Next Generation IEEE 802.11 Wireless Local Area Networks: Current Status,Future Directions and Open Challenges

Boris Bellaltaa,, Luciano Bononib, Raffaele Brunoc, Andreas Kasslerd, Alexey Vinele

aDepartment of Information and Communication Technologies, Universitat Pompeu Fabra, 08018 Barcelona, SpainbDepartment of Computer Science and Engineering, University of Bologna, 40127, Bologna, Italy

cInstitute for Informatics and Telematics (IIT) - Italian National Research Council (CNR), 56124 Pisa, ItalydComputer Science Department, Karlstad University, 65188 Karlstad, Sweden

eCentre for Research on Embedded Systems, Halmstad University,

Abstract

A new generation of Wireless Local Area Networks (WLANs) will make its appearance in the market in the forth-coming years based on the amendments to the IEEE 802.11 standards that have recently been approved or are underdevelopment. Examples of the most expected ones are IEEE 802.11aa (Robust Audio Video Transport Streaming),IEEE 802.11ac (Very-high throughput at < 6GHz), IEEE 802.11af (TV White Spaces) and IEEE 802.11ah (Machine-to-Machine communications) specifications. The aim of this survey is to provide a comprehensive overview of these noveltechnical features and the related open technical challenges that will drive the future WLAN evolution. In contrast toother IEEE 802.11 surveys, this is a use case oriented study. Specifically, we first describe the three key scenarios inwhich next-generation WLANs will have to operate. We then review the most relevant amendments for each of theseuse cases focusing on the additional functionalities and the new technologies they include, such as multi-user MIMOtechniques, groupcast communications, dynamic channel bonding, spectrum databases and channel sensing, enhancedpower saving mechanisms and efficient small data transmissions. We also discuss the related work to highlight the keyissues that must still be addressed. Finally, we review emerging trends that can influence the design of future WLANs,with special focus on software-defined MACs and the internet-working with cellular systems.

Keywords: WLANs, IEEE 802.11, video streaming, cognitive radio, M2M, Internet of Things

1. Introduction

The IEEE 802.11 standard for Wireless Local Area Net-works (WLANs), commonly known as WiFi, is a maturetechnology with more than 15 years of development andstandardisation. The earliest version of the IEEE 802.11standard was realised in 1997 as a wireless alternative orextension to existing wired LANs using Ethernet technol-ogy. However, since its appearance, the IEEE 802.11 speci-fication has continuously evolved to include new technolo-gies and functionalities, and several amendments to thebasic IEEE 802.11 standard have been developed. WLANsare currently not only the most common Internet accesstechnology; but they have also expanded across a widevariety of markets, including consumer, mobile and au-tomotive [1]. WLANs are thus widely available every-where (homes, public hotspots, enterprise environments)and IEEE 802.11-based radio interfaces are found in manytypes of devices1.

Email addresses: boris.bellalta@upf.edu (Boris Bellalta),luciano.bononi@unibo.it (Luciano Bononi), r.bruno@iit.cnr.it(Raffaele Bruno), andreas.kassler@kau.se (Andreas Kassler),alexey.vinel@hh.se (Alexey Vinel)

1According to ABI Research, in 2013 more than two billion IEEE802.11-enabled devices were shipped.

Several factors have contributed to the success of theIEEE 802.11 family of standards, interoperability, ease ofuse, and flexibility being among the most important. First,the IEEE 802.11 standards were initially designed to beused within unlicensed spectrum bands, referred to as In-dustrial Scientific and Medical (ISM) bands. More pre-cisely, most IEEE 802.11 standards work in 2.4 GHz and5 GHz frequency bands, which are globally available, al-though local restrictions may apply for some aspects oftheir use. Thus, anyone can deploy a WLAN in thosebands given that a few basic constraints, such as a maxi-mum transmission power, are satisfied. On the downside,this also means that most WLANs are deployed in an un-controlled fashion with limited or no consideration of in-terference issues. This has made it especially challengingto guarantee performance bounds and reasonable Qualityof Service (QoS) levels. This problem is further exacer-bated by network densification, i.e., the emerging trend ofdeploying a large number of base stations in hotspot areasto cope with the increase in traffic demands [2]. A sec-ond fundamental characteristic of the IEEE 802.11 stan-dards is the adoption of a media access control (MAC)protocol called Carrier Sense Multiple Access with Col-lision Avoidance (CSMA/CA). The main reason is that

Preprint submitted to Elsevier September 27, 2021

arX

iv:2

109.

1177

0v1

[cs

.NI]

24

Sep

2021

IEEE 802.11-based systems are half duplex, i.e., a stationcannot carrier-sense/receive while it is sending, and it ishence impossible to detect a collision as in the case oftransmissions over twisted copper wires (e.g., using Ether-net). A major advantage of the CSMA/CA method is thatchannel access procedures are simple and cheap to imple-ment, as they do not impose stringent timing requirementson the radio interface. Furthermore, CSMA/CA protocolsare scalable and they provide easy support for mobility anddecentralised network architectures, from classical ad hocnetworks to emerging people-centric networks [3, 4]. Onthe negative side, CSMA/CA protocols can only providea best effort transmission service and major efforts havebeen dedicated to the design of mechanisms for supportingbetter QoS, such as in the IEEE 802.11e amendment [5].

The perceived shortcomings of the first WLAN prod-ucts have driven the evolution of the IEEE 802.11 stan-dards [6]. In particular, throughput enhancements havebeen a key priority in the IEEE 802.11 technology develop-ment. The key enabler for high-throughput WLANs wasthe adoption of new physical-layer techniques. The firstof these techniques was the orthogonal frequency-divisionmultiplexing (OFDM), which allowed achieving maximumdata rates up to 54 Mb/s. However, it is only with theadoption of the IEEE 802.11n amendment in 2009 thatthe throughput performance of WLANs came close to thatof a wired Ethernet network, as a result of the introduc-tion of multiple-input multiple-output (MIMO) technolo-gies [7]. At the same time, new amendments to the originalstandard were proposed to foster a more diversified useof WLAN products in various application domains. Forinstance, the IEEE 802.11p amendment was approved in2010. This defines enhancements to the IEEE 802.11 stan-dards to support vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communication (together referred toas V2X) in the 5.9 GHz band, which is licensed for Intelli-gent Transportation Systems (ITS) [8]. Following the samediversification strategy, the IEEE 802.11s amendment wasapproved in 2011; this described how wireless mesh net-works should operate on top of the existing IEEE 802.11MAC protocol [9]. This includes the specification of newinfrastructure-based elements needed for mesh networkingand the routing protocol to establish mesh paths betweenthese elements. In an attempt to consolidate and system-atise all the adopted IEEE 802.11 enhancements, the lastIEEE 802.11 standard (identified as IEEE 802.11-2012)was finally released to incorporate in an unique specifica-tion all the amendments published from 2008 to 2011 [10].

As pointed out above, the technological development ofthe WLAN specifications is a continuously evolving pro-cess. Thus, while the IEEE 802.11-2012 major revision ofthe IEEE 802.11 standard was finalised, the IEEE 802.11working group was also rapidly moving its focus towardsnext-generation WLANs [11]. Three key drivers were fore-casted: i) Machine-to-Machine communications ii) High-Definition Multimedia Communications and iii) “Spec-trum Sharing” in licensed bands by using cognitive radio

technology. Specifically, with the emergence of the Inter-net of Things (IoT) vision, i.e., a world were all sorts ofsmart objects (ranging from home appliances to small bat-tery powered devices) are connected to the Internet [12],a low-power WLAN technology is required [13, 14]. Atthe same time, the widespread diffusion of mobile deviceswith diverse networking and multimedia capabilities, aswell as the wide adoption of advanced multimedia appli-cations, is fuelling the growth of mobile video traffic, whichwas already more than half of the global mobile data traf-fic by the end of 2013 [15]. Thus, WLANs need specificfunctions to cope with various multimedia applications,including real-time interactive audio and video, or stream-ing live/stored audio and video [16]. Finally, new regula-tions for the unlicensed usage of TV white spaces are offer-ing new opportunities for additional spectrum utilisation,which can be particularly useful to improve rural coverageof WLANs [17]. However, cognitive radio mechanisms arerequired for enabling WLAN communications in TV whitespaces. A new generation of amendments is consequentlyunder development or has been completed since 2012 toaddress these new application requirements. The mostrelevant are the IEEE 802.11aa (approved in 2012), IEEE802.11ac (approved in 2013), IEEE 802.11ad (approved in2012), IEEE 802.11af (approved in 2013), IEEE 802.11ah(in progress, expected for 2016), and IEEE 802.11ax (inprogress, expected in 2019), among others2.

In this survey we discuss the most compelling chal-lenges of the new usage models and applications forWLANs that we have identified above. Then, basedon those scenarios, we classify and review a selectedgroup of IEEE 802.11 amendments, i.e., IEEE 802.11ac,IEEE 802.11ax, IEEE 802.11aa, IEEE, 802.11ah and IEEE802.11af, by describing the new technologies and function-alities they introduce to cope with these challenges, such asmulti-user MIMO techniques, groupcast communications,dynamic channel bonding, spectrum databases and chan-nel sensing, enhanced power saving mechanisms and effi-cient small data transmissions. A summary of the mainfeatures of these amendments in provided in Table 1. It isimportant to point out that the IEEE 802.11 specificationsdo not define all mechanisms, but they typically providethe building blocks and interfaces to allow different man-ufacturers to implement compatible procedures. Thus, wealso provide a detailed review of the main research activ-ities in the various areas and we identify open technicalchallenges. Finally, we look at emerging new trends forWLANs, with a special interest in Programmable WLANsand LTE-WiFi interworking. Overall, this survey providesa comprehensive overview of the most relevant features innext-generation WLANs, which may be of interest to bothresearchers and engineers working in the field. For the sakeof completeness, in Table 2 we also list the other on-goingIEEE 802.11 amendments that have not been analysed in

2The association between the IEEE 802.11 amendments and thedifferent use cases is specified in Section 2.

2

Amendment Release Band Goal New features

802.11aa-2012 2012 2.4, 5 GHz Robust streaming of au-dio/video streams • Groupcast communication

mechanisms

• Intra-access category prioritisa-tion

• stream classification service

• overlapping BSS management

802.11ac 2014 5 GHz Very high-throughput WLANin <6 GHz band • Channel bonding

• Multi-user Downlink MIMO

• Packet aggregation

802.11af 2014 470-790 MHz (EU) WLAN in the TV White Space

• Geolocation-based spectrumdatabases

• Channel sensing

• Non-contiguous channel bond-ing

54-72, 76-88, 174-216,470-698, 698-806 MHz (US)

802.11ah 2016 902-928 MHz (US) WLAN in the Sub 1 GHz band

• Enhanced power saving mecha-nisms

• Hierarchical station organisa-tion

• Efficient small data transmis-sions

863-868 MHz (EU)755-787 (China)

916.5-927.5 MHz (JP)802.11ax 2019 2.4, 5 GHz High efficiency WLANs

(HEW) • Dynamic channel bonding

• Multi-user Uplink MIMO

• Full-duplex wireless channel

Table 1: Summary of the IEEE 802.11 amendments that are reviewed in this survey

Amendment Release Band Goal

802.11ae-2012 2012 2.4, 5 GHz Prioritisation of management frames802.11ad-2012 2012 57.05-64 GHz (US) Very high-throughput WLAN in the 60GHz band

57-66 GHz (EU)59-62.90 GHz (China)

57-66 GHz (JP)802.11ai 2016 – Fast initial link setup802.11aj 2016 45, 59-64 GHz WLAN in the Chinese Milli-Meter Wave frequency bands802.11aq 2016 – Pre-association discovery (PAD)802.11ak 2017 – Enhancements for transit links within bridged networks

Table 2: List of other on-going and upcoming IEEE 802.11 amendments.

this survey.Given the importance of WLANs, other surveys have

been published on the IEEE 802.11 standards. Earlier sur-veys primarily focused on presenting the different classesof proposed MAC protocols [18]. A complete overviewof the wealth of amendments that have been accepted orwere in the process of being standardised before 2010 isprovided in [1]. More recently, other surveys have givendetailed consideration to specific amendments (e.g., IEEE802.11s [19]), or classes of similar amendments [16, 20, 11].

However, none of the existing surveys follows our use-caseoriented approach and covers in such detail all the amend-ments that we believe will be relevant in coming years.We also include some of the latest advances and relatedresearch.

The structure of this survey is illustrated in Figure 1and explained in the following. In Section 2 we introducethe four key scenarios for WLAN technologies that are con-sidered here. In Section 3 we focus on high-throughputWLANs, presenting the IEEE 802.11ac, IEEE 802.11ax

3

!"#$%%&'()*+,-./012

!"#$%%&3()*+,-./042

!"#$%%&+()*+,-./05$%2

!"#$%%&6()*+,-./05$#2

!"#$%%&&()*+,-./05$52

7*8898&:;<=>?@/,*:/*,A.:B-/C()*+,-./0D$52

E8*6-F8*G&:HA&:*()*+,-./0D$#2

I:.C:&JJ&F8*<=>?K

()*+,-./0D$%2

L#L 7.C/-,-M* GN0L98,-J*H-&

!"#$$%&'()"*+$$,)$

-./011*'2)342)35$

6$)*7'$)$

Figure 1: Survey organisation

and IEEE 802.11aa amendments. Section 4 discusses theIEEE 802.11ah amendment to support M2M communica-tions and we review the IEEE 802.11af for WLANs oper-ating in TV white spaces. Finally, Section 6 presents someemerging trends for future WLANs.

2. Future Scenarios & New Challenges

WLANs can be found everywhere. They are common inhomes, offices, public parks in cities, shops, airports andhotels, among many different places. Today’s WLANs areable to provide a fast and reliable wireless access to In-ternet for browsing the web, exchanging files, chatting,receiving and answering e-mails, and for low-quality real-time audio/video streams, as just a few representative ex-amples of their current usage. This situation is changingrapidly however. The number of persons that use Internetapplications and objects that are connected to the Inter-net is growing every day, proportionally to the number ofnew applications and services that constantly appear. Thisclearly results in a steady increase of the Internet traffic.Two representative examples of the change in Internet useare: (i) the high demand for mobile-rich multi-media con-tent, mainly motivated by the use of smart-phones, tabletsand other multimedia portable devices; and (ii) the in-creasing interest in IoT applications driven by the almostubiquitous existence of devices able to collect data fromthe environment, ranging from low-power sensor nodesto connected cars. Therefore, WLANs must also evolveto provide effective solutions to these new upcoming sce-narios, and the challenges they pose to satisfy their re-quirements. Four of the key use cases for next-generationWLANs are discussed in the following subsections.

2.1. High-Quality Multimedia Content Delivery

Our new mobile and portable devices are designed tohandle rich multimedia contents, including high-definitionvideo and images. Table 3 reports the requirements interms of maximum data rate and latency for some of themost common real-time video applications [21]. Key sce-narios in which the support of real-time video transmis-sion is required of course include Internet TV and video

Type Max data rate Max latency

Uncompressed raw video 1.49 Gbit/s 100 msuncompressed HDTV 150 Mbit/s 150 msBlue-ray Disc 54 Mbit/s 200 msMPEG2 HDTV 19.2 Mbit/s 300 msMPEG4 HDTV 8 to 10 Mbit/s 500 ms

Table 3: Performance requirements for different HD streaming ap-plications

streaming. Similarly, scenarios in which multiple usersconnect to the same wireless network to request differ-ent multi-media content at the same time are increasingevery day. However, not all multi-media content is realtime. Stored video and image files can also be exchangedbetween different devices. Those files can have sizes rang-ing from a few Megabits to several Gigabits, hence re-quiring a high network transport capacity in order to pro-vide a good Quality-of-Experience to end users. Althoughvideo encoding schemes exist that offer substantial videocompression efficiency, such as H.264/MPEG-4 AVC [22],WLANs must be able to achieve very high transmissionrates and have content-aware mechanisms that are specif-ically designed for multi-media applications to ensure asatisfactory service for multimedia delivery. The mecha-nisms that are considered by various IEEE 802.11 stan-dardisation groups to satisfy those requirements are de-scribed in Section 3, such as group-cast communicationprotocols, single and multi-user spatial multiplexing andchannel bonding among others to make the communicationmore efficient, and offer higher transmission rates. Thereference IEEE 802.11 amendments for high-quality multi-media content delivery are IEEE 802.11aa, IEEE 802.11ac,and IEEE 802.11ax.

2.2. Machine-to-Machine (M2M) Communications

The almost ubiquitous presence of sensor/actuator devicesthat are able to interact with the environment has fosteredthe creation of new services and applications. Conceptssuch as smart cities and smart grids are being developedon the basis of the existence of those sensor/actuator net-works to achieve a more sustainable use of the environmen-tal resources and provide citizens with a higher quality of

4

life [23, 24].In a classic sense, Wireless Sensor Network (WSN)

technologies are used to collect data from spatially dis-tributed sensor nodes and to transmit the data over amulti-hop wireless network to a central sink [25]. TheM2M paradigm is broadening the scope of the WSN con-cept because it enables networked devices, wireless and/orwired, as well as services, to exchange information or con-trol data seamlessly, without explicit human intervention.Clearly, M2M communications face most of the techni-cal challenges of WSNs. One of the main limitations ofWSNs and M2M systems is that the network nodes areusually battery powered or have limited access to powersources. Designing mechanisms and protocols to reducetheir power consumption with the goal of extending thenetwork lifetime is therefore crucial for the successful com-mercial take-up of these kinds of networks. Fortunately,devices in M2M systems typically generate or consume alimited amount of data per unit of time. Thus they canspend a large fraction of their time sleeping. This facil-itates energy saving at the cost of additional complexityfor the channel access and networking protocols.

Popular wireless protocol standards for M2M commu-nications are Bluetooth, ZigBee and BT-LE [12]. An alter-native, promoted by mobile networks, is to connect devicesin M2M systems directly to the Internet by using the cellu-lar network infrastructure, for which specific protocols arebeing developed [26]. WLANs are envisioned as an alterna-tive to both multi-hop WSNs and cellular networks. How-ever, current WLANs are not able to satisfy the minimumrequirements for M2M communications [13]. Novel specificpower-saving mechanisms are required to support the longperiods of inactivity needed by the sensor/actuator devicesand to manage the thousands of nodes associated with asingle AP. These challenges will be discussed in Section 4,when presenting the IEEE 802.11ah amendment.

2.3. Efficient Use of the Spectrum

The ISM bands are used by several wireless communicationtechnologies, including IEEE 802.11, IEEE 802.15.4 andLong Term Evolution (LTE)-Unlicensed networks. Thisresults in a high spectrum occupancy. Unfortunately, wire-less networks operating in the same spectrum region cansuffer from mutual interference, which might degrade theperformance of all of them. This is exacerbated by theuncontrolled deployment of wireless networks in the ISMband, which is typically very common in urban environ-ments. For example, let us consider a building with severalapartments and a WLAN in each one. There would easilybe several WLANs operating in overlapping channels andsuffering mutual interference [27]. To deal with this issue,it is expected that new APs will increasingly incorporateDCA (Dynamic Channel Allocation) mechanisms to selectand update their operating channel at run-time.

An alternative approach to alleviate the spectrum oc-cupancy problem is to move to a different part of the spec-trum, even if the new part of the spectrum is occupied by

communication systems operating under a license. In thatcase, WLANs would be the secondary users and there-fore must avoid causing interference to the primary users.In recent years, the change from analogue to digital TVbroadcast emissions has resulted in a reorganisation of thespectrum at VHF/UHF bands. This reorganisation hasshown that there are many empty TV channels, called TVwhite spaces, that can be used for data communication,especially in rural regions [28]. Furthermore, WLANs op-erating in those TV white spaces can take advantage ofradio propagation properties in the UHF band to providelarge coverage areas. The challenges to be addressed touse CSMA/CA protocols in VHF/UHF bands, as well ashow to obtain higher transmission rates when the spec-trum is fragmented, will be discussed in Section 5, whenpresenting the IEEE 802.11af amendment.

3. High Performance WLANs for Multimedia Ap-plications

This section reviews the IEEE 802.11ac, IEEE 802.11axand IEEE 802.11aa amendments. These three amend-ments target multimedia scenarios by introducing newphysical-layer technologies and MAC functionalities to im-prove the WLAN capacity and QoS provision. Applicationexamples include home scenarios in which an WLAN APcan act as an Internet gateway and wireless media serverfor home appliances (e.g., IPTV set-top boxes, projectors,game consoles) and content storage devices. A possibleuse case is illustrated in Figure 2.

Figure 2: High-throughput demanding multimedia devices associ-ated to an IEEE 802.11ac/ax AP.

3.1. The IEEE 802.11ac amendment

IEEE 802.11ac [29] aims to provide users with a through-put close to 1 Gbps, which represents a roughly four-foldincrease with respect to IEEE 802.11n [7]. Comparedto IEEE 802.11n, IEEE 802.11ac supports larger channelwidths (up to 160 MHz), introduced a new modulationscheme, i.e., a 256-QAM modulation, and downlink mul-tiuser MIMO (DL-MU-MIMO).

5

3.1.1. Novel Features

The most relevant new features included in IEEE 802.11acare described in the following.

Channel Bonding. IEEE 802.11ac enables the use of chan-nel bandwidths of 20, 40, 80 (mandatory) and 160 MHz(optional). Channel bandwidths larger than 20 MHz arecreated by “bonding” (i.e., grouping) a group of consecu-tive 20 MHz channels, and aim to offer higher transmissionrates.

Two extensions have been proposed in IEEE 802.11acfor the basic DCF (Distributed Coordination Function)access method in order to support channel bonding: (i)the Static Bandwidth Channel Access Protocol (SBCA),which always transmits over the same group of 20 MHzchannels, and requires that all sub-channels are idle beforestarting a packet transmission; and (ii) the Dynamic Band-width Channel Access scheme (DBCA), which is able todynamically adapt the channel width to the instantaneousspectrum availability [30, 31]. As expected, in dense sce-narios the use of DBCA offer a much better performancethan SBCA due to adaptability [32].

