handsets (e.g., mobilephones) to large HD displays (e.g., LCD televi-sions). Therefore, video streams need to betransmitted at a suitable spatio-temporal (ST)resolution supported by the user’s display device.If conventional video coding technologies areused, the above mentioned issues cannot besolved efficiently. Scalable video coding (SVC)techniques [1, 2] address these problems, as theyallow “encoding a sequence once and decodingit in many different versions.” Thus, scalablecoded bitstreams can efficiently adapt to the

ABSTRACT

Scalable video delivery over peer-to-peer net-works appears to be key for efficient streamingin emerging and future Internet applications.Contrasting the conventional server-clientapproach, here, video is delivered to a user in afully distributed fashion. This is, for instance,beneficial in cases where a high demand for aparticular video content is imposed, as differentusers can receive the same data from differentpeers. Furthermore, due to the heterogeneousnature of Internet connectivity, the contentneeds to be delivered to a user through networkswith highly varying bandwidths. Moreover, con-tent needs to be displayed on a variety of devicesfeaturing different sizes, resolutions, and compu-tational capabilities. If video is encoded in ascalable way, it can be adapted to any requiredspatio-temporal resolution and quality in thecompressed domain, according to a peer band-width and other peers’ context requirements.This enables efficient low-complexity contentadaptation and interoperability for improvedpeer-to-peer streaming in future Internet appli-cations. An efficient piece picking and peerselection policy enables high quality of service insuch a streaming system.


Naeem Ramzan, Queen Mary University of London

Emanuele Quacchio, STMicroelectronics

Toni Zgaljic and Stefano Asioli, Queen Mary University of London

Luca Celetto, STMicroelectronics

Ebroul Izquierdo, Queen Mary University of London

Fabrizio Rovati, STMicroelectronics

Peer-to-Peer Streaming of ScalableVideo in Future Internet Applications

RAMZAN LAYOUT 2/18/11 3:11 PM Page 128


application requirements. The adaptation is per-formed fully in the compressed domain, bydirectly removing parts of the bitstream. TheSVC encoded bitstream can be truncated tolower resolution, frame rate, or quality. In P2Penvironments such real-time low-complexityadaptation results in a graceful degradation ofreceived video quality, avoiding the interruptionof the streaming service in case of congestion orbandwidth narrowing.

Recently, P2P scalable video streaming hasattracted significant attention from researchers.Liu et al. [3] employ layered video to accommo-date asynchronous requests from users of het-erogeneous bandwidths. Baccichet et al. [4]develop a mathematical framework to quantifythe advantage of using a scalable codec for tree-based overlays, particularly during network con-gestion. Ding et al. [5] and the MMV platform[6] present a P2P video on demand (VoD) sys-tem that utilizes SVC for delay minimization andto deal with heterogeneous user capabilities aswell as dynamic end-to-end resource availability.The possibility to exploit the flexibility given byscalable bitstreams within P2P overlays has alsobeen an important topic in large cooperativeprojects. Among others, P2P-Next [7] and SEA[8] aim to build the software infrastructureenabling high-quality and reliable P2P-based TVservices over the Internet. In the following sec-tions we explain the fundamentals of videostreaming techniques used in some of these pro-jects.

SCALABLE VIDEO CODINGIn general, a scalable video sequence can beadapted in three dimensions: temporal (framerate reduction), spatial (resolution reduction),and quality (quality reduction), by simple parsingand dropping specific parts of the encoded rep-resentation. Thus, the complexity of adaptationis very low, in contrast to the adaptation com-plexity of non-scalable bitstreams. The SVCscheme gives flexibility and adaptability to videotransmission over resource-constrained networksin such a way that, by adjusting one or more ofthe scalability parameters, it selects a layer con-taining an appropriate ST resolution and qualityaccording to current network conditions. Figure1 shows an example of video distribution throughlinks supporting different transmission speedsand display devices. At each point where videoquality/resolution needs to be adjusted, an adap-tation is performed. Since the adaptation com-plexity is very low, the video can be efficientlystreamed in such an environment.

The latest video coding standard, H.264/MPEG-4 AVC, provides a fully scalable exten-sion, SVC1 [1]. It reuses key features ofH.264/MPEG-4AVC, and also uses some othertechniques to provide scalability and improvecoding efficiency. The scalable bitstream is orga-nized into a base layer and one or severalenhancement layers. SVC provides temporal,spatial, and quality scalability with a low increaseof bit rate relative to the single-layer H.264/MPEG-4 AVC.

The SVC standard is based on a hybrid tech-nology. In principle, it uses a combination of

spatial transform based on discrete cosine trans-form and temporal differential pulse code modu-lation. An alternative approach is to use waveletsfor both temporal and spatial decorrelation. Thisapproach is commonly referred to as wavelet-based SVC (W-SVC). Several recent W-SVCsystems [2] have shown exemplary performancein different types of application scenarios, espe-cially when fine-grained scalability is required.Observe that fine-grained scalability is not sup-ported by the current SVC standard.

STREAMING OF SCALABLE VIDEOOVER P2P NETWORKS

P2P protocols have been widely deployed forsharing files over the Internet. One of the mostcommonly used P2P protocols is BitTorrent [9].However, BitTorrent is not suitable for stream-ing applications since segments of the video fileare generally not received in sequential order.Thus, substantial research has been conductedrecently to extend the BitTorrent protocol andmake it suitable for streaming. An example ofsuch an extended protocol is Tribler [10].

A generic P2P streaming architecture usingSVC is depicted in Fig. 2. At the sender side, thevideo is compressed by a scalable encoder. Thecompressed bitstream may optionally be furtherprocessed to make it more suitable for transmis-sion or add additional data into the stream. Theprocessing stage may consist of separating thescalable bitstream into files, each carrying anindividual scalable layer, or multiplexing theaudio and video streams. The corresponding bit-

Figure 1. Streaming of scalable video.

Cinemaprojector Stored

video

Adaptation

PC connected tooffice network

Laptopconnected to

phone lineMobile phone

High definitionTV

Scalable videoencoder

Adaptation

Adaptation

1 In the remainder of thisarticle, the acronym SVCis used interchangeably todenote both the standardand the concept of scal-able video coding.



stream description is created during either theencoding or processing phase. The descriptionmay contain information about the organizationof the video into scalable layers, the resolutionand quality of each layer, and so on. Finally, thetorrent file is produced, which, among otherinformation, describes the mapping of the storedvideo into chunks. A chunk represents the small-est unit of data that will be transmitted over theP2P network. Sometimes, the term piece is usedto denote a chunk. In this article both terms(chunk and piece) are used interchangeably.

At the consumer side, the peer downloadsthe torrent file and requests the video. Here,conventional P2P protocols used for file sharingrequire modifications. In BitTorrent, file chunksare downloaded in rarest-first fashion. This is anefficient strategy in file sharing applicationssince the availability of rare chunks is eventuallyincreased, and higher downloads rates can beachieved for these chunks. However, in videostreaming this can result in an interruption ofthe video playback since chunks are not receivedsequentially. Therefore, special care needs to begiven to those chunks that are close to the play-back position. An example of an algorithm thattakes into account these considerations is Give-to-Get (G2G) [11], implemented in Tribler [10].In this algorithm chunks of compressed videoare classified into three priority categories: high,medium, and low. This classification depends onthe current playback position. Chunks close tothe playback positions are marked as high-priori-ty chunks; they are downloaded first and insequential order. Medium- and low-prioritychunks are downloaded according to the stan-dard BitTorrent strategy: rarest-first.

Given a scalable encoded video unit, a peercan initially decide to recover all or just a subset

of available layers, depending on its capabilities.For instance, a mobile device can decide toretrieve the base layer providing CIF (352 × 288)resolution, while a set-top box may additionallyretrieve the 4CIF (704 × 576) enhancementlayer. Besides, in such a static selection, a peercan dynamically retrieve just a subset of layers inorder to react to a temporary narrowing of thebandwidth. Such dynamic adaptation can beachieved through a carefully designed piece-picking policy. For this purpose the G2G algo-rithm can be modified to take into account notonly the playback position, but also different lay-ers in the scalable bitstream.

To ensure continuous video playback, chunksclose to the playback position need to bereceived on time, at least for the base layer ofscalable video. Therefore, neighboring peersneed to be carefully selected. Ideally, these peersshould be able to deliver video pieces beforethey are requested by the video player.

In the following subsections we explain threerecent state-of-the-art P2P scalable video stream-ing systems based on the principles describedabove.

THE MMV PLATFORMThe system proposed in [12] is based on Tribler.The main modifications are in the G2G algo-rithm and peer selection policy. The videosequence is compressed by the W-SVC encoder[2]. The compressed sequence consists of groupsof pictures (GOPs) and scalable layers. Thenumber of frames within a GOP and the numberof layers are set during encoding and are con-stant throughout the sequence. Along with thebitstream, the description file is generated,which contains information on mapping of GOPsand layers into chunks and vice versa. The

Figure 2. P2P streaming of scalable video.

Input video sequence

Content producer side

Scalable bit-stream

SVC encoder

SVC decoder

Audio

User

User

User

User

User

Torrent file

Bit-stream description

Adaptation Peer

selection Chunk picking

Consumer side

P2P Network

To ensure

continuous video

playback, chunks

close to the playback

position need to be

received on time, at

least for the base

layer of scalable

video. To ensure this,

neighboring peers

need to be carefully

selected.



description file is transmitted with the videosequence. It has the highest priority and there-fore should be downloaded before the bitstream.

The size of a chunk in the MMV platformhas been set to 32 kbytes; however, other sizesare supported as well. If the size of a GOP inbytes is not an integer multiple of the chunksize, padding bytes are added at the end of theGOP. In this way each GOP consists of chunksthat are independent from other GOPs.

Piece Picking Strategy — At the beginning ofthe streaming session, information about GOPsand layers is extracted from the bitstreamdescription file. At this point, a sliding window isdefined, made of several GOPs (typically threeto four), and the prebuffering phase starts.Chunks are picked only from those inside thewindow unless all of them have already beendownloaded. In the latter case, the piece pickingpolicy will be rarest-first. Inside the window,chunks have different priorities, following theidea from the original G2G algorithm. First, apeer will try to download the base layer (BL),

then the first enhancement layer (EL1), and soon. Pieces from the BL are downloaded insequential order, while all other pieces aredownloaded rarest-first (within the same layer).

The window shifts every tGOP s, where tGOPrepresents GOP duration in seconds. An excep-tion is given in the first shift, which is performedafter prebuffering. The duration of the pre-buffering stage corresponds to the length of thesliding window in seconds.

Every time the window shifts, two operationsare made. First, downloaded pieces are checkedto evaluate which layers have been completelydownloaded. Second, pending requests concern-ing pieces of the GOP located just before thewindow are dropped. Fully downloaded layersfrom that GOP are sent to a video player forplayback. Note that the window shifts only if atleast the BL has been received; otherwise, thesystem auto-pauses. Figure 3 shows the behaviorof the system with a window three GOPs wide.An early stage of the prebuffering phase is shownin Fig. 3, first row. The peer is downloadingpieces from BL in a sequential way. In Fig. 3,

Figure 3. Sliding window operations. First row: prebuffering phase starts; second row: prebuffering phaseends; third row: the window shifts the first time; fourth row: the window shifts the second time.

Enhancement layers

Vid

eo

qual

ity

GOP 0

Playback position

GOP 1

Sliding window

GOP 2 GOP 3 GOP 4 GOP 5

Base layer

Enhancement layers

Vid

eo

qual

ity

GOP 0

Playback position

GOP 1 GOP 2 GOP 3 GOP 4 GOP 5

Base layer

Enhancement layers

Vid

eo

qual

ity

GOP 0

Playback position


Base layer

Enhancement layers

Vid

eo

qual

ity

GOP 0

Playback position

Time


Base layer

The size of a chunk

in MMV platform

has been set to 32

Kbytes; however,

other sizes are

supported as well.

If the size of a GOP

in bytes is not an

integer multiple of

the chunk size,

padding bytes are

added at the end of

the GOP. In this way

each GOP consists of

chunks that are

independent from

other GOPs.



second row, the first two layers have been down-loaded, and chunks are being picked from EL2according to the rarest-first policy. These piecescan belong to any GOP within the window. InFig. 3, third row, the window has shifted,although not all pieces from EL2 in GOP 0 havebeen received. This layer is discarded from GOP0. Inside the window, before downloading anyother pieces from GOP 1 or 2, the system willpick chunks from GOP 3 until the quality ofreceived layers is the same. In other words,before picking any chunks belonging to EL2,chunks belonging to the BL of GOP 3 and EL1of GOP 3 will be picked. In Fig. 3, fourth row,all GOPs within the window have the same num-ber of completed layers, and pieces are pickedfrom EL3.

Peer Selection Strategy — Occasionally, slowpeers in the swarm may delay the receiving of aBitTorrent piece even if the download band-width is high. This problem is critical if therequested piece belongs to the BL, as it mightforce the playback to pause. Therefore, thesepieces should be requested from good neighbors.Good neighbors are those peers that own thepiece with the highest download rates, whichalone could provide the current peer with atransfer rate that is above a certain threshold.Each time the window shifts, download rates ofall the neighbors are evaluated, and the peersare sorted in descending order. Pieces are thenrequested from peers providing download ratesabove the threshold.

The performance of this framework is shownin Fig. 4. Here, the download rate is not highenough to allow the transfer of all layers. More-over, the download rate is not constant; there-fore, the received video bit rate needs to matchthe behavior of the download speed. It can beseen that the video is received at a higher bit rateimmediately after the prebuffering phase and atthe end of the sequence. The higher quality after

the prebuffering phase results from the fact thatthe first couple of GOPs are in the window for aslightly longer period than other GOPs. At theend of the sequence, when the window cannot beshifted anymore, the window is shrinking. There-fore, more enhancement layers can be download-ed for GOPs at the end of the sequence.

THE NEXTSHARE PLATFORMBased on the Tribler P2P protocol, theNextShare platform has been designed and isbeing implemented in the framework of theP2PNext project [7]. The NextShare platformsupports delivery over the P2P overlay of com-pressed video contents conforming to the specifi-cations of the SVC standard [1]. At the senderside, first, the SVC bitstream is divided into sev-eral files, each carrying a single scalable layer.Only the BL is encapsulated with audio inMPEG-TS transport format. Within the torrentfile, layers are indexed as independent files, leav-ing the user free to select the one to be down-loaded. Additional metadata is included in thetorrent to provide a description of the scalablebitstream in terms of resolutions and bit ratessupported. SVC layers are downloaded in theform of data segments from different files andcomposed by the NextShare P2P engine at theconsumer side. Once a target resolution is com-posed, the corresponding data is forwarded to amedia player.

Adaptation Strategy — In NextShare themaximum target layer is selected and providedto the piece picking algorithm by matching con-text information and available scalable resolu-tions extracted from the torrent. To react to atemporary narrowing of the bandwidth, theadaptation decision is made periodically bychecking the status of the input buffer, playbackposition, and pieces availability. Adaptationactions occur in correspondence to a synchro-nization point; for a compressed SVC bitstreamthis corresponds to an instantaneous decodingrefresh (IDR) frame. IDR represents a specialintra picture, which cuts off all inter-picturedependencies from previously decoded frames.The stream is therefore divided in units of con-stant temporal duration called time slots; eachtime slot corresponds to a period of framesbetween two consecutive IDRs.

The size of a NextShare chunk has been setto 56,400 bytes in order to have multiple MPEG-TS packets in each chunk, while preserving areasonable chunk size for efficient transmission.Here, the size of each MPEG TS packet is 188bytes. Each scalable layer is encoded at a con-stant bit rate (CBR), and hence is representedby a constant amount of bits in each time slot.Since CBR is used, it is possible to predict thenumber of chunks in a time slot of each layer.For example, for an SVC layer encoded at 512kb/s and 25 frames/s, there will be a time slot of64 frames (2.56 s), and the size of the corre-sponding block of frames will be approximately164 kbytes, which corresponds approximately tothree NextShare chunks. A bit overhead is con-sidered in piece mapping for each scalable layer,in order to compensate for the drift from thetarget bit rate always present in CBR algorithms.

Figure 4. Received download rate and received video bit-rate for Crew CIFsequence.

Time (s)40

200

Dow

nloa

d ra

te (

kb/s

)

0

400

600

800

1000

1200

1400

80 120 160 200 240 280 320 0

Video download rateReceived video bit rate



Piece Picking Strategy — The procedureimplemented in NextShare to download scalabledata chunks is an extension of the G2G algo-rithm. Priorities are defined as in G2G andextended to the multiple files, as depicted in Fig.5. In the high-priority set pieces are downloadedsequentially, while in the low-priority set piecesare downloaded in a rarest-first fashion. Eachblock in the figure represents a time slot.

In Fig. 5 at time instance t (playback posi-tion), the algorithm has to decide which block todownload for time point (t + x). Here, the algo-rithm might decide to start downloading piecesinside EL1 at time t + 1 in order to improve thequality for the near future or may decide to con-tinue downloading blocks for BL to ensure thatthe playback will not stop even if network condi-tions become worse. Periodically the algorithmmakes such a decision depending on the currentstatus of the download, pieces lost, and playbackposition. The controller implemented inNextShare tries to switch to a higher quality assoon as there is enough saved buffer for the cur-rent quality. Therefore, a safe buffer of chunksdownloaded and not yet delivered to the playeris defined; the size of this buffer is a function ofthe parameter x depicted in Fig. 5. The mini-mum value for x corresponds to five time slots,and can vary depending on network perfor-mance. When enough segments for the BL havebeen downloaded, the quality is increased, andthe download of the next layer is selected. In thiscase the high-priority set is redefined, and highpriorities are assigned to blocks belonging to theupper layers (the sequence of high priorities willfollow layer dependencies H0 → H1 → H2). Inorder to guarantee a safe time alignment, foreach increase in quality the initial time slots todownload are ahead in time with respect to pre-vious layers. An offset of one time slot is addedat the beginning of each enhancement layer asdepicted in Fig. 5. When not enough pieces areavailable for a certain resolution, the controllerswitches back to a safer position by interruptingthe download of the correspondent file and reas-signing priorities to lower layers.

Peer Selection Strategy — The peer selectionstrategy is inherited from the approach imple-mented in Tribler. For each layer, pieces with anearly deadline are downloaded from good (i.e.,fast) peers. Bad peers (i.e., peers that missed agiven deadline for a particular piece) are movedfrom the high-priority set to a low-priority onestarting from the BL. A bad performancecounter is incremented each time a piece down-load fails. As a consequence, peers with a posi-tive bad performance counter are only availablefor picking low-priority pieces.

SEACAST PLATFORMIn the framework of the SEA project [8], a differ-ent P2P architecture to support scalable videostreaming has been implemented. While theNextShare and MMV platforms were both basedon full mesh P2P topology, in SEA the scalablecontent is delivered over a multitree overlay. P2Pplatforms developed in SEA are based on the Vid-Torrent protocol [13], which creates an overlay for-est of independent trees, each one carrying adifferent part of the stream. A central entity (track-er/broker) is used for creating and managing theoverlay. Nodes of a tree structure have well definedparent-child relationships. Such an approach is typ-ically push-based, that is, when a node receives adata packet, it also forwards copies of the packetto each of its children. If peers do not change toooften, tree-based systems require little overhead,since packets are forwarded from node to nodewithout the need for extra messages. However,with respect to mesh topologies in high churn envi-ronments, the tree must be continuously destroyedand rebuilt, which requires considerable controlmessage overhead. Furthermore, nodes mustbuffer data for at least the time required to repairthe tree in order to avoid packet loss.

In the SEA project two different platformshave been developed. The first platform(SEABone) targets code optimization focusingon cross-platform development for set-top boxes.The second platform (SEACast) aims at addingsupport for layered and multiple-descriptive cod-ing schemes [14].

Figure 5. Layered priorities in NextShare.

Quality switch

Quality switch

Time slotV

ideo

qua

lity

t t + x

Time Playback position

Low priority L2 High priority H2 EL2

Low priority L1 High priority H1 EL1

Low priority L0 BL High priority H0

The size of

NextShare chunk has

been set to 56400

bytes in order to

have a multiple

number of MPEG-TS

packets in each

chunk, while preserv-

ing a reasonable size

of a chunk for

efficient transmis-

sion. Here, the size

of each MPEG TS

packet is 188 bytes.



SVC Support in SEACast — The structure ofthe P2P tree generated with the SEACast appli-cation is depicted in Fig. 6.

The content is injected in the tree overlay bya root node that informs the broker about thepublished content. In case of layered video, theroot node also generates a file containing infor-mation about the content structure, in terms ofsupported resolutions and dependencies amongthe layers. Such information is formatted in a so-called Session Description Protocol (SDP) mes-sage [15].

The original VidTorrent protocol was modi-fied to deliver layered contents over multitreeoverlays. That is, for an SVC video, each layer isdelivered over a different tree. The granularityintroduced by splitting the content into multiplelayers allows peers with limited upload band-width to contribute to the swarm by uploadingdata relative to a reduced version of the scalablebitstream. Intrinsic robustness to node failures isadded, as the loss of connections carrying anenhancement layer will cause temporary degra-dation of the overall quality. Clearly, in case offailure or congestion of a connection carryingthe BL, there is no possibility to easily recoverthe quality even if enhancement layers arereceived. In such a situation, the performancemay be improved by assigning priorities to eachdifferent tree according to the importance of theSVC layer carried, and injecting and forwardingthe BL over the most reliable path.

Data Transport in SEACast — In SEACastdata packets are simply forwarded from parent tochildren nodes. As shown in Fig. 6, the publisheris connected to the SEACast root node by means

of a different Real-Time Transport Protocol(RTP) connection for each scalable layer. RTPconnections act also as interfaces between clientnodes and the media player. SVC data are encap-sulated by the publisher in RTP payload. In theSEACast root node, RTP packets are directlypushed over a different tree according to the cor-respondent layer. Each SEACast client keeps abuffer of a few seconds for each tree in which itparticipates; RTP packets are extracted from theclient buffer and again delivered to the mediaplayer over different sessions. The SVC stream isfinally reconstructed inside the media player byaggregating the sub-bitstream, and synchroniza-tion among sessions is kept by using timestampinformation contained in packet headers.

Peer Selection and Adaptation Strategy inSEACast — When a peer wants to join theswarm, it first contacts the broker to receive thelist of neighboring nodes that already have thevideo. The SDP message is also transmitted bythe broker. By parsing the SDP, the peer knowsavailable resolutions and dependencies amongthe layers for the selected video. By matchingthis information with local capabilities and localbandwidth, it selects a target resolution. Accord-ing to the layers hierarchy, the peer pings itspotential parent nodes, starting from the BL andrepeating the operation for ELs. Each SVC layeris therefore retrieved from a different node. Alocal probe technique is used to select the bestparent. In particular, a combination of round-trip time, available bandwidth, and the numberof already active connections is used. Once theparent node is selected for each layer, connec-tions are established. Periodically, available

Figure 6. SEAcast architecture.

Broker

High-end STB and laptop

Video producer

and publisher

SEAcast root node

SEAcast leaf node

SEAcast leaf node

SEAcast leaf node

SEAcast leaf node

SEAcast leaf node

Low-end smartphones

SDP(Session and scalabilityinformation)Data packets BLData packets EL

By parsing the SDP

the peer knows

available resolutions

and dependencies

among the layers for

the selected video.