To avoid hidden terminals operating in any of the 20MHz bonded channels, the IEEE 802.11ac amendment in-cludes extended RTS/CTS frames in order to signal themaximum channel width that can be used at both thetransmitter and the receiver. In case the CTS includes alower channel width than the RTS, the transmitter willadopt it. Similarly to the ACK frames, when the RTS andCTS frames are transmitted, they are duplicated over allthe 20 MHz sub-channels used. Note that this enhancedRTS/CTS mechanism is also needed to facilitate the co-existence between IEEE 802.11ac AP and other nearbylegacy APs, as the latter could transmit at overlappingtimes on different sub-channels. The operation and perfor-mance of channel bonding in WLANs is thoroughly anal-ysed in [27], showing the new interactions between neigh-bouring WLANs that may appear and their impact in thethroughput of each one.

Downlink Multiuser MIMO. The main novelty introducedby the IEEE 802.11ac amendment compared with theIEEE 802.11n one is the support of MU-MIMO transmis-sions in the downlink, hence allowing multiple simultane-ous transmissions from the AP to different STAs. In theIEEE 802.11ac amendment, the AP can be equipped witha maximum of eight antennas and send up to four spatialstreams to two different users, or up to two spatial streamsto four different users at the same time.

When an IEEE 802.11ac AP performs a multi-usertransmission it specifies the group of STAs to which thattransmission is directed. This information is contained inthe new IEEE 802.11ac PHY headers, which are broadcastomni-directionally to all STAs. The way STAs are groupedis decided by the AP after obtaining the channel state in-formation (CSI) feedback from all STAs. To gather theCSI information by the AP, IEEE 802.11ac considers only

an explicit channel sounding feedback mechanism calledExplicit Compressed FeedBack (ECFB). The channel ac-cess is governed by EDCA (Enhanced Distributed ChannelAccess). At each transmission attempt, the multiple ac-cess categories (AC) managed by the AP should contendfor the channel medium, as only one AC can be served foreach transmission attempt. In the case that the queue as-sociated with the AC that has won the internal contentiondoes not contain packets to enough different destinationsto fill all the available spatial streams, it can decide toshare the remaining ones with the other ACs.

Packet Aggregation. To increase the efficiency of eachtransmission by reducing unnecessary overheads, IEEE802.11ac allows the transmission of several MPDUs ag-gregated in a single A-MPDU. Then, to acknowledge eachMPDU individually a Block ACK packet is used, whichcontains a bitmap to indicate the correct reception of allincluded MPDUs. Thus, leveraging on the informationcontained in the Block ACK, the transmitter is able toselectively retransmit only those MPDUs that have failedinstead of the whole A-MPDU.

3.1.2. Open Challenges

Since the IEEE 802.11ac amendment has recently been fi-nalised, current research around it should cover two mainaspects: a) understanding the performance bounds ofIEEE 802.11ac, which entails the development of new mod-els, simulation tools and experimental platforms of IEEE802.11ac-based WLANs, and b) proposing specific solu-tions for those aspects that are not defined by the IEEE802.11ac amendment on purpose, such as the mechanismfor creating the groups of STAs for DL-MU-MIMO trans-missions, smart packet schedullers able to decide when theuse of DL-MU-MIMO outperforms SU-MIMO transmis-sions, and the implementation of the TXOP sharing fea-ture between several ACs. The results and conclusionsobtained in both cases will be very valuable in the devel-opment of IEEE 802.11ac technologies, as well as in theconception of the future amendments that will substituteIEEE 802.11ac in four to five years, such as the recentlyinitiated IEEE 802.11ax.

Following the first mentioned research direction, thereare several efforts that have focused on understanding boththeoretical and experimental performance bounds of IEEE802.11ac. The maximum downlink throughput that anIEEE 802.11ac AP can achieve when packet aggregation,channel bonding and different spatial stream configura-tions are considered is presented in [33]. In [34], the au-thors evaluate the IEEE 802.11ac performance experimen-tally using commodity devices, focusing on the effects thatthe use of wider channels, the 256-QAM modulation andthe number of SU-MIMO spatial streams have in terms ofthroughput and energy consumption. It is worth mention-ing that DL-MU-MIMO was not yet implemented in theequipment they were using, and that feature was there-fore not considered. The evaluation of a DL-MU-MIMO

6

implementation for WLANs using the WARP platform ispresented in [35], where a deep evaluation of the potentialbenefits of DL-MU-MIMO transmissions is done in termsof the location of the receivers, number of users, and usermobility, among other aspects. A solution that combinesboth packet aggregation and DL-MU-MIMO transmissionsis presented in [36]. Results show the need of properly di-mensioning the buffer space to achieve the full potentialof such a combination. In [37], the authors compare thethroughput achieved by IEEE 802.11n and IEEE 802.11acwhen packet aggregation is used, with and without chan-nel errors. They show that in most cases the packet ag-gregation mechanism introduced in IEEE 802.11ac out-performs the one in IEEE 802.11n. An analytical modelto evaluate the performance of the IEEE 802.11ac TXOPsharing mechanism in DL-MU-MIMO communications isdeveloped in [38]. The main goal of this study is to iden-tify how the TXOP sharing mechanism could improve thesystem efficiency while achieving channel access fairnessamong the different ACs.

How to optimally exploit the new DL-MU-MIMO ca-pabilities provided by IEEE 802.11ac is still an open chal-lenge. First, due to the need of frequent CSI exchangesbetween STAs and the AP, it is not yet clear in whichconditions DL-MU-MIMO outperforms SU-MIMO [39, 40,41, 42], or even whether MU-MIMO does or does not out-perform multi-user packet aggregation when the amountof data directed to each destination is not balanced [43].Packet aggregation can be a solution to balance the du-ration of the multi-user spatial streams as shown in [36],although it will always depend on the amount of trafficdirected to each destination and the buffer capacity at theAP. In [44], the authors compare different strategies toassign the spatial streams between the available destina-tions at each transmission in a fully connected mesh net-work, showing in ideal channel conditions the theoreticalbenefits of MU-MIMO vs. SU-MIMO.

Closely related to the previous point, a second openchallenge is the design of efficient schedulers that con-sider traffic priorities, the buffer state, the different MIMOstrategies, TXOP sharing policies, grouping of STAs andthe availability of fresh CSI feedbacks to maximise thethroughput and guarantee the required QoS for each activetraffic flow. It is important to consider that the availabil-ity of updated CSI estimates from all STAs allows the APto reduce the mutual interference between the transmittedspatial streams, which means lower packet error probabili-ties and higher transmission rates. However, the overheadsfor obtaining the CSI from all STAs is large, and increaseslinearly with the channel sounding rate and the number ofSTAs. Proposals for reducing the CSI overhead are underdevelopment. For example, in [40], the CSI overhead isreduced by inhibiting the channel sounding whenever pos-sible. Another open problem is how to group the STAs,as the goal is to find groups of STAs with compatible (i.e.,orthogonal) channels. In [45], the authors show the chal-lenges inherent to the group assignment problem, and they

propose an heuristic method to solve them. TXOP shar-ing is considered in [46] by presenting two alternative ap-proaches to enhance the considered back-off procedure forthe purpose of improving both throughput and fairness.

A third key challenge for IEEE 802.11ac is to achieve anefficient use of the spectrum when several channel widthsare used in scenarios with multiple overlapping WLANs.Increasing the channel width theoretically allows individ-ual WLANs to achieve a higher throughput. However, thepresence of other WLANs in the vicinity also increases thechances of frequency overlapping, which may cause theopposite effect as there appears inter-WLAN contention[27]. Adaptive mechanisms to select the channel cen-tre frequency and the channel width, and MAC proto-cols to choose the instantaneous channel width used foreach transmission are thus required. For instance, in [47],the authors focus on the channel selection problem whenWLANs can use multiple channel widths using a game-theoretic framework. In [48] a scheme is proposed to en-able the communication between nodes with partially over-lapping channels, which may provide stronger resilience tochannel interferences.

3.2. The IEEE 802.11ax amendment

In 2014 the High Efficiency WLANs (HEW) Task Group[49] initiated the development of a new IEEE 802.11amendment, called IEEE 802.11ax. The IEEE 802.11axamendment is expected to be released in 2019, and, tosome extent, it will be the IEEE 802.11 response to theexpected challenges of future dense WLAN and high-bandwidth demanding scenarios [50, 51].

The open challenges that are considered in the devel-opment of the IEEE 802.11ax amendment are to:

(i) Improve the WLANs performance by providing atleast a four-fold capacity increase compared to IEEE802.11ac.

(ii) Provide support for dense networks, consideringboth the existence of multiple overlapping WLANsand many STAs in each of them. Spatial reuse of thetransmission resources is a must.

(iii) Achieve an efficient use of the transmission resourcesby minimising the exchange of management and con-trol packets, revisiting the structure of the pack-ets, and improving channel access and retransmis-sion mechanisms, among others aspects.

(iv) Provide backward compatibility with previousamendments. This is achieved by the mandatorytransmission of the legacy PHY preamble in allframes, and by keeping EDCA as the basic channelaccess scheme.

(v) Introduce effective energy saving mechanisms tominimize the energy consumption.

(vi) Support multi-user transmission strategies by fur-ther developing MU-MIMO and Orthogonal Fre-quency Division Multiple Access (OFDMA) capabil-ities in both downlink and uplink.

7

In addition to the aforementioned challenges, next-generation WLANs will have to implement some otherfunctionalities beyond the raw packet transmission and re-ception. Examples are a fast, efficient and robust hand-off between APs in the same administration domain [52],device-to-device communication (D2D) [53] and coordina-tion of multi-AP networks [54]. In the first case, the IEEE802.11ai amendment, called Fast Initial Link Setup, is inprogress and expected for 2016. Its target is to complete ahandoff in less than 100 ms, including new AP discovery,user authentication and configuration. Using D2D com-munication, we can avoid the use of the AP as a relay,hence improving the overall efficiency as the number ofpacket transmissions required is reduced. Finally, the vir-tualisation of network functions adds a new dimension inthe management of multiple APs, which in dense scenarioscan contribute to notably improving the user experience.We further discuss this last topic in Section 6.

Different from the other amendments covered in thissurvey, the IEEE 802.11ax amendment is just in its ini-tial stages of development, with only very few technicalaspects consolidated at this stage. Therefore, in the fol-lowing subsection, we will overview both the new featuresand open challenges of the IEEE 802.11ax amendment si-multaneously.

3.2.1. Novel features & Open Challenges

The IEEE 801.11ax Task Group is currently working infour areas: PHY, MAC, Multi-user, and Spatial Reuse.Next, we will overview some of the topics currently underdiscussion in the IEEE 802.11 Task Group in each one.

PHY layer. The IEEE 802.11ax PHY layer will be an evo-lution of the IEEE 802.11ac one. The challenges in thedesign of the IEEE 802.11ax PHY layer are related withthe extensions required to support multi-user MU-MIMOand OFDMA transmissions, and Dynamic CCA. Also, im-provements in the supported modulation and channel cod-ing techniques will be likely considered to allow for highertransmission rates at lower SNR values. For example,IEEE 802.11ax may consider LDPC (Low-Density ParityCheck) coding, which are optional in IEEE 802.11ac, in-stead of the traditional convolutional codes, as they pro-vide a coding gain of 1-2 dB [55]. Moreover, the PHY layermay also include some functionalities to support the use ofHybrid ARQ schemes to improve the efficiency of packetretransmissions.

Medium Access Control. In order to keep backward com-patibility with previous IEEE 802.11 amendments, besidesa common PHY frame preamble, compatible MAC proto-cols are required. This means that it is likely that EDCAwill be kept as the main channel access technique in theIEEE 802.11ax amendment. Therefore, the most relevantopen challenges are related to EDCA extensions to supporta large number of STAs, improve traffic differentiation ca-pabilities, improve the energy consumption and provide

mechanisms to fairly co-exist with neighboring wirelessnetworks.

To support a large number of contenders with a lowcollision probability a simple solution is to set a largerbackoff contention window compared to the value usedin the IEEE 802.11ac amendment for all ACs. To miti-gate the extra backoff duration when using larger back-off contention windows, the slot duration can be reducedif the time required to perform the CCA, to switch be-tween reception and transmission modes, and the packetprocessing delay are reduced. Another approach is to con-sider decentralised collision-free MAC protocols to enhancethe underlying CSMA/CA mechanism in EDCA. ThoseMAC protocols are able to build collision-free schedules,thus improving the network efficiency as collisions are re-duced, while preserving backward compatibility with thedefault EDCA implementation. The benefits of decen-tralised collision-free MAC protocols can be found in [56],including a comparative evaluation of several key proto-cols, including CSMA/ECA [57]. CSMA/ECA is speciallyrelevant since it is fully compatible with EDCA and is ableto adapt to the instantaneous number of contenders. Inany case, IEEE 802.11ax WLANs can rely on the IEEE802.11aa amendment to further improve EDCA traffic dif-ferentiation capabilities with intra-AC traffic differentia-tion and groupcast communication mechanisms, amongother features. We overview the IEEE 802.11aa amend-ment in Section 3.3.

IEEE 802.11ax will likely keep the same channel widthsthat were defined in the IEEE 802.11ac amendment, i.e.,20, 40, 80 and 160 MHz. However, it is expected thatIEEE 802.11ax will extend the use of channel bonding tofurther improve the spectrum utilisation and the coexis-tence between neighbouring WLANs. For example, it hasbeen shown in [32] that the use of dynamic channel bond-ing provides significant throughput gains in dense scenar-ios compared with the static approach, while minimizingthe inter-WLAN negative interactions [27]. Furthermore,additional mechanisms are required to fully exploit the po-tentials of using wider channels, such as the use of efficientalgorithms to select the position of the primary channel,or even to consider multiple primary channels to increasethe number of bonded channel combinations that can beused for transmission.

The MAC layer in IEEE 802.11ax may work with thePHY layer to implement an efficient Hybrid ARQ mecha-nism able to retransmit short packets containing only in-cremental redundancy bits. Opportunistic piggy backingof data packets in ACKs and viceversa may further im-prove the efficiency of IEEE 802.11ax WLANs by reduc-ing the number of transmissions in a bidirectional dataexchange [58]. Finally, the packet headers can be reducedif shorter STAs identificators are used instead of MAC ad-dresses, and unnecessary fields for the given transmissionare not included.

8

Multi-User. Multi-user communications will likely be oneof the main characteristics of IEEE 802.11ax, as both up-link and downlink MU-MUMO and OFDMA are underconsideration. The use of multi-user communication tech-niques does not necessarily represent a system capacity in-crease because the available transmission resources may bethe same as in the single-user communication case. How-ever, in WLANs, the simultaneous transmission from dif-ferent users is able to parallelize the large temporal over-heads of each transmission (i.e., DIFS, SIFS, ACKs, packetheaders, etc.) which can notably improve the WLAN effi-ciency.

IEEE 802.11ax will further develop the MU-MIMO ca-pabilities of IEEE 802.11ac by allowing multiple simulta-neous transmissions in the uplink, which is known as up-link MU-MIMO. The performance benefits of uplink MU-MIMO have been already extensively studied, showing itsbenefits but also the requirements and challenges to imple-ment such a solution [59]. Similar to DL-MU-MIMO trans-missions, an open challenge to enable uplink MU-MIMOis to design a mechanism able to efficiently schedule theusers that will transmit simultaneously. In one hand, apure decentralized approach would be easy to implementwith minimal signalling overheads. However, since it re-quires that all STAs finish their backoff at the same timeit may show a very low efficiency, besides that those STAsmay not be compatible in terms of their spatial signature.In the other hand, a pure centralized approach requiresthat the AP has complete CSI and buffer occupancy infor-mation from all STAs to select the most suitable group toperform a multi-user transmission. Once a suitable groupof STAs is selected by the AP, a ”Trigger” frame may beused to notify the group of selected users that can initiatea transmission. This approach guarantees efficient multi-user transmissions but requires some extra overheads tocollect all the required information by the AP and signalthe selected STAs. In both cases, new multi-user ACKswill be likely introduced by IEEE 802.11ax to acknowl-edge all transmissions with a single control packet.

Multi-user OFDMA is also in the agenda for IEEE802.11ax. Using OFDMA, a channel can be split in sev-eral sub-channels and assigned to different users. Likely,OFDMA will be implemented in combination with chan-nel bonding, where each of the 20 MHz subchannels canbe assigned to a different user, in both downlink and up-link. Besides that, a similar operation as in the multi-userMIMO case is expected, as there are almost the same chal-lenges to solve. A survey of current OFDMA proposals forWLANs is presented in [60], showing also how the use ofOFDMA is able to significantly improve the WLAN effi-ciency.

In addition to Multi-user MIMO and OFDMA, the useof Simultaneous Transmit and Receive (STR) techniques,commonly known as full-duplex transmission, have beensuggested for IEEE 802.11ax [50, 51]. Using STR a pair ofnodes is able to transmit and receive simultaneously, whichtheoretically doubles the channel capacity. The challenge

is that both the AP and the STA involved in a full-duplextransmission start to transmit at the same time. There-fore, information about full duplex transmissions can beincluded in control packets or in the PHY headers fromthe transmission initiator.

Spatial Reuse. Dense WLAN deployments are necessaryto offer a continuous coverage with high transmission rates.To improve both the co-existence with those neighboringnetworks and the spatial reuse of the spectrum, a WLANhas two options: (i) minimise its area of influence by reduc-ing its transmit power, and (ii) accept higher interferencelevels by increasing the Clear Channel Assessment (CCA)level. Use of both techniques may increase the number ofconcurrent transmissions between neighbouring WLANs,and therefore their capacity, although it may also result inthe opposite effect since the achievable transmission ratesmay be negatively affected by the higher interference levelsobserved, which is the main challenge to be solved.

Due the high WLAN dynamics, the use of adaptive sys-tems is crucial but challenging as adaptivity requires extracomplexity in terms of computing and memory resources,and there are not guarantees that the implemented solu-tion converges due to the decentralized operation of eachWLAN. The use of DSC (Dynamic Sensitivity Control) todynamically adjust the CCA level is one of the aspects cur-rently under discussion in the IEEE 802.11ax Task Group.Initial works evaluating the performance of DSC for IEEE802.11ax WLANs show a clear improvement on the spa-tial reuse and the area throughput [61]. Another exampleof the achievable throughput gains obtained by adaptingthe CCA can be found in [62], where the authors showthat gains up to 100 % can be achieved. Moreover, trans-mit Power Control (TPC) to mitigate interference betweenWLANs in dense scenarios is studied in [63], showing theneed of jointly optimising both TPC and CCA to maximisethe network performance.

Finally, sectorization by using beamforming is alsounder consideration for the development of the IEEE802.11ax amendment as a potential solution to improvespatial reuse [64]. Using sectorization, only the nodes of agiven area are allowed to receive or transmit data, hencereducing the contention between different networks when-ever they activate non-overlapping sectors. A challengehere is to coordinate the different neighboring APs whenthey belong to different administration domains. Decen-tralized learning approaches may be implemented to findfeasible temporal patterns of non-overlapping sectors.

3.3. The IEEE 802.11aa amendment

As discussed above, legacy IEEE 802.11 standards donot provide robust and efficient delivery of audio/videostreaming services. Thus, the IEEE 802.11aa amend-ment was developed to include new features and addi-tional mechanisms to improve the performance of real-time multi-media content delivery [65]. Specifically, IEEE

9

802.11aa addresses the following five shortcomings of pre-vious 802.11 standards [66, 16]:

(i) the lack of reliable and efficient support for multicastand group communications;

(ii) the incapacity of applying traffic prioritisation todifferent multimedia streams or different types offrames from the same stream;

(iii) the absence of methods for cooperative resourcesharing among neighbouring APs;

(iv) the lack of mechanisms for graceful degradation ofaudio/video streaming quality;

(v) the non interoperability with existing IEEE 802.1standards for Audio Video Bridging (AVB).

In the following sections we present in detail the solutionsto those problems introduced in the IEEE 802.11aa amend-ment. We further discuss the research studies that haveprovided the basis for the IEEE 802.11aa design and weidentify the remaining open challenges.

3.3.1. Novel features

Groupcast communication mechanisms. In most au-dio/video streaming applications a group of clients mustreceive the same stream simultaneously. A multicastprotocol is necessary to avoid that the same content isreplicated throughout the network. In wireless networks,multicast transmission can exploit the intrinsic broadcastnature of the wireless channel, i.e., broadcast transmis-sions from an AP are physically received by all otherstations in the same collision domain. However, multicastand broadcast frames in IEEE 802.11 networks are not pro-tected by an acknowledgement mechanism as in the case ofunicast frames. Thus, layer-2 multicast transmissions de-fined by legacy IEEE 802.11 standards are unreliable andnot suitable for streaming applications. To partially ad-dress this limitation, the Direct Multicast Service (DMS)was first specified in the IEEE 802.11v amendment [67].Basically, DMS converts multicast streams into unicaststreams. In this way, frames destined to a multicast ad-dress are individually transmitted as unicast frames tothe stations that joined that multicast group. ObviouslyDMS provides the same reliability as unicast transmissionservices but the consumed bandwidth increases linearlywith the number of group members. To address this scal-ability issue, IEEE 802.11aa includes the Groupcast withRetries (GCR) service in addition to DMS. Specifically,the GCR service defines new mechanisms and the relatedmanagement frames for group formation, which allowsa set of stations to agree on a shared (non-multicast)address, called the groupcast concealment address3. Fur-thermore, the GCR service specifies two retransmissionpolicies: GCR Unsolicited Retry (GCR-UR) and GCR

3The concealment address protects legacy stations, i.e., GCR-incapable stations, from receiving duplicated group-addressedframes.

Block Ack (GCR-BA). When using GCR-UR, the APcan proactively retransmit all groupcast frames a numberof times to mitigate the impact of channel errors (seeFigure 3.a)), while receivers are not requested to sendacknowledgements. Intuitively this approach improvestransmission reliability, but it still suffers from scalabilityissues. In contrast, when GCR-BA is used the AP sends aburst of consecutive groupcast frames and it requests thereceivers to reply with a Block ACK frame, which containsa bitmap to positively or negatively acknowledge transmit-ted frames (see Figure 3.b)). The Block ACK mechanismdefined for the GCR-BA service is quite flexible becauseBlock ACK frames can be requested immediately after atransmission burst or after a randomised back-off delay.Furthermore, the AP can request the Block ACK frameto all groupcast recipients or only to a subset of them toreduce overheads and delays. The advantages of the GCRmethods over broadcast and DMS have been extensivelydemonstrated in the literature [16, 68].

(a) GCR−UR example

AP

#1 #1 #1

DATA packets, K copies DATA packets, K copies

#2 #2 #2

DATA packets, K packets

#A #B

AP

#1 #2

(b) GCR−BA example

Block ACKs

#A

#B

t

t

GCR member A

GCR member B

GCR member A

GCR member B

#K

X

X

X

X

X

X

X

X

Block ACK Requests

Figure 3: GCR service with different retransmission schemes

Intra-access category prioritisation. The IEEE 802.11eamendment only allows traffic differentiation between fourdifferent access categories (ACs) that are broadly mappedto four application classes: voice (VO), video (VD), best-effort (BE), and background (BK). However, there is avariety of streaming services, ranging from simple video-conferencing to HD streaming over IPTV systems, whichhave different QoS requirements (see Table 3). To providethe ability to differentiate among individual streams, IEEE802.11aa includes an additional scheduling layer with re-spect to IEEE 802.11e. IEEE 802.11aa splits each one ofthe transmission queues associated with voice and videoACs into a primary and an alternate queue. In this way,specialised scheduling rules can be applied to decide whichqueue to serve when the EDCA function for inter-ACcollision resolution grants an access opportunity to voiceor video ACs. To facilitate the management of servicelevel agreements, IEEE 802.11aa follows the default map-pings between user priority values and traffic types thatare defined in the IEEE 802.1D standard [69]. It is thenstraightforward to further map traffic types onto transmis-

10

sion queues and ACs (see Figure 4). Finally, it is impor-tant to point out that the intra-AC differentiation func-tionality can be used to provide more sophisticated trafficdifferentiation than simple stream prioritisation. For in-stance, most video applications use Scalable Video Cod-ing (SVC) schemes that enable the partitioning of a videosequence into multiple layers with different qualities andrates [70]. Typically, an SVC-based video stream containsa base layer, which provides a basic level of quality, andmultiple enhancement layers, which can only be decodedtogether with the base layer to improve the video quality.Thus, the different layers of the same encoded video steamcan be easily mapped to different transmission queues toreceive differentiated QoS [71].

Stream classification service. The stream classificationservice (SCS) is an optional service that can be providedby an AP to the associated stations to classify multimediastreams based on arbitrary rules that are established di-rectly by the stations instead of the conventional 802.1Duser priorities. To this end the station requesting the useof SCS must initiate an SCS session by sending an SCSrequest frame to the AP, which contains an identifier forthe SCS stream and the descriptor of the classificationrule. The AP may accept or reject the requirements spec-ified by the station. Once accepted, the AP must assignall frames that match the classification rule to a specificAC. When intra-access category prioritisation is enabled(see Section 3.3.1), matching frames can also be assignedto the primary or alternate transmit queues so that finergrained prioritisation can be applied. Finally, there is alsoa Drop Eligibility Indicator (DEI) bit in the SCS descrip-tor that indicates whether frames from this traffic streamcan be dropped in the case that there are insufficient re-sources. Specifically, frames with the DEI bit set to onehave a higher probability of being discarded because theirmaximum number of allowed retries is smaller than thedefault. Note that how to combine intra-AC queues andframe dropping settings to achieve graceful degradation ofthe audio/video stream quality in case of bandwidth short-age is beyond the scope of the IEEE 802.11aa specification.

Overlapping Basic Service Set (OBSS) management. Net-work densification, i.e., a denser deployment of wireless in-frastructure nodes, is one of the key strategies that is usednowadays to easily increase the capacity of wireless sys-tems, even for indoor WLANs [72]. However, IEEE 802.11networks have a limited number of orthogonal channelsavailable and, even if optimised frequency planning is ap-plied, it might happen that neighbouring APs are mutuallyinterfering and a station may affect multiple overlappingBSSs. In this case, congestion not only increases but it isalso likely to observe an unfair usage of wireless capacitywith the channel retained by one AP for long time inter-vals. This is mainly due to the neighbourhood captureeffect, i.e., hidden terminal phenomena among APs [73].To address this issue, IEEE 802.11aa specifies a new func-

tionality, called Overlapping BSS (OBSS) management,which is based on two new mechanisms. The first definesa set of parameters to quantify the load and interferenceamong neighbouring BSSs, such as medium occupancyfraction, number of admitted audio/video streams, datatraffic volumes, and the number of BSSs that are usingthe same channel as the target one. Note that the traf-fic load consists of two components: the allocated traffic,which is derived on the basis of the TSPEC values of ad-mitted streams4, and predicted traffic, which is evaluatedby tracking the maximum value of the allocated EDCAand HCCA traffic over seven-day periods. Once load mea-surement reports are exchanged among the APs, a secondOBSS component is responsible for coordinated admissioncontrol procedures on the basis of two suggested shar-ing schemes: proportional sharing and on-demand shar-ing. The purpose of both schemes is to keep the totalallocated traffic below a maximum value in order to pro-vide some QoS protection to admitted multimedia streams.Finally, IEEE 802.11aa recommends implementing addi-tional OBSS management procedures for channel selectionand cooperatively creating HCCA schedules that do notcollide.

Interworking with IEEE 802.1AVB. Audio Video Bridg-ing (AVB) is a term commonly used to denote a set of tech-nical standards developed by IEEE to support real-timestreaming services with bounded latency through IEEE802 networks [74]. This objective is achieved by specify-ing mechanisms to allow the synchronisation of multiplestreams (IEEE 802.1AS [75]) and traffic shaping (IEEE802.1Qav [76]), and to reserve network resources for spe-cific audio/video streams traversing a bridged local areanetwork by using a signalling protocol called the StreamReservation Protocol (SRP) (IEEE 802.1Qat [77]). IEEE802.11aa integrates the SRP operations with the EDCAadmission control procedures. Specifically, the SRP Re-quest/Response messages are encapsulated in the manage-ment frames that are used to carry the traffic character-istics and the QoS requirements during admission control.This enables the end-to-end management of resource reser-vation for QoS guaranteed streams even when one or moreIEEE 802.11 links are part of a path from the stream pro-ducers (called IEEE 802.1Q talkers) and the stream con-sumers (called IEEE 802.1Q listeners).

3.3.2. Open challenges

In recent years several MAC enhancements have been in-vestigated to improve QoS guarantees for real-time mul-timedia applications in IEEE 802.11 networks [20], andthe IEEE 802.11aa standard, which was finalised in 2012,included several of these proposed improvements. Sig-nificant research efforts have focused on improving the

4TSPEC is a traffic specification sent from a QoS capable wirelessclient that requests a certain amount of network traffic from the APfor the traffic stream it represents.

11

Voice (VO)Best Effort

(BE)

Primary

VO

AC_VO

Alternate

VO

AAC_VO

802.1Q scheduler

Video (VI)

Primary

VI

AC_VI

Alternate

VI

AAC_VI

Classification to ACs and Transmission Queues

AC_BE

Background

(BK)

AC_BK

802.1Q scheduler

Intr

a-A

cce

ss C

ate

go

ry

Prio

ritiza

tio

n

Inte

r-A

cce

ss C

ate

go

ry

Prio

ritiza

tio

n

Highest

Priority

Lowest

Priority Data Packets, 802.1Q User Priorities, Traffic Classes

Frame Transmission

ED

CA

fu

nctio

ns

with

in

tern

al co

llisio

n

reso

lutio

n

80

2.1

1a

a

Tra

nsm

issio

n

Qu

eu

es

Figure 4: Stream classification and inter-AC traffic prioritisation.

transmission reliability of multicasting by integrating ARQmechanisms in IEEE 802.11-based multicast transmis-sions. Modifications to the legacy MAC protocol wereproposed in [78] to enable the RTS/CTS option in mul-ticast mode and to select one or more multicast receivers(called leaders) for acknowledging multicast data packets.However, these enhancements require changes to the stan-dard specifications. The main problems of leader-basedARQ schemes are leader election and the trade-off betweenscalability and reliability. The authors in [79] proposeselecting the multicast recipient operating in the worstchannel conditions as the unique leader but this approachmay perform poorly in lossy environments. In the Batchmode multicast MAC (BMMM) [80] all multicast recip-ients are polled by the multicast originator to send in-dividual ACKs, but this scheme is not suitable for largemulticast groups. The Enhanced Leader Based Protocol(ELBP) is proposed in [81] on the basis of multiple ACK-leaders and block acknowledgement techniques. Analyticalmodels are then developed to help select optimal ACK-leaders to meet application QoS requirements. However,the models apply only to saturated traffic while multime-dia streams are typically bursty. Another class of reliablemulticast protocols relies on busy tones to reduce packetlosses due to collisions [82], but the additional radio in-terface needed for the busy tone limits the practicality ofsuch solutions. An alternative approach to avoid collisionsof multicast packets is the multicast collision prevention(MCP) scheme [83], which is based on the use of a shorterwaiting time for transmitting multicast packets. An in-teresting approach is also proposed in [84], to retransmitlost packets using an online linear XOR coding algorithm.However, a modification to the standard MAC protocolis required to enable simultaneous ACK transmissions. Insummary, several different methods have been proposed to

improve multicast transmission reliability by integratingARQ schemes into the protocol architecture, but there arenot conclusive results on which is the best solution. Thechoice of the most efficient mechanism depends on a vari-ety of interdependent factors, such as loss ratios, channelcongestion, multicast group size, and QoS requirements ofmultimedia streams. A comprehensive analytical frame-work is needed to optimise the setting of the parametersfor each scheme and to dynamically select the best one.

As discussed above one main difference between uni-cast services and multicast services in the legacy IEEE802.11 standard was the lack of acknowledgements. An-other critical difference is that multicast frames must betransmitted using a fixed rate in the basic rate set whilethe transmission rate of unicast frames can be dynami-cally adapted to the channel and traffic conditions [85].Thus, a group of research papers has investigated the useof rate adaptation to improve the throughput of multicastservices in IEEE 802.11 networks [86, 87, 88, 71, 89]. Forinstance, the authors in [86] propose using RTS frames toallow group members to estimate channel conditions. Eachmember will then send a dummy CTS frame with a lengthinversely proportional to channel quality. In this way, themulticast transmitter can use the collision duration to pre-dict the lowest data rate that can be used for group trans-missions. The overhead introduced by this mechanism isquite high, however. The solution proposed in [88], calledARSM, also relies on feedback messages sent by the multi-cast receivers, called multicast response frames, to identifythe group member exhibiting the poorest channel condi-tions. However, in this case a different back off timer isassociated with each multicast receiver depending on theSNR of previously received feedback messages in order toprevent collision. An approach similar to the one employedin the Auto Rate Fallback (ARF) protocol, a rate adap-

12

tation scheme originally proposed in [90], is used in [87].Specifically, the number of successful consecutive trans-missions and consecutive transmission failures are used todecide when to increase or decrease the transmission datarate, respectively. A modified ARF scheme is also pro-posed in [71], which can be applied to videos that are en-coded into two layers, namely the base and enhancementlayers. However, how to integrate rate adaptation with thedifferent retransmission policies that are defined in IEEE802.11aa is still an open issue.

One research area that is expected to be crucial in thesuccessful development of IEEE 802.11aa-based productsis the design of efficient scheduling algorithms for sup-porting voice/video traffic. Almost all research work inthis field has been triggered by the IEEE 802.11e amend-ment that enhanced the original IEEE 802.11 MAC withtwo new QoS-aware access mechanisms, i.e., EDCA andHCCA [91]. In principle, with a well-designed admis-sion control and scheduling scheme, HCCA is able to pro-vide hard QoS guarantees to traffic flows [92, 93]. How-ever, HCCA is rarely implemented in IEEE 802.11e-basedWLANs owing to its higher complexity and cost concerns.Instead, EDCA is widely adopted. Most papers have thusfocused on improving EDCA performance. Many papershave proposed analytical models for various subsets ofEDCA functionalities. For instance, a saturation-basedperformance analysis is conducted in [94] by differentiatingthe minimum back-off window size, the back-off window-increasing factor, and the retransmission limit. The au-thors of [95, 96] also model AIFS differentiation, while themodel in [97] jointly captures all the four EDCA param-eters for traffic differentiation. More recent papers haveanalysed the EDCA performance for non-saturated condi-tions and for arbitrary buffer sizes [98]. The authors in [99]have developed an analytical model to predict the QoSlevels that can be achieved once a new voice/video flowis introduced in the WLAN. A Kalman filter is proposedin [100] to obtain estimates on the number of active trans-mission queues of each Access Category in EDCA. Theseanalytical models can then be exploited to derive the op-timal configuration of the EDCA parameters to achievegiven performance criteria, or to design admission cotnrolschemes that preserve QoS constraints. For instance, ascheme that assigns contention-window values to achievepre-defined weighted-fairness goals is proposed in [101].A control-theoretic scheme is designed in [102] with thegoal of minimising the video traffic delay. However, mostof these solutions rely on non-realistic assumptions aboutvideo traffic dynamics. An alternative class of solutionsdynamically updates the EDCA parameters based on theobserved network conditions. In [103], the EDCA param-eters are optimised considering a WLAN with rigid andelastic traffic simultaneously, analysing the interactionsbetween both types of traffic. The authors in [104] spec-ify several bandwidth-sharing mechanisms with guaran-teed QoS for voice and video traffic. Measurement-basedadmission control schemes are proposed in [105]. A TXOP

adaptation method is described in [106] that takes into ac-count video frame sizes and transmit queue lengths. How-ever, the main drawback of these solutions is that theyare based on heuristics and hence do not ensure optimaland guaranteed performance. Finally, a third categoryof research papers tries to improve video performance bydesigning cross-layer scheduling approaches. Specifically,these works take advantage of multi-layer video encodingto classify the frames according to their importance and as-sign them to different access categories [107]. For instance,the authors in [108] define classifiers and waiting time pri-ority schedulers that dynamically change the packet prior-ities according to end-to-end delay measurements. A dis-advantage of this approach however is that an additionaladaptation layer may be needed to implement the complexinteractions that are typically required between the videocoding applications and the MAC layer. We conclude thissection by pointing out that existing studies provide thebasic design principles and techniques for improving mul-timedia streaming performance in IEEE 802.11 networks.Still the IEEE 802.11aa standard poses new research chal-lenges that have not been sufficiently explored and thatwill require innovative solutions. For instance, schedul-ing between primary and alternate queues is still an openresearch area, as the mapping of individual frames to mul-tiple queues in order to achieve graceful degradation ofvoice/video quality [16].

4. Sensor Networks & Machine-Type Communica-tions

As discussed in Section 2, M2M communications re-fer to any communication technology that enables sen-sor/actuator devices to exchange information and performactions without the manual assistance of humans. Thissection reviews the main features currently under consider-ation in the development of the upcoming IEEE 802.11ahamendment, which targets the main challenges of thosenetworks, such as the energy consumption or the manage-ment of many devices.

4.1. The IEEE 802.11ah amendment

The IEEE 802.11ah amendment [109] aims to provideWLANs with the ability to both manage a large numberof heterogeneous STAs within a single BSS, and minimisethe energy consumption of the battery-powered STAs.

The initial design requirements of the IEEE 802.11ahamendment are detailed in [110]; these entail the sup-port of up to 8192 STAs associated with a single AP,the adoption of efficient power saving strategies, a min-imum network data rate of 100 Kbps, the operation in thelicense-exempt Sub 1 GHz band, and a coverage up to 1km in outdoor areas (see Figure 5 for an illustrative ex-ample). A preliminary assessment of performance of theIEEE 802.11ah technology, in terms of the number of STAsthat can be effectively supported in a WLAN, as well astheir energy consumption, is presented in [111].

13

1

2

AP

i−th

Server

N−1

N

Sensor/Actuator Devices

The Cloud

Figure 5: WLANs for M2M communications. STAs represent sensorand actuator devices.

IEEE 802.11ah operates over different sub-1GHz ISMbands depending on country regulations: 863-868 MHz inEurope, 902-928 MHz in the US and 916.5-927.5 MHz inJapan. China, South Korea and Singapore also have spe-cific channelisations. Channel widths of 1 MHz and 2 MHzhave been adopted, although 4, 8 and 16 MHz are alsosupported in some countries. IEEE 802.11ah furthermoreproposes new PHY and MAC layers. The IEEE 802.11ahPHY layer can be considered to some extent a sub-1GHzversion of the IEEE 802.11ac one. At the physical layerOFDM is the chosen modulation method using 32 or 64tones/sub-carriers that are spaced by 31.25 kHz. The sup-ported modulations include BPSK, QPSK and from 16 to256-QAM. A broad range of antenna technologies, rang-ing from single-user beam-forming to MIMO and DL-MU-MIMO, which was first introduced in the IEEE 802.11acamendment, are also included in the IEEE 802.11ah spec-ification. Similarly, the IEEE 802.11ah MAC protocol in-clude most of IEEE 802.11 main characteristics, furtherextending its power saving mechanisms.

4.1.1. Novel features

This section gives an overview of how IEEE 802.11ah fur-ther extends the IEEE 802.11 PS mechanisms to accountfor the specific characteristics of resource-constrained sen-sor and actuator devices, with the aim to offer to the readera concise vision of the most relevant IEEE 802.11ah fea-tures. A more detailed review can be found in [110], in-cluding a performance assessment of IEEE 802.11ah in sev-eral of the key scenarios for M2M communications, suchas agriculture and animal monitoring, smart metering, andindustrial automation plants. In addition, a detailed sur-vey of the IEEE 802.11ah is reported in [112], which ad-dresses aspects not considered in [110] such as the use ofsectorisation to avoid overlapping between multiple IEEE802.11ah WLANs, and the use of relays to directly in-terconnect IEEE 802.11ah WLANs, among other aspects.Finally, in [113], another overview of the IEEE 802.11ahnovel features is presented, introducing both PHY andMAC characteristics with emphasis on the benefits of us-

TIM

DATA

PS−POLL

ACK

STA

AP

ACK ACK

PS−POLL

DATA

t

TIM

DTIM

Page

DTIM Period

TIM period

t

TIM period

PRAWRAW

Backoff

Downlink Uplink

slots

Figure 6: IEEE 802.11ah PS mode for TIM-STAs.

ing the sub-1GHz ISM band in terms of link budget, thelocation of the OFDM pilots for outdoors operation, theuse of bidirectional transmission opportunities, and thenew packet formats to minimize overheads, among others.

Enhanced Power Saving Mechanism. Power saving (PS)mechanisms for WLANs were already considered in the de-velopment of the first IEEE 802.11 standard with the goalof improving the lifetime of battery equipped devices [114].In PS mode, STAs keep the transceiver in sleeping modeas much time as possible. They periodically wake up tolisten to the beacons transmitted by the AP. Those bea-cons indicate whether an STA has packets waiting for it atthe AP. In the positive case, the STA remains awake andrequests the delivery of those packets. Otherwise, giventhat the STA has nothing to receive, it returns to sleepmode until the next beacon is expected.

In the IEEE 802.11ah amendment, time is divided intopages, DTIM (Delivery Traffic Indication Map) periods,TIM (Traffic Indication Map) periods, and slots. DTIMand TIM periods begin with the corresponding DTIM andTIM beacons sent by the AP. The functions of DTIM andTIM beacons are described below:

1. DTIM Beacons: They inform as to which TIMGroups (i.e., the group of STAs assigned to the sameTIM period) have pending packets at the AP.

2. TIM (Traffic Indication Map) Beacons: Each TIMmessage informs a TIM Group about which specificSTA has pending data in the AP. Between two con-secutive DTIMs, there are as many TIM beacons asTIM Groups.

Using this DTIM/TIM-based approach, any STA canenter into a power saving state if it does not have pack-ets pending for transmission and one of two conditions ismet: 1) it observes in the DTIM beacon that there is nodownlink traffic addressed to its TIM Group or 2) it ob-serves in the DTIM beacon that there is some downlinktraffic addressed to its TIM Group but where this doesnot explicitly appear in the TIM beacon. Compared to

14

the preliminary IEEE 802.11 PSM this approach reducesthe size of the Traffic Indication Map in each TIM bea-con, thus reducing the overhead and the time STAs needto listen and process them. In addition, TIM periods canbe organised in pages, which further increases the numberof TIM groups between two DTIM beacons.

The temporal organisation of pages, DTIM and TIMperiods is reported in Figure 6, which shows the distribu-tion of TIM periods in RAW (Restricted Access Window)and PRAW (Periodic RAW). The RAW is a time inter-val in each TIM period where TIM stations can transmit(see below the different types of STAs defined in IEEE802.11ah). In addition, it can be divided into several down-link and uplink slots for further granularity. In the down-link, the slots are assigned to a single STA or a group ofSTAs, while the uplink slots are randomly selected by theSTAs with packets ready for transmission. The PRAW isthe period of time in each TIM where non-TIM stationscan transmit.

Types of STAs. IEEE 802.11ah supports three types ofSTAs: TIM, non-TIM and Unscheduled STAs. A TIMstation is assigned to a TIM Group (i.e., the group of STAsassigned to the same TIM beacon). Their data transmis-sions must be performed within a RAW (Restricted Ac-cess Window) period. Non-TIM stations do not have tolisten to beacons to transmit data. During the associa-tion process, non-TIM devices directly negotiate with theAP to obtain a transmission time allocated in a PRAW.The following channel access can either be renegotiated oroccurs periodically, depending on the requirements set bythe station. Unscheduled stations do not need to listen toany beacons, similar to non-TIM stations. Even inside anyrestricted access window, unscheduled station can send apoll frame to the AP asking for immediate access to thechannel. The response frame indicates an interval (outsideboth restricted access windows) during which unscheduledstations can access the channel. This procedure is meantfor STAs that transmit data very sporadically.

Hierarchical Station Organisation. To support a largenumber of STAs and their organisation in pages, DTIMand TIM periods, IEEE 802.11ah assigns to each associ-ated STA a unique identifier of 13 bits, which is calledthe Association Identifier (AID). Using this new AID, themaximum number of supported STAs is increased fromthe original 2007 in IEEE 802.11 to 8191 (= 213 − 1) inIEEE 802.11ah. However, it also allows categorising STAsaccording to the type of application they are executing,their power level or even their desired QoS by assigningthem to different TIM groups.