By matching this

information with

local capabilities and

local bandwidth,

it selects a target

resolution.



bandwidth is checked, and in case of congestionor connection failure, the peer contacts the bro-ker again to rebuild the tree.

CONCLUSIONSIn P2P networks video is streamed to the user ina fully distributed fashion. Network resources aredistributed among users instead of handled by asingle entity. However, due to the diversity ofusers’ displaying devices and available bandwidthlevels in the Internet, the underlying coding andtransmission technology needs to be highly flexi-ble. Such flexibility can easily be achieved bySVC, where bitstreams can be adapted in thecompressed domain according to available band-width or user preferences. When using SVC inP2P streaming, special care needs to be given tohandling different bitstream layers according tothe current playback position. Since it is highlyimportant that pieces close to playback positionarrive on time, these need to be downloadedfrom good peers. In this article we have present-ed several advanced P2P systems supportingstreaming of scalable video and designed to sup-port future Internet applications. Considering theflexibility given by scalable bitstreams within P2Poverlays, it is clear that P2P streaming systemssupporting SVC technology will play an impor-tant role in the Internet of the future.

ACKNOWLEDGMENTThe authors wish to thank Simone Zezza fromthe Department of Electronics, Politecnico diTorino, for his help with revising this manuscript.This research has been partially funded by theEuropean Commission under contract FP7-247688 3DLife, FP7-248474 SARACEN, andFP7-248036 COAST.

REFERENCES[1] H. Schwarz, D. Marpe, and T. Wiegand, “Overview of

the Scalable Video Coding Extension of the H.264/AVCStandard,” IEEE Trans. Circuits Sys. Video Tech., vol. 17,no. 9, Sept. 2007, pp. 1103–20.

[2] M. Mrak et al., “Performance Evidence of Software Pro-posal for Wavelet Video Coding Exploration Group,”Tech. Rep ., ISO/IEC JTC1/SC29/WG11/MPEG2006/M13146, 2006.

[3] Z. Liu, Y. Shen, and K. Ross, “LayerP2P: Using LayeredVideo Chunks in P2P Live Streaming,” IEEE Trans. Multi-media, vol. 11, no. 7, 2009.

[4] P. Baccichet et al., “Low-Delay Peer-to-Peer StreamingUsing Scalable Video Coding,” Proc. Packet Video ‘07,2007, Lausanne, Switzerland, pp. 173–81.

[5] Y. Ding et al., “Peer-to-Peer Video-on-Demand withScalable Video Coding,” Comp. Commun., 2010.

[6] PetaMedia Project; http://www.petamedia.eu.[7] P2PNext Project; http://www.p2pnext.org.[8] SEA Project; http://www.ist-sea.eu.[9] B. Cohen, “Incentives Build Robustness in BitTorrent,”

Proc. 1st Wksp. Economics Peer-to-Peer Sys., 2003.[10] J. A. Pouwelse et al., “Tribler: A Social-Based Peer-to-

Peer System,” 5th Int’l. Wksp. Peer-to-Peer Sys., Feb.2006; http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.60.8696.

[11] J. J. D. Mol et al., “Give-to-Get: Free-Riding-ResilientVideo-on-Demand in P2P Systems,” Proc. SPIE Multime-dia Comp. Net., vol. 6818, San Jose, CA.

[12] S. Asioli, N. Ramzan, and E. Izquierdo, “A Novel Tech-nique for Efficient Peer-To-Peer Scalable Video Trans-mission,” Proc. Euro. Sig. Process. Conf., Aalborg,Denmark, Aug. 23–27, 2010.

[13] VidTorrent Protocol; http://web.media.mit.edu/~vyzo/vidtorrent/index.html.

[14] S. Zezza et al., “SEACAST: A Protocol for P2P VideoStreaming Supporting Multiple Description Coding,” Proc.IEEE Int’l. Conf. Multimedia Expo, New York, NY, 2009.

[15] T. Schierl and S. Wenger, “Signaling Media DecodingDependency in the Session Description Protocol (SDP),”IETF RFC 5583, July 2009; http://tools.ietf.org/html/rfc5583.

BIOGRAPHIESNAEEM RAMZAN ([email protected]) is a post-doctoral researcher at the Multimedia and Vision Group,Queen Mary University of London. His research activitiesfocus around multimedia search and retrieval, image andvideo coding, scalable video coding, surveillance-centriccoding, multimedia transmission over wireless, and P2Pnetworks. Currently, he is a senior researcher and coremember of the technical coordination team in the EUfunded projects PetaMedia and SARACEN. He is the authoror co-author of more than 40 research publications.

EMANUELE QUACCHIO ([email protected]) receivedan M.S. degree in electronic engineering from the Poly-technic University of Turin, Italy, in 2003. He worked twoyears as a researcher in the Department of Electronics ofthe same university and joined STMicroelectronics in 2006.His activities are focused on embedded software develop-ment for STB/mobile platforms and multimedia communi-cation. He has published several papers in the principaljournals of engineering and conferences. Since 2006 he hasparticipated in a number of EU funded projects.

TONI ZGALJIC ([email protected]) received a Ph.D.degree from Queen Mary University of London in 2008. Heis currently a research assistant in the Multimedia andVision Group in the School of Electronic Engineering andComputer Science at the same university. His research inter-ests include scalable video coding and transmission, univer-sal multimedia access, surveillance-centric coding, and videotranscoding. He has published more than 20 technicalpapers in these areas, including chapters in books.

STEFANO ASIOLI ([email protected]) received anM.Sc. degree in telecommunications engineering from theDepartment of Information Engineering and Computer Sci-ence (DISI), University of Trento, Italy, in 2009. He is cur-rently pursuing a Ph.D. degree in electronic engineering atthe Multimedia & Vision Group, Queen Mary University ofLondon. His research interests include peer-to-peer net-works and scalable video coding.

LUCA CELETTO ([email protected]) received a Master’sdegree in electronic engineering from the University of Pado-va, Italy, in 1998. He joined STMicroelectronics in 1999,where he contributed to research projects in video compres-sion and streaming. He has published and co-authored sev-eral papers in the principal journals of engineering andconferences, and has been granted several patents on videotechnologies. As an MPEG committee member, he partici-pated in the standardization of the H.264 specifications. Hehas collaborated on EU-funded projects.

EBROUL IZQUIERDO ([email protected]) isChair of Multimedia and Computer Vision and head of theMultimedia and Vision Group in the School of ElectronicEngineering and Computer Science at Queen Mary Univer-sity of London. He received a Ph.D. from Humboldt Univer-sity, Berlin, Germany. He has been a senior researcher atthe Heinrich-Hertz Institute for Communication Technology,Berlin, and the Department of Electronic Systems Engineer-ing of the University of Essex. He holds several patents inmultimedia signal processing and has published over 400technical papers, including chapters in books.

FABRIZIO SIMONE ROVATI ([email protected]) received hiselectronic engineering degree from Politecnico of Milano in1996. In 1995 he joined STMicroelectronics in the AST Sys-tem R&D group working on digital video processing algo-rithms and architectures. He currently leads the corporateR&D group in the field of networked multimedia. Duringhis career he has authored or co-authored 15 British, Euro-pean, and U.S. granted patents, and 10 publications inconferences or technical journals.

Considering the

flexibility given by

scalable bitstreams

within P2P overlays,

it is clear that P2P

streaming systems

supporting SVC

technology will play

an important role in

the Internet of

the future.



INTRODUCTION

In recent years there has been a trend for moreuser participation in Internet-based applications.There has been an explosion of user-generated,tailored, and reviewed content, while social net-working is beginning to replace traditional com-munications technologies such as email andwebsites. Typical examples of popular applica-tions that only exist for and because of signifi-cant user participation are Facebook, YouTube,Flickr, Digg, eBay, Second Life, and Wikipedia.However, even though content is being created,

modified, and consumed by a large number ofparticipants, almost all of these applications stillrely on servers adequately dimensioned andcarefully positioned by service providers in largedata centers at strategic locations across theInternet to ensure an adequate quality of experi-ence (QoE) for their users. These deploymentsrequire significant investment to maintain; theycannot expand beyond the selected locations andhave limited flexibility to adapt to demand varia-tions over time.

The next generation of applications will con-tinue the trend of user-centricity where users arenot just seen as consumers of a product or ser-vice, but are active participants in providing it.They will be more interactive and distributed,putting prosumers at the center of a massivelymultiparticipant communications environmentwhere they can interact in real time with otherusers and provider resources, to provide andaccess a seamless mixture of live, archived, andbackground material. Furthermore, future net-worked media environments will be high-quality,multisensory, multi-viewpoint, and multi-streamed, relying on HD and 3D video. Theseapplications will place unprecedented demandson networks for high-capacity, low-latency, andlow-loss communication paths between unpre-dictable and arbitrarily large meshes of networkendpoints, distributed around the entire globe,putting additional pressure for upload capacityin access networks.

If the entire burden of supporting high vol-umes of HD/3D multimedia streams is pushed tothe Internet service providers (ISPs) with highlyconcurrent unicast flows, this would requireoperators to upgrade the capacity of their infra-structure by several orders of magnitude toensure end-to-end quality of service betweenarbitrary endpoints.

Rather than simply throwing bandwidth atthe problem, we advocate the development ofintelligent cross-layer techniques that, on onehand, will mobilize network and user resources

ABSTRACT

In recent years there has been a trend formore user participation in Internet-based ser-vices leading to an explosion of user-generated,tailored, and reviewed content and social-net-working-based applications. The next generationof applications will continue this trend and bemore interactive and distributed, putting theprosumers at the center of a massively multipar-ticipant communications environment. Further-more, future networked media environments willbe high-quality, multisensory, multi-viewpointand multistreamed, relying on HD and 3D video.These applications will place unprecedenteddemands on networks between unpredictableand arbitrarily large meshes of network end-points. We advocate the development of intelli-gent cross-layer techniques that, on one hand,will mobilize network and user resources to pro-vide network capacity where it is needed, and,on the other hand, will ensure that the applica-tions adapt themselves and the content they areconveying to available network resources. Thisarticle presents an architecture to enable thislevel of cooperation between the applicationproviders, the users, and the communicationsnetworks so that the quality of experience of theusers of the application is improved and networktraffic optimized.


Bertrand Mathieu and Selim Ellouze, Orange Labs

Nico Schwan, Bell Labs

David Griffin and Eleni Mykoniati, University College London

Toufik Ahmed, University Bordeaux-1

Oriol Ribera Prats, Telefonica I&D

Improving End-to-End QoE via Close Cooperation betweenApplications and ISPs

MATHIEU LAYOUT 2/18/11 3:13 PM Page 136


to provide network capacity where it is needed,and, on the other hand, will ensure that theapplications adapt themselves and the contentthey are conveying to available networkresources, considering core network capacity aswell as the heterogeneity of access network andend device capabilities. Meeting these challengesrequires a previously unseen amount of coopera-tion between application providers, users, andthe communications networks that will transportthe application data. While previous work, main-ly in the telco domain, such as IP MultimediaSubsystem (IMS) [1] or Parlay/X [2] and morerecently ALTO, has made progress in this direc-tion, their use is confined to walled-garden envi-ronments or limited in terms of services theapplications can request. Furthermore, they donot address the specific exchange of informationin both directions so as to enable overlay appli-cations and networks to be optimized. This arti-cle presents an architecture that enables thislevel of cooperation between the actors andelaborates on the related interactions.

The next section presents some exampleapplications to illustrate the benefits of collabo-ration between the applications and the network.We then present an overall picture of ourapproach. We then elaborate on the cross-layerinteractions. The functional architecture whichcaptures the high-level building blocks of oursystem is detailed in the following section. Wethen describe research and standardization ini-tiatives related to our approach. Finally, a sum-mary of the article and a discussion of futurework is presented.

EXAMPLE APPLICATIONSThe environment assumed in this article is onewhere applications are decoupled from underly-ing communications networks and organized asoverlay networks. The application logic will bedistributed over numerous overlay nodes provid-ed by end users, application providers, and eventhe ISPs themselves. The complexity of thenodes and the application logic being executedat each one depends on the sophistication of theapplication. The architecture and cooperativeapproach presented in this article has beendesigned to be of benefit to a wide spectrum ofapplications. In the simplest cases the overlayalgorithms may be streaming live video throughswarms, while more complex applications mayrequire on-the-fly multistream 3D video process-ing or real-time discovery and interactive track-ing of user-generated content. Novel applicationsdelivering rich user experiences will present newbusiness opportunities for the consumer elec-tronics and user applications industries beyondincremental changes to today’s server-centricand web-based content retrieval services. Theinteraction with underlying networks will alsopresent new opportunities for ISPs to cooperatein the delivery of media applications.

We present the following examples to illus-trate the type of advanced services that can ben-efit from increased cooperation with underlyingISPs. Their features include low-latency andhigh-capacity content dissemination; manipula-tion of distributed and streaming content such as

the interpolation of multiple audio-visual streamsfrom different viewpoints; exchange of live, mul-tisensory, and contextual information betweenparticipants; and the discovery and navigation ofdistributed content, information and users. Thecommon denominator is the collaborative pro-duction, processing, and consumption of a mix-ture of live, archived, and backgroundhigh-quality media from multiple sources withdemands than can outstrip the capabilities of theunderlying networks unless the applicationsadapt themselves, and content is tailored to meetnetwork capacity and performance constraints,and/or specific network services such as multicastor in-network caching facilities are provisioned tosupport their efficient distribution.

One example of such an application is themulti-viewpoint coverage of sporting events,such as a bicycle race like the Tour de France(Fig. 1). In this scenario numerous fixed andmobile sources, such as professional media orga-nizations, trackside spectators, as well as thecyclists themselves can generate live audio-visualstreams. As each stream shows a specific view ofa potentially different subject, a potentially largenumber of streams with overlapping content aregenerated. Consumers of this content, who maybe distributed around the globe using variousfixed and mobile end devices, can tailor theirviewing experience by selecting from manystreams according to their preference, or navi-gate between streams in real time to zoom orpan around, or follow particular cyclists. Thepopularity of individual streams is difficult topredict and may change rapidly; therefore, thecreation and adaptation of efficient distributiontrees or meshes to transmit the content to inter-ested sets of users at the required quality levelspresent problems that cannot easily be solved byindividual ISPs or the application overlay in iso-lation. ISPs are unaware of the popularity ofdynamically changing content sources, the loca-tions of the consumers of that content, or theheterogeneous end terminal capabilities inremote networks. From the ISPs’ perspective theapplications are simply generating large quanti-

Figure 1. Bicycle use case.



ties of traffic between unpredictable locations.On the other hand, application logic can trackand match content sources and consumers, butefficient distribution overlays can only be builtwith knowledge of underlying network capabili-ties so that caching and adaptation functions, forexample, can be placed where they are requiredand are most effective, or advanced network ser-vices such as regional multicast distribution canbe invoked where most needed to relieve con-gestion and improve the QoE for users.

A second example is a virtual meeting suchas a 3D virtual conference, where a large num-ber of participants, represented by virtualavatars, can meet and communicate via voiceand avatar gestures, as well as share additionalmultimedia data such as live video, 3D models,text, and presentation slides (Fig. 2). Users willbe both consumers and generators of content —custom avatars, user-generated environments,and sources of interactive video streams. Partici-pants can move around the virtual meetingspace, attend presentations, establish specialinterest discussion groups, socialize in coffeebreaks, and so on. Tracking the participants andmanaging their interests and participation invarious activities is the responsibility of theapplication overlay only; however, the distribu-tion of content to various groups of users issomething that benefits from cooperation withthe underlying networks. Users in the same vir-tual meeting room with a similar point of viewneed access to similar data, such as static back-ground material as well as dynamically changingobjects that need to be synchronized betweenmany consumers. The efficiency of the systemcan therefore be greatly improved by organizingthe overlay with regard to the position of con-tent within the virtual space and making use ofISP-provided network services such as localizedin-network content caching or multicast for dis-tributing state changes of common objects toreduce latency in live updates, network load,and therefore costs.

It can be seen that both of the above exam-ples could generate huge amounts of data to betransmitted to sets of receivers that range fromsmall groups to many hundreds or even millions

of consumers, with some specific constraints ofquality levels such as maximum latency. Our pro-posed solution of increased cooperation betweenISPs and the overlays will assist in exchangingrich information between the application and thenetwork so that the overlays can be organizedefficiently and can adapt to network constraintsby avoiding high cost or highly congested areas,or adapting the quality of the streams to matchavailable network capacity, or enabling specificnetwork services such as multicast distribution orcaching to be invoked in areas of densely popu-lated receivers.

OVERALL APPROACHBecause applications will be more participatoryand interactive, today’s model of centralized orreplicated servers in large data centers is likelyto be replaced by a highly distributed modelwhere processes run in user equipment andinterwork with one another in an overlay layerand can be enhanced via the invocation of net-work services. In our approach, we advocateclose and strong cooperation between ISPs andoverlay applications for optimized delivery ofcontent to end users.

This cooperation is achieved via the compre-hensive, media-aware and open CollaborationInterface between Network and Applications(CINA), which bridges the gap between ISPsand application overlays and aims at:• Increasing the degree of cooperation

between the network layer and the applica-tions through mutual exchange of informa-tion

• Optimizing application overlay networkswith respect to the capabilities of the under-lying networks and the participant endusers

• Providing the means by which service pro-viders can request the activation of special-ized network services or resources toachieve efficient distribution of highlydemanding content streams

• Enabling dynamic adaptation of the contentto meet the abilities of the underlying net-works and user requirementsThe overall picture of the system is illustrated

in Fig. 3, which highlights the CINA interface.The overlay application network consists ofnodes provided by one or more service providers(SPs), the users themselves, and, optionally, ISPnodes. There will be separate overlay networksfor each application. The different applicationsmay be more or less dependent on SP nodes,with peer-to-peer (P2P) applications runningentirely on user nodes being the extreme case.

Given that the applications are global in cov-erage and require end-to-end traffic optimizationinvolving multiple hops in different networks, it isnecessary to collect information from manyunderlying networks via CINA. Since data fromone network may conflict with that provided byanother, or the quantity and quality of the infor-mation may differ from ISP to ISP, the harmo-nization of the information gleaned from theISPs is required in the overlay. The overlay couldalso aggregate the information collected fromdifferent ISPs, with additional data collected by

Figure 2. 3D virtual conference.



measurements of the overlay itself for the globaloptimization of the application. The use of thisinformation will benefit algorithms that are need-ed for optimizing the distribution of the content.These algorithms determine which applicationresources need to be involved, and how to bestinterconnect the participants and distribute theload and the content to achieve the best QoEgiven the available resources.

Also, content adaptation services will profitfrom the cooperation. Until now digital codingand encoding systems have been designed fol-lowing the client/server paradigm, but now appli-cations will have to deal with the fact thatcontent may come from several sources and ter-minal devices with different capabilities, residingin networks that offer different service levels.Applications need to adapt and select quality lay-ers with a brand new set of constraints and cir-cumstances. Pre-adaptation of content usingoffline pregeneration strategies on the subset ofthe most popular content is inefficient and can-not meet the continuously increasing number ofusers’ changing interests and heterogeneous ter-minal capabilities.

Content adaptation has two dimensions: per-sonalizing and tailoring the content for the sub-jective viewpoint of the user(s); and encodingcontent in a flexible way to match the capabili-ties of the network. The latter could also allowthe upload capacity of participants, especiallythose that act as content sources, to be boostedby making use of parallel connections across sev-eral available access networks.

Adaptation is achieved through mechanismsthat dynamically adjust the content via the use ofscalable video coding, video layer coding, oradaptive streaming being performed at thesource or end-user side, or by intermediatenodes as appropriate to match the requirementsof the set of users receiving the content whileadhering to network capabilities and restrictionssuch as congested access links.

CROSS-LAYER INTERACTIONS

The Internet Engineering Task Force (IETF)ALTO working group defined an architectureand a protocol for communication betweenALTO clients and ALTO servers. PossibleALTO information to be exchanged are anabstracted network map and an associated costmap or list of peers ranked according to ISPpreferences. In our study we aim to go further,allowing network operators to provide any kindof network information it wishes or that itaccepts to provide to overlay applications, afteragreement between the two actors. For instance,the ISP could inform the application about thecapabilities of its network: access networks (type,link capabilities, coverage, etc.), current networkservices status (availability of multicast groups,caches, etc.), as well as other possible metrics(load of routers, bandwidth, delay, etc.). Sincesome information is critical for network opera-tors and they do not want to reveal it (e.g., inter-nal detailed topology or BGP policies), theCINA interface is designed having in mind theagreements between the applications and theISP, and is adaptable to allow any kind of agreedinformation to be exchanged.

Furthermore, via the CINA interface, thenetwork operators can also get information fromthe overlay so that they can optimize the trafficin their networks, mobilize resources, and adaptto the overlay applications, eventually transpar-ently; this is not covered by ALTO. Typically,the application could inform the ISP about itstraffic demand: information related to users(e.g., user location and estimated traffic matrix)or content (quantity of sources, their bit rates,adaptive coding, etc.).

This information exchanged between theapplication and the ISP goes further than infor-mation reflecting the preferences and policies ofthe involved business entities as currentlydefined in the ALTO working group.

Figure 3. Overview of the system and its relationship with users, ISPs, and overlay applications.

Application #2 overlay

Server equipmentEnd-user device

Customerpremises

Customerpremises

Serviceprovider

site

Serviceprovider

site

CINA

ISP ISP Cacheserver

ISP

Application #1 overlay

Content adaptation

has two dimensions:

personalizing and

tailoring the content

for the subjective

viewpoint of the

user(s); and

encoding content

in a flexible way

to match the

capabilities of the

network.



CINA also goes further than ALTO in theway that it enables future networked mediaapplications to make use of advanced networkservices in a dynamic and flexible way to achievecost-efficient delivery of high QoE for theirusers. It is known that an ISP could offer infor-mation to applications such as the location ofusers or some user profile information, but inour approach we go further via the offering ofadvanced network services. For example, suchpossible network services can be:• Multicasting: Possibly with hybrid applica-

tion layer and native IP multicast since theapplications will usually be spread over sev-eral ISPs, or the use of high fan-out nodeslocated in the network.

• Caching: Via the use of specialized nodes,provided either by ISPs or third-party enti-ties, to optimize delivery and save band-width in the network.

• Bandwidth on demand: To enable deliverytoward end users over multiple access net-works simultaneously, and provide band-width on demand over aggregated accessnetworks.

• Dynamic quality of service (QoS) mapping:Invocation and mapping of application QoSrequirements, network capabilities, end-user devices, and access networks.

• Ad/text insertion: In order to offer added-value services that might be monetized bynetwork operators.

• Content adaptation: The presence of het-erogeneous end-user devices and networkinfrastructures will require multiple ver-sions of the same resource that can be effi-ciently generated using content adaptation.Our work mainly relates to the open interface

that lets applications request the dynamic activa-tion of such services and does not focus onredefining them. It is the ISP’s responsibility tolist the network services it can offer and makeagreement with an application to decide if it isfor free or provided with appropriate billing.The final decision of activating the serviceremains under the control and management ofthe network operator. No one can force theoverlay applications and network operators tocollaborate, but with the possible win-winapproach of both (optimized QoE for applica-tion, reduction of traffic load, as well as moneti-zation of network services), both actors willbenefit from cooperation.

To illustrate the potential benefits of coopera-tion between the application overlay and theunderlying ISPs, consider the following example.The bicycle race use case introduced earlierdepends on the streaming of live audio-visualstreams from multiple sources to many consumers.The efficiency of the delivery scheme depends onthe popularity of the streams and the distributionof the consumers across the Internet. While low-popularity feeds can be delivered efficiently byunicast streaming techniques, the source nodeneeds significant upload network capacity for thisto scale to a large number of consumers. OverlayP2P swarming techniques reduce the need forhigh-bandwidth links at the source node, but theydo not reduce the load on the underlying network,and can even increase traffic.

However, if the application overlay is able toinvoke network services such as multicast orcaching services through CINA, the load on theunderlying network can be reduced significantly.In addition, there is less overhead in terms ofswarming control operations required at theoverlay.

As an example, the Abilene1 core networktopology of 11 nodes and 14 links was assumedfor the ISP. Assuming equal link metrics, astream bandwidth of S, a single source located atone of the Abilene nodes and stream consumersat each of the other nodes, for unicast streamingthe maximum link load ranges from 4S to 8S andthe total traffic (sum of link load) ranges from21S to 30S depending on the node to which thesource is attached. In all cases the source requiresan upload capacity of 10S. If the overlay uses aP2P swarming distribution scheme such as Bit-Torrent, the source node requires an uploadcapacity of no more than S resulting in a totalload on the network (sum of link load) of approx-imately 26.8S, each peer retrieves an equal frac-tion of the stream from each of the others. If,however, the overlay is able to cooperate withthe ISP through CINA and invoke network-layermulticast to distribute the stream to all con-sumers, the total load on the network is reducedto 11S, a reduction of 58.9 percent of the totaltraffic. If we assume there are even more con-sumers at each of Abilene’s core routers, thetotal traffic on the ISP’s network increases linear-ly with the average number of peers per corerouter to which they are attached, while the loadin the multicast case is not increased.

OVERALL FUNCTIONALARCHITECTURE

In this section we describe the functional archi-tecture defined to model the new interfaces andinteractions required to enable the users, thirdparties, application providers, and ISPs to con-tribute and allocate resources in a dynamic andcoordinated way, allowing the cross-layer opti-mization between the independent applicationand the network processes. The proposed archi-tecture is built in order to define the boundariesof responsibility between the involved actors andthe required interactions that are required acrossthese boundaries.

The ISPs (lower layer blocks in Fig. 4) oper-ate the network and provide Internet connectiv-ity services at particular locations defining theirnetwork domains. Within their domains, theISPs can provide enhanced network services,including multicast and prioritized traffic treat-ment, and they can operate application layerservices such as content caching on behalf of theapplication, or for improving their particulartraffic optimization objectives. On top of thenetwork infrastructure, application providers(middle layer blocks in Fig. 4) operate their ser-vice infrastructure, which may span several loca-tions and the network domains of many ISPs.Application providers may be assisted by third-party providers who would operate additionalinfrastructure at additional locations or providespecialized added-value services, offered to1 http://abilene.internet2.edu/

No one can force

both the overlay

applications and the

network operators to

collaborate, but with

the possible win-win

approach of both

(optimized QoE for

application,

reduction of traffic

load as well as

monetization of net-

work services), both

actors will benefit

from cooperation.



application providers under flexible agreements.Finally, the end users (upper layer block in Fig.4) access the applications and offer theirresources to enhance the application infra-structure in a dynamic way, similar to third-party providers but on a smaller scale andpossibly under the full control of the applicationproviders.

The rest of this section describes the blocksand interfaces that model the functionality andthe interactions between these actors. The inter-actions between the network and the applicationproviders in particular (interfaces between themiddle and bottom layer blocks) define CINA.

The first functional block, end-user applica-tion management, models the functionality at theend user. It includes functions for:• Content generation, consumption, search,

and so on• Data flow handling (e.g., transmission,

reception, synchronization)• Interest and profile management• Providing QoE feedback to the application

At the application level, the overlay manage-ment block includes the application optimizationlogic, and communicates with services control todynamically invoke services and resources wherethey are required, and with data management,which is responsible for maintaining up-to-dateinformation about the network and the application.

The data management block collects, consoli-dates, maintains, and provides network andapplication level information. It allows the appli-cation to access the information provided byISPs regarding network performance and capa-bilities, and the network and application servicesthey may provide, and the ISPs to access appli-cation information regarding the traffic demandand quality requirements at particular locations.

The services control block manages the basicfunctions and value-added services that build theapplication overlay following the instructionsfrom the overlay management block. Such basicservices include data uploading to multiplereceivers, content adaptation, caching, contentpersonalization like picture-in-picture, ad place-ment, and so on. These services could be provid-ed using dedicated application providerequipment, by third-party infrastructure or serviceproviders, or by the end users themselves, con-tributing their own resources to the application.

The overlay management block implementsthe overlay optimization algorithms, taking intoaccount information regarding the particularcontent characteristics and adaptation options,the end-user access means and QoE require-ments, the available network resources andoffered network services (e.g., multicast) and theavailable application services (caching, contentadaptation, etc.). This block decides which QoEis sustainable for the users and how to allocateresources to their service requests. It dynamical-ly invokes resources and services, requestingmulticast transmission within a particular ISP,performing content adaptation for a set of endusers, activating a set of cache servers, and soon. Finally, it interacts with the underlying ISPsto coordinate the use of the network resourcesin a mutually beneficial way (e.g., by reducingthe data rate for a specified region under con-gestion conditions) or to indicate where cachingresources could be allocated to reduce the loadon the network.

At the network level, similar blocks modelthe network optimization, service control, anddata management functionality. In particular,the network data management block maintainsinformation about the network topology, offered

Figure 4. Functional architecture.

D0

C8

C2M2

C4C5

M5

M1

M7

M0

M6

C6

C3

M3

C0 D1D10

C10

C7

C1

M4

End-user application management1

Overlay management

5

Network management

9

Datamanagement

4Servicescontrol

3

Networkdata

management

8

OverlayAAA

Control flowMetadata flowData flow

2

NetworkAAA

7Network services

control

6



network services, ISP policies, and preferencesto be communicated to the application, the sta-tus of the network resources, the current net-work performance, and so on. This blockinteracts with the application to provide infor-mation regarding the particular network domainto the data management block.

The network services control block operatesthe available network services, including multi-cast, caches, ad insertion, and so on. These ser-vices are invoked by the network managementblock to optimize the data distribution as aresponse to explicit requests received by overlaymanagement, or independently to optimize net-work optimization objectives.

The network management block is responsi-ble for the management of the network resourceswithin the ISP network domain. It receives infor-mation about the network from network datamanagement, information about the trafficdemand and quality requirements of differentapplications from data management, explicitrequests for allocation of resources or invocationof services from overlay management, and final-ly, feedback regarding the performance of net-work services from network services control.Based on this information, the network manage-ment block determines the appropriate alloca-tion of resources and invocation of networkservices, and specifies the preferences that arecommunicated to the overlay application.

Finally, the interactions between differentactors create security considerations that neces-sitate the introduction of authentication, autho-rization, and accounting (AAA) functionality.The overlay AAA block handles the authentica-tion of users joining the overlay, ISPs cooperat-ing with the overlay, and potentially third-partyservice providers. The network AAA blockenables the authentication of overlays cooperat-ing with the ISP. Both blocks offer standardAAA functions, such as accounting facilities,access authorization, and profile managementwith security mechanisms, and can be controlledby the overlay management and network man-agement blocks, respectively.

RELATED WORKOverlay applications are currently agnostic of theunderlying network infrastructure and thus per-form end-to-end measurements to gain someknowledge [3], but as this is not in cooperationwith the ISP, it can lead to undermining the rout-ing policies of ISPs [4, 5]. To avoid this, severalinitiatives have promoted cooperation betweenoverlay applications and underlying networks.

The P4P initiative [6] and later the IETFALTO working group have investigated howoverlay networks and ISPs can cooperate tooptimize traffic being generated by P2P applica-tions and transported over the ISP’s infra-structure. In their approach the ISP is able toindicate preferences on which peers shouldexchange data. Our solution is related to ALTO;however, it proposes a much richer interfacethat will allow true cross-layer cooperation, interms of both information exchange and the pos-sibility for overlays to dynamically request ISPs’network services. Since our approach can be

seen as an extension or evolution of ALTO, weinvestigate if solutions defined in this group canbe applied to ours. Typically, the use of a REST-ful HTTP-based interface that uses JSON encod-ing is under consideration.

Some research work, such as that investigatedin [7, 8], aims at building a P2P framework or adelivery platform for live TV. In contrast to ourapproach, they do not consider multi-viewpointapplications and highly interactive applicationsbetween participants in the network; also, [7]does not deal with content adaptation.

Research work studied P2P support in thecontext of massively multiparticipant virtualenvironments such as [9] for virtual environ-ments and [10] for 3D streaming. Those applica-tions are close to the ones we focus on, but inthese solutions the volume of the content israther small compared to HD video from poten-tially a large number of sources we address inour approach. Also dynamic activation of net-work services via the cooperation between over-lay and ISPs for improving QoE is not takeninto consideration.

The definition of interfaces between theunderlying network and the application or con-trol level has been investigated for some yearsnow, and some solutions are deployed such asthe well-known IMS [1]. IMS designs an inter-face between the control entity and the networkentities. However, this interface is still closedand only usable by the ISPs in a walled-gardenfashion. Also, the possible network services thatmight be dynamically activated are limited. Inanother initiative, the Parlay/OSA frameworkwas created some years ago for the telecommu-nications networks and telecom services (callcontrol, call redirection, etc.), and has morerecently been adapted for the Internet (web ser-vices) with the Parlay/X specification [2]. Theobjectives of this group were similar to ours;however, the defined interfaces depend on thenetwork service to be activated, do not permitthe exchange of dynamic information betweenthe network and the application (in both direc-tions), and the supported services were limitedto more traditional telecom applications ratherthan the media applications we aim to support.Furthermore, those two initiatives do not reallyaddress issues for the massive multiparticipantdistributed applications we envision.

CONCLUSIONS AND FUTURE WORKIn this article we have presented a new architec-ture, fostering cooperation between overlayapplications and ISPs for optimized delivery ofservices to end users. Overlay algorithms areoptimized thanks to information provided byISPs, and the delivery QoE is further improvedvia the activation of network services providedby ISPs.

While the approach presented in this articlehas been developed to support interactive, multi-party, high-capacity media applications such asthose presented earlier, the architecture, theinterface, and the principles of cross-layer coop-eration can also benefit existing applications.Content distribution networks (CDNs) providingvideo on demand, for example, could have

Since our approach

can be seen as an

extension or

evolution to ALTO,

we investigate if

solutions defined in

this group can be

applied to ours.

Typically the use of a

RESTful HTTP based

interface that uses

JSON encoding is

under consideration.



greater awareness of network capabilities andalso make use of other network services provid-ed by CINA, such as capacity reservation forbackground distribution of content to CDNnodes.

Ongoing work is focused on the internal func-tions of each block, including the developmentof overlay optimization algorithms, source selec-tion algorithms according to context informa-tion, dynamic activation logic for networkservices such as multicast (and multi-ISP multi-cast), and caching/adaptation functions. Finally,evaluation through both simulation and testbeddeployments is underway.

ACKNOWLEDGMENTSThis work was supported by the ENVISIONproject (http://www.envision-project.org), aresearch project partially funded by the Euro-pean Union’s 7th Framework Program (contractno. 248565). The authors wish to thank all pro-ject participants for their valuable comments andcontributions to the work described in this arti-cle.

REFERENCES[1] 3GPP TS 23.228, “IP Multimedia Subsystem (IMS); Stage

2 (Release 7)”; http://www.3gpp.org.[2] OSA/Parlay X, “3GPP TS 29.199 Release 7 Specifica-

tions,” Sept. 2007.[3] V. Gurbani et al., “ A Survey of Research on the Appli-

cation-Layer Traffic Optimization Problem and the Needfor Layer Cooperation,” IEEE Commun. Mag., vol. 47 ,no. 8, 2009, pp. 107–12.

[4] R. Keralapura et al., “Can ISPs Take the Heat from Over-lay Networks?,” ACM HotNets ‘04, Nov. 15–16, 2004,San Diego, CA.

[5] T. Karagiannis, P. Rodriguez, and K. Papagiannaki,“Should Internet Service Providers Fear Peer-AssistedContent Distribution?,” Internet Measurement Conf.,Oct. 19–21, 2005, Berkeley, CA.

[6] H. Xie et al., “P4P: Explicit Communications for Cooper-ative Control Between P2P and Network Providers,”http://www.dcia.info/ documents/ P4P_Overview.pdf.

[7] R. Fortuna et al., “QoE in Pull Based P2P-TV Systems: Over-lay Topology Design Tradeoffs,” Proc. 10th Int’l. Conf.Peer-to-Peer Comp., Delft, The Netherlands, Aug. 2010.

[8] R. Jimenez, L. E. Eriksson, and B. Knutsson, “P2P-Next:Technical and Legal Challenges”; tslab.ssvl.kth.se

[9] D. Frey et al., “Solipsis: A Decentralized Architecture forVirtual Environments,” 1st Int’l. Wksp. Massively Mul-tiuser Virtual Environments, 2008.

[10] S.-Y. Hu et al., “FloD: A Framework for Peer-to-Peer 3DStreaming,” INFOCOM 2008.

BIOGRAPHIESBERTRAND MATHIEU [SM] ([email protected]) is a senior researcher at France Telecom,Orange Labs since 1994. He received a Diploma of Engi-neering in Toulon, an M.Sc. degree from the University of

Marseille, and a Ph.D. degree from the University Pierre etMarie Curie, Paris. His research activities are related todynamic overlay networks, P2P networks, and information-centric networking. He has contributed to several nationaland European projects, and published more than 30papers in international conferences, journals, and books.He is a member of several conferences’ Technical ProgramCommittees and an SEE Senior Member.

SELIM ELLOUZE received his M. Eng degree in telecommuni-cation from Ecole Nationale Supérieure d'ElectroniqueInformatique et Radiocommunications de Bordeaux, Uni-versité Bordeaux 1. He is currently working toward a Ph.D.degree as a research engineer in Orange Labs Lannion,France. His research interests include the evolution of thecontrol protocols for IP networks.

NICO SCHWAN is a research engineer with Bell Labs ServiceInfrastructure Research Team in Stuttgart, Germany. He isalso an academic at the University of Cooperative Educa-tion Stuttgart. His current research interests are in theareas of content delivery networks, multimedia Internetoverlay applications, and content-centric networking. Hehas a B.Sc. degree in applied computer science from theUniversity of Cooperative Education Stuttgart.

DAVID GRIFFIN is a principal research associate in the Depart-ment of Electronic and Electrical Engineering, UniversityCollege London (UCL). He has a B.Sc. from LoughboroughUniversity and a Ph.D. from UCL, both in electrical engi-neering. His research interests are in the planning, man-agement, and dynamic control for providing QoS inmultiservice networks, P2P networking, and novel routingparadigms for the future Internet.

ELENI MYKONIATI ([email protected]) received a B.Sc.in computer science from Piraeus University, Greece, in1996 and a Ph.D. degree from the National Technical Uni-versity of Athens (NTUA), Greece in 2003. She has workedas a research associate for Telscom S.A. Switzerland, theNTUA DB Lab and Telecom Lab, and Algonet S.A. Greece.Since 2007 she is a senior research associate in the Depart-ment of Electronic and Electrical Engineering at UniversityCollege London. Her research interests include business-driven traffic engineering in IP networks, QoS, and peer-to-peer networking.

TOUFIK AHMED is a professor at ENSEIRB-MATMECA Schoolof Engineers in Institut Polytechnique de Bordeaux (IPB)and performing research activities in CNRS-LaBRI Lab-UMR5800 at University Bordeaux 1. His main research activitiesconcern QoS management and provisioning for multimediawired and wireless networks, media streaming over P2Pnetworks, cross-layer optimization, and end-to-end QoSsignaling protocols. He has also worked on a number ofnational and international projects. He is serving as TPCmember for international conferences including IEEE ICC,IEEE GLOBECOM, and IEEE WCNC. He is currently teamleader of the COMET group at LaBRI Lab.

ORIOL RIBERA PRATS is a business consultant specialized inquantitative market analysis at the Telefonica R&D Centerin Barcelona, Spain. He received his TelecommunicationsEngineer degree from the Universitat Politecnica of Catalo-nia and has had a consolidated career of ten years as prod-uct manager in the mobile Internet industry beginning atNokia, and continuing with co-founding Genaker (a tech-nology start-up) and now at Telefonica.

Ongoing work is

focused on the inter-

nal functions of each

block, including the

development of

overlay optimization

algorithms, source

selection algorithms

according to context

information, dynamic

activation logic for

network services,

such as multicast

(and multi-ISPs

multicast) and

caching/adaptation

functions.



INTRODUCTION

The future media Internet is a research area thataims to improve the mechanism by which userscommunicate using the Internet and their expe-rience. The media Internet supports professionaland novice content producers, and is at thecrossroads of digital multimedia content andInternet technologies. It encompasses two mainaspects: media being delivered through Internetnetworking technologies (including hybrid tech-nologies), and media being generated, con-sumed, shared, and experienced on the web.

Our system and its architecture provide asolution that enhances the delivery, sharing,experience, and exchange by providing methodsto:• Semantically describe contents with a multi-

lingual-multimedia-multidomain ontology

• Annotate the content against this ontology• Process the content and adapt it to the net-

work and the network status, user behavior,and the terminal that is to consume thecontent

• Enrich the content in every process anditeration

• Apply to content-oriented network architec-ture based on standardized interfaces

The system offers several advantages compared toothers as it is able to enrich the content and meta-data in every process by improving the search,retrieval, and content distribution operationssimultaneously, improve the personalization of theresults to the users’ profiles, and take advantage ofcontent-oriented network architectures to workwith content as a native information piece.

This system architecture aims to improve thelogical process dealing with semantic audiovisualcontent search, the development of new tech-nologies for the future media Internet, and theintegration of a broad spectrum of high-qualityservices for innovative search and retrieval appli-cations.

The major novelty of the system lies in thecreation of an architecture consisting of differentmodules around a media component (the multi-media component) that joins together the infor-mation needed to perform the proposedoperations (semantic search, automatic selection,composition of media, etc.), dealing with boththe user context and preferences or smart con-tent adaptation for existing network architec-tures. Within the scope of this media componentis the data storage from all the processesinvolved in the semantic search (as low- andhigh-level descriptors from media inputs), a vec-tor repository from user queries, and a variety ofuser-related data as well.

The outcomes may have extremely highimpact on users, producers, and content cre-ators, which will be able to exploit the givencapabilities for new business models in the futuremedia Internet.

ABSTRACT

This article describes a novel system and itsarchitecture to handle, process, deliver, person-alize, and find digital media, based on continu-ous enrichment of the media objects through theintrinsic operation within a content orientedarchitecture. Our system and its architectureprovide a solution that enhances the delivery,sharing, experience, and exchange by providingthe methods to semantically describe contentswith a multilingual-multimedia-multidomainontology; annotate the content against thisontology; process the content and adapt it to thenetwork and the network status, taking intoaccount the user behavior and the user terminaldevice to consume the content; and enrichingthe content at any additional iteration or pro-cess, over a content-oriented architecture basedon standardized interfaces. The article presentsthe architecture, modules, functionalities, andprocedures, including the system applicationmodel to the future media Internet concepts forcontent-oriented networks.


María Alduán, Faustino Sánchez, Federico Álvarez, David Jiménez, and José Manuel Menéndez, Universi-

dad Politécnica de Madrid

Carolina Cebrecos, Indra Sistemas S.A.

System Architecture for Enriched Semantic Personalized Media Search and Retrieval in theFuture Media Internet

ALVAREZ LAYOUT 2/18/11 3:05 PM Page 144

The remainder of the article is organized asfollows. An overview of the related work is pro-vided in the next section. We then describe thesystem architecture in detail. We explain theapplication of the architecture to the futuremedia Internet, and finally, some conclusions aregiven.

RELATED WORKSeveral systems have been proposed in the pastthat can partially solve the issues of media pro-cessing, handling, distribution, and so on, such asmedia assets management systems or mediadelivery platforms. New architecture proposalsthat go beyond the classical concepts of net-worked media are [1], which proposes a newarchitecture for content-centric networks basedon named content/content chunks instead ofnamed hosts. This new approach decouples con-tent routing/forwarding, location, security, andcontent access, departing from IP in a criticalway. The new trend focuses on the contentinstead of on the hosts where it is stored; in thisway, the new routing of content is achieved byname, not by host address [2–4].

Regarding media content production, thePISA project [5] uses standardized metadatastandards and data models to construct a coher-ently integrated production platform. The differ-ent types of metadata are exchanged betweenthe parts of the system, which makes feasible theimplementation of an entire production work-flow, providing seamless integration between dif-ferent components.

Finally, the Content-Aware Searching

Retrieval and Streaming (COAST) project [6]proposes a future content-centric network (FCN)overlay architecture able to intelligently and effi-ciently link billions of content sources to billionsof content consumers, and offer fast content-aware retrieval, delivery, and streaming.

This related work offers some solutions to themanagement, storage, transmission, or routing ofmedia, but does not offer a framework that canbe adapted to different content-oriented net-works, able to enrich the media and metadata,improving the result of the processes at everystep, and with a user-centric approach to theprocesses (where personalization is applied tothe media objects recommendation, and adapta-tion to the user context and behavior).

SYSTEM ARCHITECTUREThe main target of the system design is to han-dle, process, deliver, personalize, and find digitalmedia, based on continuous enrichment in a con-tent-centric architecture, by means of a three-dimensional ontology, and to adapt multimediacontent to any network technology, device, lan-guage, context, or user (professional or not). Forthis reason, the system consists of the followingmodules, as depicted in Fig. 1: annotation, ontol-ogy, search, personalization, automatic contentgeneration, and context adaptation.

SYSTEM MODULES AND FUNCTIONALITIESThe system and its architecture are composed ofdifferent modules, which perform the aforemen-tioned operations cooperating with each other,and with external systems and user inputs.


Figure 1. Global system architecture.

Non-structuredtext

Text Audio

M3

External

Ontology

Contextcaption

Userprofile Recommender Narrative syntactic

automatic production

Interactive personalizedcontent selection

Automatic generationPersonalization

QoE

Context

Video

Indexation

Multimediacomponent

Fusion

AnnotationSearch

Image Content universe(text, audio,

image and video)

Overontology

Specificquestions

Image-image(query byexample)

Audio-audio(query byexample)

Text

Inpu

t in

terf

ace

Out

put

inte

rfac

eImage

Video

Voice

Audiofile

Distribution

Distribution

Adaptation(network,

terminal, user)

The main target of

the system design is

to handle, process,

deliver, personalize,

and find digital

media, based on

continuous enrich-

ment in a content-

centric architecture,

by means of a

three-dimensional

ontology, and to

adapt multimedia

content to any

network

technology, device,

language, context,

or user.