Long Sleeping Periods. IEEE 802.11ah offers TIM, Non-TIM and Unscheduled STAs the possibility to set verylong doze times (up to months). The corresponding clockdrift produced by such long doze times must be taken intoconsideration, however, as the higher the time an STA has

been asleep, the further in advance it should wake up toavoid possible synchronisation problems with the network.

Efficient Small Data Transmission. Three new enhance-ments have been proposed to reduce the overhead whenthe data packet size is small. First, while IEEE 802.11contains a 28-byte MAC header, IEEE 802.11ah proposesa short 18-byte version by using AIDs instead of MAC ad-dresses. Second, IEEE 802.11 has defined several null datapacket (NDP) frames, which consist only of a PHY header.These frames can be used to create short ACKs, shortBlock ACKs, short CTSs and short PS-Polls. Finally, aFast Frame Exchange mechanism has been developed, sothat, if an STA has data to transmit, it can notify a suc-cessful reception by transmitting its data frame instead ofan ACK.

Sectorisation. Since the PHY layer is based on the IEEE802.11ac amendment, single and multi-user beam-formingare also supported by IEEE 802.11ah. This allows thetransmission of data to multiple STAs simultaneously inthe downlink, increasing the system capacity. The use ofthe beam-forming capability of IEEE 802.11ah APs is alsoconsidered as a means to group STAs into different inde-pendent antenna sectors with the main goal of reducinginterference issues. This would be particularly useful inthe case of overlapping with other IEEE 802.11ah WLANsor in the presence of hidden nodes.

4.1.2. Open Challenges

A first open challenge is to understand the coverage andachievable transmission rates in IEEE 802.11ah WLANsin both indoor and outdoor scenarios. Propagation mod-els for WLANs working at frequencies lower than 1 GHzare evaluated in [115]. The authors compare two pathloss propagation models proposed by the IEEE 802.11ahTask Group (one for macro, and one for pico/hotzone de-ployments) [116] with Lee and Hata-Okumura propagationmodels. Results show that the IEEE 802.11ah channelmodels underestimate path loss with respect to Lee andHata models. Moreover, in a comparison with empiricaldata, it is observed that the IEEE 802.11ah channel mod-els also underestimate the initial loss and the slope of thepath-loss function. A new model parameterisation is thusproposed by the authors. In [117], the feasibility of anIEEE 802.11ah deployment is also evaluated in terms ofthe achievable range and bitrate, computed on the basisof the link budget using the same path loss propagationmodels as in [115]. Results show that the transmissionpower limitations in the uplink can limit the overall net-work performance. Finally, in [118] the authors evaluatethe achievable transmission range for the different trans-mission rates, and the achievable throughput for differentcombination of transmission power values and transmis-sion rates. Further studies are required due to the hetero-geneity of scenarios in which IEEE 802.11ah WLANs can

15

be deployed to characterize obstacles and effective cover-age areas, with special attention to outdoors and in pres-ence of mobile nodes.

Single-hop uplink transmissions from distant STAs re-quire high transmission power to reach the AP. Since notall STAs may be able to transmit at the required power,a solution could be the introduction of nodes able to relaytransmissions from them. However, the use of relays has tobe efficiently harmonised with the operation in power sav-ing mode, allowing temporal periods in which the relayscan gather the data from their associated STAs and thenperiods in which they forward the data to the AP, whichis still an open challenge for IEEE 802.11ah WLANs. Incase relays are used, data prediction and aggregation tech-niques could be implemented to make more efficient trans-missions.

Another challenge is the assessment of the efficiencyof IEEE 802.11ah PS mechanisms and find their optimalconfiguration. In [111], the authors evaluate the capacityof the TIM and page segmentation mechanisms in terms ofthe maximum number of nodes supported and the energyconsumed given a certain network traffic profile. Resultsconfirm that a large number of STAs can be supportedwith low energy consumption. The impact of the numberof nodes in wake mode on both the energy consumptionand the delay performance is analysed in [119]. A detailedanalysis of the TIM group-based channel access adoptedby the IEEE 802.11ah task group is made in [120], wherethe STAs assigned to each TIM are uniformly distributedbetween the slots of the RAW. Since only one group of sta-tions is allowed to transmit in each RAW slot, the channelcontention is minimised. The number of downlink and up-link slots in a RAW, as well as the duration, is optimisedon the basis of the number of active STAs and the net-work traffic profile in [121]. In [122], the authors proposean algorithm to determine the optimal size of the RAWinterval for uplink transmissions based on the estimationof the number of STAs transmitting and the duration ofthe RAW. Finally, in [123], the authors address the con-tention problems in Smart Grid communication networkswith many nodes and periodic traffic when using a IEEE802.11ah WLAN. It is still a challenge to consider smartsystems able to adapt the IEEE 802.11ah parameters tothe instantaneous system state in order to save as muchenergy as possible.

Since the PS mechanisms affect only the operation ofTIM STAs, the performance of non-TIM and UnscheduledSTAs has not yet been considered in the literature. Toidentify the scenarios in which the use of non-TIM andUnscheduled STAs are of interest is still an open challenge,as further investigating how non-TIM STAs negotiate overwhich PRAWs they can use to transmit and receive data.Co-existence issues between the three types of STAs is alsoan open challenge that requires further studies.

The design of strategies to distribute the STAs betweenall the TIM groups based on their specific traffic profile,battery level and application priority is another important

challenge that is still completely open. EDCA is the de-fault channel access scheme included in the IEEE 802.11ahamendment and provides some basic traffic differentiationcapabilities at the packet level. Besides the different accesscategories (ACs) needing to be renamed and their param-eters (Arbitration Inter-Frame Spacing, or AIFS, CWmin,TXOP duration) updated to fit the specific traffic profilesof M2M communications, other mechanisms can be im-plemented. A mechanism to assign the downlink RAWslots when packets from multiple priority levels are wait-ing for transmission is required, for example. Also, uplinkRAW slots are currently selected randomly by the STAsthat want to transmit a packet. A mechanism to reservesome slots at each uplink RAW for high priority STAs maytherefore be an option, although this may severely degradethe performance of the low priority STAs. Lastly, differ-ent repetition patterns for different priority TIM groupsmay also allow priority STAs assigned to those priorityTIM groups to access the channel more often. We expectthis challenge will receive much attention in the upcomingyears because of the heterogeneity of sensors that will beconnected though a single IEEE 802.11ah AP, specially inurban scenarios.

The design of mechanisms to avoid and resolve conges-tion situations when the same event is detected by mul-tiple sensors is another open challenge for IEEE 802.11ahWLANs with many nodes. Similar to what has been dis-cussed for LTE cellular networks [124], the development ofmechanisms aware that overload situations may happen isnecessary. For instance, a mechanism in which STAs haveto wait a random number of DTIM periods before they canstart a transmission may be implemented. This approachwould distribute the traffic over a longer period of timein overload conditions but, otherwise, would unnecessarilyincrease the access delay.

Finally, since IEEE 802.11ah will compete with 4G/5Gcellular networks and WSNs to provide M2M connectivityin many different application domains, such as smart citiesor e-health, comparative performance studies to determinethe strong and weak points of each technology are required,including also aspects such as the cost of the devices andthe system reliability.

5. Cognitive Radio Technology for TV WhiteSpaces

This section gives an overview of the IEEE 802.11afamendment, introducing its most relevant features andopen challenges. Figure 7 shows the basic components andfeatures considered for WLANs operating in the TV WhiteSpaces (TVWS), which include local spectrum sensing bythe AP and STAs, and the use of geolocation data baseswith information on channel availability.

5.1. The IEEE 802.11af amendment

Thanks to the transition from analog TV to digital TV,several VHF and UHF spectrum channels used for decades

16

f

N

N−1

1

2

InternetAP

TV White Spaces

Database

Space database (PAWS)

Protocol to Access White.

i−th

List of available TV channels

Satellite

GPS positioning

Figure 7: WLAN operating in TVWS: basic elements and function-alities.

for analog TV broadcasting are now unused or are under-utilised in many geographical areas [125][126][127]. This“digital dividend” of spectrum has suggested national reg-ulators, such as the FCC in the U.S., and Ofcom in theU.K., to discuss how to reuse these channels for unlicenseddevices’ communications [128] [129]. These potentially va-cant channels in the VHF and UHF bands are referredto as TVWS [130] and include spectrum portions like the470-790 Mhz in Europe, and non contiguous 54-72, 76-88MhZ, 174-216 MhZ, 470-698 MhZ and 698-806 MhZ inUSA. A snapshot of the TV spectrum occupancy in thecity of Barcelona in 2012 is shown in [131].

The attractive characteristics of TVWS (not only forWLANs) include the ability to penetrate through wallsand other obstacles much more effectively than otherwidely used spectrum bands, such as the 2.4 and 5.7 GHzISM bands [132] [133] [134] [135]. This fact, along with aprogressive spectrum scarcity in ISM bands, has suggestedthe birth of the IEEE 802.11af amendment, published inFebruary 2014 [136], which provides the IEEE 802.11 op-erational characteristics for TVWS access of unlicensedWhite Space Devices (WSD). A good summary of IEEE802.11af can be found in [135]. The main advantage of op-erating IEEE 802.11 WLANs in the TVWS comes from anincreased coverage range, which can reach up to one kilo-metre in rural areas and open fields [137], and less energyneeded to transmit. However, this comes at the cost of anincreased interference risk to other WSDs, which createscoexistence problems, and demands new PHY and MAClayers to efficiently support channel access and operationspreserving licensed users’ devices.

WLANs operated in the TVWS could cover a numberof interesting and emerging use cases and scenarios, e.g.,Internet access in rural or sparsely populated areas, SmartGrid, sensor aggregation, metering and control, Internet ofThings, advanced WLAN operations and TVWS traffic of-floading in indoor environments [138] [139]. Nevertheless,the concept could be realised to create or extend commer-cial or municipality WiFi services offered to citizens, withcoverage at the whole city level realised with reasonablylimited network infrastructures [140] [141].

IEEE 802.11af will use a PHY layer derived from IEEE

802.11ac. It will adopt concepts such as the OFDM, multi-user beam-forming, contiguous and non contiguous chan-nel bonding and packet aggregation. Among the manda-tory and innovative behavioural and operational parame-ters, the most notable one is the channel acquisition sup-port realised through remote geolocation-based spectrumallocation databases, which maintain the channels’ avail-ability information in any given area and time of day, pro-viding upon request the list of free channels available foruse.

In the following section, we highlight and summarisethree main novelties that IEEE 802.11af introduces. Insection 5.1.1 we explain how the access infrastructure tothe remote spectrum database is designed and what therequirements are; we discuss the coexistence issues andmethodologies adopted in IEEE 802.11af, and we also dis-cuss the novel concept of non-contiguous channel bonding.The following sections also contain an illustration of re-lated works. Open research challenges will be summarisedin Section 5.1.2.

5.1.1. Novel features

This section provides an illustration of novel features intro-duced and discussed in the path to IEEE 802.11af and thedirections provided in related works and standardisationinitiatives to resolve classical problems, properly charac-terised in the new framework of TVWS technologies.

Channel acquisition: spectrum database and channel sens-ing. Much of the attention in operating in the TVWS isgiven to the protection of the primary licensed users in theTVWS spectrum band. In general, the primary users wereconsidered as the Digital TV (DTV) broadcasters and re-ceivers. In fact receivers would suffer the wasteful effect ofcollisions during the TV broadcast reception, in the caseof secondary users’ (SU) transmission interference (Figure8).

Figure 8: Hidden terminal problem in TVWS when the secondarynetwork uses only instantaneous channel sensing to decide whetheror not a TV channel is occupied. When the STA is not able todetect the TV signal due to the presence of obstacles, it may initiatea transmission and create interference to the primary users.

It is a well-known problem that the definition of a co-herent wireless channel status (idle/busy), from the view-point of distributed primary users (e.g., DTV receivers)

17

is made difficult by the difference between transmittersand receivers interpretation of signals in a variable time-space collision domain, resulting in hidden/exposed ter-minals concepts. Solutions proposed to reduce the hid-den/exposed terminals included the RTS/CTS mecha-nisms and beaconing. However, it is difficult to implementsolutions that would effectively determine the channel sta-tus of TVWS channels based on localised distributed sens-ing activities performed by multiple secondary users in thesame collision domain. On the other hand, the use ofRTS/CTS mechanism would not be effective in this casebecause most of the DTV primary users (and specifically,the receivers) were not originally conceived to transmitsignals or implement the RTS/CTS handshake. In conclu-sion, the actions to identify free (and busy) TVWS chan-nels should not be based on primary users’ involvement,but should be almost totally implemented by specificallydesigned secondary users’ methodologies.

There has been much debate on the methodology thatIEEE 802.11af secondary devices should adopt to getknowledge of free channels at a given time on a given tar-get communication area, in an effort to guarantee verified,optimised and reliable TVWS spectrum use. The threemain methodologies discussed include: sensing solutions,geolocation databases (DB) and beaconing. Moreover, twoapproaches are possible, in general: distributed and cen-tralised, with their well known tradeoffs in terms of ef-fectiveness, required infrastructure deployment and coor-dination overhead. The distributed methodology that isreceiving more attention in the literature is based on dis-tributed channel sensing solutions [142] [143] [125]. Thesesolutions include the observation of spectrum use, and pos-sible cooperative aggregation of spectrum sensing informa-tion to increase the accuracy of detection and reduce thevulnerability of primary users’ transmissions. The spec-trum sensing capability is required in all the secondarydevices, with a -114 dBm sensitivity. Dynamic transmis-sion power control must be provided and the upper limit onemissions is 100 mW EIRP (20dBm) for portable devices(further limited in the case of adjacent channel use to re-duce out-of-band interference). However, database (DB)coordination of TVWS combined with spectrum sensingis considered the most promising and effective technique,compared with spectrum sensing alone. The DB spectruminformation has more chances to be effective (i.e., identi-fying all free channels opportunities) and reliable (i.e., notprone to attacks or erroneous interpretation of channelsstatus). In fact, regulators decided to push for the conser-vative solution of a remotely accessible, centralised spec-trum DB that maintains the information on the availabil-ity of all TVWS channels at any given point in time andlocation (with a target accuracy around 50 m), under theirresponsibility. For this reason, the spectrum databaseshave become a mandatory part of many spectrum sharingsystems, and the dominant technical solution to supportTV white spaces.

The IEEE 802.11af standard adheres to this vision and

also has the role of a common regulatory framework fordifferent spectrum DB implementations, defining a gener-alised coordination architecture, protocols and interfacesfor spectrum queries and local spectrum information man-agement. The protocol explicitly considers out of scopethe communication protocol and technology adopted in thebackground, leaving the freedom to those players respon-sible for the deployment to select any suitable Internet-based and local access network technology. Secondarydevices that need to access the TVWS spectrum shouldget information on which channels they can effectivelyuse, and which parameters to adopt, without impactingprimary users’ transmissions (and receptions), by query-ing the available spectrum DB. The query must containthe transmission characteristics and accurate geographi-cal position of the secondary device, which can be deter-mined by means of a geo-positioning system (GPS), reg-ularly updated in the case of movement (or manually setfor static devices). The IEEE 802.11af reference systemarchitecture for DB spectrum access is composed of mul-tiple Geolocation-based Spectrum information Databases(GDBs) entities connected via Internet to Registered Lo-cation Secure Servers (RLSS). RLSSs work as local prox-ies of the GDB for a localised group of Basic Service Sets(BSS). RLSS are connected via a secure protocol architec-ture with the equivalent of multiple Access Points (AP) indifferent BSS, realising a trusted DB infrastructure. APslocally coordinate the exchange of information and chan-nel access management between the GDB (via RLSS) andthe secondary users’ end stations (STA).

Mode I devices are those under the control of a devicethat employs geo-location database access, while Mode IIdevices are those employing geo-location database accessby themselves. The information exchange provided be-tween the GDB and secondary user STA can be providedin both an open-loop (e.g., adopted by FCC) and a closed-loop (e.g., adopted by ETSI) implementation. In an openloop implementation daily spectrum availability informa-tion is provided by the GDB and no feedback on the spec-trum information received is provided by STAs; thus thesystem approach for spectrum access is much more conser-vative, leading to low channels’ utilisation potential. In aclosed loop implementation the STA can provide feedbackto the GDB, and there are more communication overheadsdue to the high granularity of updates, however, the sys-tem is more effective and reliable in the exploitation ofTVWS on behalf of STAs. The typical information pro-vided by the GDB includes 1) the updated White SpaceMaps (WSM) of frequencies allowed for secondary use atthe time/space of the querying STA, and 2) the device-dependent power limitations for transmission (in general,conservative and accurate enough to avoid relevant inter-ference effects on primary users identified in the area).

The basic IEEE 802.11af mechanisms regulating thecommunication between STA and GDB can be found in[135] and [136].

In parallel with IEEE 802.11af, standards such as the

18

IEEE 1900.6 have been created that consider interfacesand data structures supporting spectrum sensing informa-tion exchange, applicable to spectrum sensing (and par-ticularly distributed spectrum sensing) scenarios. In par-ticular, in October 2014 the IEEE 1900.6 WG initiated aproject for a new standard, called IEEE 1900.6b [144], con-cerning the use of spectrum sensing information to supportand optimise the effectiveness, reliability and robustnessof spectrum database solutions. The aim is to enhancethe performance and capabilities of spectrum databasesthrough the use of spectrum sensing information.

Co-existence. Support for co-existence mechanisms sothat multiple technologies can effectively utilise the TVWSspectrum is important. Self co-existence between networkdevices of a common technology (e.g. deployed by differ-ent operators in the same area), and co-existence amongdifferent technologies, are relevant topic of research forCognitive Radio (CR) systems, and specifically for IEEE802.11af. Many solutions appeared on the research scene,but no one was so far finalized as the target solution forIEEE 802.11af. Specific standardisation has been startedto regulate coexistence between wireless standards of un-licensed devices, including the IEEE 802.19.1 [145]. Thepurpose of the IEEE 802.19.1 standard is to enable thefamily of IEEE 802 Wireless Standards to most effec-tively use TV White Space by providing standard coex-istence methods among dissimilar or independently oper-ated TVWS devices. Early examples of generalized coexis-tence mechanisms included Dynamic Frequency Selection(DFS), Transmission Power Control (TPC), listen beforetalk (e.g., for contention based IEEE 802.11, 802.15), timedivision multiplexing (also among different techniques suchas the IEEE 802.16, 802.20, 802.22), and Message-basedSpectrum Contention (that is, beaconing messages thatcarry coexistence information). Opportune metrics mustbe defined to assess the measurable coexistence achievedamong different technologies: as an example, the hiddennode probability for a target scenario, or the estimateof percentage variation in normalised network throughputand latency (before and during the SU transmissions). Onthe other hand, a centralised coexistence control mecha-nism could be effectively realised by a central manager (orcoexistence-DB, like the GDB) in critical scenarios. Tothis end, IEEE 1900.4 [146] (a standard for heterogeneousnetworks in dynamic spectrum context, part of IEEE Stan-dards Coordinating Committee 41) aims to standardise theoverall system architecture and information exchange be-tween the network and mobile devices, which will allowthese elements to optimally choose from available radioresources.

There are three possible classification of co-existencearchitectures and inter-network coordination channels forCR systems: centralized, coordinated and autonomous co-existence mechanisms [147]. In centralized co-existenceschemes, the co-existence is administered by a central en-tity (e.g like in IEEE 802.19.1). This solution could be

applied in both homogeneous IEEE 802.11af systems (e.g.incarnated by the centralized spectrum DB entity) andin heterogeneous inter-networks, without requiring modi-fications to existing standards, under the assumption thatall the involved co-existing network entities would adhereto the same centralized coordination scheme and inter-network coordination protocol. On the other hand, thisassumption could be hard to satisfy in some practical sce-narios. The centralized schemes are effective in providingminimization of inter-network interference, based on theavailability and quality of centralized co-existence informa-tion. In coordinated co-existence schemes, the centralizedentity could be present, but not taking decisions, in gen-eral. The central entity could provide a co-existence DBwith required information which can be used via in-bandand out-band signalling for co-existence decisions takenby the cluster heads of many co-existing inter-networks,for implementing a proper common coordination proto-col (e.g. out-band busy tone signalling). These are oftenhybrid solutions realizing a compromise between effective-ness and suitability. In autonomous co-existence schemes,all the decisions and coordination are implemented in adistributed way by the involved network entities. In theseschemes, many policies can be adopted to realize a suffi-cient level of co-existence, under a best effort approach.Solutions that could be implemented include the use ofbusy tones, beaconing and other signalling protocols indedicated control channels, dynamic distributed frequencyselection schemes, listen before transmit, token-based anddynamic reservation schemes. All the above mentioned so-lutions have goods and bad, and a general illustration canbe found in [147]. Other proactive co-existence techniquestry to early detect and recover/mitigate co-existence issueswith relaxed inter-network coordination, e.g. realized viaspectrum sensing and interference avoidance/suppressiontechniques.

It must be clear that different aims exist, regarding theco-existence, for primary TVWS users protection againstsecondary users, and for secondary devices mutual inter-ference avoidance. Many co-existence problems have beenanalysed in the literature for TVWS technologies, in par-ticular between the primary (licensed) and secondary (un-licensed) devices. The coexistence management in TVWSis complex because primary users of DVB transmissionsare intended to be the pure receivers, rather than thebroadcasters. In other words, potential protected userscould be everywhere, and the hidden terminal problemcould arise for these systems. The co-existence problemis also complicated by many factors that cause asym-metry and dynamicity in the TVWS, such as the mo-bility, variable density, power asymmetry, and heteroge-neous MAC/PHY layers. TVWS co-existence problemsoriginate between the IEEE 802.11af technologies and theIEEE 802.22 technologies for wireless regional area net-works (WRAN). Results of analysis of TVWS usage inEurope show that white spaces are typically present andfragmented. They are typically more abundant in rural ar-

19

eas, where larger contiguous blocks of unused channels areavailable, due to broadcast network planning giving prior-ities linked to population density. However, the exploita-tion of TVWS in urban areas is possible, e.g., some recentresearch was realized under the assumption of exploitingshadowing effects created in the communication environ-ment (e.g., by buildings, obstacles) in favour of frequencyreuse [148].

Since different standards for opportunistic communi-cation in the TV White Spaces have now been published,such as IEEE 802.22 [149], IEEE 802.15.4m [150], Weight-less [151], and of course IEEE 802.11af, improved methodsto guarantee the coexistence of different devices, operatingon several protocols in the same bands, must be deployed.We highlight here some work, e.g., [152] [153] [154] [155],[148], that focuses on the coexistence between differenttechnologies operating in the same bands. In particular,[153] studies the performance degradation of IEEE 802.22when an IEEE 802.11af network is operated in the samearea. The problem becomes even worse when the IEEE802.11af network is located near 802.22 user equipment,causing both networks to perceive a strong interferencefrom each other. One proposed solution is the CoexistenceBeacon Protocol [156], studied for 802.22 networks, whichforesees the exchange of a periodic beacon to identify theneighbouring, and possibly interfering, networks.

For what concerns the co-existence problems, researchhas addressed other important related aspects of newWLANs on TV White Spaces. For instance, the highercoverage range makes them attractive for applications inthe smart grid, and also for M2M on TVWS [132] [148][157]. Here, the increased range compared to standardtechnologies in the ISM bands can have beneficial effectsfor indoor mobile devices [148], and M2M [157]. Specifi-cally, indoor devices communicating on TV White Spacesoffer better propagation characteristics and penetrationthrough obstacles, making it easier to realise scalable homeconnectivity [133], although interference and co-existenceproblems could be exacerbated and must be resolved, e.g.see [137]. However, studies in the literature show how thesignal coming from an indoor transmitter on TV WhiteSpaces remains constrained inside the house, making it dif-ficult to deploy indoor-to-outdoor networks. On the otherhand this shadowing limit can be seen as an opportunity,lowering the interference outside the building in which theTVWS network is locally operated [132] [148]. Under thisapproach, HDTV streaming has recently been consideredas a possible use case of WLAN networks operating in theTV White Space [158].

Non-contiguous channel bonding. A relevant novel featureof IEEE 802.11af is the potential for contiguous and noncontiguous channel bonding, which permits aggregatingbasic channels (also non adjacent ones), and leveragingthe possible large frequency spread between multiple avail-able channels. Due to the rather static nature of primaryDTV transmissions, where a busy channel is unlikely to be-

come free in the near future, and state changes are coarse-grained in general, it is crucial to exploit the time-localityeffect and exploit the maximum physical channel availabil-ity that could be aggregated at any given location. Withthe methodology inherited from IEEE 802.11ac, IEEE802.11af is capable of bonding together two up to four ba-sic channels grouped in up to two different non-contiguouschunks [135]. The spectrum bandwidth of a DVB-T basicchannel can be either 6, 7, or 8 MHz, depending on thecountry in which the service is operated. As an example,this creates a 144 to 168 OFDM channels’ bandwidth po-tential when up to four 6-7-8 MhZ channels are bonded[135].

5.1.2. Open Challenges

The deployment of WLANs in the TVWS at their maxi-mum potential still requires the resolution of open prob-lems. Generally speaking, as mentioned in previous sec-tions, the challenges for such networks can be divided intothree main categories: 1) spectrum sharing between oppor-tunistic (secondary) and heterogeneous devices, 2) spec-trum sensing/management and maximum exploitation ofavailable spectrum (potentially enabling the spectrum-on-demand concept), and 3) co-existence and interferencemitigation to the primary network (primary user protec-tion) and secondary networks [152].

The overall aim of spectrum sharing techniques is tomaximise separation to avoid overlapping or contiguouschannels operations, provided that sufficient channels areavailable in the TVWS spectrum in a given space/timescenario [152]. In general, important research contribu-tions have to be realised for the design of proper spectrumallocation methodologies and techniques satisfying multi-factorial QoS requirements at the system-level and theuser-level. Novel ideas include the spectrum sharing andspectrum sensing techniques being further divided into co-operative and non-cooperative.

Non-cooperative solutions attempt to realise spectrumsharing and sensing on a local basis, without direct co-operation between devices. Examples include the ”listen-before-transmit” approach, and transmission power con-trol to limit the interference, and spectrum allocation poli-cies that are based on local node feedback only. Coopera-tive solutions rely on the existence of a common commu-nication channel, through which devices can tentativelyagree on the spectrum allocation that provides the desiredspectrum separation and QoS. Major issues include thedefinition of a methodology to identify a common channel(or multiple channels mutually shared in space by pairs ofheterogeneous and distributed devices) and to efficientlyshare common information. For the access to the spectrumDBs, the identification of proper common access channelsis still under discussion: out-of-TVWS-band cellular com-munications can be widely used, given their high coverage,also indoor.

Trying to avoid the use of complex and resource-hungry coordination schemes, and dedicated control chan-

20

nels, rendezvous protocols play a significant role, bothfor co-existence and for establishing dynamic communi-cation channels, based on spectrum sharing. Rendezvousis considered a fundamental problem in generalised CRnetworks, at the basis of many communication processes,e.g. including neighbour discovery, routing and broad-cast. Rendezvous protocols attempt to establish a newlink for communication on a selected and agreed frequencyband (channel) identified within a set of available resources(channels), thus creating the basis for communication be-tween two or more SUs. This problem is often exacer-bated by the high (and unknown) numbers of contendingSUs, high number of available channels and their ambigu-ous boundaries and identifiers, and tough co-existence re-quirements satisfaction in cognitive radio systems [159].Preliminary approaches for rendezvous were based on theexistence of a dedicated Common Control Channel (CCC)agreed among all the SUs, or the existence of a centralisedagent (decision maker) assigning channels to any pairs ofrequesting users. This approach could be considered in thecase of IEEE 802.11af solutions based on the spectrum DBimplementation, however, the problem is complicated bythe fact that the common channel must be identified andmade available for all the requesting SUs at the same time,even when these are spread over multiple collision domainsin space. On the other hand, the scalability and reliabilityissues which are caused by the congestion of CCCs, andthe vulnerability risks of centralised DBs, motivate theresearch of alternative distributed solutions, called blindrendezvous protocols [160]. Distributed solutions havebeen mainly based on beaconing and/or slot-based channelhopping algorithms exploiting randomisation, sensing anddiscovery/synchronisation messages spread in slotted timeover a set of candidate available channels [159]. These pro-tocols aim to minimise the convergence time, to maximisethe use of resources, and to avoid collisions among SUs,by identifying and agreeing on the use of a given commonchannel among N non-overlapping and unambiguously la-beled channels. The main complication in IEEE 802.11afsystems is given by the wide availability of many differ-ent non-contiguous spectrum bands, which cannot be uni-formly quantised in single channels and unambiguously la-beled in space, in order to realise a common reference do-main for all the IEEE 802.11af devices. In fact, the knowl-edge of the number of channels, their common labels, theunique IDs of SUs and the number of SUs contending forthe channel assignment are the key factors determining theeffectiveness and good properties of blind rendezvous pro-tocols. Recently, new classes of rendezvous protocols havebeen proposed which are distributed, blind and oblivious.Oblivious means that available channels may be identifiedwith different labels on behalf of the different involvedSUs. A number of these distributed blind and obliviousrendezvous protocols have been proposed and have beenreferenced and analysed in [160]. To conclude, solutionsfor dynamic spectrum handoff issues (e.g. in case PUs sud-denly appear in a region) must be resolved in advance to

minimise the latency of re-establishing a functional com-munication channel, e.g. by implementing a proactive al-ternative rendezvous channel selection in background toongoing communication processes [161].

Another challenge is to provide cooperative tech-niques in a distributed vs. centralised implementation,by analysing their convergence under variable conditions,overheads and tradeoffs. In some cases, as in vehicularnetworks, the exploitation of scenario characteristics andfactors such as the constrained mobility could help to re-alise effective dissemination of sensing information overa extended and predictable horizon under a cooperativeapproach [162], [163], [164]. Cooperation could be help-ful in the realisation of the cooperative spectrum sensingin combination with the Geolocation DB approach. An-other interesting direction of research is the investigationof mutual effects of coexisting cooperation-based vs. non-cooperative techniques. In cooperation-based devices theexchange of information could allow a quick convergence toa solution, and avoidance of further interference [143]. Cer-tainly, cooperative techniques have desirable advantagesover non-cooperative ones. However, tight time synchroni-sation and high coordination overheads are required. Thecommon channel for cooperation purposes could becomea bottleneck and reduce the potential advantages of dy-namic spectrum allocation. The IEEE 802.19 foresees thepresence of a shared common control channel (CCC), towhich devices need to tune in order to gather the state ofthe network. Solutions such as [139] propose the use ofTVWS for the CCC.

A promising direction for research is the adoptionof clustering schemes for secondary devices, associatedwith a hybrid distributed/cooperative vs. centralised/non-cooperative approach, in an effort to reduce overheadswhile maximising advantages. In this way, secondarynodes could implement cooperation for sensing functionsusing properly identified common channels, and delegatethe centralised decisions and spectrum-DB management towell instrumented cluster-leader nodes. This concept is in-carnated to some extent by RLSS in IEEE 802.11af. Morerecently there have been proposals to build distributeddatabases that refresh their contents periodically by query-ing the remote spectrum database [165]. Here, Master de-vices (according to Ofcom terminology) periodically cachethe query replies from the spectrum database in order toreply to and broadcast the spectrum availability to neigh-bour Slave devices.

Another challenging direction for research is the defi-nition of accurate models and efficient simulation tools en-abling dynamic spectrum analysis for both rural and urbanareas. Recent developments of digital maps, simulationtools and propagation models theoretically allow improv-ing the capability of predicting radio propagation effects incomplex scenarios with an acceptable computation time.This enabling technology could be used in parallel to spec-trum DBs to provide more accurate forecasting of dynamicfrequency allocation in time/space scenarios. Another is-

21

sue to be taken into account is the accuracy of the remotespectrum database regarding the channel availability esti-mation. Several works have already shown the inaccuracyof propagation models, due to the complexity of param-eters to be considered and differences between modelledand real scenarios [166, 167, 132, 130]. Better propagationmodels will make it possible to more efficiently estimatethe interference between devices, and build Radio Envi-ronment Maps (REM) or White Space Maps (WSM) thatwill possibly lead to a more accurate sharing of the radiospectrum [142]. To this end, current research could focuson how to leverage simulation models in order to buildmore accurate frequency DBs.

In general, the adoption of remote spectrum DBsmethodology pushed by national regulators has made thespectrum sensing process optional and conservatively re-alised (e.g., power control for sensing-only devices is lim-ited to below 17 dBm EIRP). Thus, much of the researchhas focused on techniques and solutions to successfullyquery the remote database, also in challenging environ-ments, such as rural areas and indoor. However, sincethe queries must provide the position of the mobile de-vice, the accuracy of positioning for mobile and handhelddevices still needs more appropriate solutions. Triangula-tion can be used to estimate the position of mobile deviceswith the accuracy needed by the current regulations out-door (50m). However, a low GPS accuracy is possible in-doors [168]. Certified manual positioning can be providedby technicians for static and indoor devices, although lim-iting mobility. Alternatives to costly infrastructures forpositioning (e.g., based on short range beaconing devices)must be deployed in indoor scenarios.

Finally, regarding spectrum sensing and interferencemitigation, new challenges come from the heterogeneityand different characteristics that networks operating inthe TV White Spaces can have, such as different trans-mitting power (4 W for static devices, 100 mW for mobileand portable devices, and 40 mW for communication inchannels adjacent to occupied ones), different bandwidths(5, 10, 15, and 20 MHz are currently foreseen for IEEE802.11af; 802.15.4m can have narrow bandwidths or wideones), and different medium access schemes (protocols thatuse either CSMA or TDMA are required to co-exist in theTV White Spaces). With all this in place, the problem ofdetermining whether another secondary device is currentlytransmitting in the same channel is still an open issue.

Spectrum sensing is a very hot topic in the literature,with many proposals that leverage the cooperation be-tween devices to raise the accuracy of the detection pro-cess, e.g., [142, 169]. The current literature also offersroom for proposals that suggest exploiting sensing to buildmore accurate spectrum DBs [162] [170] [171]. Lastly, re-garding the interference mitigation to the primary net-work, regulators provide a conservative approach basedon a remote spectrum database to avoid high risk for theprimary users. However, in particular for the indoor sce-nario, the concept of secondary users’ transmission in TV

Grey Spaces has been proposed. TV Grey Space identifiesbusy channels that are formally used by primary users inthe area of interest (e.g., at the rooftop level of a build-ing for DVB-T), but which could be considered usable in-doors (e.g., at the basement level), without causing ef-fective interference to the primary users [132, 148, 172].This concept is based on the generalised spectrum under-lay paradigm [173]. The remaining research challenge isto provide TV Grey space access guaranteeing a sufficientamount of protection to the primary user, while providinga satisfactory QoS to opportunistic devices, e.g., throughthe use of customised and validated propagation modelsfor indoor TVWS networks, and feedback mechanisms toconstantly monitor the interference generated to the pri-mary network.

6. Emerging new trends and technologies

In this last section we review three emerging trends that,in our opinion, will have a large influence in the conceptionof future WLANs as they change the way WLAN protocolsand functionalities are developed, implemented, tested andintegrated with other wireless networks.

6.1. Programmable Wireless LANs

Especially in the enterprise environment, WLAN deploy-ments need to support a wide range of functionalities andservices. This is intrinsically difficult because of the largenumber of APs that must be managed, which calls forscalable solutions. Typical services include channel assign-ment, load balancing among APs, authentication, autho-risation and accounting (AAA), policy management, sup-port for client mobility and interference coordination. An-other problem is the fact that WLAN clients autonomouslytake several decisions such as which APs to associate with,when to hand-over, etc. Therefore, supporting roamingclients requires the management of a large number of as-sociation states across several APs, which is a challenge ifsupport for real-time hand-over is desired. Typically, suchmanagement schemes are centralised and most of themare proprietary, such as WLAN controller solutions fromAruba [174] or Cisco [175], although the 802.11u amend-ment has been released to allow mobile users to seamlessroam between WiFi networks with automatic authentica-tion and handoff [176]. For example, Dyson [177] enablesSTAs to send information such as radio channel condi-tions to a centralised controller based on a custom API(e.g., based on Python). As the controller has a cen-tralised view of the network, it can enforce a rich set ofpolicies to control the network also using historical in-formation. A demo system has been implemented alongwith applications such as airtime reservations for specificclients or optimised handoffs. However, Dyson requiresSTAs to be modified in order to use those new servicesoffered by the centralised controller. TRANTOR [178] isanother example of a centralised management system that

22

requires changes of clients in order for the infrastructure togather information from them in terms of e.g., interferencemeasurements. In addition, control commands enable theinfrastructure to exercise control such as modifying thetransmit power or influencing the association procedure.Clients still use standard CSMA MAC layer for data trans-mission. In contrast to DenseAP [179] or DIRAC [180],which are also based on a centralised controller but do notmodify clients, TRANTOR exercises much more controlon clients by the use of a dedicated API, which enables asignificantly higher gain in coordination and thus capac-ity. CENTAUR [181] is another example of a centralisedcontroller that centrally schedules hidden and exposed ter-minals in the downlink while using standard CSMA MACfor uplink traffic and legacy downlink traffic. To achievegood performance, it uses a fixed back-off, packet stagger-ing techniques and a hybrid data path that only schedulesdownlink transmissions towards hidden and exposed ter-minals centrally. All other traffic is sent using standardDCF with a standard back-off procedure. In contrast toprevious approaches, it does not require any modificationsto the clients but does require data plane centralisationand it remains questionable how scalable such a solutionis for large WLAN deployments with high PHY layer datarates beyond 100 Mbps. However, new approaches for fastdata plane processing that are currently explored in thevirtualisation community which move all the packet pro-cessing into user space may be an option to significantly in-crease speed. Designing such management systems, raisesseveral interesting research questions: (a) Centralized in-frastructures may incur high latency but are more simpleto program and maintain while designing a completely dis-tributed approach that operates close to WLAN devicesmay result in significant distributed coordination and con-sistency problems; (b) what level of distributed control isrequired in order to support the flexibility needs of futureWiFi based networks supporting a high level of mobility.

Traditionally, WLAN APs are built on proprietary op-erating systems that are tightly coupled with the hard-ware. This design makes it hard to create new applicationson top of such networking devices. Despite the fact thatprotocols and mechanisms are available that would greatlyimprove the utility of existing networks, those new proto-cols are not deployed, because the closed system designmakes it very difficult to extend their functionality. Animportant aspect for all such new approaches is thereforehow to provide open interfaces and open source to speedup innovation. As an example, Linux-based devices arecompletely open source but in order to increase flexibility,much more work is needed in the area of open drivers andfirmware. In fact the Atheros based ath9k driver has beenthe main driver of innovation in the open source commu-nity because it’s the only driver that interworks with openfirmware. In general solutions that modify the OS driverfocus mainly on programmable network level solutions forWLANs, such as channel switching or handovers betweenAPs. While the ath9k driver is an excellent example in

how to speed up innovation, it is still very cumbersometo support flexible MAC engine reprogrammability withath9k. In contrast, MAClets and Wireless MAC proces-sors (WMP) [182] allow a much more flexible reprogram-ming of MAC functionalities. A MAC processor is an en-tity able to execute general MAC commands that specifythe MAC operations through a software-defined state ma-chine. The behaviour of the MAC protocol can thereforebe updated at run-time by simply changing the sequencein which those commands are executed (i.e., the MAClets).As a proof-of-concept, the authors implement the proposedMAC processor solution in a commodity WLAN hardwarecard, extending the basic DCF in three directions: piggy-backing ACKs, a pseudo TDMA, and the use of multi-ple channels. In [183] a control framework for this WMPsystem is also proposed to support MAClet code mobility,i.e., for moving, loading and activating MAC software pro-grams embedded into ordinary data packets (akin to thecapsule model of traditional active networks).

The difficulty in re-programming networking hardwarehas also led to the concept of Software Defined Networks(SDNs) based on the OpenFlow protocol [184]. The mainidea of SDNs is to extend networking devices with stan-dardised APIs that allow third-party programmers to flex-ibly control the data path. In addition, SDNs providehigher level abstractions to network designers and pro-grammers through the use of a centralised control planeoffered by a network controller such as NOX [185], whichallows reuse of components such as topology discovery ornetwork access control for different applications. The SDNconcept has recently been applied to WLAN architecturein order to enable fine grained control over mobility man-agement and data forwarding focusing on programmableenterprise WLAN architecture. An important part of theSDN architecture is the network controller, which providesa centralised view of (parts of) the SDN enabled networkand uses the OpenFlow protocol to install the forwardingrules on SDN-enabled devices (routers, switches, accesspoints, cellular base stations, etc.).

ODIN [186] is designed to support programmabilityin enterprise WLAN architecture by separating the as-sociation state from the physical AP. They implementLight Virtual Access Points (LVAP) using a Split-MACapproach, where the infrastructure controls the handoverprocedure among different WLAN APs. By managing as-sociations through SDN controllers, ODIN enables proac-tive mobility management and load balancing within theSDN enabled WLAN enterprise network without the needfor changes in the client WLAN stack or IEEE 802.11 MAClayer. While ODIN requires agents to reside on the APsthat communicate with the Odin Master within the SDNcontroller, CLOUDMAC [187] is based purely on the con-cept of SDN and virtual APs. Similar to ODIN, CLOUD-MAC implements a Split-MAC approach but in additionenables the processing of WLAN MAC layer frames withina co-located Cloud using so-called Virtual APs (VAPs).The physical APs in CLOUDMAC are lightweight WLAN

23

VAP1 VAP2 VAPN

Internet

SDN

(Open Flow)

WT1 WT2 WTN

STA

Dummy AP

Host

Figure 9: The CLOUDMAC architecture. VAP stands for VirtualAccess Point, and WT stands for Wireless Termination.

APs that are responsible only for sending out IEEE 802.11based MAC layer ACKs to standard WLAN clients andtunnel WLAN MAC layer frames through an SDN-enabledenterprise WLAN towards the VAPs. As association statesare kept in the VAPs, fast mobility is supported using sim-ple OpenFlow forwarding rules. Because of the additionalprocessing of IEEE 802.11 WLAN MAC frames in the co-located Cloud, CLOUDMAC has higher latency than astandard WLAN deployment. However, CLOUDMAC of-fers a Webservice based API to third party applicationsin order to program the enterprise WLAN and enable newservices. The architecture of CLOUDMAC is depicted inFigure 9 and has been extended in [188] to support QoSmanagement and in [189] to support flexible MAC man-agement frame prioritisation based on 802.11. A systembased on OpenFlow has been recently proposed in [190] toallow the station to be associated with multiple APs simul-taneously and to switch between APs with low overhead.

An important aspect to consider in order to enableprogrammability is backwards compatibility. Several ap-proaches (e.g., [178]) require clients to be modified forthem to utilise the features provided. This is difficult todo in practice because it requires changes in the operatingsystem software of all clients. If not all clients can utilisethose APIs, it is questionable how usable the new archi-tecture will be and how much benefit in terms of aggregateperformance such architecture will allow. In contrast, theSDN based approaches are interesting in the sense thatthey do not require changes in the client WLAN stack andwork with the standard IEEE 802.11 MAC layer deployedwithin the clients. However, it remains questionable howscalable such solutions are. For example, an interestingopen research topic is to evaluate the scalability of ap-proaches such as CLOUDMAC [187]. In addition, pro-cessing IEEE 802.11 MAC frames within co-located pri-vate Cloud requires low latency support from local Cloudsolutions (such as OpenStack) in order to reduce the IEEE

802.11 MAC processing time, which is an area of active re-search.

6.2. Prototyping and testing IEEE 802.11 enhancements

Most of the new proposals for next-generation WLANsare currently only evaluated using mathematical analysisand simulation. While both analysis and simulation arenecessary to characterise and study those enhancementsin the initial design phase or to consider large-scale sce-narios, it is difficult to consider all practical aspects of areal-world scenario. This can sometimes cause significantdifferences between what simulations and real experimentsshow. However, real experiments are challenging becauseof the high complexity and costs of building the new hard-ware and software for each specific solution to test.

To mitigate implementation complexity, there are sev-eral flexible hardware platforms where low-level MACmechanisms can be completely implemented in software.Examples based on FPGAs are the USRP (Universal Soft-ware Radio Peripheral) [191] and the WARP (WirelessOpen-Access Research Platform) [192]. As and alterna-tive, SORA (Microsoft Research Software Radio) [193]works on general purpose computers by taking some ad-vantage from modern multiple core systems.