The annotation module’s main functionality isto allow the classification of multimedia objects(images, video, audio, text, or a combination ofthese), which belong to a so-called Universe ofContents, by automatically or semi-automaticallyproviding descriptive metadata to improve theknowledge about the object. The resulting meta-data can be divided into two levels: low andhigh. Low-level metadata describe the nature ofthe content itself (e.g., in an audio track, theloudness), whereas high-level metadata definethe properties of the content extracted after aspecific processing (e.g., the music genre playedin the audio track). High-level metadata areenriched by the ontology to obtain a semanticdescription (e.g., the audio track is a rock song).If the annotation is only made over each mediatype (text, audio, images, or video content), itcan be ambiguous, and the metadata precision isreduced. This disadvantage can be minimized byusing the existing synergies between the differenttypes of information in media content. To avoidthese problems, the annotation module imple-ments multimodal fusion. Thus, the results ofthe annotation process turn out enriched, morerobust, and with higher accuracy. After metadataextraction, the annotation module performs anindexation of the semantic enriched content tobe accessed by other modules.

The purpose of the ontology is to create for-mal models to represent the multimediaresources, taking into account the multidomain,multimedia, and multilanguage dimensions (Fig.2). The ontology provides the system with asemantic nature, allowing it to store higher-levelfeatures to help to bridge the semantic gap.

The search module includes the search andretrieval tasks, using indexed multimedia objects.This module responds to search queries with anymedia nature: text, audio, video, images, or acombination of them. These queries suffer a dif-ferent processing, using the 3D ontology and dif-ferent semantic processes, depending on theindividual search technique to be applied: overnon-structured text or ontology, based on specif-ic questions and query by example techniques(video and audio). Query by example techniquesallow image to image or audio to audio searchrather than classic text (metadata) searches.Other functionalities of the search module areinterpretation of multilingual queries by voice,

and interactive and hybrid (combination of textand a textual representation of any multimediaelement) search.

As the objective is to refine the search resultsto get the most suitable content according touser expectations, the main target of the person-alization module is to adapt the multimedia con-tent to the specific needs and contexts of users.To achieve this, the module generates user pro-files that are enriched with information explicitlysupplied by the users by modeling their behaviorin the system. Another functionality of this mod-ule is to measure the quality of the user experi-ence, which is also used to enrich the generateduser profiles. The personalization module hashybrid recommendation engines that combinecontent-based filtering techniques and collabora-tive filtering algorithms to increase the efficiencyand solve typical problems derived from theusage of pure filtering techniques. Therefore,the recommendations performed by this modulepresent a great advantage: they consider notonly the user profiles and their context, but alsothe semantic information extracted from themultimedia content.

With the aim to offer an audiovisual summa-ry of a group of multimedia objects that matchesthe users’ interests, the content generation mod-ule automatically generates an audiovisual narra-tive. For this purpose, domain knowledgenarrative discourses’ characteristics are analyzed,and adequate syntactic patterns are established.The efficiency of this audiovisual compendiumresides in good content selection and correctinterpretation of the content. On the other hand,this efficiency depends on syntactic coherenceaccording to the synopsis’s purpose.

The adaptation module’s target is to providemultimedia objects with a mechanism to adaptthe content format. This module adapts the for-mats of the original objects to create audiovisualsummaries with a common format. Besides, themodule plays an important role within the wholesystem due to the fact that content coding is nor-mally network- or/and device-dependent.

In order to get higher-quality final results,previous modules have to work in a cooperativemanner. The element in charge of being the linkamong them is the multimedia component, whichmakes possible the interoperability between themodules to enrich the multimedia objects.

THE MULTIMEDIA COMPONENTThe multimedia component, as can be seen inFig. 3, is composed of four different metadataabstraction layers (identification layer, technicalmetadata layer, descriptive metadata layer, andfunctional metadata layer) to provide flexibility inthe communications between the different mod-ules of the system. Each of these layers is relatedto one or several modules. The objective of thislayered design in the multimedia component isto provide the system with a multidimensionalindex, which allows complex queries with themultimedia component.

The identification layer’s main target is to givea single identifier to each multimedia object.This identifier must be persistent since the objectinformation, including location, may change overtime. The structure design of the multimedia

Figure 2. Multidimensional ontology.

Multilingual

Multidomain

Multimedia

In order to get

higher-quality final

results, previous

modules have to

work in a coopera-

tive manner. The

element in charge of

being the link

among them is the

multimedia compo-

nent, which makes

possible the interop-

erability between the

modules to enrich

the multimedia

objects.



component identifiers is based on the DigitalObject Identifier System (DOI) [7], which is per-sistent, unique, resolvable, interoperable, andvery useful for the management of content overdigital networks in an automated and controlledway. The main advantages of DOI names for thesystem are, on one hand, the difference fromcommonly used Internet pointers to materialsuch as the URL (because they identify an objectas a first-class entity, not only the place wherethe object is located) and, on the other hand,that a DOI is not intended as a replacement forother identifier schemes, although other recog-nized identifiers may be integrated into the DOIname syntax.

The technical metadata layer contains metada-ta that can be directly extracted from the multi-media objects since they are related toproduction features (format, bit rate, etc.).These metadata become especially useful whentransmitting the content through the networkand adapting it to the user terminal or prefer-ences.

The descriptive metadata layer stores two dif-ferent types of metadata: structural and seman-tic. The structural metadata describe the spatialand temporal elements of the multimediaobjects, and the semantic metadata give a high-level description of multimedia elements. Thedefinition of both metadata types is based onthe MPEG-7 standard [8]. The annotationmodule fills in the descriptive metadata layerand uses the ontology to extract the semanticmetadata of the multimedia objects. The col-lection of these descriptors is used by the func-tional blocks (searching , personalization ,generation) of the system to complete theirindividual applications.

The fourth and higher layer of the multime-dia component is the functional metadata layer.It is composed of high-level metadata generat-ed in the functional modules, which are classi-fied in three different categories: narrativemetadata, syntax metadata, and affinity metada-ta. The narrative metadata express narrativecharacteristics of certain multimedia objects,which are deduced from the descriptive meta-data layer. Narrative metadata include accuratedescriptions of the way to articulate a discourseor storytelling through text or a group of pic-tures. These metadata are composed by itemssuch as the narrative facts, characters, narrativespace, narrative time, communicative purpose,or edition (in an audiovisual object). On theother hand, syntax metadata define how toshow the narrative information; for example, ina video object, the number of shots inside ascene, the shot type (close-up, medium shot,long shot), or the style of the scene transitions(cut to, fade in, fade out, cross fade or othereffects). Finally, affinity metadata are generat-ed to model the subjective perception of a user.These type of metadata come from the manipu-lation of the descriptive metadata, and theyhave the objective of establishing affinity rela-tions between users and multimedia objects.For instance, they include the measurement ofthe inherent geometry of a picture or somecomposite rules that can be influenced by theperception of a user.

In the present system design, the multimediacomponent is the interoperable main core of thearchitecture that allows the multimedia objectsto be continuously enriched. For this reason,every module within the system not only worksfor its specific application, but also for the globalobjective of the system, which becomes a wide-ranging tool. In this way, the content generationmodule uses the second-layer descriptive meta-data for its own tasks of generation and produc-tion of new audiovisual objects. Throughoutthese processes, which are necessary to makepossible the creation of original compositions,new metadata are generated so that the objects’narrative and syntax are learned. After this step,the new metadata, obtained in the modeling pro-cedure, are stored in the fourth multimedia com-ponent abstraction layer in order to be used byother system tasks. With these new availabledata, for example, the search module will beable to locate resources departing from narrativeor syntactic information, and users will have thepossibility to use this information to do refinedfiltering of the provided results. On the otherhand, the personalization module recommenda-tion engines can take into account the particularaffinity of users with particular ways of tellingstories or ideas (narrative and syntax). The affin-ity elements are filled by the personalizationmodule, and they can be used, for example, bythe content generation module to create person-alized new audiovisual objects according to thestyle/taste of the user.

USE OF THE ARCHITECTURE FORPERSONALIZED SEARCH AND RETRIEVAL

OF MULTIMEDIA OBJECTS

Due to the flexibility of the proposed architec-ture, there is a large number of use cases wheredifferent system modules can be combined toachieve a common goal; besides, these modulesoffer additional independent applications.

Figure 4 presents the search and multimediaobjects retrieval application as an example of

Figure 3. Multimedia component layer-structure.

Technical metadata layer

Narrativemetadata

Space Time

Syntaxmetadata

Affinitymetadata

Functional metadata layer

Identification layer

Descriptive metadata layer

Structural metadataSemanticmetadata



system functionality. As depicted in the picture,in this process not only the search module takespart, and the search is enriched.

As can be observed in Fig. 4, first the usersends a query to the system by inputting text,image, video, audio, or a combination of those.These input multimedia objects are semanticallyenriched by the ontology, which takes intoaccount the three dimensions (multilanguage,multidomain, and multimedia). Depending onthe nature of the user query, the annotationmodule applies the most suitable metadataextraction procedure; then the search moduleuses the generated metadata and, according totheir nature, starts a specific search process overthe multidimensional index stored in the multi-media component. Depending on the user search

target (the nature of the obtained metadata andthe user preference), the process looks in thesuitable multimedia component layer in order toretrieve the multimedia elements according tothe user search.

Thus, the results of the process are theobjects’ identifiers found by the search methods(metadata and query by example techniques),and a weight factor that indicates their potentialimportance.

Nevertheless, the process is not finished atthis stage. The rest of the modules within thesystem allow the enrichment of the search, aswell as an important increase of its accuracy.The results of the search module are sent tothe personalization module, which combinesthe information of a multimedia object (fromthe multimedia component) with the user infor-mation. In this way, the recommendationengines modify and adapt the weight accuracyindex provided by the search module to theuser preferences. These engines sort the resultsdepending on the user preferences and searchcontext (search query location, time, etc.). Asthe recommendation engines have beendesigned to exploit all available information,the input data does not always need to be thesame. Besides, the personalization moduleenriches and stores the metadata in the multi-media component in a structured way. Theend-to-end personalization process is shown inFig. 5.

At this stage, the system has a list of the mul-timedia objects related to the user search queryand his/her preferences. This list may be adaptedto user permissions depending on the scenario.

In the next step, the content generationmodule enriches both the results and the mul-timedia objects. This module adds a new pieceof content to the previous results, which isautomatically generated using the user prefer-ences and considering the desired narrativeand syntactic characteristics. This process isperformed to coherently generate pieces ofaudiovisual content. This new generatedresource can be seen as a multimedia summaryof the obtained results that emphasizes theuser interests.

Finally, the adaptation module distributes theresults to the user, allowing him/her to consumethe audiovisual content regardless of the currentlocation, network, or device.

The first part of the overall process ends withthe distribution stage. However, it is possible toperform iterative queries to refine the results orhelp the user to extend the search. In this itera-tive procedure, the same modules participateagain.

A point of major importance is the synchro-nization of the different processes. For this rea-son, a handle process has been included in orderto lower the latencies and manage the synchro-nization. As some modules, such as the contentgeneration module which manages a large num-ber of audiovisual sources, can have higherlatencies, the handle process may offer the usera version of the results without the automaticsummarization. In this case, the user is informedabout the possibility of receiving a synopsis whenavailable.Figure 4. Use case example: personalized search and retrieval.

Yes

No

Refinement?

Adaptation

Distribution

Personalization

Contentgeneration

Annotation (video, audio, text or image)

Query sending

Metadataextraction

Search

Multimediacomponent

Ontology

Yes

No

Query byexample?

Ontologyenrichment

Query byexample



APPLICATION OF THE ARCHITECTURETO THE FUTURE MEDIA INTERNET

Internet evolution includes new paradigms tooffer better content delivery services to users.Users give more value to the content, while theInternet, due to its original conception, givesmore value to the localization of that content.For this reason, a major change is needed in theInternet of Contents orientation: from where towhat. This new paradigm is called the content-oriented network (CON), which addresses thebasic needs of Internet design to cope with con-tent as a native element.

The system proposed in this article is intend-ed to work over a CON in a possible design ofthe future media Internet. This CON implemen-tation will allow the proposed system to imple-ment all the functionalities associated with thepreviously described modules at the networklevel, avoiding the development of ad hoc solu-tions to obtain the same results.

Providing media content with searchable andaccessible (metadata generation and structuring)capabilities is one of the major challenges of thefuture Internet. The proposed system becomes apowerful solution due to the architecture andfunctionalities of the described modules, whichprovide improved functionalities for not only thecontent-centric but also the user-centricapproach followed in the personalization andcontent generation modules. In addition, theproposed multimedia component provides anenriched description of objects thanks to the lay-ered metadata structure, which is continuouslyenriched in every process. This layered metadatastructure adds the necessary interaction and isperfectly adaptable to the concept of a CON.

In order to implement our system over aCON, modules can be allocated in the networkcloud. The multimedia component is linked tospecific nodes called content nodes, which allowthe network to access and route both mediaessence and metadata information.

Figure 6 depicts a high-level approximationof the system configuration over a CON. Theapproximation we followed is based on a CONsearch, with the advantage of being protocolagnostic (messages, naming, etc.), and adaptableto some of the content-centric networking(CCN) protocols already proposed [1].

The multimedia search process in Fig. 6 con-sists of six steps:

1) A user starts a search query composed ofone or more multimedia objects (text, image,video or audio).

2) These multimedia objects are annotated bythe annotation module service. The result of theannotation service is a metadata vector with thefeatures of a multimedia object. This vectorforms a new search query.

3) This metadata query is flooded to theCON and will reach every content node. When acontent node receives a metadata search query,it has to perform search algorithms over themultimedia component to find the objects thatsatisfy the user query. The content node returns,as a reply, the metadata of the list of multimediaobjects.

4) The CON router sends this reply to thecontent generation module service. First, thecontent generation module creates a discursiveor narrative structure (depending on the com-municative purpose) using only the descriptiveand functional metadata of the objects (receivedas an entry). Second, once the narrative struc-ture has been created and decided which multi-media objects are necessary to create it, thegeneration module throws a query to get onlythe necessary objects for the final edition.

5) In parallel to step 4, the CON router gen-erates a request to the personalization moduleto obtain a list of recommendations over thesearch results. This module does not need towork with the original multimedia objects; it isenough to take advantage of the descriptive andfunctional metadata, stored in the multimedia

Figure 5. Use case example: detailed personalization process in search andretrieval.

Contextenrichment

User preferencesanalysis

Recommendation (video, audio, text or image)

Search resultsanalysis

QoEmeasurement

User profileenrichment

Context analysis

User profile

Contextontology

Multimediacomponent

User profileenrichment



component, and the user information, stored inthe user profiles. The personalization moduleimplements a hybrid recommender (content-based and social), which is possible because ofthe presence of the multimedia component andthe CON architecture. The multimedia compo-nent allows the existence of a sophisticated con-tent-based recommender module, because alarge amount of information is available forevery multimedia element. The CON contributesknowledge of the multimedia objects users’ con-sumption, due to the fact that these objects canbe unambiguously identified. This knowledgeallows the development of automatic and trans-parent recommendation algorithms based onsocial techniques such as collaborative filtering.Content-based and social-techniques-based algo-rithms make up the hybrid recommender system.

6) The final recommended results and theautomatic summary are sent back to the user.

The search process described is supported bythe previous classification procedures which areperformed through the content nodes providingthe features collected inside the multimediacomponent. Metadata are enriched dynamicallyby the different modules as described in this arti-cle.

CONCLUSIONA system architecture for enriched semantic per-sonalized media search and retrieval that hasbeen adapted to future media Internet conceptshas been described. The system provides a usefulsolution for the future media Internet, applica-ble to both evolutionary or purely content-ori-

ented networks, able to handle, process, deliver,personalize, find, and retrieve digital media.

The advantages of our system go further withthe provision of a complete framework to dealwith media objects in a smarter way by propos-ing a multimedia component centered modulararchitecture that is able to enrich the contentdynamically and improve the performance of themedia processes described. Another meaningfulfeature relevant to the future media Internet isthe user-centric approach. Our system offers anintegral solution that involves the user from thebeginning and is able to personalize media assetsaccording to their needs, tastes, and preferences.

ACKNOWLEDGMENTSThis publication is based on work performed inthe framework of the Spanish national projectBUSCAMEDIA (CEN- 20091026), which is par-tially funded by the CDTI — Ministry for Sci-ence and Innovation and the project AMURA(TEC2009-14219-C03-01). The authors wouldlike to acknowledge the contributions of col-leagues from the partners of the BUSCAME-DIA project.

REFERENCES[1] V. Jacobson et al., “Networking Named Content,” Proc. 5th

ACM CoNEXT ‘09, Rome, Italy, Dec. 1–4, 2009, pp. 1–12.[2] T. Koponen et al., “A Data-Oriented (and Beyond) Net-

work Architecture,” ACM SIGCOMM, 2007.[3] D. Lagutin, K. Visala, and S. Tarkoma, “Publish/Sub-

scribe for Internet: PSIRP Perspective,” Towards theFuture Internet — A European Research Perspective, G.Tselentis et al., Eds., IOS Press, 2010, pp. 75–85.

[4] M. Caesar et al., “ROFL: Routing on Flat Labels,” ACMSIGCOMM, 2006.

Figure 6. The system over a content-oriented network.

Multimodel fusion

Textannotation

Image/video

annotation

Annotation

CONrouter

Contentnode

MC

GenerationPersonalization

Audioannotation

Narrativestructureanalysis

Communicativepatterns

Logiccomposition

Socialtechniques

Content-based

techniques

Metadataanalysis

Hybrid recommendation

Userprofiles

Editionmodule

User n

6

1

33

3

3

54

2

CONrouter

Contentnode

MC

Contentnode

MC



[5] D. V. Rijsselbergen et al., “How Metadata EnablesEnriched File-Based Production Workflows,” SMPTEMotion Imaging J., May/June 2010.

[6] COAST Consortium, “End-to-End Future Content Net-work Specification”; http://www.coast-fp7.eu/public/COAST_D2.2_BM_FF_20100825.pdf.

[7] N. Paskin, “Digital Object Identifier (DOI) System,” Ency-clopedia of Library and Information Sciences, 3rd ed.

[8] MPEG-7 Standard, “ISO/IEC 15938-3:2002, InformationTechnology — Multimedia Content Description Inter-face — Part 3: Visual.”

BIOGRAPHIESMARIA ALDUAN MELLADO ([email protected]) receivedher computer engineer degree in 2009 from CentroPolitécnico Superior, Universidad de Zaragoza. Since 2009she has been a Ph.D. candidate and research assistantwith the Visual Telecommunications Application Group inthe Signals, Systems and Radio Communications Depart-ment of E.T.S. Ingenieros de Telecomunicación, Universi-dad Politecnica de Madrid, where she has beencollaborating since 2008 . She is currently leading thearchitecture research work in the national project BUSCA-MEDIA. She has several publications in national and inter-national conferences.

FAUSTINO SANCHEZ ([email protected]) received his tele-com engineer degree (Hons.) in 2008 and telecom systemsMaster’s degree (Hons.) in 2010, both from the Universi-dad Politécnica de Madrid. Since 2007 he has worked forthe research group of the Visual TelecommunicationsApplications Group (G@TV) of Universidad Politécnica deMadrid. Currently, he is a Ph. D. candidate in the samegroup, and his professional interests include interactivitytechnologies, audience measurement techniques, userbehavior modeling, and recommendation systems. Aboutthis, he has been participating with technical responsibili-ties in several national projects, and he is author and co-author of several papers and scientific contributions ininternational conferences and journals.

FEDERICO ALVAREZ [M‘07] ([email protected]) received hisTelecom Engineer degree (Hons.) in 2003 and Ph.D. degree(cum laude) in 2009, both from Universidad Politécnica deMadrid, where he is currently an assistant professor. Since2003 he has worked for the research group in G@TV ofUniversidad Politécnica de Madrid. He has been participat-ing with different managerial and technical responsibilitiesin several national and EU projects, being the coordinatorof ARENA (IST-024124) and currently the coordinator ofnextMEDIA (ICT-249065), and with a relevant role in pro-jects such as SEA, AWISSENET, and SIMPLE. He had partici-pated in national and international standardization fora(DVB, CENELEC TC206, etc.), is a member of the programcommittees of some scientific conferences and is author

and co-author of 30+ papers and several books, bookchapters, and patents in the field of ICT networks andaudiovisual technologies.

DAVID JIMÉNEZ ([email protected]) received the TelecomEngineer degree (Hons.) in 2004 from Universidad Politéc-nica de Madrid. Since 2002 he has been a member of theSignals, Systems and Radio Communications Departmentof E.T.S. Ingenieros de Telecomunicación, where he is cur-rently a Ph.D. candidate.His professional interests includeimage processing, digital video broadcasting, coding, com-pression formats, and very high resolution and immersiveTV. His Master’s thesis was on the software emulation ofDV format codecs, SMPTE 306M (D7), SMPTE 314M (DV),and SMPTE316M (D9). He joined the European UniversityElite Program of Texas Instruments in order to develop areal-time DSP-based multiformat codec. He has been chair-ing the Standardization Group within Foro Español de AltaDefinición promoted by Ministerio de Industria, Turismo yComercio. Currently he is working on visual quality assess-ment and quality of experience analysis.

JOSÉ MANUEL MENÉNDEZ ([email protected]) is an associ-ate professor (tenured) of signal theory and communica-tions since 1996 at the Signals, Systems, andCommunications Department of E.T.S. Ingenieros de Tele-comunicación, Universidad Politécnica de Madrid. Directorof the Visual Telecommunication Application ResearchGroup (G@TV) of the same university since 2004, he hasmore than 60 international publications about signal pro-cessing and communications, in both international journalsand conferences, and many national publications, includinga book (in Spanish) for the undergraduate engineeringlevel. He has participated or led more than 80 R+D pro-jects, including public funding (Spanish or European) andprivate. He has been a regular reviewer for the IEEE SignalProcessing Society since 2000 for several journals and inter-national conferences, and for IET Image Processing since2009; he also collaborates with Spanish national andregional entities, as well as the European Commission inthe evaluation and review of R+D projects, and with sever-al national telecommunication and broadcasting compa-nies as a consultant.

CAROLINA CEBRECOS DEL CASTILLO ([email protected])obtained her Telecom Engineer degree at Carlos III Univer-sity, Madrid. She has worked for Indra, one of the biggestSpanish companies, focused on investigation and develop-ment of new technologies, since 2008. She works as a con-sultant and has participated in different research projectsin the information technology field, mainly in the audiovi-sual sector, in both national and international environ-ments. She has participated with different technicalresponsibilities in the CBDP European project and Inmer-siveTV national project, among others. She is currentlyleading ‘s responsibilities in the BUSCAMEDIA project.



INTRODUCTION

In the new era of the media Internet where multi-media content is overflowing the networks andusers are usually near a networked device, theavailable forms of communication seem inade-quate. IP has changed the way we communicateand entertain ourselves, using voice over IP (VoIP),videoconferencing, IPTV, email, instant messaging,social networks, and so on. However, now is thetime users really need a completely new user expe-rience that transcends these communication forms.

In this new form of communication, 3D virtu-al environments will be the places for meeting,conversation, and entertainment among friendsand colleagues. Moreover, these places will beconstructed dynamically from each user basedon the exact needs of the communication ses-sion. In order to do so in an easy and intuitiveway, users should have the ability to search,retrieve, and use the deluge of multimedia con-tent available in the networks as building com-ponents of a new 3D environment. However, inthe current Internet architecture search andretrieval of multimedia content is not addressed

sufficiently since the need for such servicesappeared rather late in the design process. Onthe other hand, the new trend of content-awarenetworks aims to solve exactly theses issues byfocusing on the content, not on the host comput-ers and their network addresses. This approachallows the user to retrieve content by its namewithout knowing where this content resides.