A different approach is provided by OpenFWWF [194].OpenFWWF provides an open CSMA/CA firmware forspecific models of Broadcom chipsets, so the resultingfirmware can be uploaded and tested in real commercialhardware. OpenFWWF implements a simple State Ma-chine (SM) for controlling the hardware in real time. TheSM evolution is driven by a main loop that reacts to eventsby executing specific handlers. When a packet, originallyprepared by the Linux kernel, is ready in the NIC mem-ory, handler Packet Ready sets up the radio hardwareaccording to the packet metadata (e.g., it fixes rate, modu-lation format, and power level), schedules the transmissionand jumps back to the main loop. The Transmission En-gine (TXE) then takes care of accessing the channel, i.e.,it decrements the back-off counter according to the DCFrules and eventually starts the actual transmission. Thistriggers the execution of the TX frame now event thatprepares the ACK time-out clock and finalises the MACheader (as the transmission has already started, these ac-tions must be completed before the end of the physicalpreamble). If the ACK-frame is received or if the ACKtime-out expires and the maximum number of attemptsfor this packet is reached, handler Update Params re-sets the contention window to its minimum value, or oth-erwise doubles it. Finally, it loads the back-off counterwith a fresh value. In the following, we will review a listof selected papers that use the OpenFWWF firmware andthe WARP or USRP platforms to test new proposals forWLANs in real scenarios.

A first implementation of the groupcast mechanismsdefined in IEEE 802.11aa is presented in [195]. The au-thors modify the OpenFWWF firmware to include thosefunctionalities and evaluate them using a 30 STAs testbed.

24

A second example of the use of OpenFWWF is thecollision-free MAC protocol presented in [196]. In [196],the authors design and implement in OpenFWWF a MACprotocol that is able to achieve a collision-free operationby waiting for a deterministic timer after successful trans-missions. This work shows the experimental results of acollision-free MAC (CF-MAC) protocol for WLANs usingcommercial hardware. Testbed results show that the pro-posed CF-MAC protocol leads to a better distribution ofthe available bandwidth among users, a higher throughputand lower losses than the unmodified WLANs clients usinga legacy firmware.

In [197], the authors use the WARP platform to testa variant of the DCF – called IEEE 802.11ec – that sub-stitutes control packets such as the RTS, CTS and ACKwith short detectable sequences. Since those sequences areshorter than the control packets, and can be detected cor-rectly even at lower SNRs values, a significant gain in per-formance is achieved. MIDAS (Multiple-Input DistributedAntenna Systems) [198] implementation using the WARPplatform shows the benefits of distributing the antennasover the area to cover instead of co-locating them at theAP in terms of capacity when MU-MIMO is employed.The authors also propose a new MAC protocol to benefitfrom the spatial reuse that the DAS allows, taking as abasis the IEEE 802.11ac. Results in a testbed show thattheir proposal is able to achieve up to 200 % gains versusthe traditional approach where all antennas are co-locatedat the AP.

The USRP platform has recently been proposed to de-velop a first implementation of IEEE 802.11ah in orderto obtain a preliminary performance assessment of sucha technology, because no commercial off-the-shelf hard-ware is yet available as the amendment is still in progress[199]. The USRP platform is also used in [200] to evaluateTIMO (Technology Independent Multi-Output) a solutionto deal with high-power non-IEEE 802.11 interferers inISM bands.

To conclude this section, it is worth mentioning herealso to MAClets and WMP, as given such flexibility toimplement and distribute new MAC protocols and otherIEEE 802.11 functionalities, we believe that if a use-friendly implementation of such a MAC processors frame-work is provided, it would significantly contribute to thedevelopment and testing of new MAC enhancements bythe research community.

6.3. Cellular/WLAN interworking

Public hotspots that offer Internet access over a WLANusing IEEE 802.11 technology are now nearly ubiquitous.It is forecasted that the cumulative installed base of WiFihotspots worldwide will amount to 55.1 millions by 2018,excluding private hotspots (e.g., WiFi access points de-ployed at home) [201]. The sharp increase in the avail-ability of public WiFi was initially perceived by mobilecellular operators as a threat due to the additional compe-tition from wireline Internet service providers or emerging

WLAN

LTE

Internet

LTE cell

WLAN

cell

offloading path

Contents

Mobile

Figure 10: LTE to WIFI offloading.

crowdsourced WiFi networks, such as FON5. However, ascellular operators are fighting to cope with the explosionof mobile data traffic created by the rising use of multime-dia content traffic over mobile devices [15], they are alsostarting to use WLANs based on the IEEE 802.11 technol-ogy to offload data from their core and access networks.In general, mobile data offloading refers to the use of com-plementary network technologies (in licensed or unlicensedspectrum) for delivering data originally targeted to cellularnetworks. Intuitively, the simplest type of offloading con-sists of exploiting connectivity to existing co-located WiFinetworks and transferring data without any delay (Figure10). Thus, this offloading technique is know as on-the-spotoffloading. As a consequence of this new trend, the seam-less integration of cellular (e.g., 3G/LTE) and WiFi tech-nologies has attracted significant research interest in recentyears (see [202] for a survey), a few solutions have alreadybeen standardised [203, 204], and roaming between cellu-lar and WiFi is becoming increasingly transparent to endusers. Cellular/WLAN interworking is also fostered by thesupport in the evolving 4G standards of heterogeneous net-work deployments (HetNets), in which the existing macrocells are complemented with a number of small, low-powerbase stations with the goal of increasing capacity in highlycongested areas [205, 206]. It is envisaged that small cellswill be based on 4G standards (e.g., pico and femto cells)as well as IEEE 802.11 technologies, and multimode basestations that work simultaneously with LTE and WiFi arealready entering the market.

It is important to point out that LTE standards alreadysupport a variety of mechanisms that enable data offload-ing. However, most of the existing solutions, such as LocalIP Access (LIPA) and Selected Internet IP Traffic Offload(SIPTO), focus on data offloading at the core of the net-work [203]. For instance, LIPA allows an IP-enabled mo-bile device to transfer data to another device in the samepico or femto cell without passing through the cellular ac-cess network, while SIPTO enables the routing of selected

5https://corp.fon.com/en.

25

IP data flows through different gateways. The only LTEoffloading mechanism that supports seamless interwork-ing with IEEE 802.11-based WLANs is IP Flow Mobilityand Seamless Offload (IFOM). Specifically, IFOM relies onMobile IPv6 technologies to allow a user terminal to simul-taneously route selected IP flows over different radio accesstechnologies [207, 208]. In this way, a user terminal canoffload selected flows to a WLAN based on some operator-defined policy while keeping the LTE connection running.However, to enable such an approach it is necessary tosupport an entity in the cellular core network that cancommunicate with the user terminal to exchange informa-tion about the availability and quality of neighbouring ac-cess networks, as well as to provide the user terminal withpredefined rules to manage the handover process. This en-tity in current LTE standards is called the Access NetworkDiscovery and Selection Function (ANDSF) server [209].Note that the problem of selecting the best communicationtechnology in an heterogeneous wireless network has beenextensively investigated in the literature both with cen-tralised as well as decentralised approaches (see [210] for asurvey). However, the use of multiple interfaces in parallel,as well as per-flow offloading are relatively new concepts.Thus, the design of scalable and efficient network selec-tion strategies for the ANDSF framework is still an openissue. For instance, three offloading methods suitable forthe ANDSF framework are proposed in [211], which arebased on coverage, SNR, and system load. A reinforce-ment learning approach is designed in [212] that allowsmultimode base stations to autonomously steer their traf-fic flows across different access technologies depending onthe traffic type, the users’ QoS requirements, the networkload, and the interference levels. A number of studies havealso tried to quantify the potential capacity gain of WiFioffloading in real-world WiFi deployments. For instance,the authors in [213] evaluate offloading efficiency usingtrace-based urban mobility patterns and WiFi connectiv-ity distributions. The authors in [214] instead analysethe offloading performance in an indoor scenario in whichfemto cells and WiFi access points coexist. On the otherhand, mathematical models are needed to derive perfor-mance bounds and guide the design of optimal offloadingstrategies. A simple queuing model for the analysis of theoffloading efficiency is developed in [215] by assuming thatthe WiFi network availability is exponentially distributed.A more general scenario is considered in [216] by assum-ing multiple classes of access points that differ in transmitpower, deployment density and bandwidth. The optimalassociation strategy is then derived to maximise the frac-tion of time that a typical user in the network is servedwith a rate greater than its minimum rate requirement.A somehow related problem consists of deciding how tooptimally deploy WiFi hotspots to maximise offloading ef-ficiency. Traditionally, this problem has been consideredfrom the point of view of coverage maximisation, e.g., toensure continuous WiFi connectivity by taking into con-sideration user mobility characteristics [217]. On the con-

trary, in the context of mobile data offloading, AP de-ployment is tackled to maximise the throughput perfor-mance in an heterogeneous wireless network. For instance,a heuristic algorithm is proposed in [218] that selects as APlocations the cells with the higher frequency of downloadrequests. A graph-theoretical solution for AP deploymentis developed in [219] by consorting a time-dependent graphthat describes the interdependencies between users’ mobil-ity trajectories, points of interest and traffic demands. Anopen issue in this field of research concerns the design ofmore adaptive traffic steering mechanisms between cellu-lar and WiFi. Furthermore, the increase in the number ofwireless infrastructure nodes with the dense deployment ofsmall cells will make the future network deployments quasistochastic. Thus, new methodologies, such as stochasticgeometry, have to be explored to model the performancebounds of heterogenous wireless networks that allow theinter-working between cellular systems and WLANs [220].Finally, the use of historical data to predict network con-ditions and user locations may also become infeasible dueto the scale of the network in terms of infrastructure nodesand users. To deal with this issue limited measurementscould be coupled with statistical inference methods.

On-the-spot offloading is the dominant but not theonly form of interworking between cellular networks andWLANs. More recently, delayed offloading has also beenproposed for delay-tolerant traffic. Basically, if a user iswilling to accept a delayed content reception (e.g., thedownload of a YouTube video), the cellular operator mayintentionally postpone the content transfer in order to waitfor WLAN availability or better transmission conditions.The cellular network is then used to complete the datatransfer only if the content reception can not be guaran-teed within a user-specified deadline. A number of stud-ies have explored the feasibility of delayed offloading fordifferent delay deadlines using trace-based WLAN usagepatterns. In particular, results in [213] confirm that in-creasing the delay-tolerance of content significantly im-proves the fraction of traffic that can be offloaded. Theoffloading performance clearly depends on several factors,including the location of 802.11 hotspots and the ability toaccurately predict the future availability of WLAN cover-age. Thus, several studies have addressed the problem offorecasting mobile connectivity. For instance, the solutionproposed in [221], called BreadCrumbs, tracks the move-ments of the mobile device’s owner and maintains a his-tory of observed networking conditions to train a forecastmodel of near-term connectivity. More recently, a time-based prediction model of visited locations derived frommovement traces of mobile users is given in [222]. Then,the authors in [223] propose a system, called Wiffler, thatallows mobile users to decide whether or not to wait for afuture WiFi offloaded opportunity based on the predictedWLAN capacity and the total data that needs to be trans-ferred. However, the design of location prediction modelsthat have low computation complexity and are suitablefor short-term mobility is an open issue. Also related to

26

the problem of delayed offloading feasibility is the optimalplacement of APs. The authors of [224] develop the Hot-Zones algorithm, which selects the cells with the highestnumber of daily visits as the location of additional WiFiaccess points. A similar solution, called Drop Zones, isproposed in [225]. In the context of vehicular networks,the optimal deployment of RoadSide Units (RSU) over agiven road layout to maximise the overall system through-put is analysed in [226]. In [227] it is analysed the dataoffloading gain in a vehicular sensor network as a functionof the percentage of equipped vehicles, of the number ofdeployed road side units, and of the adopted routing pro-tocol. Although existing work established the potential ofdelayed offloading, there is still a need for considerable re-search to design incentive mechanisms for motivating usersto leverage their delay tolerance for cellular traffic offload-ing. For instance, an auction-based pricing framework isproposed in [228] to give priority to users with high delaytolerance and a large offloading potential.

We conclude this section by discussing a third type ofmobile data offloading, known as opportunistic offloading,which does not rely on a WLAN infrastructure but ex-ploits direct communications between mobile devices, e.g.,through the emerging WiFi Direct standard [229]. Op-portunistic offloading schemes allow saving significant cel-lular bandwidth because the content spreads through theopportunistic network formed by the users, while the cel-lular network is mainly used for signalling and triggeringthe content dissemination. Clearly, the performance of op-portunistic offloading solutions depends on several factors,including the user mobility patterns, the user density, thedelay tolerance for content reception, and the popularity ofthe content that needs to be transferred. Note that oppor-tunistic offloading requires a less controlled type of inter-working between LTE and WLANs because user devicescan setup direct connections and start ad hoc communi-cation autonomously with little or no intervention fromthe operator. For instance, the authors of [224] propose asimple algorithm, called MixZones, to allow the cellularoperator to decide when the mobile users should be noti-fied to switch their wireless interface for data transfer withpotential other users that they are predicted to encounter.In addition, most of the papers consider that content mustbe delivered to users within a given deadline. Most of theresearch in this context addresses the problem of select-ing the best (e.g., the smallest) set of users, called seeds,that should receive the content from the cellular networkand help to disseminate it over the opportunistic network.Three simple algorithms for initial seed selections are pro-posed in [230]. The authors of [231] propose using socialnetwork properties, e.g., betweenness or degree central-ity, to select the most useful seeds for offloading the cellu-lar network. A similar social-aware approach is also usedin [232, 233]. It is assumed in [234] that a centralised en-tity keeps tracks of the speed of the content disseminationprocess to decide when the cellular network should directlytransmit the content to the interested users to guarantee

that the delivery deadline is met with high probability. Anactor-critic learning framework is designed in [235] to un-derstand when and to how many users the cellular networkshould directly transmit the content. In this context, animportant area of research is the design of scalable andefficient network-aided offloading schemes, where the cel-lular network guides the mobile users in the connectivitymanagement and dissemination phase. Finally, offloadingdata traffic when the requests for popular content are notsynchronised is still an open issue [236].

7. Summary

In this paper we have described the main scenarios,novel functionalities and mechanisms that will characterisethe use, operation and performance of next-generationWLANs and provided an extensive and thorough reviewof the IEEE 802.11ac, IEEE 802.11aa, IEEE 802.11ah,IEEE 802.11p and IEEE 802.11af amendments. The pa-per also provides an up-to-date survey of the most repre-sentative work in this research area, summarising the keycontributions to the current status and future evolutionof WLANs. Differently from other 802.11-related surveys,this overview is structured with regards to the emergingWLAN application scenarios, such as M2M, cognitive ra-dios and high-definition multimedia delivery. We also de-scribe some open challenges that require further researchin coming years [237], with special focus on software-defined MACs and the internet-working with cellular sys-tems.

Acknowledgements

We would like to thank the anonymous reviewers for theirinsightful comments on the paper, as these comments ledus to an improvement of the work. A special thank alsoto Toke Høiland-Jørgensen and Ognjen Dobrijevic for thetime they spent reviewing the manuscript.

References

References

[1] G. Hiertz, D. Denteneer, L. Stibor, Y. Zang, X.P. Costa, andB. Walke. The IEEE 802.11 Universe. IEEE CommunicationsMagazine, 48(1):62–70, 2010.

[2] N. Bhushan, Junyi Li, D. Malladi, R. Gilmore, D. Brenner,A. Damnjanovic, R. Sukhavasi, C. Patel, and S. Geirhofer.Network densification: the dominant theme for wireless evolu-tion into 5G. IEEE Communications Magazine, 52(2):82–89,February 2014.

[3] M. Conti and S. Giordano. Mobile ad hoc networking: mile-stones, challenges, and new research directions. IEEE Com-munications Magazine, 52(1):85–96, 2014.

[4] M. Conti, C. Boldrini, S. Kanhere, E. Mingozzi, E. Pagani,P.M. Ruiz, and M. Younis. From MANET to people-centricnetworking: Milestones and open research challenges. Com-puter Communications, pages –, 2015.

27

[5] H. Zhu, M. Li, I. Chlamtac, and B. Prabhakaran. A surveyof quality of service in IEEE 802.11 networks. IEEE WirelessCommunications, 11(4):6–14, August 2004.

[6] B. Bellalta, A. Vinel, P. Chatzimisios, R. Bruno, and C. Wang.Research advances and standardization activities in WLANs.Computer Communications, 39:1–2, 2014.

[7] IEEE Std 802.11n-2009. Part 11: Wireless LAN Medium Ac-cess Control (MAC) and Physical Layer (PHY) Specifications– Amendement 5: Enhancements for Higher Throughput, Oc-tober 2009.

[8] IEEE Std 802.11p-2010. Part 11: Wireless LAN Medium Ac-cess Control (MAC) and Physical Layer (PHY) Specifications– Amendment 6: Wireless Access in Vehicular Environments,July 2010.

[9] IEEE Std 802.11s-2011. Part 11: Wireless LAN Medium Ac-cess Control (MAC) and Physical Layer (PHY) Specifications,September 2011.

[10] IEEE 802.11-2012. Part 11: Wireless LAN Medium Ac-cess Control (MAC) and Physical Layer (PHY) Specifications,March 2012.

[11] W. Sun, O. Lee, Y. Shin, S. Kim, G. Yang, H. Kim, andS. Choi. Wi-Fi could be much more. IEEE CommunicationsMagazine, 52(11):22–29, November 2014.

[12] E. Borgia. The Internet of Things vision: Key features, appli-cations and open issues. Computer Communications, 54:1–31,2014.

[13] S. Tozlu, M. Senel, Wei Mao, and A. Keshavarzian. Wi-Fienabled sensors for internet of things: A practical approach.IEEE Communications Magazine, 50(6):134–143, June 2012.

[14] C.W. Park, D. Hwang, and T.-J. Lee. Enhancement of IEEE802.11ah MAC for M2M Communications. IEEE Communi-cations Letters, 18(7):1151–1154, July 2014.

[15] Cisco. Cisco Visual Networking Index: Global Mobile DataTraffic Forecast Update, 2013-2018. Technical report, Cisco,February 2014.

[16] K. Kosek-Szott, M. Natkaniec, S. Szott, A. Krasilov,A. Lyakhov, A. Safonov, and I. Tinnirello. What’s new forQoS in IEEE 802.11? IEEE Network, 27(6):95–104, Novem-ber 2013.

[17] C.-S. Sum, G.P. Villardi, M.A. Rahman, T. Baykas, H.N. Tran,Z. Lan, C. Sun, Y. Alemseged, J. Wang, C. Song, C.-W. Pyo,S. Filin, and H. Harada. Cognitive communication in TV whitespaces: An overview of regulations, standards, and technology.IEEE Communications Magazine, 51(7):138–145, July 2013.

[18] A.C.V. Gummalla and J.O. Limb. Wireless medium accesscontrol protocols. IEEE Communications Surveys & Tutori-als, 3(2):2–15, Second Quarter 2000.

[19] R.C. Carrano, L.C.S. Magalhaes, D.C.M. Saade, and C.V.N.Albuquerque. IEEE 802.11s Multihop MAC: A Tutorial. IEEECommunications Surveys & Tutorials, 13(1):52–67, First 2011.

[20] E. Charfi, L. Chaari, and L. Kamoun. PHY/MAC En-hancements and QoS Mechanisms for Very High ThroughputWLANs: A Survey. IEEE Communications Surveys & Tuto-rials, 15(4):1714–1735, Fourth 2013.

[21] ITU-T, International Telecommunication Union. Recommen-dation G.114: One-way transmission time, ITU-T StudyGroup 12, May 2003.

[22] H. Schwarz, D. Marpe, and T. Wiegand. Overview of theScalable Video Coding Extension of the H.264/AVC Standard.IEEE Transactions on Circuits and Systems for Video Tech-nology, 17(9):1103–1120, September 2007.

[23] E. Ancillotti, R. Bruno, and M. Conti. The role of communica-tion systems in smart grids: Architectures, technical solutionsand research challenges. Computer Communications, 36(17–18):1665–1697, 2013.

[24] ETSI. Applicability of M2M architecture to Smart Grid Net-work. Technical Report 102 935 V2.1.1, September 2009.

[25] Ian F Akyildiz, Weilian Su, Yogesh Sankarasubramaniam, andErdal Cayirci. Wireless Sensor Networks: A Survey. Computernetworks, 38(4):393–422, 2002.

[26] Shao-Yu Lien, Kwang-Cheng Chen, and Yonghua Lin. Toward

ubiquitous massive accesses in 3GPP machine-to-machinecommunications. Communications Magazine, IEEE, 49(4):66–74, 2011.

[27] B. Bellalta, A. Checco, A. Zocca, and J. Barcelo. On the Inter-actions between Multiple Overlapping WLANs using ChannelBonding. IEEE Transactions on Vehicular Technology, Febru-ary 2015.

[28] M. Nekovee. A Survey of Cognitive Radio Access to TV WhiteSpaces. International Journal of Digital Multimedia Broad-casting, pages 1–8, 2010.

[29] IEEE Std P802.11ac. Part 11: Wireless LAN Medium AccessControl (MAC) and Physical Layer (PHY) specifications: En-hancements for Very High Throughput for Operation in Bandsbelow 6 GHz., 2013.

[30] M. Gong, B. Hart, L. Xia, and R. Want. Channel Boundingand MAC Protection Mechanisms for 802.11ac. In Proc. ofIEEE GLOBECOM’11, 2011.

[31] M. Park. IEEE 802.11 ac: Dynamic Bandwidth Channel Ac-cess. In Proc. of IEEE ICC’11. IEEE, 2011.

[32] A. Faridi, B. Bellalta, and A. Checco. Analysis of Dy-namic Channel Bonding in Dense Networks of WLANs.arXiv:1509.00290, 2015.

[33] J. Ong, E.H.and Kneckt, O. Alanen, Z. Chang, T. Huovinen,and T. Nihtila. IEEE 802.11ac: Enhancements for very highthroughput WLANs. In Proc. of IEEE PIMRC’11, pages 849–853, 2011.

[34] Y. Zeng, P.H. Pathak, and P. Mohapatra. A First Look at802.11 ac in Action: Energy Efficiency and Interference Char-acterization. In Proc. of IFIP Networking’14, 2014.

[35] E. Aryafar, N. Anand, T. Salonidis, and E.W. Knightly. De-sign and experimental evaluation of multi-user beamforming inwireless LANs. In Proc. of ACM MOBICOM’10, pages 197–208, 2010.

[36] B. Bellalta, J. Barcelo, D. Staehle, A. Vinel, and M. Oliver. Onthe performance of packet aggregation in IEEE 802.11 ac MU-MIMO WLANs. IEEE Communications Letters, 16(10):1588–1591, 2012.

[37] O. Sharon and Y. Alpert. MAC level Throughput compari-son: 802.11ac vs. 802.11n. Physical Communication, 12:33–49,2014.

[38] M. Yazid, A. Ksentini, L. Bouallouche-Medjkoune, and D. Ais-sani. Performance Analysis of the TXOP Sharing Mechanismin the VHT IEEE 802.11ac WLANs. IEEE CommunicationsLetters, 18(9):1599–1602, September 2014.

[39] G. Redieteab, L. Cariou, P. Christin, and J.-F. Helard.PHY+MAC channel sounding interval analysis for IEEE802.11 ac MU-MIMO. In Proc. of IEEE ISWCS’12, pages1054–1058, 2012.

[40] O. Bejarano, E. Magistretti, O. Gurewitz, and E.W Knightly.MUTE: Sounding Inhibition for MU-MIMO WLANs. In Proc.of IEEE SECON’14, 2014.

[41] Q. Wang, L. Greenstein, L. Cimini, D. Chan, and A. Hedayat.Multi-User and Single-User Throughputs for Downlink MIMOChannels with Outdated Channel State Information. IEEEWireless Communications Letters, 3(3):321–324, 2014.

[42] Takefumi Hiraguri and Kentaro Nishimori. Survey of Trans-mission Methods and Efficiency Using MIMO Technologies forWireless LAN Systems. IEICE Transactions on Communica-tions, 98(7):1250–1267, 2015.

[43] J. Cha, H. Jin, B.C. Jung, and D.K. Sung. Performancecomparison of downlink user multiplexing schemes in IEEE802.11ac: Multi-user MIMO vs. frame aggregation. In Proc.of IEEE WCNC’12, pages 1514–1519, 2012.

[44] R. Liao, B. Bellalta, J. Barcelo, V. Valls, and M. Oliver. Perfor-mance analysis of IEEE 802.11ac wireless backhaul networksin saturated conditions. EURASIP Journal on Wireless Com-munications and Networking, 2013(1):1–14, 2013.

[45] O. Aboul-Magd, U. Kwon, Y. Kim, and C. Zhu. Managingdownlink multi-user MIMO transmission using group mem-bership. In Proc. of IEEE CCNC’13, pages 370–375. IEEE,2013.

28

[46] C. Zhu, C. Ngo, A. Bhatt, and Y. Kim. Enhancing WLANbackoff procedures for downlink MU-MIMO support. In Proc.of IEEE WCNC’13, pages 368–373. IEEE, 2013.

[47] K. Hanada, K. Yamamoto, M. Morikura, K. Ishihara, andK. Riichi. Game-Theoretic Analysis of Multibandwidth Chan-nel Selection by Coordinated APs in WLANs. IEICE Trans-actions on Communications, 96(6):1277–1287, 2013.

[48] W.-S. Jung, K.-W. Lim, and Y.-B. Ko. Utilising partially over-lapped channels for ofdm-based 802.11 {WLANs}. ComputerCommunications, 63:77–86, 2015.

[49] IEEE 802.11 TGax. Status of IEEE 802.11 HEW Task Group.http://www.ieee802.org/11/Reports/tgax update.htm, 2014.

[50] Michelle X Gong, Brian Hart, and Shiwen Mao. Advancedwireless lan technologies: Ieee 802.11 ac and beyond. ACMSIGMOBILE Mobile Computing and Communications Re-view, 18(4):48–52, 2015.

[51] Boris Bellalta. IEEE 820.11 ax: High-EfficiencyWLANs. IEEE Wireless Communications (arXiv preprintarXiv:1501.01496), Accepted July 2015.

[52] E.H. Ong. Performance analysis of fast initial link setup forIEEE 802.11ai WLANs. In Proc. of IEEE PIMRC’12, pages1279–1284. IEEE, 2012.

[53] D. Camps-Mur, A. Garcia-Saavedra, and P. Serrano. Device-to-device communications with Wi-Fi Direct: overview and ex-perimentation. IEEE Wireless Communications, 20(3), 2013.

[54] L. Suresh, J. Schulz-Zander, R. Merz, A. Feldmann, andT. Vazao. Towards programmable enterprise WLANS withOdin. In Proc. of ACM HotSDN’12, pages 115–120. ACM,2012.

[55] M.S. Gast. 802.11ac: A survival guide. O’Reilly Media, Inc.,2013.

[56] M. Fang, D. Malone, K.R. Duffy, and D.J. Leith. Decentralisedlearning MACs for collision-free access in WLANs. WirelessNetworks, 19(1):83–98, 2013.

[57] L. Sanabria-Russo, A. Faridi, B. Bellalta, J. Barcelo, andM. Oliver. Future evolution of CSMA protocols for the IEEE802.11 standard. In Proc. of IEEE ICC’13, pages 1274–1279,2013.

[58] Yang Xiao. IEEE 802.11 performance enhancement via con-catenation and piggyback mechanisms. IEEE Transactions onWireless Communications, 4(5):2182–2192, 2005.

[59] R. Liao, B. Bellalta, M. Oliver, and Z. Niu. MU-MIMO MACProtocols for Wireless Local Area Networks: A Survey. IEEECommunications Surveys & Tutorials, 2015.

[60] B. Li, Q. Qu, Z. Yan, and M. Yang. Survey on OFDMA basedMAC protocols for the next generation WLAN. In Proc. ofIEEE WCNCW’15, pages 131–135, 2015.

[61] M.S. Afaqui, E. Garcia-Villegas, E. Lopez-Aguilera, G. Smith,and D. Camps. Evaluation of dynamic sensitivity control algo-rithm for IEEE 802.11 ax. In Proc. of IEEE WCNC’15, pages1060–1065, 2015.

[62] I. Jamil, L. Cariou, and J.-F. Helard. Improving the capacityof future IEEE 802.11 high efficiency WLANs. In Proc. IEEEICT’14, pages 303–307, 2014.

[63] V.P. Mhatre, K. Papagiannaki, and F. Baccelli. Interfer-ence mitigation through power control in high density 802.11WLANs. In Proc. of IEEE INFOCOM’07, pages 535–543,2007.

[64] X. Liu, A. Sheth, M. Kaminsky, K. Papagiannaki, S. Seshan,and P. Steenkiste. Pushing the envelope of indoor wirelessspatial reuse using directional access points and clients. InProc. of ACM MOBICOM’10, pages 209–220, 2010.

[65] IEEE. Specific requirements Part11: Wireless LAN MediumAccess Control (MAC) and Physical Layer (PHY) Specifica-tions Amendment 2: MAC Enhancements for Robust AudioVideo Streaming. IEEE Std 802.11aa-2012, May 2012.

[66] K. Maraslis, P. Chatzimisios, and A.C. Boucouvalas. IEEE802.11aa: Improvements on video transmission over wirelessLANs. In Proc. of IEEE ICC’12, pages 115–119, 2012.

[67] IEEE. Part 11: Wireless LAN Medium Access Control (MAC)and Physical Layer (PHY) specifications Amendment 8: IEEE

802.11 Wireless Network Management. IEEE Std 802.11v-2011, February 2011.

[68] A. Banchs, A. De La Oliva, L. Eznarriaga, D.R. Kowalski, andP. Serrano. Performance Analysis and Algorithm Selection forReliable Multicast in IEEE 802.11aa Wireless LAN. IEEETransactions on Vehicular Technology, 63(8):3875–3891, Oc-tober 2014.

[69] IEEE. IEEE Standard for Local and metropolitan area net-works: Media Access Control (MAC) Bridges. IEEE Std802.1D-2004, June 2004.

[70] P. Pancha and M. El Zarki. MPEG Coding for Variable BitRate Video Transmission. IEEE Communications Magazine,32(5):54–66, May 1994.

[71] M.A. Santos, J. Villalon, and L. Orozco-Barbosa. A NovelQoE-Aware Multicast Mechanism for Video Communicationsover IEEE 802.11 WLANs. IEEE Journal on Selected Areasin Communicatons, 30(7):1205–1214, August 2012.

[72] N. Bhushan, J. Li, D. Malladi, R. Gilmore, D. Brenner,A. Damnjanovic, R.T. Sukhavasi, C. Patel, and S. Geirhofer.Network densification: the dominant theme for wireless evolu-tion into 5G. IEEE Communications Magazine, 52(2):82–89,February 2014.

[73] M. Benveniste. Wireless LANs and ’neighborhood capture’. InProc. of IEEE PIMRC’02, volume 5, pages 2148–2154, 2002.

[74] IEEE. IEEE Standard for Local and metropolitan areanetworks–Audio Video Bridging (AVB) Systems. IEEE Std802.1BA-2011, September 2011.

[75] G.M. Garner and Hyunsurk Ryu. Synchronization of au-dio/video bridging networks using IEEE 802.1AS. IEEE Com-munications Magazine, 49(2):140–147, February 2011.

[76] IEEE Standard for Local and Metropolitan Area Networks- Virtual Bridged Local Area Networks Amendment 12:Forwarding and Queuing Enhancements for Time-SensitiveStreams. IEEE Std 802.1Qav-2009, January 2009.

[77] IEEE. IEEE Standard for Local and Metropolitan AreaNetworks—Virtual Bridged Local Area Networks Amendment14: Stream Reservation Protocol (SRP). IEEE Std 802.1Qat-2010, September 2010.

[78] J. Kuri and S.K. Kasera. Reliable Multicast in Multi-AccessWireless LANs. Wireless Networks, 7(4):359–369, September2001.

[79] D. Dujovne and T. Turletti. Multicast in 802.11 WLANs: AnExperimental Study. In Proc. of ACM MSWiM’06, pages 130–138, 2006.

[80] M.-T. Sun, L. Huang, A. Arora, and T.-H. Lai. Reliable MAClayer multicast in IEEE 802.11 wireless networks. In Proc. ofIEEE ICCP’02, pages 527–536, 2002.

[81] A. Lyakhov and M. Yakimov. Analytical Study of QoS-Oriented Multicast in Wireless Networks. EURASIP Journalon Wireless Communications and Networking, 11:1–13, 2011.

[82] S. K S Gupta, V. Shankar, and S. Lalwani. Reliable multicastMAC protocol for wireless LANs. In Proc. of IEEE ICC’03,volume 1, pages 93–97, 2003.

[83] M.A. Santos, J. Villalon, L.O. Barbosa, and F. Ramirez-Mireles. A new ARQ mechanism for multicast traffic overIEEE 802.11 WLANs. In Porc. of IEEE WMNC’11, pages1–8, 2011.

[84] Jalaluddin Qureshi, Chuan Heng Foh, and Jianfei Cai. On-line {XOR} packet coding: Efficient single-hop wireless multi-casting with low decoding delay. Computer Communications,39:65–77, 2014.

[85] E. Ancillotti, R. Bruno, and M. Conti. Design and PerformanceEvaluation of Throughput-aware Rate Adaptation Protocolsfor IEEE 802.11 Wireless Networks. Performance Evaluation,66(12):811–825, December 2009.

[86] A. Basalamah, H. Sugimoto, and T. Sato. Rate Adaptive Re-liable Multicast MAC Protocol for WLANs. In Proc. of IEEEVTC-Spring 2006, volume 3, pages 1216–1220, 2006.

[87] S. Choi, N. Choi, Y. Seok, T. Kwon, and Y. Choi. Leader-Based Rate Adaptive Multicasting for Wireless LANs. In Proc.of IEEE GLOBECOM ’07, pages 3656–3660, 2007.

29

[88] J. Villalon, P. Cuenca, L. Orozco-Barbosa, Y. Seok, andT. Turletti. ARSM: a cross-layer auto rate selection multicastmechanism for multi-rate wireless LANs. IET Communica-tions, 1(5):893–902, October 2007.

[89] G.-H. Liaw, K.-H. Tsai, S.-Y. Wang, T.-L. Kao, L.-C. Hwang,and Y.-C. Lin. Adaptive rate control for broadcasting mul-timedia streams in IEEE 802.11 networks. In Proc. of IEEEISNE’13, pages 296–300, 2013.

[90] M. Lacage, M.H. Manshaei, and T. Turletti. IEEE 802.11 RateAdaptation: A Practical Approach. In Proc. of ACM MSWiM’04, pages 126–134, 2004.

[91] IEEE. Specific requirements Part 11: Wireless LAN MediumAccess Control (MAC) and Physical Layer (PHY) Specifica-tions. IEEE Std 802.11-2012, March 2012.

[92] G. Boggia, P. Camarda, L.A. Grieco, and S. Mascolo.Feedback-based control for providing real-time services withthe 802.11e mac. IEEE/ACM Transactions on Networking,15(2):323–333, April 2007.

[93] K.-Y. Lee, K.-S. Cho, and W. Ryu. Efficient QoS SchedulingAlgorithm for Multimedia Services in IEEE 802.11e WLAN.In Proc. of IEEE VTC-Fall’11, pages 1–6, 2011.

[94] Y. Xiao. Performance analysis of priority schemes for IEEE802.11 and IEEE 802.11e wireless LANs. IEEE Transactionson Wireless Communications, 4(4):1506–1515, July 2005.

[95] H. Zhu and I. Chlamtac. Performance analysis for ieee 802.11eedcf service differentiation. IEEE Transactions on WirelessCommunications, 4(4):1779–1788, July 2005.

[96] Z. Tao and S. Panwar. Throughput and delay analysis forthe IEEE 802.11e enhanced distributed channel access. IEEETransactions on Communications, 54(4):596–603, April 2006.

[97] D. Xu, T. Sakurai, H.L. Vu, and T. Sakurai. An Access DelayModel for IEEE 802.11e EDCA. IEEE Transactions on MobileComputing, 8(2):261–275, February 2009.

[98] Q. Zhao, D.H.K. Tsang, and T. Sakurai. A Scalable and Ac-curate Nonsaturated IEEE 802.11e EDCA Model for an Arbi-trary Buffer Size. IEEE Transactions on Mobile Computing,12(12):2455–2469, December 2013.

[99] N.D. Taher, Y.G. Doudane, B. El Hassan, and N. Agoulmine.Towards voice/video application support in 802.11e WLANs:A model-based admission control algorithm. Computer Com-munications, 39:41–53, 2014.

[100] I. Kadota, A. Baiocchi, and A. Anzaloni. Kalman Filter-ing: Estimate of the numbers of active queues in an 802.11e{EDCA} {WLAN}. Computer Communications, 39:54–64,2014.

[101] R.-G. Cheng, C.-J. Chang, C.-Y. Shih, and Y.-S. Chen. Anew scheme to achieve weighted fairness for WLAN supportingmultimedia services. IEEE Transactions on Wireless Commu-nications, 5(5):1095–1102, May 2006.

[102] P. Patras, A. Banchs, and P. Serrano. A Control TheoreticScheme for Efficient Video Transmission over IEEE 802.11eEDCA WLANs. ACM Transactions on Multimedia Comput-ing, Communications and Applications, 8(3):29:1–29:23, July2012.

[103] C. Cano, B. Bellalta, A. Sfairopoulou, and J. Barcelo. Tuningthe EDCA parameters in WLANs with heterogeneous traffic:A flow-level analysis. Computer Networks, 54(13):2199–2214,2010.

[104] Y. Xiao, F. Haizhon Li, and B. Li. Bandwidth Sharing Schemesfor Multimedia Traffic in the IEEE 802.11e Contention-BasedWLANs. IEEE Transactions on Mobile Computing, 6(7):815–831, July 2007.

[105] Y. Xiao and L. Li. Voice and video transmissions with globaldata parameter control for the IEEE 802.11e enhance dis-tributed channel access. IEEE Transactions on Parallel andDistributed Systems, 15(11):1041–1053, November 2004.

[106] H. Liu and Y. Zhao. Adaptive EDCA Algorithm Using VideoPrediction for Multimedia IEEE 802.11e WLAN. In Proc. ofICWMC ’06, pages 1–10, July 2006.

[107] A. Ksentini, M. Naimi, and A. Gueroui. Toward an improve-ment of H.264 video transmission over IEEE 802.11e through

a cross-layer architecture. IEEE Communications Magazine,44(1):107–114, January 2006.

[108] W. He, K. Nahrstedt, and X. Liu. End-to-end Delay Con-trol of Multimedia Applications over Multihop Wireless Links.ACM Transanctions on Multimedia Computing, Communica-tions and Applications, 5(2):16:1–16:20, November 2008.

[109] IEEE 802.11 TGah. Status of Project IEEE 802.11ah.http://www.ieee802.org/11/Reports/tgah update.htm, 2015.

[110] T. Adame, A. Bel, B. Bellalta, J. Barcelo, and M. Oliver.IEEE 802.11ah: The WiFi Approach for M2M Communica-tions. IEEE Wireless Communications Magazine, 2014.

[111] T. Adame, A. Bel, B. Bellalta, J. Barcelo, J. Gonzalez, andM. Oliver. Capacity Analysis of IEEE 802.11 ah WLANs forM2M Communications. In Proc. of MACOM’13, pages 139–155. Springer, 2013.

[112] E. Khorov, A. Lyakhov, A. Krotov, and A. Guschin. A sur-vey on IEEE 802.11ah: An enabling networking technology forsmart cities. Computer Communications, 58(53–69), March2015.

[113] Minyoung Park. IEEE 802.11 ah: sub-1-GHz license-exemptoperation for the internet of things. Communications Maga-zine, IEEE, 53(9):145–151, 2015.

[114] M. Meo, E. Le Rouzic, R. Cuevas, and C. Guerrero. Researchchallenges on energy-efficient networking design. ComputerCommunications, 50:187–195, 2014.

[115] S. Aust and T. Ito. Sub 1GHz Wireless LAN propagationpath loss models for urban smart grid applications. In Proc.of IEEE ICNC’12, pages 116–120. IEEE, 2012.

[116] J. Lansford. TGah Channel Models. IEEE 802.11-11/498r0,April 2011.

[117] A. Hazmi, J. Rinne, and M. Valkama. Feasibility study of IEEE802.11 ah radio technology for IoT and M2M use cases. In Proc.of IEEE GC Wkshps’12, pages 1687–1692. IEEE, 2012.

[118] W. Sun, M. Choi, and S. Choi. IEEE 802.11 ah: A Long Range802.11 WLAN at Sub 1 GHz. Journal of ICT Standardization,1(1):83–108, 2013.

[119] D. Jung, R. Kim, and H. Lim. Power-saving strategy for bal-ancing energy and delay performance in WLANs. ComputerCommunications, 50:3–9, 2014.

[120] L. Zheng, L. Cai, J. Pan, and M. Ni. Performance Analysis ofGrouping Strategy for Dense IEEE 802.11 Networks. In Proc.of IEEE GLOBECOM’13, pages 219–224, 2013.

[121] A. Bel, T. Adame, B. Bellalta, J. Barcelo, J. Gonzalez, andM. Oliver. CAS-based channel access protocol for IEEE 802.11ah WLANs. In Pros. of European Wireless’14, pages 1–6.VDE, 2014.

[122] C.W. Park, D. Hwang, and T.-J. Lee. Enhancement of IEEE802.11ah MAC for M2M Communications. IEEE Communi-cations Letters, 18(7):1151–1154, 2014.

[123] R.P. Liu, G.J. Sutton, and I.B. Collings. Power save withOffset Listen Interval for IEEE 802.11 ah Smart Grid commu-nications. In Proc. of IEEE ICC’13, pages 4488–4492. IEEE,2013.

[124] Mohammed Hasan, Ekram Hossain, and Dusit Niyato. Ran-dom access for machine-to-machine communication in lte-advanced networks: issues and approaches. CommunicationsMagazine, IEEE, 51(6):86–93, 2013.

[125] R. Davies and M. Ghosh. Field trials of DVB-T sensing for TVWhite Spaces. In Proc. of IEEE DySPAN’11, pages 285–296,May 2011.

[126] M. Nekovee. Quantifying the Availability of TV White Spacesfor Cognitive Radio Operation in the UK. In Proc. of IEEEICC’09, pages 1–16, 2009.

[127] A. Domingo, B. Bellalta, and M. Oliver. White Spaces inUHF band: Catalonia case study and impact of the DigitalDividend. In Proc. of EUNICE’12, pages 33–40. Springer,2012.

[128] Unlicensed Operation in the TV Broadcast Bands - THIRDMEMORANDUM OPINION AND ORDER, 2012.

[129] Ofcom. TV white spaces – A consulta-tion on white space device requirements.

30

http://stakeholders.ofcom.org.uk/consultations/whitespaces/,November 2012.

[130] J. van de Beek, J. Riihijarvi, A. Achtzehn, and P. Mahonen.TV White Space in Europe. IEEE Transactions on MobileComputing, 11(2):178–188, February 2012.

[131] L. Sanabria-Russo, J. Barcelo, A. Domingo, and B. Bellalta.Spectrum Sensing with USRP-E110. In Proc. of MACOM’12,pages 79–84. Springer, 2012.

[132] L. Bedogni, A. Achtzehn, M. Petrova, and P. Mahonen. Smartmeters with TV gray spaces connectivity: A feasibility studyfor two reference network topologies. In Proc. of IEEESECON’14, 2014.

[133] T. Novlan, K. Rele, and S. Srikathyayani. Coverage and Den-sity Study of Wi-Fi in the TV White Spaces, 2010.

[134] T. Rappaport. Wireless Communications: Principles andPractice. Prentice Hall PTR, second edition, 2001.

[135] A.B. Flores, R.E. Guerra, and E.W. Knightly. IEEE 802.11af:a standard for TV white space spectrum sharing. IEEE Com-munications Magazine, 51(10):92–100, October 2013.

[136] IEEE 802.11af. 802.11af-2013 Local and metropolitanarea networks Part 11: Wireless LAN Medium AccessControl (MAC) and Physical Layer (PHY) SpecificationsAmendment 5: Television White Spaces (TVWS) Opera-tion. http://standards.ieee.org/findstds/standard/802.11af-2013.html, 2013.