The Internet has evolved from a network ofcomputers to a limitless resource of multimediacontent with critical societal and commercialimpacts. This fact led the academic community tostudy different new architectures that put the con-tent in focus. In the 4WARD project a system forflexible, modular network components is exam-ined [1], and an information-centric paradigmfocuses on information objects rather than end-to-end connections. In the ANA frameworkgeneric abstractions of networking protocols arepresented in order to support network hetero-geneity and coexistence of clean slate and legacyInternet technology and protocols [2]. Koponen etal. propose a replacement of DNS with flat, self-certifying files in the DONA architecture [3]. VanJacobson et al. [4] proposed the content-centricnetworking (CCN) approach, where the content isrouted based on hierarchically named compo-nents. The CCN protocol is based on two packettypes, Interest and Data. The consumer transmitsan Interest packet, which is propagated in thenetwork only to the nodes with available content.As a result, Data packets return to the consumerfrom the path the Interest packet passed. Basedon the CCN architecture, Daras et al. [5] pro-posed an extension that introduces similarity mul-timedia content search in such networks. Insearch-enabled CCN, the user is able not only toretrieve content objects by name but also querythe network for similar multimedia content.

At the same time, there are a plethora of aca-demic as well as commercial attempts to build aworkflow that target on virtual 3D environments.Many of them aim at corporate communications,

ABSTRACT

In this article a complete and innovative sys-tem for automatic creation of 3D environmentsfrom multimedia content available in the networkis presented. The core application provides aninterface where the user sketches in 2D the scenethat s/he aims to build. Moreover, the GUI appli-cation exploits the similarity search and retrievalcapabilities of search-enabled content-centricnetworks to fetch 3D models that are similar tothe drawn 2D objects. The retrieved 3D modelsact as the building components for the automati-cally constructed 3D scene. Two CCN-basedapplications are also described, which performthe query routing and similarity search on eachnode of the CCN network.


Theodoros Semertzidis, Informatics and Telematics Institute and Aristotle University of Thessaloniki

Petros Daras, Informatics and Telematics Institute

Paul Moore, Atos Research & Innovation

Lambros Makris, Informatics and Telematics Institute

Michael G. Strintzis, Informatics and Telematics Institute and Aristotle University of Thessaloniki

Automatic Creation of 3D Environmentsfrom a Single Sketch UsingContent-Centric Networks

DARAS LAYOUT 2/18/11 3:07 PM Page 152


events, and meetings, while others aim at e-learning applications or entertainment.

Products such as Assemb’Live, 3Dxplorer,web.alive, and Second Life are some of the com-mercial software available as services or appli-ances. All of them have their advantages anddisadvantages based on the targeted audiencedue to the different background of each firm.Unfortunately, none of them provide a fast andeasy-to-use interface for constructing on the fly a3D environment from the media content thatexists on the network nearby.

Sketch as a user interface has been studied invarious works in recent years, since drawing isone of the primary forms of communicatingideas. Humans tend to draw, with simple featurelines, the objects or places they want to describe.Chen et al. use sketch input along with someword tags to retrieve relevant images from adatabase and montage a novel image [6]. Howev-er, the search queries are in fact word tags, whilethe sketch helps only with the positioning of theobjects and refinement of the search results. ThePhotoSketch system presents another approachfor achieving sketch to image retrieval, based onthe direction of the gradients of the sketchimages [7]. Pu et al. [8] present a sketch-baseduser interface for retrieving 3D computer-aideddesign (CAD) models. Various papers proposedifferent descriptors and methods for extractingfeatures from sketch images for 2D image or 3Dmodel retrieval. However, to the best of ourknowledge, none of them is targeting the cre-ation of 3D worlds using such asymmetricretrieval of 3D models from 2D sketches.

In this work we present a complete workingenvironment for automatic construction of such3D virtual places by exploiting the search-enabled CCN architecture and applying multi-media retrieval principles for the retrieval of 3Dmodels from 2D sketches. The user sketches a2D drawing of the scene s/he wants to construct.The objects, drawn in the 2D sketch, form thequeries dispatched in a CCN to find similar 3Dmodels. The similar media content are retrievedfrom the CCN through a gateway, and, finally, a3D environment is automatically constructed.

The rest of the article is organized as follows.The next section presents the overall systemarchitecture and the subsystems of the proposedframework. We then give a detailed descriptionof the proposed sketchTo3D application. Next,we describe the search components of the CCNnetwork and present the experimental setup.The final section draws conclusions on the cur-rent work and gives insights for the future.

SYSTEM ARCHITECTUREThe main components of the proposed systemare an application named sketchTo3D, whichprovides the sketch user interface and displaysthe 3D virtual environment; the searchGatewayapplication, which acts as a gateway betweenTCP/IP and the CCN network; and finally, thesearchProxy application that runs on everysearch-enabled CCN party in order to perform asimilarity search in their local repository.

Figure 1 presents the overall architecturewith the search-enabled CCN, the CCN searchgateway, and the end user’s PC that hosts thesketchTo3D application. All these componentsare explained in detail in the following sections.

THE SKETCHTO3D APPLICATIONQuerying by example (QBE) is a very commontechnique for querying multimedia databases.However, a prerequisite for the user is to havean object that is similar to what s/he is lookingfor. From this point of view, the QBE approachmay not be that practical for a single-modalitysearch. On the other hand, in a multimodalsetup where objects from one modality arealways available to the user (e.g., sketch) for theformation of queries, the searching in such a sys-tem grants a greatly enhanced user experience.

The sketchTo3D application aims to providethe user with an easy-to-use interface for search-ing 3D models and building a 3D virtual world.

The application is coded in C++ using the QT 4.6library for the graphical user interface (GUI), multi-threading, and networking components. For the ren-dering of the 3D models, the openGL library is used.

Figure 1. The system's architecture.

TCP/IP

CCN search toIP gateway

CCN router

CCN router

CCN router

CCN router

CCN search-enabled party

CCN search-enabled party

CCN search-enabledparties

PC hostingsketchTo3Dapplication

Querying by example

(QBE) is a very com-

mon technique for

querying multimedia

databases. However,

a prerequisite for the

user is to have an

object that is similar

to what s/he is

looking for. From

this point of view,

the QBE approach

may not be that

practical for a single-

modality search.



The sketch area is locked at the center of theGUI, and all other widgets are flexible to bedocked at the left, right, top, or bottom of theapplication allowing the user to build a workingenvironment that fulfills his/her needs (Fig. 2).At the left pane of the GUI (as depicted in Fig.2) there is the basic toolbar with actions such asbrush size or clearing the sketch. The submitbutton is pressed every time the user finishesdrawing an object in order to start the search forsimilar 3D objects. The search process runs on aseparate thread; thus, the user can continuedrawing the next object in his/her scene.

To the right side of the GUI the 3D virtualworld is depicted as it is seen from a single cam-era pointing at (0,0,0) in Cartesian coordinates.The user may translate, rotate, and zoom the vir-tual camera by applying left, middle, or rightclick-and-drag mouse gestures, respectively, onthe widget.

Finally, at the bottom of the GUI the tabulat-ed widget for the presentation of the searchresults is locked. Each search session is separat-ed in a different tab in which the results are pre-sented in ranked lists from the most to leastsimilar objects.

By double clicking on an object in the resultspane, the 3D model is placed in the scene at thecorrect position. When all the results have beenretrieved and the user has selected the appropriate3D models from the ranked lists, the 3D environ-ment is completed. By applying the same mouseactions described above, the user may navigateinside the 3D world and explore the object set.

SUBMITTING A SEARCH QUERYAfter the user clicks on the search button of thetoolbar, a new search session is initiated. Sincethe search has to be conducted on each object ofthe scene alone, a segmentation step has to take

place first. Our segmentation technique makesthe assumption that the user does not draw over-lapping objects. Taking this into account theapplication keeps a history of the sketchedimages between each search. For isolating thenew object from the scene, a simple subtractionof the current sketch image from the previousone is automatically made. Then a boundingsquare of the object is found. If the boundingsquare is larger than 100 × 100 pixels, a scalingstep takes place in order to have a 100 × 100 pix-els image of the object. The procedure is pre-sented in Fig. 3. The scaling of the query imagehelps the descriptors extraction algorithm runfaster, and thus be efficient for real-time usage,without crucially affecting the retrieved results.

After the isolation of the sketched object fromthe scene, a new thread is initiated that handlesall the procedures concerning this search withoutblocking the whole application from workingproperly. The query image extracted from thesegmentation process is fed in the CMVDdescriptor extractor [9], and a descriptor vectorof low-level features of size 212 is computed.This descriptor vector is the actual query foreach search. The next step is to make a requestfor connection to searchGateway using a TCPsocket. When the connection is established andafter a handshake process, the descriptor vectoris sent to searchGateway for searching the CCNnetwork. Finally, the thread enters in an idlestate waiting for the gateway to reply with results.

FROM SKETCH TO 3DFor the insertion of a 3D model into the 3D vir-tual world, three basic parameters should beknown: the positioning of the object in the 3Dspace, the scale of the object with respect to theother objects of the scene, and the orientation ofthe 3D model in order to view it from the right

Figure 2. The GUI of the sketchTo3D application.

The 3Dvirtualworldwidget

The sketch area

Logs pane

The resultspane. Eachsearch hasa separate

tab.

Toolbar withbasic functions

For the insertion of a

3D model into the

3D virtual world,

three basic

parameters should

be known: the

positioning of the

object in the 3D

space, the scale of

the object with

respect to the other

objects of the scene,

and the orientation

of the 3D model in

order to view it from

the right angle.



angle. The following subsections describe ourapproach to these issues.

3D Model Positioning — The position of the3D object in the 3D environment has to beinferred from the position of the object in the2D sketch image as well as with respect to otherobjects in the sketch.

In our approach we assume that all theobjects of the scene are attached to the ground,so the Y coordinate of the 3D world is always Y= 0 (we have no elevation). Thus, by using as areference point the lower-most point of thesketched object, we map the X-Y coordinates ofthe 2D to the X-Z coordinates of the 3D world,as seen in Fig. 4.

A heuristic stretch factor needs to be takeninto account in order to map the height of thesketch image to the required depth in the 3Dworld. The user interface provides the user theability to change the stretch factor and experi-ence a deeper or narrower 3D scene, after theinsertion of the objects in the scene (Fig. 2).

Scale — Taking into account the fact that all the3D models are normalized at the unite sphere asin [9], the ratio of the 2D sketches may be usedto apply the scale in the 3D environment. Indeed,the sketchTo3D application keeps an array of allobjects that are drawn in the 2D sketch and theirrelevant sizes based on the size of the first objectdrawn. Moreover, in order to keep the ratiointact, we calculate the relevant sizes with abounding square of every object at its centroid.Finally, based on this information a scaling factoris applied to every 3D model that is entered inthe 3D virtual environment.

Orientation — The orientation of 3D models isstill an open issue, and much research is con-ducted toward an efficient solution. However,for this application we use a simple solution thatworks efficiently.

We make the assumption that if we have alarge unbiased database of 3D objects and their2D view descriptors, and query this database witha sketch of a 2D view, the most similar resultswould be not only the most similar objects, butalso the exact views of these similar objects.

Moreover, by using the rotation matrix weused for the extraction of the 2D views of eachmodel, we may orient the 3D models in the 3Dworld with respect to the virtual camera coordi-nates. Since the descriptor extraction techniquewe use extracts descriptors for 18 views of each3D model, the application is able to orient the3D model in question in 18 different angles inthe scene [9]. For finer tuning of the orientationof the 3D model, the user interface of theSketchTo3D application allows for manual in-place rotation of the model.

SIMILARITY SEARCH INCONTENT-CENTRIC NETWORKING

The CCN architecture as described in [4] andimplemented in the CCNx project [10] permitsretrieval of content, provided that the consumerknows all or at least a prefix of the name of the

desired content object. Although this design hassome serious advantages, it does not solve theproblem of similar content searching, which isone of the most critical issues for future mediaInternet architectures. Our previous work in [5] isa first attempt to face this issue by proposing asearch protocol as an extension to the CCN archi-tecture. In the current work we use the aforemen-tioned protocol to build a searchProxy daemonthat works on the user space in the Linux system,as well as a gateway that works on the edge of theCCN network in order to interface it with otherapplications over TCP/IP networks.

The search-enabled CCN network as depictedin Fig. 1 has two basic components: the CCNgateway, which is responsible for the intercon-nection of the CCN with the TCP/IP network,and the CCN party, which is a node of the net-work that acts as a consumer and producer ofdata. The CCN search gateway computer shouldhave the searchGateway application and aCCND [10] daemon that is the actual CCNrouter. For the CCN party the applications need-ed are the CCND router, the searchProxy appli-cation that implements the search protocol andconducts the searches for similar objects, and afile proxy application. The file proxy applicationis an implementation of a CCN repository avail-able as a demo application with the CCNx pro-ject distribution [10].

The searchGateway and searchProxy subsys-tems are presented in the following two subsec-tions.

THE SEARCHGATEWAYThe CCN searchGateway is a Java applicationthat runs on a Linux computer and interfacesthe CCN with the TCP/IP network. The basicoperation of the application is to receive search

Figure 3. The steps for the extraction of the query image from the sketch: a) 800× 745 pixels. The current sketch and the image to subtract for the next sketch;b) 800 × 745 pixels. The Isolated object; c) 100 × 100 pixels. The final queryimage the is fed into the CMVD algorithm; d) 800 × 745 pixels. The currentsketch and the image to subtract for the next sketch; e) 800 × 745 pixels. TheIsolated object that resulted by subtraction of image 3a from image 3d; f) 100× 100 pixels. The final query image is fed into the CMVD algorithm.

Sear

ch #

1Se

arch

#2

a) b) c)

d) e) f)



requests from outside the CCN network (viaTCP connections), form the Interest queries,and submit the search Interests to the search-enabled CCN network. If there are similarobjects in the CCN network, the CCN search-Gateway first receives and caches the similarcontent from the CCN parties, and then commu-nicates with the client that started this search tosend the requested content using FTP.

The gateway in idle time listens on a TCPsocket for new clients from the IP side. When anew client requests a connection, a new thread isinitiated that handles the connection. After theconnection is established, a handshake proce-dure is followed to confirm that this is a validclient, and finally, the descriptor vectors fromthe query object are sent to the CCN gateway toform the search Interest. Figure 5 presents themessages exchanged by the CCN gateway andthe sketchTo3D application for the initiation ofa search session.

Upon reception of the descriptors vectorfrom the sketchTo3D client, the searchGatewayforms a search Interest, described in detail in[5]. In short, the search Interest name containsthe descriptor vectors as well as the local nameprefix in order for the CCN parties to refer tothe CCN gateway that expressed the Interest.

While the main loop of the gateway waits fornew search requests, the thread that expressedthe Interest enters in a wait state, waiting for theCCN search-enabled parties to answer with simi-lar multimedia objects. As a result of a success-ful search in a CCN party, a list of content namesis sent to the gateway. The first record of the listrefers to a file containing the ranking of the suc-cessfully retrieved content and the distance ofeach one from the query.

The CCN searchGateway waits for a prede-fined time window for responses and finally usesthe ranked lists from the CCN parties to re-rankthe available content. The re-ranking is based onthe Euclidean distance (L2) of each contentdescriptor vector to the descriptor vectors of thequery object. The result of this procedure is anew ranked list from which the top K most simi-lar 3D models are retrieved and cached in theCCN searchGateway. Moreover, for every filethat is cached in the temporary FTP directory ofthe gateway, a message is sent to the sketch-

To3D application informing it that a result isavailable as well as the ranking of this result.This message is transmitted through the TCPsocket that was initially established for the trans-mission of the query descriptors. As describedabove, the sketchTo3D application gets theresulting 3D model files using the FTP client.

THE SEARCHPROXY APPLICATIONThe searchProxy is also a Java application. Eachsearch-enabled CCN party must have a search-Proxy running in the background in order tosupport the CCN search protocol. A searchProxyinstance is responsible for indexing the contentthat is available on the party’s local repositoryand reply to search queries if similar contentexists in its index.

For each 3D object in the local repository(file proxy) that has to be indexed, CMVD [9]descriptors (212 values for each view, 18 views intotal) are extracted and saved on a localdatabase. The searchProxy application uses a kd-tree indexing structure implementation in orderto organize the records and perform fast exactsearches or nearest neighbor searches. Thesearch process is as follows.

First, a nearest neighbor search is performedto find the 10 most similar records in thedatabase. Since the database consists of thedescriptor vectors of the views of the 3D models,sometimes there are more than one views of thesame 3D model that match the query and appearin the returned list. As a result, in a second steppossible double records of 3D models areremoved from the returned list. Third, theEuclidean distance is calculated between thequery descriptors and each nearest neighbor inorder to have the exact distances. Next, the 3Dobjects with distances greater than a thresholdare also discarded from the results list. Based onthe remaining objects and their distances, aranked list file is created in the local repositorycontaining information such as each object’sname, distance from the query, and the rotationmatrix of the winning view of the 3D object.Then a collection of objects is compiled with thefirst one being the name of the ranked list file.

After the compilation of the collection of 3Dmodels that successfully passed the similaritymatching process, a reply Interest is expressed toadvertise the available results. When this Inter-est reaches the searchGateway, it is on thesearchGateway’s side to collect the desired con-tent from the file proxy that serves these contentobjects as described in the previous subsection.

EXPERIMENTAL SETUPThe experimental setup consisted of one Win-dows PC that hosted the sketchTo3D applicationand two Windows PCs that hosted in total fourvirtual machines running the Ubuntu Linux oper-ating system in order to form the CCN network.

The first virtual box (VB1) worked as both aCCN search gateway and a CCN party. In otherwords, both searchGateway and searchProxyapplications were running on the VB1 virtualmachine. All the other virtual boxes (VB2, VB3,and VB4) played only the CCN party role of thenetwork.

Figure 4. Map X-Y coordinates from the sketch image to X-Z coordinates ofthe 3D world.

Z

X

Y

X

Y



For the 3D models database we used theSHREC 2008 generic models track, which con-tains 1814 3D models of various objects (humans,vehicles, plants and flowers, etc.). The databasewas manually split into four overlapping parts,and each part was stored on a different virtualbox of the CCN network in order to have differ-ent records on the different nodes of the network.However, we introduced a small overlap in orderto test how duplicates would be handled from there-ranking process of the gateway application.

CONCLUSIONS AND FUTURE WORKIn this article we have presented a search andretrieval scheme that uses a single sketch toretrieve 3D models and compile a 3D environ-ment by using the available multimedia contenttraveling in a content-centric network. By expand-ing the content-centric network to support multi-media similarity content search, we provide usersthe ability to retrieve multimedia content alreadyavailable in the network without knowing wherethis content is stored or the name of each con-tent in question. On the other hand, the sketch-based user interface is an intuitive UI thatprovides a greatly enhanced user experience formultimedia content search while helping usersexpress in detail their thoughts and ideas.

For the future, as far as the user interface isconcerned, we plan to rebuild it in a web applica-tion in order to have a wider group of testers andextend the user’s actions available in the currentversion. Also, we are considering inserting real 3Dvideo streams so as to create real-time on-the-fly3D immersive environments from simple sketches.

ACKNOWLEDGMENTSThis work was supported by the EU FP7 project3DLife, ICT-247688.

REFERENCES[1] N. Niebert et al., “The Way 4WARD to the Creation of a

Future Internet,” IEEE 19th PIMRC ‘08, Sept. 15–18,2008, pp. 1–5.

[2] G. Bouabene et al., “The Autonomic Network Architec-ture (ANA),” IEEE JSAC, vol. 28, no. 1, Jan. 2010, pp.4–14.

[3] T. Koponen et al., “A Data-Oriented (and Beyond) Net-work Architecture,” Proc. ACM SIGCOMM ‘07, Aug.2007.

[4] V. Jacobson et al., “Networking Named Content,”CoNext, 2009.

[5] P.Daras et al., “Similarity Content Search in ContentCentric Networks” ACM Multimedia, 2010, Firenze,Italy.

[6] T. Chen et al., “Sketch2Photo: Internet Image Mon-tage,” ACM Trans. Graphics, vol. 28, no. 5, Dec. 2009,pp. 1–10.

[7] M. Eitz et al., “PhotoSketch: A Sketch Based ImageQuery and Compositing System,” ACM SIGGRAPH ’09,Aug. 3–7, 2009, New Orleans, LA.

[8] J. Pu, K. Lou, and K. Ramani, “A 2D Sketch-Based UserInterface for 3D CAD Model Retrieval,” Comp. AidedDesign App., vol. 2, no. 6, 2005, pp. 717–27.

[9] P. Daras and A. Axenopoulos, “A Compact Multi-ViewDescriptor for 3D Object Retrieval,” 7th Int’l. Wksp.Content-Based Multimedia Indexing, 2009, pp. 115–19.

[10] Project CCNx, accessed July 2010; http://www.ccnx.org/.

BIOGRAPHIESTHEODOROS SEMERTZIDIS received a Diploma degree in electri-cal and computer engineering from Democritus Universityof Thrace (2004) and an M.Sc. degree in advanced com-puter and communication systems from Aristotle University

of Thessaloniki, Greece (2009), where he is now a Ph.D.candidate. He has worked for the Informatics and Telemat-ics Institute as a research associate since 2006. His researchinterests include distributed systems, and multimediasearch and retrieval.

PETROS DARAS [M‘07] ([email protected]) is a senior researcher atthe Informatics and Telematics Institute. He received aDiploma degree in electrical and computer engineering, anM.Sc. degree in medical informatics, and a Ph.D. degree inelectrical and computer engineering from Aristotle Univer-sity of Thessaloniki in 1999, 2002, and 2005, respectively.His main research interests include computer vision, searchand retrieval of 3D objects, and medical informatics.

PAUL MOORE is a graduate in computer business systems ofRyerson University, Toronto, Canada, and also holds adegree in economics from the University of Toronto. Hehas more than 20 years of experience in IT systems, includ-ing six years as technical director or coordinator of differ-ent European projects. He is head of the Media unit inAtos Research & Innovation, and is the representative forAtos Origin on the Steering Committee of NEM.

LAMBROS MAKRIS is a research associate at the Informaticsand Telematics Institute, Greece. He received his Diplomaand Ph.D. in electrical engineering from Aristotle Universityof Thessaloniki in 1994 and 2007, respectively. His researchinterests include applications of local area and wide areanetworks, distributed information systems, databases, elec-tronic commerce, data security, and encryption.

MICHAEL GERASSIMOS STRINTZIS [M’70, SM’80, F‘04] received aDiploma degree in electrical engineering from the NationalTechnical University of Athens, Greece, in 1967, and M.A.and Ph.D. degrees in electrical engineering from PrincetonUniversity, New Jersey, in 1969 and 1970, respectively. Heis a professor of electrical and computer engineering at theUniversity of Thessaloniki. He has served as Associate Editorfor IEEE Transactions on Circuits and Systems for VideoTechnology since 1999. In 1984 he was awarded one ofthe Centennial Medals of the IEEE.