[137] L. Simic, M. Petrova, and P. Mahonen. Wi-Fi , but not onSteroids : Performance Analysis of a Wi-Fi-like Network Oper-ating in TVWS under Realistic Conditions. In Proc. of ICC’12,pages 1533–1538, 2012.

[138] M. Nekovee. Cognitive Radio Access to TV White Spaces:Spectrum Opportunities, Commercial Applications and Re-maining Technology Challenges. In Proc. of IEEE DyS-PAN’10, pages 1–10, 2010.

[139] L. Bedogni, A. Trotta, M. Di Felice, and L. Bononi. Machine-to-Machine Communication over TV White Spaces for SmartMetering Applications. In Proc. of IEEE ICCCN13-SEP’13,2013.

[140] Karol Andersson, Carlson Wireless Technologies. Super Wi-Fi.White Paper, March 2011.

[141] T.X. Brown and D.C. Sicker. Can Cognitive Radio SupportBroadband Wireless Access? In Proc. of IEEE DySPAN’07,pages 123–132, 2007.

[142] I.F. Akyildiz, B.F. Lo, and R. Balakrishnan. Cooperative spec-trum sensing in cognitive radio networks: A survey. PhysicalCommunication, 4(1):40–62, March 2011.

[143] K. Chowdhury, R. Doost-Mohammady, W. MEleis, M. Di Fe-lice, and L. Bononi. Cooperation and communication in Cog-nitive radio networks based on TV spectrum experiments. InProc. of IEEE WoWMoM’11, pages 1–9, 2011.

[144] IEEE 1900.6. IEEE 1900.6 Working Group on Spectrum Sens-ing Interfaces and Data Structures for Dynamic SpectrumAccess and other Advanced Radio Communication Systems.http://grouper.ieee.org/groups/dyspan/6/, 2011.

[145] IEEE 802.19.1. IEEE 802.19 Task Group 1-Wireless Coexistence in the TV White Space.http://ieee802.org/19/pub/TG1.html.

[146] IEEE 1900.4. IEEE 1900.4 Architectural Building Blocks En-abling Network-Device Distributed Decision Making for Opti-mized Radio Resource Usage in Heterogeneous Wireless AccessNetworks. http://grouper.ieee.org/groups/dyspan/4/.

[147] K. Bian, Park J.-M., and B. Gao. Cognitive radio networks:medium access control for coexistence of wireless systems. InSpringer Book, pages 1–170. Springer, 2014.

[148] L. Bedogni, M. Di Felice, F. Malabocchia, and L. Bononi. In-door Communication over TV Gray Spaces based on SpectrumMeasurements. In Proc. of IEEE WVNC’14, 2014.

[149] C. Stevenson, G. Chouinard, S. Shellhammer, and W. Cald-well. IEEE 802.22: The first cognitive radio wireless regionalarea network standard. IEEE Communications Magazine,47(1):130–138, January 2009.

[150] R. Funada, F. Kojima, and H. Harada. IEEE 802.15.4m: The

first low rate wireless personal area networks operating in TVwhite space. In Proc. of IEEE ICON’12, pages 326–332, De-cember 2012.

[151] Cambridge University Press, editor. Understanding Weight-less. Webb, W., 2012.

[152] C. Ghosh, S. Roy, and D. Cavalcanti. Coexistence challengesfor heterogeneous cognitive wireless networks in TV whitespaces. IEEE Wireless Communications, 18(4):22–31, August2011.

[153] S. Kulac, A. Eksim, and M.H. Sazli. Effective Cooperativespectrum sensing in IEEE 802.22 standard with time diversity.In Proc. of IEEE ACTEA’09, pages 528–531, July 2009.

[154] G. Villardi, Y. Alemseged, C. Sun, C.-S. Cum, T. Nguyen,T. Baykas, and H. Harada. Enabling coexistence of multiplecognitive networks in TV white space. IEEE Wireless Com-munications, 18(4):32–40, August 2011.

[155] J. Wang, T. Baykas, S. Filin, M.A. Rahman, C. Song, andH. Harada. Coexistence protocol design for autonomousdecision-making systems in TV white space. In Proc. of IEEEWCNC’12, pages 3249–3254, April 2012.

[156] T. Henderson, G. Pei, R. Groves, T. Bosaw, M. Rush,C. Ghosh, and S. Roy. Wireless Network Coexistence - BoeingProject Report. Technical Report, 2009.

[157] Y. Zhang, R. Yu, M. Nekovee, Y. Liu, S. Xie, and S. Gjessing.Cognitive machine-to-machine communications: visions andpotentials for the smart grid. IEEE Network, 26(3):6–13, 20122012.

[158] M. Fadda, M. Murroni, and V. Popescu. A cognitive radioindoor HD1T V multi-vision system in the TV white spaces.In Proc. of IEEE ICCE’12, pages 27–28, 2012.

[159] W.S. Hoi-Sheung, J. Walrand, and J. Mo. McMAC: A ParallelRendezvous Multi-Channel MAC Protocol. In Proc. of WCNC2007, pages 334–340. IEEE, 2007.

[160] Z. Gu, Q.-S. Hua, and W. Dai. Fully Distributed Algorithmsfor Blind Rendezvous in Cognitive Radio Networks. In Proc.of ACM MobiHoc 2014, pages 155–165. IEEE, 2014.

[161] P. Ren, Y. Wang, Q. Du, and J. Xu. A survey on dynamicspectrum access protocols for distributed cognitive wirelessnetworks. In EURASIP Journal on Wireless Communicationsand Networking, pages 1–21. EJWCN, 2012.

[162] M. Di Felice, A.J. Ghandour, H. Artail, and L. Bononi. Inte-grating Spectrum Database and Cooperative Sensing for Cog-nitive Vehicular Networks. In Proc. of IEEE VTC’13, LasVegas, Nevada, USA, September 2013.

[163] M. Di Felice, K.R. Chowdhury, and L. Bononi. CooperativeSpectrum Management in Cognitive Vehicular Ad Hoc Net-works. In Proc. of IEEE VNC’11, pages 14–16, Amsterdam,Netherlands, November 2011.

[164] M. Di Felice, A.J. Ghandour, H. Artail, and L. Bononi. Analyz-ing the potential of Cooperative Cognitive Radio Technologyon Inter-Vehicle Communications. In Proc. of IFIP WirelessDays’10, pages 20–22, Venice, Italy, October 2010.

[165] L. Bedogni, M. Di Felice, A. Trotta, and L. Bononi. Dis-tributed Mobile Femto-Databases for Cognitive Access to TVWhite Spaces using PAWS. In Proc. of IEEE VTC-Fall’14,Vancouver, Canada, September 2014.

[166] M.A. McHenry, P.A. Tenhula, D. McCloskey, D.A. Roberson,and C.S. Hood. Chicago spectrum occupancy measurements& analysis and a long-term studies proposal. In Proc. ofTAPAS’06, 2006.

[167] P. Avez, P. Van Wesemael, A. Bourdoux, A. Chiumento,S. Pollin, and V. Moeyaert. Tuning the Longley-Rice propa-gation model for improved TV white space detection. In Proc.of IEEE SCVT’12, pages 1–6, November 2012.

[168] P.A. Zandbergen. Accuracy of iPhone locations: a comparisonof assisted GPS, WiFi and Cellular Positioning. Transactionsin GIS, 13:5–26, 2009.

[169] M. Di Felice, K.R. Chowdhury, W. Meleis, and L. Bononi. ToSense or To Transmit : A Learning-based Spectrum Manage-ment Scheme for Cognitive Radio Mesh Networks. In Proch ofIEEE WiMESH’10, pages 1–6, June 2010.

31

[170] R. Doost-Mohammady and K.R. Chowdhury. Design of Spec-trum Database Assisted Cognitive Radio Vehicular Networks.In Proc. of ICST CROWNCOM’12, 2012.

[171] D. Gurney, G. Buchwald, L. Ecklund, S.L. Kuffner, andJ. Grosspietsch. Geo-Location Database Techniques for In-cumbent Protection in the TV White Space. In Proc. of IEEEDySPAN’08, pages 1–9, October 2008.

[172] J.M. Peha. Spectrum sharing in the grey space. Telecommu-nications Policy, 37(2-3):167–177, March 2013.

[173] L.B. Le and E. Hossain. Resource Allocation for SpectrumUnderlay in Cognitive Radio Networks. IEEE Transactions onWireless Communications, 7(12):5306–5315, December 2008.

[174] ARUBA: Mobility Controllers.http://www.arubanetworks.com/products/mobility-controllers/, March 2014.

[175] CISCO: Cisco Wireless Control System.http://www.cisco.com/c/en/us/products/wireless/wireless-control-system/index.html, March 2014.

[176] Y. Cui, X. Ma, J. Liu, L. Wang, and Y. Ismailov. Policy-based flow control for multi-homed mobile terminals with{IEEE} 802.11u standard. Computer Communications, 39:33–40, 2014. Research advances and standardization activities in{WLANsResearch} advances and standardization activities in{WLANs}.

[177] R. Murty, J. Padhye, A. Wolman, and M. Welsh. Dyson:An Architecture for Extensible Wireless LANs. In Proc. ofACM USENIX’10, USENIXATC’10, pages 1–15, Berkeley,CA, USA, 2010. USENIX Association.

[178] R. Murty, J. Padhye, A. Wolman, and M. Welsh. An Ar-chitecture for Extensible Wireless LANs. In Proc. of ACMHotNets’08, October 2008.

[179] R. Murty, J. Padhye, A. Wolman, and B. Zill. Designing HighPerformance Enterprise Wi-Fi Networks. In Proc. of ACMUSENIX’08, April 2008.

[180] P. Zerfos, G. Zhong, J. Cheng, H. Luo, S. Lu, and J.J.-R. Li.DIRAC: A Software-based Wireless Router System. In Proc.of ACM MOBICOM’03, MobiCom ’03, pages 230–244, NewYork, NY, USA, 2003. ACM.

[181] V. Shrivastava, N. Ahmed, S. Rayanchu, S. Banerjee, S. Ke-shav, K. Papagiannaki, and A. Mishra. CENTAUR: Realiz-ing the Full Potential of Centralized Wlans Through a HybridData Path. In Proc. of ACM MOBICOM’09, pages 297–308,2009.

[182] I. Tinnirello, G. Bianchi, P. Gallo, D. Garlisi, F. Giuliano, andF. Gringoli. Wireless MAC processors: programming MACprotocols on commodity hardware. In Proc. of IEEE INFO-COM’12, pages 1269–1277. IEEE, 2012.

[183] G. Bianchi, P. Gallo, D. Garlisi, F. Giuliano, F. Gringoli, andI. Tinnirello. MAClets: Active MAC Protocols over Hard-coded Devices. In Proc. of ACM CoNEXT’12, pages 229–240,2012.

[184] N. McKeown, T. Anderson, H. Balakrishnan, and G. Parulkar.OpenFlow: Enabling Innovation in Campus Networks. ACMSIGCOMM Comput. Commun. Rev., 38(2):69–74, 2008.

[185] N. Gude, T. Koponen, J. Pettit, B. Pfaff, M.ın Casado,N. McKeown, and S. Shenker. NOX: towards an operatingsystem for networks. ACM SIGCOMM Comput. Commun.Rev., 38(3):105–110, July 2008.

[186] L. Suresh, J. Schulz-Zander, R. Merz, A. Feldmann, andT. Vazao. Towards Programmable Enterprise WLANS withOdin. In Proc. of ACM HotSDN’12, pages 115–120, 2012.

[187] J. Vestin, P. Dely, A. Kassler, N. Bayer, H. Einsiedler, andC. Peylo. CloudMAC: Towards Software Defined WLANs.ACM SIGMOBILE Mob. Comput. Commun. Rev., 16(4):42–45, February 2013.

[188] J. Vestin and A. Kassler. Qos enabled wifi mac layer processingas an example of a nfv service. In Network Softwarization(NetSoft), 2015 1st IEEE Conference on, pages 1–9, April2015.

[189] J. Vestin and A. Kassler. QoS Management for WiFi MAClayer processing in the Cloud - Demo Description. In Proc. of

IEEE MSWiM ’15, 2015.[190] P. Dely, A. Kassler, L. Chow, N. Bambos, N. Bayer, H. Ein-

siedler, and C. Peylo. BEST-AP: Non-intrusive estimationof available bandwidth and its application for dynamic accesspoint selection. Computer Communications, 39:78–91, 2014.

[191] USRP: Universal Software Radio Peripheral .http://www.ettus.com/.

[192] WARP: Wireless Open-access Research Platform.https://warpproject.org/trac.

[193] SORA: Microsoft Research Software Radio.http://research.microsoft.com/en-us/projects/sora/.

[194] F. Gringoli and L. Nava. OpenFWWF website.http://www.ing.unibs.it/ openfwwf/, 2009.

[195] P. Salvador, L. Cominardi, F. Gringoli, and P. Serrano. A firstimplementation and evaluation of the IEEE 802.11aa groupaddressed transmission service. ACM SIGCOMM Comput.Commun. Rev., 44(1):35–41, 2013.

[196] L. Sanabria-Russo, F. Gringoli, J. Barcelo, and B. Bellalta.Implementation and Experimental Evaluation of a Collision-Free MAC Protocol for WLANs. Proc. of IEEE ICC 2015,2015.

[197] E. Magistretti, O. Gurewitz, and E.W. Knightly. 802.11ec:collision avoidance without control messages. In Proc. of ACMMOBICOM’12, pages 65–76. ACM, 2012.

[198] J. Xiong, K. Sundaresan, K. Jamieson, M.A. Khojastepour,and S. Rangarajan. MIDAS: Empowering 802.11 ac Networkswith Multiple-Input Distributed Antenna Systems. In Proc. ofACM CoNEXT’14, pages 29–40. ACM, 2014.

[199] S. Aust, R. Prasad, and I. Niemegeers. Advances in Wire-less M2M and IoT: Rapid SDR-prototyping of IEEE 802.11ah.Proc. of IEEE LCN’14, pages 290–292, 2014.

[200] S. Gollakota, F. Adib, D. Katabi, and S. Seshan. Clearing theRF smog: making 802.11n robust to cross-technology interfer-ence. ACM SIGCOMM Comput. Commun. Rev., 41(4):170–181, 2011.

[201] C. Gabriel. Wireless Broadband Alliance Industry Report2013: Global Trends in Public Wi-Fi. WBA - Maravedis-Rethink, November 2013.

[202] F. Rebecchi, M. Dias de Amorim, V. Conan, A. Passarella,R. Bruno, and M. Conti. Data Offloading Techniques in Cel-lular Networks: A Survey. IEEE Communications Surveys &Tutorials, 2014.

[203] K. Samdanis, T. Taleb, and S. Schmid. Traffic Offload En-hancements for eUTRAN. IEEE Communications Surveys &Tutorials, 14(3):884–896, Third 2012.

[204] D. Astely, E. Dahlman, G. Fodor, S. Parkvall, and J. Sachs.LTE release 12 and beyond. IEEE Communications Magazine,51(7):154–160, July 2013.

[205] S. Landstrom, A. Furuskar, K. Johansson, L. Falconetti, andF. Kronestedt. Heterogeneous Networks (HetNets) – an ap-proach to increasing cellular capacity and coverage. EricssonReview, February 2011.

[206] J.G. Andrews. Seven ways that hetnets are a cellular paradigmshift. IEEE Communications Magazine, 51(3):136–144, March2013.

[207] 3GPP. 3GPP TS 23.261: IP flow mobility and seamless Wire-less Local Area Network (WLAN) offload (Rel. 10), 2011.

[208] A. de la Oliva, C.J. Bernardos, M. Calderon, T. Melia, and J.C.Zuniga. IP flow mobility: smart traffic offload for future wire-less networks. IEEE Communications Magazine, 49(10):124–132, October 2011.

[209] 3GPP. 3GPP TS 24.312: Access Network Discovery and Selec-tion Function (ANDSF) Management Object (MO) (Rel. 10),2011.

[210] Lusheng Wang and G.-S.G.S. Kuo. Mathematical Modeling forNetwork Selection in Heterogeneous Wireless Networks – A Tu-torial. IEEE Communications Surveys Tutorials, 15(1):271–292, First 2013.

[211] D.H. Hagos and R. Kapitza. Study on performance-centricoffload strategies for LTE networks. In Proc. of IEEE/IFIPWMNC’13, pages 1–10, April 2013.

32

[212] M. Bennis, M. Simsek, A. Czylwik, W. Saad, S. Valentin,and M. Debbah. When cellular meets WiFi in wireless smallcell networks. IEEE Communications Magazine, 51(6):44–50,June 2013.

[213] Kyunghan Lee, Joohyun Lee, Yung Yi, Injong Rhee, andSong Chong. Mobile data offloading: How much can wifi de-liver? IEEE/ACM Transactions on Networking, 21(2):536–550, April 2013.

[214] L. Hu, C. Coletti, Huan. N., I.Z. Kovacs, B. Vejlgaard,R. Irmer, and N. Scully. Realistic Indoor Wi-Fi and FemtoDeployment Study as the Offloading Solution to LTE MacroNetworks. In Proc. of IEEE VTC Fall’12, pages 1–6, Sept2012.

[215] F. Mehmeti and T. Spyropoulos. Performance Analysis of“On-the-spot” Mobile Data Offloading.pdf. In Proc. of IEEEGLOBECOM’13, pages 1577–1583, Dec 2013.

[216] S. Singh, H.S. Dhillon, and J.G. Andrews. Offloading inHeterogeneous Networks: Modeling, Analysis, and DesignInsights. IEEE Transactions on Wireless Communications,12(5):2484–2497, May 2013.

[217] T. Wang, W. Jia, G. Xing, and M. Li. Exploiting Statisti-cal Mobility Models for Efficient Wi-Fi Deployment. IEEETransactions on Vehicular Technology, 62(1):360–373, Jan-uary 2013.

[218] E. Bulut and B.K. Szymanski. WiFi Access Point Deploy-ment for Efficient Mobile Data Offloading. In Proc. of ACMPINGEN’12, PINGEN ’12, pages 45–50, New York, NY, USA,2012. ACM.

[219] E.M.R. Oliveira and A.C. Viana. From routine to networkdeployment for data offloading in metropolitan areas. In Proc.of IEEE SECON’14, pages 126–134, June 2014.

[220] H. Elsawy, E. Hossain, and M. Haenggi. Stochastic Geometryfor Modeling, Analysis, and Design of Multi-Tier and Cogni-tive Cellular Wireless Networks: A Survey. IEEE Communi-cations Surveys & Tutorials, 15(3):996–1019, Third 2013.

[221] A.J. Nicholson and B.D. Noble. BreadCrumbs: ForecastingMobile Connectivity. In Proc. of ACM MOBICOM’08, pages46–57, 2008.

[222] S. Gatmir-Motahari, H. Zang, and P. Reuther. Time-clustering-based Place Prediction for Wireless Subscribers.IEEE/ACM Transactions on Networking, 21(5):1436–1446,October 2013.

[223] A. Balasubramanian, R. Mahajan, and A. Venkataramani.Augmenting Mobile 3G Using WiFi. In Proc. of ACM Mo-biSys’10, MobiSys ’10, pages 209–222, New York, NY, USA,2010. ACM.

[224] N. Ristanovic, J.-Y. Le Boudec, A. Chaintreau, and V. Er-ramilli. Energy Efficient Offloading of 3G Networks. In Proc.of IEEE MASS’11, pages 202–211, Oct 2011.

[225] I. Trestian, S. Ranjan, A. Kuzmanovic, and A. Nucci. Tam-ing the Mobile Data Deluge With Drop Zones. IEEE/ACMTransactions on Networking, 20(4):1010–1023, August 2012.

[226] F. Malandrino, C. Casetti, C. Chiasserini, and M. Fiore. Op-timal Content Downloading in Vehicular Networks. IEEETransactions on Mobile Computing, 12(7):1377–1391, July2013.

[227] A. Bazzi, B.M. Masini, A. Zanella, and G. Pasolini. {IEEE}802.11p for cellular offloading in vehicular sensor networks.Computer Communications, 60:97–108, 2015.

[228] X. Zhuo, Gao. W., Cao. G., and H. Hua. An Incentive Frame-work for Cellular Traffic Offloading. IEEE Transactions onMobile Computing, 13(3):541–555, March 2014.

[229] L. Lei, Zhong. Z., C. Lin, and X. Shen. Operator controlleddevice-to-device communications in LTE-advanced networks.IEEE Wireless Communications, 19(3):96–104, June 2012.

[230] B. Han, P. Hui, V.S.A. Kumar, M.V. Marathe, Jianhua Shao,and A. Srinivasan. Mobile Data Offloading through Op-portunistic Communications and Social Participation. IEEETransactions on Mobile Computing, 11(5):821–834, May 2012.

[231] M.V. Barbera, A.C. Viana, M. Dias de Amorim, and J. Stefa.Data offloading in social mobile networks through {VIP} del-

egation. Ad Hoc Networks, 19:92–110, 2014.[232] X. Wang, M. Chen, Z. Han, D. Wu, and T. Kwon. TOSS: Traf-

fic Offloading by Social Network Service-Based OpportunisticSharing in Mobile Social Networks. In IEEE INFOCOM, pages2346–2354, April 2014.

[233] W. Peng, F. Li, X. Zou, and J. Wu. The virtue of patience:Offloading topical cellular content through opportunistic links.In Proc. of IEEE MASS ’13, pages 402–410, Oct 2013.

[234] J. Whitbeck, Y. Lopez, J. Leguay, V. Conan, and M. Dias deAmorim. Push-and-track: Saving infrastructure bandwidththrough opportunistic forwarding. Pervasive and Mobile Com-puting, 8(5):682–697, 2012.

[235] L. Valerio, R. Bruno, and A. Passarella. Cellular traffic offload-ing via Opportunistic Networking with Reinforcement Learn-ing. Computer Communications, pages –, 2015.

[236] R. Bruno, A. Masaracchia, and A. Passarella. Offloadingthrough Opportunistic Networks with Dynamic Content Re-quests. In Proc. of IEEE CARTOON’14, 2014.

[237] M. Conti. Computer communications: Present status and fu-ture challenges. Computer Communications, 37:1–4, 2014.

33