Figure 5. Handshake messages between the CCN gateway and a sketchTo3Dclient submitting query descriptors.

t

CCN gateway

“ccnGateway”

“ACK”

“ACK”

Descriptors vector

“sid”

sid_number

“desc”

sketchTo3D


158

SERIES EDITORIAL

The second article presents a large-scale testbed, Panlab, forfuture Internet applications. Panlab is a testbed to pilot futureInternet applications over the existing Internet. It is a distributedtestbed with platform resources contributed by the Panlab part-ners, coordinated and configured by the Panlab office, and offeringtesting services to Panlab customers to test deployed applicationsor their control modules. With the shared Panlab testbed, pilotinga new application becomes easier than constructing a global-scaletestbed of one’s own to experiment with an application. The articleillustrates how to use Panlab through a case study on testing adap-tive admission control and resource allocation algorithms for weband database applications, where traffic generators are configuredat two testbeds, and the web and database applications, along withthe algorithms under test, reside at a third testbed.

BIOGRAPHIESYING-DAR LIN () is a professor of computer science at National Chiao Tung Uni-versity, Taiwan. He received his Ph.D. in computer science from the Universityof California at Los Angeles in 1993. He spent his sabbatical year, 2007–2008,as a visiting scholar at Cisco Systems, San Jose, California. Since 2002 he hasbeen the founder and director of the Network Benchmarking Laboratory(NBL, www.nbl.org.tw), which reviews network products with real traffic. Healso cofounded L7 Networks Inc. in 2002, which was later acquired by D-LinkCorp. His research interests include design, analysis, implementation, andbenchmarking of network protocols and algorithms, quality of service, net-work security, deep packet inspection, P2P networking, and embedded hard-ware/software co-design. His work on multihop cellular has been cited over500 times. He is currently on the editorial boards of IEEE CommunicationsMagazine, IEEE Communications Surveys and Tutorials, IEEE CommunicationsLetters, Computer Communications, and Computer Networks.

ERICA JOHNSON is the director of the University of New Hampshire InterOper-ability Laboratory. In this role, she manages and oversees over 20 differentdata networking and storage technologies providing all aspects of adminis-tration, including coordination of high profile testing events, coordinationwith different consortiums, and working with various industry forums. She isalso a prominent member of organizations both internally and externally, Sheenjoys a powerful mix of technology and business related activities. At theUniversity of New Hampshire she participates in the UNH Steering Committeefor Information Technology, Senior Vice Provost for Research Working Group,and Computer Science Advisory Board. In the industry she was appointedtechnical representative of North America for the IPv6 Ready Logo Committeeand was also chosen to be an IPv6 Forum Fellow. Passionate about the labo-ratory and its possibilities, she continues to work with many industry forums,commercial service providers, network equipment vendors, and other univer-sities in order to further the InterOperability Laboratory’s mission.

EDUARDO JOO is software project leader at Empirix, Inc., Bedford, Mas-sachusetts. He received his M.S. in computer system engineering, computercommunications and networks, from Boston University in 2006. He joinedEmpirix, Inc., in 2001 and has led the successful development of networktesting and emulation systems, including PacketSphere Network Emulator,PacketSphere RealStreamer, Hammer NxT, and Hammer G5. He is currentlyleading the development of next-generation mobile broadband data net-work monitoring and testing tools. His areas of interest include voice anddata protocols, and wired, wireless, and mobile network communications.

he objective of the Network Testing Series of IEEE Commu-nications Magazine is to provide a forum across academia and

industry to address the design and implementation defects unveiledby network testing. In the industry, testing has been a mean to eval-uate the design and implementation of a system. But in academia, amore common practice is to evaluate a design by mathematical anal-ysis or simulation without actual implementations. A less commonpractice is to evaluate a design by testing a partial implementation.That is, academia focuses more deeply on algorithmic design evalua-tion, while industry has broader concerns on both algorithmic designand system implementation issues. Often an optimized algorithmiccomponent cannot guarantee optimal operation of the whole systemwhen other components throttle the overall performance.

This series thus serves as a forum to bridge the gap, where thedesign or implementation defects found by either community canbe referred to by another community. The defects could be foundin various dimensions of testing. The type of testing could be func-tionality, performance, conformance, interoperability, and stabilityof the systems under test (SUT) in the laboratory or field. TheSUT could be black-box without source code or binary code, grey-box with binary code or interface, or white-box with source code.For grey-box or white-box testing, profiling would help identify anddiagnose system bottlenecks. For black-box testing, benchmarkingdevices of the same class could reflect the state of the art. The SUTcan range from link-layer systems such as Ethernet, WLAN,WiMAX, third-/fourth generation (3G/4G) cellular, and digital sub-scriber line (xDSL) to mid-layer switches and routers, and upper-layer systems such as voice over IP, Session Initiation Protocol(SIP) signaling, multimedia, network security, and consumerdevices such as handhelds, to even a large-scale network system.

Our first call received nine submissions, and we selected two ofthem. The selection is based on several factors: how relevant to networktesting the work is, how new the test methodology or result is, how infor-mative the article is, and so on. The selected two happen to representtwo extremes in terms of scale, with the first on a specific issue of a net-work component and the second on a large-scale distributed testbed.Both articles have authors at universities or research organizations. Weshall continue to solicit contributions from industry, although industrypeople usually do not have publication as their priority job.

The first article answers the question of how adjacent channelinterference (ACI) affects the performance of IEEE 802.11a,which was believed to be ACI-free due to better channelizationcombined with orthogonal frequency-division multiplexed(OFDM) transmissions. They present a modified model for thesignal-to-interference-plus-noise ratio (SINR) and quantify thethroughput degradation due to ACI. The model is verified by theemulated wireless medium with cables and attenuators to isolatethe affected 802.11 mechanisms. The result implies that even witha large number of channels in 802.11a, careful channel selection isstill needed in order to achieve higher throughput.


T

TOPICS IN NETWORK TESTING

Ying-Dar Lin Erica Johnson Eduardo Joo

LYT-SERIES EDIT-Lin 2/22/11 5:07 PM Page 158


1 A transmit spectral maskis the power contained ina specified frequencybandwidth at certain off-sets relative to the totalcarrier power.

INTRODUCTION

The IEEE 802.11a standard amendmentdescribes an OFDM-based physical layer for802.11 wireless stations operation in the 5 GHzband. Due to the poor channelization of802.11b/g that left only three of the availablechannels non-overlapping, the channelizationscheme of 802.11a was over-advertised to offer19 non-overlapping channels in the EuropeanTelecommunications Standards Institute (ETSI)regulatory domain and 20 in the Federal Com-munications Commission (FCC) domain. Thisimplied that no adjacent channel interference(ACI) was to be expected in 802.11a, and there-fore no performance degradation would be

observed for neighboring links operating inneighboring channels. Indeed, the 52 orthogonalfrequency-division multiplexing (OFDM) subcar-riers defined in the 802.11a amendment appearto lie well within the channel bandwidth of 20MHz, and each channel’s central frequency has aspacing of 20 MHz from the next/previous adja-cent channel. Still, examining the transmit spec-tral mask1 required for compliance in thespecification [2], we see that some transmittedpower is allowed to leak not only to the immedi-ately adjacent channels, but also as far as twochannels away from the communication channel.

Meanwhile, the wireless network researchcommunity endorsed 802.11a as the standard ofchoice for multiradio nodes and dense wirelessLAN (WLAN) deployments. Two main reasonswere behind this: First the 2.4 GHz band hadbeen already overcrowded as the 802.11b/g com-pliant devices had been in the market longbefore the 802.11a ones and second because itwas widely believed that 5 GHz capacity prob-lems due to interference would be mitigated bythe non-overlapping channels promised by thestandard and the vendors. Unfortunately,although the power allowed to leak into theneighboring channels is indeed quite low com-pared to the transmitted signal power, it is suffi-cient to cause ACI effects, especially whenneighboring radio interfaces use nearby chan-nels, or when the signal-to-interference-plus-noise ratio (SINR) observed at the receiver ofnode is marginally larger than the thresholdrequired to support a required rate.

RELATED WORKThe authors of [3] performed experiments on atestbed with Atheros-based 802.11a interfaces toexamine the effect of potential ACI on a dual-radio multihop network. Their work includesboth laboratory and outdoor experiments usingomnidirectional antennas. The former indicatedthat the Atheros AR5213A-chipset interfacesthey employed were indeed compliant with the

ABSTRACT

Wireless LAN radio interfaces based on theIEEE 802.11a standard have lately found wide-spread use in many wireless applications. A keyreason for this was that although the predeces-sor, IEEE 802.11b/g, had a poor channelizationscheme, which resulted in strangling adjacentchannel interference (ACI), 802.11a was widelybelieved to be ACI-free due to a better channel-ization combined with OFDM transmission. Weshow that this is not the case. ACI does exist in802.11a, and we can quantify its magnitude andpredict its results. For this, we present minormodifications of a simple model originally intro-duced by [1] that allow us to calculate boundingvalues of the 802.11a ACI, which can be used inlink budget calculations. Using a laboratorytestbed, we verify the estimations of the model,performing experiments designed to isolate theaffected 802.11 mechanisms. This isolation wasenabled by not using the wireless medium, andemulating it over cables and attenuators. Ourresults show clear throughput degradationbecause of ACI in 802.11a, the magnitude ofwhich depends on the interfering data rates,packet sizes, and utilization of the medium.

TOPICS IN NETWORK TESTING

Vangelis Angelakis, Linköping University

Stefanos Papadakis, Foundation for Research and Technology — Hellas (FORTH)

Vasilios A. Siris, FORTH and Athens University of Economics and Business

Apostolos Traganitis, FORTH and University of Crete

Adjacent Channel Interference in802.11a Is Harmful: Testbed Validationof a Simple Quantification Model

ANGELAKIS LAYOUT 2/18/11 3:04 PM Page 160


spectral requirements of the 802.11a specifica-tion. Their testbed was based on a single boardLinux-based PC that hosted two interfaces andused the opensource MadWifi driver. The out-doors experiments in that work were the first toprovide evidence of ACI. They report observingno board crosstalk or interference other thanthat caused by operating neighboring links onadjacent channels. They were the first to suggestincreasing channel separation and antenna dis-tance as well as using directional antennas inorder to mitigate the effects of 802.11a ACI onthe reduction of throughput. These were the firstreports of 802.11a ACI, which, however, did notinclude any insight or attempts at some solidhypothesis as to why this ACI exists. However,ACI effects were clearly demonstrated, leavingno doubts about the existence of ACI in 802.11a.

In [1] the authors introduced a simple modelto theoretically quantify ACI caused by overlapsin neighboring channels. Their key idea wasfocused on taking an integral over the wholeoverlapping region of the interfering channelsspectral masks. They applied it to the spectralmasks of 802.11b/g which have had known over-lap issues due to poor channelization design, andalso that of 802.16. They claim that the use ofpartially overlapped channels is not harmful,provided that higher layers take it into consider-ation and adapt accordingly. Furthermore theyshow that a careful use of some partially over-lapped channels can often lead to significantimprovements in spectrum utilization and appli-cation performance, with respect to the interfer-ing nodes’ distances.

In [4, 5] we introduced minor modificationsto the limits of the integral used for the ACIquantification model introduced by [1] and werethe first to apply it to the 802.11a spectral mask;we produced results on a testbed where thewireless channel was emulated using attenuators,and on a testbed with real outdoor mid-rangewireless links using directional antennas. Thosetwo works verified:• Our hypothesis that ACI observed in

802.11a is caused by the overlap of thechannel sidelobes allowed by the IEEEspecifications

• That ACI can be caused by channels thatare not only directly adjacent

• That ACI can be harmful if not taken intoaccount during system and resource plan-ningThe testbeds in all the above papers used

Atheros-based wireless interfaces and the open-source MadWifi driver. This choice was madeprimarily because MadWifi was at the time thede facto reference driver for the vast majority oftestbeds in the literature. Since then the Mad-Wifi driver has been rendered obsolete, declaredlegacy, and is no longer supported by the Linuxkernel.

In this work we provide new evidence that802.11a ACI can be quantified and its effectspredicted. In particular, we first demonstrate theexistence of 802.11a ACI on a testbed withAtheros-based interfaces driven by the newlydeveloped ath5k open source driver. Second, wequantify the ACI effect in terms of goodput,completely isolating the medium access control

(MAC) and physical (PHY) layer mechanismsthat are susceptible to its effect, using a wirelesslink emulation testbed.

SYSTEM MODEL/ACIQUANTIFICATION

The SINR criterion for data reception (Eq. 1)requires that the signal of interest power arrivingat a receiver, over the sum of the interferenceand the thermal noise powers must be above athreshold which is defined with respect to thetransmission parameters (modulation scheme)and the quality of service requirements (datarate of transmission and reception bit error rate[BER]). In Eq. 1 we assume k interfering trans-mitters operating on the same channel as thesignal of interest transmitter, with powers Pi andPtx, respectively.

(1)

Typically, the SINR criterion is applied, as inEq. 1, in single-channel systems, where the inter-fering transmissions are assumed to occupy theentire bandwidth of the used channel and areconsidered noise. In a channelization schemewhere more than one channels are used withsome partial overlap on their bandwidth [1], anACI factor Xi,rx is introduced for each of theinterferers, which can be used in the SINR cal-culations. This factor depends on the spectralproperties of the channels and the transmittedsignals, and the separation between the channelsof an interferer i and the receiver rx. Specifically,the affecting properties are the interchannel spec-tral distance, channel bandwidth, spectral mask,and receiver filter. This factor takes values in [0,1], with 0 indicating no overlap (i.e., completeorthogonality) and 1 indicating that the interfer-er is using the same channel as the receiver. Forour work we calculate this interference factor bynormalizing the spectral mask S(f) within a fre-quency width w that should be at least equal tothe nominal channel width, and then filter thisnormalized S′(f) over the frequencies that will bewithin the bandpass filter of the receiver. Ideally,for the case of 802.11a, the spectral mask shouldbe a flat bandpass 20 MHz filter, but for thesake of being more realistic we assume that theinterfaces employed use a single imperfect,wider than nominal, bandpass filter both fortransmission and reception. In the general casewe could use Eq. 2 to obtain the factor Xi,rx foran interferer i and a receiver rx, as a function ofR′(f), the normalized receiver filter transferfunction in [–w/2, w/2];

(2)

where we have denoted fint the frequency offset atwhich the interfering channel is centered (Fig. 1).

x R f S f f dfi rx iw

w

, int ,= ′( ) ′ −( )−∫2

2

P PathLoss

N P PathLoss

tx tx rx

rx i i rxi

⋅

+ ⋅

( )

( )=

,

,1

kk SINR

∑≥ θ

Due to the poor

channelization of

802.11b/g that left

only three of the

available channels to

be non-overlapping,

the channelization

scheme of 802.11a

was over-advertised

to offer 19 non-over-

lapping channels in

the ETSI regulatory

domain and 20 in

the FCC domain.



Finally, in a system where all radio interfacesadhere to the same protocol, it is reasonable toassume that all nodes have the same S(f); andfurthermore, that this output filter matches thereceiver filter, and so: S(⋅) = R(⋅). Under thesetwo assumptions, Eq. 2 becomes

(3)

where we have denoted fint the frequency offsetat which the interfering channel is centered (Fig.1).

Using this model and the spectral mask for802.11a mandated by the standard in [2], we cal-culated the maximum compliant power leakagebetween two neighboring 802.11a channels.Table 1 shows the results of our calculations ofthe ACI Xi,rx factor expressed in dB (essentiallythe attenuation of the transmitted power due tothe frequency offset) that may be used directlyin any link budget calculation.

Table 1 indicates that the interference factorX is sufficient for a single transmitter to injectACI power at a receiver which will be well abovethe thermal noise in an 802.11a system underconditions enabled by proximity or power alloca-tion, even if the interfering transmitter wereusing the next adjacent channel to that of thereceiver. For example, assuming the typical ther-mal noise of –101 dBm and a 20 MHz 802.11achannel centered at 5600 MHz (channel 120)and zero antenna gains, an interferer on an adja-cent channel (say channel 124 at 5620 MHz)transmitting at only 1 mW (0 dBm) withinapproximately 40 m of the receiver would bereceived above noise, reducing the perceivedSINR by at least 3 dB within that range.

Therefore, because of the channel design inIEEE 802.11a, ACI will be observed and, if notproperly considered, will cause degradation of asystem’s performance. We must note, though,

that our calculations in Table 1 use the spectralmask mandated by the standard, which is anenvelope for the actual implementations as ven-dors compete to achieve better specifications fortheir cards.

In order to experimentally verify our calcula-tions we developed a testbed with off-the-shelfequipment. As in our previous work, we chose toemulate the wireless medium rather than usingthe air in order to remove its non-deterministiccharacteristics, avoid unknown interference, andeliminate the inherent wireless medium uncer-tainty from our investigation. This led us to a lab-oratory testbed where nodes’ antenna connectorswere interconnected using coaxial cables, attenua-tors, signal splitters, and combiners. We separatedthe MAC and PHY mechanisms that affect theefficiency of the protocol in the presence of ACIin order to obtain bounds for the worse cases.

THE COMMUNICATION MECHANISMSAFFECTED BY ACI IN 802.11

DATA RECEPTION MECHANISM ERRORS

Assuming that the interference caused by802.11a stations can be modeled as white Gaus-sian noise, we can determine whether the SINRrequirements for the 802.11a transmission rates,given in the specifications for 10 percent packeterror rate, can be met under the presence ofACI. Since each interfering node produces someACI at the receiver, interesting interface topolo-gies can be observed, arising by poor systemdesign, where the total ACI will bring the SINRat the receiver below the threshold.

Multiradio Nodes — In a multiradio node,assignment of neighboring channels on inter-faces that have their antennas close together hasbeen shown to cause reduced performance [3].In such a scenario the interference arriving at areceiver can be sufficiently high to be harmfuldue to proximity, which causes low interferencepath losses.

Single-Radio Nodes — In dense topologieswhere channel allocation may inevitably providenearby links of adjacent channels, if the pathlosses to some receivers are high, or the numberof concurrent interfering transmissions is large,their aggregate power may be high enough tobring the SINR below threshold.

CLEAR CHANNEL ASSESSMENTFALSE NEGATIVES

IEEE 802.11 employs a distributed coordinationfunction (DCF), which essentially is a carriersense multiple access with collision avoidance(CSMA/CA) MAC protocol with binary expo-nential backoff. The DCF defines a basic accessmechanism and an optional request-to-send/clear-to-send (RTS/CTS) mechanism. Letus consider just the basic access mechanism. Inthe DCF a station has to sense the channel asclear (i.e., idle) for at least a duration of DIFS +CWmin (both defined in [6]) in order to gainaccess to it. The 802.11a standard requires that a

x S f S f f dfi rxw

w

, int ,= ′( ) ′ −( )−∫2

2

Figure 1. Graphical representation of the calculation of Eq. 2.

S‘(f-fint)S‘(f)

w/2fint0-w/2

Table 1. Theoretically calculated Xi,rx in dB.

Receiverbandwidth

Immediately adjacentchannel power leakage

Xi,(i±1)

Next adjacent channelpower leakage

Xi,(i±2)

20 MHz –22.04 –39.67

∞ –19.05 –36.67



clear channel assessment (CCA) mechanism beprovided by the PHY layer. The CCA mecha-nism that will provide this information is to pro-claim a channel as busy when it decodes a PHYlayer preamble at a power at least equal to thatof the basic rate of the 6 Mb/s sensitivity,2 ordetects any signal with power 20 dB above thebasic transmission rate 6 Mb/s sensitivity.

Interference can cause the CCA to misreportin the case of nearby located interfaces. A chan-nel may be sensed as busy due to high receivedpower from a neighboring channel that is inter-fering. This can occur when two nearby 802.11atransmitters contend over different channels,such as in a poorly designed multiradio node; forexample, in a multiradio mesh node that has twoor more interfaces using nearby channels, withomni- or directional antennas, and with insuffi-cient spatial separation, or EM shieldingbetween them.

EXPERIMENTAL VERIFICATION

TESTBED DESCRIPTIONThe testbed of our experiments consisted of fournodes interconnected using cables and attenua-tors, eliminating the unpredictable wirelessmedium and thus fully controlling the transmis-sion and reception paths and losses.

Each node is an EPIA SP13000 mini-ITXmotherboard with a 1.3 GHz C3 CPU and 512Mbytes RAM, running Gentoo Linux with thewireless testing tree kernel v.2.6.31-rc8-wl. With aRouterBoard miniPCI to PCI adapter, an AtherosAR5213A chipset miniPCI wireless interface card(CM9-GP) was used on each node, running onthe ath5k 802.11a/b/g driver and hostapd v0.6.8.The ath5k driver was modified to independentlyuse the two antenna connectors of the wirelesscard, one only for transmission and the other onlyfor reception. This was the primary key enablerfor the design of the experiments conducted.

We also used the AirMagnet Laptop Analyz-er v6.1 software to monitor the wireless trafficand a Rohde & Schwarz FSH6 spectrum analyz-er for channel power and bandwidth verification.The interconnectivity of the nodes was routedthrough coaxial cables, four-way HyperLinkTech splitters/combiners, Agilent’s fixed attenua-tors (of 3, 6, 10, 20, and 50 dB), and pro-grammable attenuators by Aeroflex/Weinschelwith 0 to 55.75 dB attenuation range and stepsof 0.25 dB. For the traffic generation andthroughput measurements, we used the iperfv2.0.4 with pthreads enabled.

Before any measurement a bootstrap proce-dure was followed where the wireless interfacewas completely reset before applying the newsettings, as a precaution catering to the unstablenature of the ath5k driver, as our experience hasshown. We generated UDP traffic both in theinterfering link and the link under test, to avoidthe flow control mechanism of TCP and thus getresults for the maximum goodput at the receiver.

EXPERIMENTS’ SETUPWe set up just two links to realize the scenariosof Fig. 2. One is the test link (link, Fig. 3a) andthe second is the interference link (interferer,Fig. 3a) to be tuned at a channel neighboring

the one used first. With these two links we wereable to generate the topologies of Fig. 2 andconducted two experiments. To avoid confusion,for both the link and interferer we use the termssource and destination for the nodes that pro-duce and consume the iperf traffic, respectively.Note that the interferer was made completelyunaware of the link by proper power assignmentat the link’s sender transmitter, and the lossesand isolation along the paths leading to both thereceivers in the interferer.

First we tested the ACI effect on the datareception mechanism. To do this, we injectedthe traffic from the interferer’s sender to thereceiving connector of the link destination.Using the values of Table 1, we calculated thetransmission power required for the interferenceand the attenuator values in order to bring theSINR at the receiver below threshold.

In the second experiment set the effect ofACI on the CCA mechanism was examined. Forthis the testbed interconnection was slightlyaltered so that the interferer’s sender was cou-pled with the receiver of the link’s sender. Thevalues for the attenuators again were calculatedusing Table 1 and taking in mind the CCArequirements.

HARDWARE, SOFTWARE, AND 802.11 ISSUESTransmission Power Instability — A majoraspect of the interference mechanism is the ini-tial transmission power, which together with thepath losses determine the received interferencepower. Unfortunately, early experiments showedthat the power control in the ath5k driver is notyet stable enough; for given power settings theactual output power depended on the data rateof the transmitted data, which appeared to bearbitrary and not due to an expected power cut-off at higher rates. To deal with this instability,

Figure 2. The SINR effect on packet reception: a) Rx2 will not be able to cor-rectly decode the data transmitted by Tx2 due to high interference from nearbychannel transmission of Tx1; b) Tx2 may falsely report the channel as busy ifchannel Tx2 → Rx2 is adjacent to channel Tx1 → Rx1.

Tx2Tx1

Rx2

Rx1

(a)

Rx2

Channelbusy

Rx1Tx2 Tx1

(b)

2 Sensitivity is the mini-mum input power level atwhich decoding can beachieved at a desired BERin a given rate.



at each data rate setting we measured thereceived power at some fixed point of ourtestbed during a calibration run, and based onthe measured value we compensated accordinglyby adjusting the attenuators.

In order to verify the theoretical assumptionspresented earlier we coupled the measuredSINR with the achievable throughput. In eachdata rate the expected throughput is relative tothe SINR as for each constellation and codingrate the BER is directly linked to the signal-to-noise ratio (SNR) [7]. With the use of the pro-grammable attenuator we were able to controlthe signal attenuation per dB; therefore, weobtained measurements that cover in detail awide SNR space for each data rate.

Antenna Isolation Instability — Anothermajor problem for designing and conducting theexperiments was the separation of transmissionand reception to the two antenna connectors onthe wireless card. Unfortunately, disabling theantenna diversity option in the ath5k driver wasnot enough. The result was to have some sparsetransmissions from the antenna connector desig-nated for reception, reducing the accuracy ofmeasurements. The problem was solved by mak-ing the appropriate corrections in the sourcecode for the antenna connector handling in theath5k driver.

Another interesting observation was that theath5k/Atheros chipset combination had a periodi-cal 3 s timeout during transmission. It was easilyobserved as a gap in the channel utilization graphduring heavy traffic generation. This behaviorcould be attributed to the periodical recalibrationof the radio frequency (RF) front-end the wire-less cards perform. The result was a lack of inter-ference during that period, giving a chance forunhindered communication in the link.

The Role of Channel Utilization — One keyaspect of the interference generated by 802.11devices is that it is not constant in time. It followsthe timings of the 802.11 DCF, and can be con-

sidered noise that is there only during the time inwhich the interfering transmitter is active. There-fore, its effect will depend on the utilization ofthe interfering link. As seen in [8] the utilizationof the wireless medium is inversely proportionalto the data rate being used for a given achievablethroughput. Quite an interesting observation isthat even at the maximum achievable throughput,the utilization of the 54 Mb/s data rate is farlower than that of the 6 Mb/s rate. Therefore, onecan expect that a 6 Mb/s interfering sender run-ning at full throughput will be more harmful thana full throughput 54 Mb/s interferer.

With the interface connectivity explained ear-lier, we managed to produce an 802.11a jammerthat does not sense the channel prior to trans-mitting, and therefore can transmit as frequentlyas a single user DCF allows, thus maximizing theutilization of the medium for its respective datarate.

EXPERIMENTAL RESULTSPacket Capture — The rate in the link underinvestigation was always fixed, having disabledthe automatic rate selection mechanism. Foreach rate we increase the attenuation level atthe programmable attenuator X (Fig. 3a) by onedB per measurement run. Essentially wedecrease the SINR by one dB in each step andrecord the average throughput for a 3 min mea-surement run period. For each data rate we havea signal strength measurement for an attenua-tion value in order to have the differences intransmission powers between them. The differ-ences in the transmission power of each rate areused to compensate the attenuation levels sothat the results of the throughput are directlycomparable. The results are presented in Fig. 4.It is obvious that the throughput curves closelyfollow the expected SINR — throughput degra-dation (e.g., as computed in [9]).

Interferer’s Rate Effect — As already stated inthe previous section, all the experiments wereconducted with fixed rate and given utilization at

Figure 3. a) The reception mechanism testbed layout schematic representation; b) the actual testbed of the experiments.

LinkS

Rx

Tx

I

Rx

JRx

D

Rx

50

16

Interferer

16

(a) (b)

x

x

Tx

Tx

Tx

x



the interferer. In this experiment we investigatedwhether the data rate of the interferer has anyimpact on the interference experienced by thereceiver. In order to have comparable results foreach data rate, we adjust the packet size to keepthe utilization fixed. The results revealed thatthe data rate of the interferer has a strongimpact, with increasing intensity at higher datarates. The reasons may be the constellation,which becomes denser at higher rates, and thetransmission-idle time distribution. Since theath5k driver always keeps the basic rate at 6Mb/s to verify the above assumption, we alsoused a Cisco 1240 AP where different basic rates(6, 12, and 24 Mb/s) can be defined. With theCisco AP we saw similar behavior, but with aneven greater degradation in throughput.Although all the parameters were the same andit was expected that the devices should behavethe same, we noticed that different devices dohave some minor differences that result in dif-ferent ACI. The differences can be attributed toprotocol timing parameters as the Cisco AP con-sistently achieves higher throughput than theAtheros/ath5k in an interference-free environ-ment.

In order to further investigate the transmit-ting-idling time distribution theory we used aseries of packet sizes (1–1472 bytes) for thethroughput measurement. We took a baselinemeasurement with absence of ACI and then pro-duced ACI, keeping the same utilization withthree different data rates (6, 24, 54 Mb/s). In 6Mb/s we had a throughput degradation of 55–58percent of the baseline, in 24 Mb/s 63–67 per-cent, and in 54 Mb/s about 80–91 percent withthe greater degradation in larger packet sizes.This result verifies that the mechanism behindthe rate effect of the ACI is the distribution oftransmission and idle times, as the long-termmedium utilization may be constant but theinterleaved transmitting-idle periods are denserin higher data rates.

CCA — In this experiment we have set channel60 for the test link and performed a baselinemeasurement of the achieved throughput using a1000-byte UDP payload, without any interfer-ence and for all possible data rates (Table 2, col-umn 1). The values of the attenuators wereproperly calculated so that an interferer withoutput power 0 dBm tuned to the adjacent chan-nel would trigger false positives in the CCAmechanism of the test link sender. A secondseries (Table 2, column 2) was recorded with theinterferer at the same channel as the test linkwhere the CCA mechanism is expected to betriggered — this marks the results of a collocat-ed and non-contending transmitter at the samechannel as the link. Tuning the interferer atchannel 56 resulted in a throughput loss rangingfrom 55 percent at 6 Mb/s to 85 percent at 54Mb/s, which of course is due to the busy mediumstate that the transmitter is frequently sensing.As is obvious from Table 2, for the same trans-mission power level two channels away (channel52), the CCA mechanism is not affected. InTable 1 we observe that the difference in powerleakage between adjacent and next adjacentchannels is 18 dB. With that in mind, we raised

the transmission power by 18 dB, and observedthat the CCA mechanism was again triggered,verifying once more our model of Fig. 1. Thesimilar results of all the columns where the CCAwas triggered indicate the binary nature of themechanism: if the received power exceeds thethreshold, regardless of the channel distance, themedium is sensed as busy.

CONCLUSIONSDespite the general belief that 802.11a is free ofACI, due to the use of non-overlapping chan-nels, we have shown that the need for carefulchannel selection is also present in the 5 GHzband. Through the use of the emulated wirelessmedium, we have isolated the affected 802.11mechanisms and quantified the throughputdegradation due to ACI. The two main mecha-nisms that are affected are data reception andclear channel assessment. In the first case theSNR is degraded, making reception impossible,

Figure 4. Results of the packet capture experiment.

Attenuation (dB)10

5

0

Thro

ughp

ut (

Mb/

s)

10

15

20

25

30

0 20 30 40 50 60

54 Mb/s48 Mb/s36 Mb/s24 Mb/s18 Mb/s12 Mb/s9 Mb/s6 Mb/s

Table 2. ACI effect on the throughput, in Mb/s, due to CCA false positives.

Link at channel 60 Interferer at channel

Tx rate (Mb/s) Baseline 60 56 52 52 (+18 dB)

6 4.86 2.21 2.54 4.94 2.23

9 6.92 3.03 3.10 7.91 2.32

12 8.92 3.12 2.67 9.01 2.12

18 11.81 3.15 2.71 11.72 2.29

24 14.30 3.20 2.78 14.10 2.46

36 18.11 3.40 2.81 18.21 2.65

48 21.13 3.00 2.82 21.19 2.33

54 23.11 3.26 2.95 22.10 2.71



and in the second case the transmitter stalls itsdata as it incorrectly senses the medium busy.Nevertheless, the large number of channelsavailable in 802.11a, taking into considerationthe facts identified and justified in this article,provide the opportunity and motivation formeticulous channel selection in order to achievehigh throughput.

ACKNOWLEDGMENTSThis work was supported by the General Secre-tariat for Research and Technology, Greece,through project 05-AKMON-80, and by theEuropean Commission in the 7th FrameworkProgramme through project EU-MESH(Enhanced, Ubiquitous, and Dependable Broad-band Access using MESH Networks), ICT-215320, http://www.eu-mesh.eu.

The authors acknowledge the help of Mr.Nick Kossifidis on the testbed setup and themodifications of the ath5k code.

REFERENCES[1] A. Mishra et al., “Partially Overlapped Channels Not

Considered Harmful,” SIGMetrics/Performance ‘06,Saint Malo, France, June 2006.

[2] IEEE 802.11a, “Supplement to IEEE 802.11 Standard —Part 11: Wireless LAN, Medium Access Control (MAC),and Physical Layer (PHY) Specifications: High-SpeedPhysical Layer in the 5 GHz Band,” Sept. 1999.

[3] C. M. Cheng et al., “Adjacent Channel Interference inDual-Radio 802.11 Nodes and Its Impact on MultihopNetworking,” IEEE GLOBECOM ‘06, San Francisco, CA,Nov. 2006.

[4] V. Angelakis et al., “Adjacent Channel Interference in802.11a: Modeling and Testbed Validation,” 2008 IEEERadio Wireless Symp., Orlando, FL, Jan. 2008.

[5] V. Angelakis et al., “The Effect of Using DirectionalAntennas on Adjacent Channel Interference in 802.11a:Modeling and Experience with an Outdoors Testbed,”WiNMee 2008, Berlin, Germany, Mar. 2008.

[6] IEEE 802.11, “Wireless LAN Medium Access Control(MAC) and Physical Layer (PHY) Specifications,” Aug.1999.

[7] J. Proakis, Digital Communications, 4th ed., McGrawHill, 2001.

[8] J. Jun, P. Peddabachagari, and M. Sichitiu, “TheoreticalMaximum Throughput of IEEE 802.11 and Its Applica-tions,” IEEE NCA ‘03, Cambridge, MA, Apr. 2003.

[9] D. Qiao, S. Choi, and K. G. Shin, “Goodput Analysis andLink Adaptation for IEEE 802.11a Wireless LANs,” IEEETrans. Mobile Comp., vol. 1, no. 4, Oct.–Dec. 2002, pp.278–92.

ADDITIONAL READING[1] J. Robinson et al., “Experimenting with a Multi-Radio

Mesh Networking Testbed,” 1st WiNMee ‘05, Italy, Apr.2005.

BIOGRAPHIESVANGELIS ANGELAKIS [S‘05, M‘09] ([email protected])is a postdoctoral research fellow at the Mobile Telecommu-nications Group of the Department of Science and Technol-ogy, University of Linköping, Sweden. He received his B.Sc.,M.Sc., and Ph.D. degrees from the Department of Comput-er Science at the University of Crete, Greece, in 2001,2004, and 2008, respectively. In the summer of 2005 hewas a visiting researcher at the Institute of SystemsResearch of the University of Maryland at College Park. In2004 he began working as a research assistant and post-doctoral research fellow with the Telecommunications andNetworks Laboratory at FORTH-ICS. His research interestsinclude wireless network planning and resource allocationoptimization and management.

STEFANOS PAPADAKIS [S‘07, M‘10] ([email protected]) is aresearch and development engineer at the Institute ofComputer Science of FORTH. He received his degree inphysics (2001), and his M.Sc. (2004) and Ph.D. (2009)degrees in computer science from the University of Crete.Since 2001 he has been working as a research assistant inthe Telecommunications and Networks Laboratory atFORTH-ICS. His research interests include position locationtechniques in wireless networks, radio propagation model-ing, and wireless network planning.

VASILIOS A. SIRIS [M‘98] ([email protected]) is an assistant profes-sor in the Department of Informatics, Athens University ofEconomics and Business, and a research associate at the Insti-tute of Computer Science of FORTH. He received his degreein physics (1990) from the National and Kapodistrian Univer-sity of Athens, his M.S. (1992) in computer science fromNortheastern University, Boston, Massachusetts, and his Ph.D.(1998) in computer science from the University of Crete. Inspring 2001 he was a visiting rresearcher at the StatisticalLaboratory of the University of Cambridge, and in summer2001 and 2006 he was a research fellow at the research lab-oratories of British Telecommunications (BT), United King-dom. His research interests include resource management inwired and wireless networks, traffic measurement, and analy-sis for QoS monitoring and anomaly detection.

APOSTOLOS TRAGANITIS joined FORTH-ICS in 1988 and sincethen has coordinated and participated in a number of EUfunded projects in the Communications and Health Caresector. He is head of the Telecommunications and Net-works Laboratory and a professor in the Department ofComputer Science, University of Crete, where he alsoteaches and does research in the areas of digital communi-cations and wireless networks. He holds M.Sc. and Ph.D.degrees from Princeton University, New Jersey.

The large number of

channels available in

802.11a, taking into

consideration the

facts identified and

justified in this

article, provide the

opportunity and

motivation respec-

tively for meticulous

channel selection in

order to achieve

high throughput.



INTRODUCTION

The Internet has grown dramatically in only afew decades, and the growth in many areas con-tinues to be exponential. Such areas includeInternet connectivity, web page impressions,social network users, video content bandwidth,as well as the value of transactions conductedby enterprises that use the Internet inherentlyfor their business. This pace requires a numberof evolutionary or even revolutionary steps con-cerning the supporting infrastructures, regard-less of whether this affects the transportnetwork architecture, the supporting platformsin the wider sense, or the processing of infor-mation.

In most cases, emerging networking con-cepts, services, and business models need to be

piloted in a large-scale environment that mimicsas far as possible the current Internet and itsfuture version. We can expect that the futureInternet, a very large-scale constellation of sys-tems and processes, might not be too differentfrom an aggregation of the best ideas currentlyin research on how to construct and operatenetworks, platforms, and systems. For this rea-son, large-scale testbeds emerge that aim notonly at providing an experimental environmentfor such new ideas to be investigated, but, mostimportant, at providing a high degree of realismin supporting experiments executed under real-life conditions. To this end, testing as a disci-pline that evaluates the design andimplementation of components and systems inorder to reveal their flaws is undergoing pro-found transformations.

Figure 1 shows the impact of increasing exper-iment realism on the associated cost [1]. In orderto enable experimentation on a very large scale,federated systems are required. In addition, manyexperiments target the interaction of the intend-ed end user of a service or device (a product)and under which conditions users engage in abusiness relationship using the product. There-fore, if a business dimension is added to experi-mentation activities, we speak of a pilot. Wedefine a pilot as the “execution of an experimentor test including business relationship assump-tions, exemplifying a contemplated added valuefor the end user of a product.”

The Pan-European Laboratory (Panlab) con-cepts and architecture outlined in this articleaspire to facilitate and support the needs forlarge-scale testing, experimentation, and piloting.In the following we provide a description of themost relevant initiatives in future Internet exper-imentation followed by the key concepts andobjectives of Panlab. We then present the Pan-lab architecture and its main components thatallow Panlab users to create a federated testingenvironment comprising widely dispersedtestbeds.

0163-6804/11/$25.00 © 2011 IEEE

ABSTRACT

Networks, services, and business modelsneed to be piloted in a large-scale environmentthat mimics as far as possible the Internet andits future version, although we certainly do notknow what this will look like. In this highlydemanding and dynamic context, new frame-works and (meta-) architectures for supportingexperimentation are proposed, aiming at discov-ering how existing and emerging testbeds andexperimental resources can be put together insuch a way that testing and experimentationmay be carried out according to specific require-ments (industry, academia, etc.). The work pre-sented in this article addresses a number of thefundamental principles and their correspondingtechnology implementations that enable theprovisioning of large-scale testbeds for testingand experimentation as well as deploying futureInternet platforms for piloting novel applica-tions. The proposed concepts and architectureof Panlab, a pan-European testbed federation,aspires to facilitate and support user needs inthese new areas.

NETWORK TESTING SERIES

Sebastian Wahle, Fraunhofer FOKUS

Christos Tranoris and Spyros Denazis, University of Patras

Anastasius Gavras, Eurescom GmbH

Konstantinos Koutsopoulos, Bluechip Technologies SA

Thomas Magedanz, Technische Universität Berlin

Spyros Tompros, University of Patras

Emerging Testing Trends and the Panlab Enabling Infrastructure

WAHLE LAYOUT 2/18/11 3:21 PM Page 167


EXPERIMENTAL FACILITIES ANDRESOURCE FEDERATION

Currently, a number of future Internet initiativessuch as FIRE (Europe), FIND/GENI (UnitedStates), AKARI (Japan), and their relatedresearch projects rely on federated experimentalfacilities/testbeds to test and validate their solu-tions.

The U.S. GENI projects are organized in so-called spirals where the findings of each spiralare assessed toward the end of a spiral anddefine the requirements for the next spiral phase.The general high-level GENI architecturedefines several entities and functions alignedwith the Slice-Based Facility Architecture (SFA).In its second version the SFA draft specificationdefines a control framework and federationarchitecture as a lowest common denominatorfor some of the GENI clusters (ProtoGENI,PlanetLab, etc.).

In Europe, several projects (e.g., Onelab2,Panlab/PII and others) are contributing to theFuture Internet Research & Experimentation(FIRE) experimental facility. The number ofFIRE facility projects was recently increasedwith new projects (BonFIRE, TEFIS, etc.) join-ing in. The FIRESTATION support action coor-dinates a FIRE Office as well as a FIREArchitecture Board that is foreseen to drive thespecification of mechanisms on how to federatethe diverse facilities. This is currently ongoingwork.

The main countries active in Asia areJapan, Korea, and China. Joint Asian activitiesare carried out under the Asia-Pacif icAdvanced Network (APAN) initiative as wellas PlanetLab CJK (China, Japan, Korea), ajoint PlanetLab cooperation by those threecountries.

A full overview of future Internet testbedsand their control frameworks was published ear-lier in [2].

THE PANLAB FRAMEWORK: ROLES,CONCEPTS, AND ARCHITECTURE

The Panlab framework and its correspondingarchitecture aim at provisioning and managingdistributed testbeds enabling experimenters tocarry out various testing and experimentationactivities. Such activities are supported by a Pan-lab office, a coordination center that facilitatesinteractions between Panlab’s customers/experi-menters and the different test sites. This givesrise to the following main roles in Panlab:• A Panlab partner is a provider of infra-

structure resources/testbed components(e.g., hardware, software, virtualizedresources) necessary to support the testingservices requested by the customer. Part-ners interact with the Panlab office to offerrequested testbed resources to customers.Collectively, the Panlab partners representthe Panlab federation.

• A Panlab customer has access to specificinfrastructure and functionality necessary toperform testing and experimentationaccording to its needs. Customers are typi-cally interested in carrying out R&D activi-ties using resources provided by Panlabpartners. They are supported by the Panlaboffice in order to implement and evaluatenew technologies, products, or servicesdrawing upon the large resource pool avail-able through the entire Panlab federation.

• A Panlab office realizes a brokering service,serving Panlab partners and customers bycoordinating legal and operational processes,the provisioning of the infrastructures andservices to be used for testing and experi-mentation, and the interconnectivity of thevarious partner test sites and customers.Entities taking on those roles interact with

each other forming various consecutive views(Fig. 2) that correspond to a series of mappingsranging from generic customer testing require-ments to detailed provisioning and managementoperations. In Panlab, we have identified threedistinct views that correspond to different levelsof abstraction with respect to testing features.The different Panlab roles are shown in Fig. 3.

The customer view captures the testbedrequirements of the user, namely the abstractdesign of the desired testbed: the virtual cus-tomer testbed (VCT). The design can be eitherspecific or agnostic with regard to the physicallocation of the resources and how the underlyingfederation of testbeds will allocate and provisionthem. In this view, the customer will search andrequest available abstract resources across thePanlab federation. The customer view can becompiled in many ways and using various tools;for example, textually (by filling out forms),graphically (by using a graph tool), using adomain-specific language (DSL), or a combina-tion of those. The main goal is to empower thecustomer to define the requirements for hisVCT in a consistent and unambiguous way. Forinstance, the customer may explicitly select com-puting resources, type of connectivity and net-working, and specific services that must beinstalled and configured on each resource.

Figure 1. New areas for testing and experimentation and their impact on thecost.

Formalmodel

Simulation Emulation Real systems Log(realism)As the opposite of the abstractionlevel

Log(cost)Cost=f(complexity, resource,environmental conditions)

Modelssys, apps,platforms,conditions

Real OSReal applications“In-lab” platformsSynthetic conditions

Real OSReal applicationsReal platformsReal conditions

Real federated systemReal systems and appsReal conditionsDistributed resources

Heterogeneousfederation

Homogeneousfederation

Models for key OS mechanismsAlgorithms and kernel appsAbstracted platformsSynthetic conditions

Loss of experimentalconditions, reproducibility,repeatability, etc.

Loss of real experimentalconditions



In the federation view, a federated infra-structure is expressed that involves the Panlabpartner testbeds that will participate in the cre-ation of a VCT along with the testbed resourcesto be provisioned. The federation frameworksmaps customer requests and takes actions on rel-evant orchestration of services for this view. Thisview can be either specific or agnostic withregard to the internal infrastructure and topolo-gies of the individual testbeds. The scope of thefederation view covers the connectivity require-ments across the various testbeds meeting thecustomer’s requirements. Accordingly, provision-ing requests are submitted to each participatingtestbed, for example, to create a virtual privatenetwork (VPN) between all involved test sites, todeploy virtualized resources, to install operatingsystems and applications, and to configure theentire environment. This view is more detailedthan the customer view but less detailed than thetestbed view. It is required on the federationplatform level for brokering between customersand the testbeds.

Finally, the testbed view contains the actualrealization and deployment of a VCT and itsassociated experimental resources across theselected testbeds. It meets the requirements ofthe federation view and exhibits all internalinfrastructure elements, e.g., internal gateways,switches, routers, computers, etc. that will partic-ipate in the testing/experiment execution. It ismore detailed than the federation view for eachindividual domain. However, on adomain/testbed level, information concerning theother domains is usually meaningless. In thissense, global information is lost (or at leastmeaningless) taking the testbed view.

The aforementioned views are necessary tocreate a taxonomy of Panlab functionality andorchestrate the selection and provisioning ofPanlab resources across various testbeds. Thisorchestration is carried out through the Panlaboffice and its corresponding architectural ele-ments aiming at automating various Panlabrelated operational processes [3]. Using the Tea-gle framework, Panlab customers may assemble adesired VCT and request its deployment. Teagletakes the customer request as well as testbedand resource availability as input to coordinatethe provisioning of the desired environment.

As Panlab controlled resources reside in dis-tributed testbeds, configuration operations arecarried out by an architectural component (localto each testbed), called the Panlab testbed man-

ager (PTM), which interacts with Teagle sup-porting management operations. Also, as theresources controlled by Teagle are highly hetero-geneous components (e.g., hardware, software,abstract services), the Panlab framework intro-duces the concept of resource adaptors (RAs)that abstract management capabilities (referencepoint T2, Fig. 3), which are then offered througha common application programming interface(API) (reference point T1, Fig. 3). Finally, inter-connection gateways (IGWs) interconnect thefederated testbeds and components inside thetestbeds with remote peers that are part of theconfiguration through automatic VPN (cross-site, reference point I1, Fig. 3) and VLAN (intra-site, reference point I2, Fig. 3) deployment.

TEAGLEThe Panlab architectural element Teagle allowsbrowsing through the Panlab Federation offer-ings, enables the definition of VCTs, and exe-cutes their provisioning. Teagle relies on thePanlab federation model and framework thathandles all the technical, operational, and legalaspects of generic resource federation [4]. Cur-rently, Teagle implements the following func-tions (Fig. 4):• A model-based repository: collectively con-

sists of several registries for users,resources, configurations, and so on.

• A creation environment (VCT tool): allowsthe setup and configuration of VCTs. Thetool can make use of all available resourcesacross the Panlab federation, but is restrict-ed by policies that can be set on a global,per-domain, or per-resource level. Theassociated request processor exposes anAPI that is called by the VCT tool or othertools.

• An orchestration engine: generates an exe-cutable workflow for resource provisioning.The engine receives a VCT definition fromthe request processor.

• A Web portal: exposes search and configu-ration interfaces, as well as general infor-mation.

• A policy engine: allows the evaluation ofpolicies that resource providers can define.The engine also allows for global federationpolicies.

• A gateway: handles bidirectional communi-cation between domain managers (PTMs)and Teagle internal entities.

Figure 2. The three views of a testbed.

Customerview

Federationview

Testbedview

NetworkQoS: best effort

Application Application

Application

Application

Testbed ATestbed A

Customer

Testbed BTestbed B

IGW

VPNSUTCustomer

Computingresource

Computingresource

Application

Computingresource

App

Computingresource

Application

Computingresource

Application

Computingresource

IGW



The repository holds data about availableresource types and instances, and can be queriedby other Teagle components such as the VCTtool. The Panlab customer can launch the VCTtool (a Java Web Start application, see Fig. 5)from the Teagle portal and use it to define aVCT. From a list of available resources, selectedelements can be dropped, connected, and config-ured on the tool workbench. During the VCTdesign, the tool interacts with the policy engineto indicate impossible or forbidden testbed lay-outs or configurations. When the VCT design

has been finished, the tool stores the VCT defi-nition in the repository, and the booking/schedul-ing procedure can be initiated via the VCT toolor the Teagle portal.

Upon booking, the request processor retrievesthe VCT definition from the repository, triggerspolicy evaluation, and sends it to the orchestra-tion engine. The orchestration engine assemblesa workflow, executes it, and sends individualprovisioning (CRUD: create, read, update,delete) requests to the involved PTMs via theTeagle gateway on interface T1. The PTMs are

Figure 3. Panlab roles and architectural components.

PTM

Resource

RARA

IGWIGW

A1

Panlabresource

registry andrepository

Panlab office

TEAGLE

Customer Teagle

Panlab partner testbed

U1

U2

Panlab customer Panlab partner

Configure andbook resources

Controlresources

Offer and maintain resources

I2

I2

T1

T2

Testbed resources

3G

Figure 4. Teagle architecture.

U1

T1

T2

POLPOL

PortalREP

REP

Policies

Repositorygatew

ay

Repository

Policy engine

VCT tool andrequest

processorOrchestration

engine

SPECTEAGLE

Teaglegateway

PTM

RA RRA RRA R

Customer

TG

Testsuites

Results Registries:VCTs

resourcesPTMs IDMInfo/data

models



responsible for conducting the provisioning ofresources according to the incoming requests.From the perspective of the Teagle platform,this operation is opaque and could be performedin a number of ways. However, the currentlyavailable implementation of a PTM usesresource adapters as an abstraction layer for het-erogeneous resources. The PTMs report on thestatus of resource operations (success/failure/configuration), and the responses are aggregatedat the orchestration engine, stored in the reposi-tory, and finally presented to the user.

More details on specific components provid-ed by the Teagle framework can be found in adedicated Panlab Wiki [5] and the Teagle portal[6]. The portal also provides tutorial videos thatdemonstrate the usage of Teagle componentssuch as the VCT tool. In addition, Panlab regu-larly offers training events to give further insightsand allow for hands-on experience. The next sec-tion focuses on the resource description mecha-nism used by Teagle.

THE TEAGLE RESOURCE REGISTRYAND RESOURCE DESCRIPTION

MODEL

Given that the federation system needs to dealwith a great number of highly heterogeneousand a priori unknown resources, the model usedto structure and describe the resources neededto be extensible. An existing information modelwas extended for our purposes to representcharacteristics of the resources and their rela-tionships in a common form independent of aspecific repository implementation. Resourcescan be modeled as concrete physical entities

such as a physical machine, or abstract logicalresources such as an administrative domain.

The DEN-ng information model [7] that isrooted in the area of network management andautonomic networking was used and extended toallow the description of resources as managedentities, their life cycle, as well as associated poli-cies. In terms of DEN-ng, resource entities pro-vide a service and have a certain configurationattached that can be defined and altered usingthe federation tools that are exposed to the exper-imenter via Teagle. Resources can exist as physi-cal or logical resources where resource providerscan define a list of resource instances as specificsubtypes based on the model to represent theirfederation offerings. The repository implementa-tion has been realized as a number of applicationsrunning as contexts on an application server.Each application has its own data storage facilityand exposes a HTTP-based RESTful interfacewith a number of Representation State Transfer(REST) resources. The repository only deals withstorage and retrieval of data on behalf of clientapplications. This allows the architectural entitiesthat collectively represent the Teagle framework,to develop independently of the repository but torely on a common data model.

Figure 6 shows a snapshoot of the resourcemanagement part of the entire model on whichthe Teagle repository is based. We differentiatebetween resource types (ResourceSpec class) andresource instances (ResourceInstance class)where every instance is of a certain type. Thisallows the instantiation of multiple resourceinstances of the same type at a given Panlabpartner testbed (e.g., virtual machines) wherethe different instances can have different config-urations. Resources can also be offered in a pre-defined configuration (pre-existing instances).

Figure 5. The Teagle VCT tool.



THE PANLAB TESTBED MANAGER

Each Panlab partner test site exposes via thePanlab Testbed Manager (PTM), a domain man-ager interface T1 (Fig. 3) that is currently imple-mented as a SOAP web service interface. ThePTM exposes resources as services to the federa-tion and provides the mapping from federationlevel commands to resource specific communica-tion using resource adaptors (RAs). The PTMresource adaptation layer (RAL) implementsmechanisms that aid resource discovery andmonitoring by generating specific change eventsto communicate the resource status as well ascontrol information to the PTM core. RAs areregistered with the runtime framework of thePTM and make their presence known to the restof the platform so that responsible PTM mod-ules can keep track of the corresponding eventsregarding each specific resource representationand consequently the actual resource.

RAs plug into a PTM the same way devicedrivers plug into an operating system to controlspecific devices. Since the RAs are the only ref-erence towards an actual resource, several eventsregarding status of both adaptor and resourcecan be reported:• Pending acknowledgments with respect to

requested configuration actions• Failures in the operation of a resource adap-

tor• Communication loss or failure with

resources• Resource failures• Resource reset• Upgrades (resource or adaptor)• Busy states

The operation of RAs and the RAL in gener-al is characterized by the fact that the PTM hasa common view of all the resources for which itis responsible. No resource-specific communica-tion protocols are exposed to or need to be dealtwith by the PTM modules. Every resource is

integrated in the PTM runtime environment inan agnostic manner with respect to what aresource may be from the point of view of net-working procedures and topology. The PTM isdeployed as installable components in a Glass-Fish ESB v2.1 application server. It consists oftwo JBI service assemblies that implement thecore logic of the PTM and a Web applicationthat is the administration interface of the PTM.

RAs are implemented in Java and areinstalled in the PTM as Java/OSGi bundles. Toease the development of RAs, we defined theResource Adapter Description Language(RADL). RADL is a concrete textual syntax fordescribing an RA based on an abstract syntaxdefined in a meta-model. RADL’s textual syntaxallows an easy description of the resource con-figuration parameters and how the RA shouldreact upon receiving CRUD requests. TheRADL support tools generate Java skeletoncode ready to be plugged into the PTM.

THE INTERCONNECTION GATEWAYThe IGW, as shown in Fig. 3, is responsible forproviding and controlling connectivity betweenPanlab partner test sites. The federated distributedresources are interconnected by means of meshedIGWs that provide a separated layer 2 tunnel perVCT. This allows building large scale virtual over-lay networks dynamically over the public Internet.

IGWs are ingress-egress points to each sitefor intra-VCT communication via one automati-cally configured multi-endpoint tunnel per virtu-al testbed. It is able to act as a dynamicallyconfigurable hub and allows isolation of localtestbed devices. One VPN per VCT instance isconfigured between all neighbor IGWs andenforces isolation of local resources by dynami-cally configured collision domain isolation. Acollision domain is an isolated network segmenton a partner’s physical test-site where data pack-ets are sent on a shared channel.

Figure 6. Resource configuration model.

ResourceSpec

state

ResourceInstance

ConfigurationInfo

ResourceInstanceState

ConfigParam

configurationParamComposite

configParams

Configlet

*

* *

*

1

ConfigParamCompositeConfigParamAtomic

String defaultParmValueString confiParamType

resourceSpec

configurationParameters

Boolean shared

Geometry

ConfigurationBase

configurationBases

Configuration

Integer wInteger yInteger hInteger x

*

1

The PTM is deployed

as installable

components in a

GlassFish ESB v2.1

application server.

It consists of two JBI

service assemblies

that implement the

core logic of the

PTM and a Web

application that is

the administration

interface of the PTM.



IGWs automatically establish connections toother peer IGWs. An important design criterionwas to make them as self-configuring as possible.For such meshing of all IGWs that are part of aspecific VCT, a stateless low-overhead tunnelingwas chosen. The IGW can be exposed as anyother resource in Teagle tools like the VCT toolor not, depending on the level of configurationgranularity that is requested by the experi-menter.

On the IGWs internal connection statemachine, active VCTs are lists of tuples consist-ing of the other IGW’s external address and thecollision domain(s) associated with the specificVCT behind it. Each interconnection state canbe expanded by adding more interconnections.New interconnection states do not interfere withexisting states. They use the same VPN tunnelbut are separated during the routing and filter-ing process. This guarantees an on-demand auto-matic resource interconnection across Panlabsites without using proprietary inter-IGW proto-cols.

An experimenter is able to connect singledevices (e.g., test clients) to his/her VCT using acustomer dial-in (reference point U2, Fig. 3) fea-ture. This layer 2 tunneling protocol (L2TP)-based on-demand tunnel delivers direct access toa specific VCT as if the experimenter were work-ing within a partner domain that had a localIGW.

The main functionalities provided by an IGWare to interconnect, keep, and protect the map-ping of local collision domain communication toexternal VPN interconnection. Therefore, it func-tions like an IP-based trunking device for testbedcomponents communicating on data planes sepa-rated by collision domains on the internal sideand VPN-based access on the external side. Rout-ing of data plane packets in between these securechannels is done by an IGW internal module, theinterconnection engine.

If requested via the domain manager, qualityof service (QoS) rules may be enforced on rout-ing decisions, for instance limiting connectionsbetween test sites to a certain maximum through-put rate. In front and in the back of the inter-connection engine, the secure channels arede-encapsulated/decrypted and filtered by astateful IP-based firewall. This makes sure thataccess to specific resources can be restricted asdefined by the Panlab partner as a resourceprovider. On the external side of the IGW, theremay also be generic collision domains bridged totestbeds that are not publicly accessible. In thisway it is possible to perform real QoS reserva-tions such as asynchronous transfer mode(ATM) or fiber optic links.

As collision domain channel isolation isrequired for connecting the federated resources,IEEE 802.1Q VLAN-based systems have beenadded as a mandatory requirement and prereq-uisite for running separated experiments in par-allel. Since several VLANs may be used as ashared medium to connect multiple resources ina single test site, the experimenter has full con-trol over the network topology to be deployed. Avirtualized host resource may act as a softwarerouter within a VCT. However, this flexibilitycomes with significant complexity in configuring

the network layer. On one hand, the VCT toolprovides the means to abstract from such com-plexity; on the other hand, some Panlab cus-tomers require this level of configurationgranularity. The Panlab mechanisms allow satis-fying such diverse user requirements.

CASE STUDY: A SETUP FOR TESTINGADAPTIVE ADMISSION CONTROL

AND RESOURCE ALLOCATIONALGORITHMS

In this section we demonstrate how Panlabmechanisms and prototypes can be used in termsof an example experiment setup targeting adap-tive admission control and resource allocationalgorithms.

The experimenter sets up a desired VCT asshown in Fig. 7 using resources from Panlabcontrolled test sites. The setup contains RUBiS,an auction site prototype modeled aftereBay.com that is used to evaluate applicationdesign patterns and application server perfor-mance scalability. It provides a virtualized dis-tributed application that consists of threecomponents: an application server, a database,and its workload generator, which produces theappropriate requests. It is deployed in a virtual-ized environment using XEN server technology,which allows regulating system resources such asCPU usage and memory.

The adaptive admission control and resourceallocation algorithm, which is the system undertest (SUT) in this specific setup, is a proxy-likecomponent for admission control using XENserver technology to regulate CPU usage. RUBiSclients produce requests pushing the RUBiSserver side components to their limits. The adap-tive admission control and resource allocationalgorithm is tested against network metrics likeround-trip time and throughput.

The experimenter uses the VCT tool to con-figure (e.g., IP addresses, bindings, max. clientrequests, VPN access) all resources such as theRUBiS HTTP traffic generators, web servers,and XEN machines. The screenshot provided byFig. 5 shows a simplified VCT setup for thisexperiment at the highest configuration granu-larity. The experimenter also needs to providethe algorithm under test by logging into theproxy unit and installing it using an SSH accountand a private key to access the VCT through aVPN.

The equipment/resources used in this setupare:• XEN servers which host virtual machines

with RUBiS based work load generatorresources

• A virtual machine for hosting the algorithmunit, based on a Linux image capable ofcompiling C and Java software

• XEN servers, which host virtual machineswith the RUBiS Web app and databaseinstalledWhile running the experiment, there is a

need to reconfigure resources and get monitor-ing data from the resources. To achieve this, ourcase study scenario uses the Federation Comput-

As collision domain

channel isolation is

required for

connecting the

federated resources,

IEEE 802.1Q VLAN

based systems have

been added as a

mandatory

requirement and as a

prerequisite for

running separated

experiments in

parallel.



ing Interface (FCI) API [8] provided by Panlab.The FCI can be used for accessing federatedresources and embedded into an application/SUT in order to gain control of the requestedresources during the experiment. Via a model-to-code transformation using the VCT definitionstored in the Panlab repository, VCT resourcescan be exported as Java classes that allow theexperimenter to work with the resources asobjects programmatically. This allows the userapplication/SUT to access the testbed resourcesduring execution of the experiment in order tomanage and configure various environmentparameters or obtain resource status information.

CONCLUSIONOne possible evolution path of the Internetcould be that it will be based on a core platformconcept that prescribes a number of fundamen-tal principles for integration. This core platformmight enumerate generic functions that supporta diverse number of application areas and allowthe integration of optional application-specificfunctions. From the set of generic and optionalfunctions, and following the fundamental princi-ples, it will be possible to deploy platforminstances that best serve target application areas.In this scenario, it will be crucial to be able toflexibly compose and evolve platform instances.Flexible composition, instantiation, and federa-tion require the functions to be described interms of their functional and non-functionalaspects, facilitating dynamic discovery and com-position of higher-layer applications.

The testbeds that are currently used to vali-date emerging concepts and technologies mighteventually evolve into the future Internet them-selves. As Panlab provides an extensive platformto flexibly compose abstracted resources to beconsumed as services, it positions itself on the

Internet evolution path as outlined above. Thecore part of the Panlab federation platform is theTeagle framework. We will continue to extendPanlab in a user-demand-driven way, targetingthe provisioning of large-scale testbeds for testingand experimentation as well as deploying futureInternet platforms for piloting novel applications.Also, interoperability with related initiatives andplatforms has already been addressed, but will beextended in the near future.

ACKNOWLEDGMENTParts of this work received funding by the Euro-pean Commission’s Sixth Framework Pro-gramme under grant agreement no. 224119. Wewould like to thank the Panlab and PII consortiafor good cooperation, Dimitri Papadimitriou,Alcatel-Lucent Bell Labs, for his contributionsto the FIRE white paper serving as input for ourFig. 1, as well as Prof. Dr. Paul Müller, TUKaiserslautern/G-Lab, for the discussions aroundfederation.

REFERENCES[1] A. Gavras (Ed.), “Experimentally Driven Research White

Paper,” v. 1, Apr. 2010; http://www.ict-fireworks.eu/fileadmin/documents/Experimentally_driven_research_V1.pdf.

[2] T. Magedanz and S. Wahle, “Control Framework Designfor Future Internet Testbeds,” Elektrotechnik und Infor-mationstechnik, vol. 126, no. 7, July 2009, pp. 274–79.

[3] S. Wahle et al., “Technical Infrastructure for a Pan-Euro-pean Federation of Testbeds,” TridentCom ‘09, Apr.2009, pp. 1–8.

[4] S. Wahle, T. Magedanz, and A. Gavras, “ConceptualDesign and Use Cases for a FIRE Resource FederationFramework,” chapter in Towards the Future Internet —Emerging Trends from European Research, IOS Press,Apr. 2010, pp. 51–62.

[5] Panlab; http://www.panlab.net.[6] Teagle Portal; http://www.fire-teagle.org.[7] J. Strassner, Policy-Based Network Management, Mor-

gan Kaufmann, 2003.[8] FCI; http://trac.panlab.net/trac/wiki/FCI.

Figure 7. The setup of the case study and participating testbeds.

Experimenter(Panlab customer)

Internet

Traffic generatorrubis_cl

Web Applicationrubis_app

Databaserubis_db

Web Applicationrubis_app

Databaserubis_db

Algorithm (SUT)rubis_proxy

Testbed AProviding Panlab controlled resources

Providing Panlab controlled resourcesTestbed B

Testbed CProviding Panlab controlled resources




We will continue to

extend Panlab in a

user-demand-driven

way, targeting the

provisioning of

large-scale testbeds

for testing and

experimentation as

well as deploying

future Internet

platforms for piloting

novel applications.



BIOGRAPHIESSEBASTIAN WAHLE ([email protected])leads the Evolving Infrastructure and Services group atNGNI within the Fraunhofer FOKUS institute in Berlin. Thegroup is active in a number of national and internationalR&D projects in the future Internet field and supports thecommercial Fraunhofer NGN testbed deployments at cus-tomer premises worldwide. He received a Diploma-Engi-neer degree in industrial engineering and managementfrom Technische Universität Berlin. His personal researchinterests include resource federation frameworks and ser-vice-oriented architectures.

CHRISTOS TRANORIS holds a Ph.D. since 2006 from the Electri-cal and Computer Engineering Department of the Universi-ty of Patras in the area of software processes on modelingand design of industrial applications. Currently he is amember of the Network Architectures and ManagementGroup of the same department, which carries out researchin the areas of future Internet, peer-to-peer, and networkmanagement while currently participating in related EUprojects.

SPYROS DENAZIS received his B.Sc. in mathematicsfrom theUniversity of Ioannina, Greece, in 1987, and in 1993 hisPh.D. in computer science from the University of Bradford,United Kingdom. He worked in European industry for eightyears, and is now an assistant professor in the Departmentof Electrical and Computer Engineering, University ofPatras, Greece. His current research includes P2P and futureInternet. He works in the PII, VITAL++, and AutoI EU pro-jects. He has co-authored more than 40 papers.

ANASTASIUS GAVRAS has more than 20 years of professionalexperience in academic and industry research. He joined

Eurescom, the leading organization for managing collabo-rative R&D in telecommunications, more than 10 years agoas program manager. His current interests are large-scaletestbed federations for future Internet research and experi-mentation. He is actively involved in several future Internetinitiatives and projects in Europe (FIA, FIRE, PII), and hasco-authored several papers and articles in the area.

KONSTANTINOS KOUTSOPOULOS received his degree of electricalengineer and Ph.D. in the field of personal and mobiletelecommunications from the National Technical Universityof Athens. He has participated in various IST projects since1998. He has experience in mobile communications, securi-ty, networking, and software development. His researchinterests include networking, embedded systems, security,and software techniques. He has been working for BCTsince March 2006.

THOMAS MAGEDANZ is a full professor in the electrical engi-neering and computer sciences faculty at Technische Uni-versität Berlin, Germany, holding the chair fornext-generation networks (http://www.av.tu-berlin.de) since2003. In addition, he is director of the Next GenerationNetwork Infrastructure (NGNI) competence center at theFraunhofer Institute FOKUS (http://www.fokus.fraunhofer.de/go/ngni).

SPYRIDON TOMBROS received his Ph.D. in broadband commu-nications from the National Technical University of Athensand his Master’s from the same faculty of the University ofPatras. His research interests are in the field of protocolsand physical communication systems design for mobile,wireless, and home networks. He has many years of work-ing experience with network test floors and test tools man-ufacturing. He has over 30 scientific publications andbooks.



Company Page

Agilent Technologies.................................................................................3, 13

CTIA...........................................................................................................22

GL Communications.................................................................................12

IEEE...........................................................................................................53

IEEE Technology Time Machine.............................................................100

IMS 2011 ....................................................................................................29

JFW Industries ..........................................................................................1

Omicron Lab..............................................................................................15

Remcom .....................................................................................................18

RT Logic.....................................................................................................5

Samsung .....................................................................................................Cover 4

SK Telecom ................................................................................................11

Synopsys .....................................................................................................Cover 2

Wiley-Blackwell .........................................................................................Cover 3

ADVERTISING SALES OFFICESClosing date for space reservation:1st of the month prior to issue date

NATIONAL SALES OFFICEEric L. Levine

Advertising Sales Manager IEEE Communications Magazine

3 Park Avenue, 17th FloorNew York, NY 10016Tel: (212) 705-8920Fax: (212) 705-8999

Email: [email protected]

SOUTHERN CALIFORNIAPatrick Jagendorf

7202 S. Marina Pacifica DriveLong Beach, CA 90803

Tel: (562) 795-9134Fax: (562) 598-8242


NORTHERN CALIFORNIAGeorge Roman

4779 Luna Ridge CourtLas Vegas, NV 89129Tel: (702) 515-7247Fax: (702) 515-7248Cell: (702) 280-1158


SOUTHEASTScott Rickles

560 Jacaranda CourtAlpharetta, GA 30022Tel: (770) 664-4567Fax: (770) 740-1399


EUROPERachel DiSanto

Huson International MediaCambridge House, Gogmore Lane

Chertsey, Surrey, KT16 9APENGLAND

Tel: +44 1428608150Fax: +44 1 1932564998


ADVERTISERS’ INDEX

APRIL 2011 EDITORIAL PREVIEW

Integrated Circuits for CommunicationsIETF Standards UpdateRecent Progress in Machine to Machine Communications

MAY 2011 EDITORIAL PREVIEWInternational Microwave (IMS 2011) Show Issue

Advances in Cooperative Wireless NetworkingTopics in Optical CommunicationsHybrid Networking: Evolution Towards Combined IP Services and Dynamic Circuits

Network CapabilitiesAutomotive Networking and Applications

LYT-ADINDEX-March 2/24/11 12:35 PM Page 176

IEEE Communications Magazine â€¢ March 2011 Volume 49 Issue 3

Documents

Transcript of IEEE Communications Magazine â€¢ March 2011 Volume 49 Issue 3