Cross-layer Optimized Networking For Next-generation 5g Ad Hoc …m... · 2021. 4. 12. ·...
Transcript of Cross-layer Optimized Networking For Next-generation 5g Ad Hoc …m... · 2021. 4. 12. ·...
Cross-layer Optimized Networking for Next-Generation 5G Ad Hoc
Networks
A Dissertation Presented
by
Jithin Jagannath
to
The Department of Electrical and Computer Engineering
in partial fulfillment of the requirements
for the degree of
Doctor of Philosophy
in
Electrical Engineering
Northeastern University
Boston, Massachusetts
August 2019
To my family for all the support and all the sacrifice they have made to empower me to achieve
everything in life.
i
Contents
List of Figures iv
List of Tables vi
List of Acronyms vii
Acknowledgments xiii
Abstract of the Dissertation xiv
1 Introduction 1
2 DRS: Deadline Based Routing and Spectrum Allocation for Tactical Ad Hoc Network 42.1 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62.2 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102.3 Deadline-Based Routing and Spectrum Allocation . . . . . . . . . . . . . . . . . . 11
2.3.1 Network Utility Function . . . . . . . . . . . . . . . . . . . . . . . . . . . 112.3.2 Distributed Deadline-based Routing and Spectrum Allocation Algorithm . 14
2.4 Performance Evaluation Through Simulation . . . . . . . . . . . . . . . . . . . . 172.4.1 Scenario 1: Network performance as the number of session increases (All
sessions started at random time) . . . . . . . . . . . . . . . . . . . . . . . 172.4.2 Scenario 2: Network performance as the data rate of the sessions increase . 192.4.3 Scenario 3: Examining the effect of different components of DRS . . . . . 20
2.5 Testbed Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222.5.1 Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222.5.2 Adaptive Cross-Layer (AXL) . . . . . . . . . . . . . . . . . . . . . . . . 24
2.6 Testbed Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292.6.1 Establishing Baseline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292.6.2 DRS and ROSA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3 HELPER: Heterogeneous Efficient Low Power Radio for Enabling Ad Hoc EmergencyPublic Safety Networks 343.1 Related Works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
ii
3.2 Concept of Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403.2.1 Types of HELPERs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403.2.2 Deployment Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.3 HELPER Design and Implementation . . . . . . . . . . . . . . . . . . . . . . . . 453.4 HELPER’s Cross-Layer Protocol Stack . . . . . . . . . . . . . . . . . . . . . . . 45
3.4.1 HELPER Packet Handling . . . . . . . . . . . . . . . . . . . . . . . . . . 543.4.2 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 563.4.3 End-User Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . 563.4.4 Emergency Response Center Dashboard . . . . . . . . . . . . . . . . . . . 58
3.5 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 593.5.1 Operational Proof-of-Concept . . . . . . . . . . . . . . . . . . . . . . . . 593.5.2 Testbed Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
3.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4 VL-MAC: Opportunistic MAC Protocol for Visible Light ad Hoc Network 674.1 LANET: Visible-Light Ad Hoc Networks . . . . . . . . . . . . . . . . . . . . . . 69
4.1.1 Envisioned Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . 694.1.2 LANETs vs Traditional MANETs . . . . . . . . . . . . . . . . . . . . . . 724.1.3 Main Design Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.2 Related Works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 764.3 Neighbor discovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 794.4 Design of VL-MAC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 844.5 Performance Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 884.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
5 VL-ROUTE: A Cross-Layer Routing Protocol for Visible Light Ad Hoc Network 915.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 915.2 Design of VL-ROUTE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 945.3 Performance Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 995.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
6 Conclusion 104
Bibliography 106
iii
List of Figures
2.1 Tactical ad hoc network. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.2 Scenario 1: η vs No. of sessions. . . . . . . . . . . . . . . . . . . . . . . . . . . . 182.3 Scenario 1: ρ vs No. of sessions. . . . . . . . . . . . . . . . . . . . . . . . . . . . 182.4 Scenario 2: η vs No. of sessions. . . . . . . . . . . . . . . . . . . . . . . . . . . . 192.5 Scenario 2: ρ vs No. of sessions. . . . . . . . . . . . . . . . . . . . . . . . . . . . 192.6 Scenario 3: η vs No. of sessions. . . . . . . . . . . . . . . . . . . . . . . . . . . . 212.7 Scenario 3: η vs Parameter τ. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212.8 Scenario 3: η vs Parameter α. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222.9 Scenario 3: η vs No. of sessions (Deadline=50 s). . . . . . . . . . . . . . . . . . . 222.10 Layout of the five-node grid topology. . . . . . . . . . . . . . . . . . . . . . . . . 242.11 Adaptive cross-layer (AXL) framework. . . . . . . . . . . . . . . . . . . . . . . . 242.12 FSM of MAC protocol. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272.13 Five-node USRP testbed. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292.14 End-to-End Delay vs Number of Session. . . . . . . . . . . . . . . . . . . . . . . 302.15 Test 1: η vs No. of sessions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322.16 Test 1: ρ vs No. of sessions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322.17 Test 1: η vs No. of sessions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322.18 Test 1: ρ vs No. of sessions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322.19 Test 3: η vs No. of sessions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.1 HELPER development prototype. . . . . . . . . . . . . . . . . . . . . . . . . . . 363.2 Static HELPER design. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413.3 Envisioned final mobile HELPER design. . . . . . . . . . . . . . . . . . . . . . . 423.4 Aerial HELPER (Erle copter). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433.5 HELPER’s cross-layer protocol stack design and implementation. . . . . . . . . . 463.6 FSM of the MAC protocol. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 473.7 Network diagram. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493.8 Packet formats. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 533.9 Website application. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 563.10 ERC application. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 583.11 HELPER protoype. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 593.12 6-node HELPER network. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 603.13 Local text messaging. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
iv
3.14 Neighbor text messaging. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 613.15 ERC’s ALERT. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 613.16 Distress message. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 613.17 6-node HELPER network for quantitative evaluation. . . . . . . . . . . . . . . . . 623.18 Maximum energy consumed by a node. . . . . . . . . . . . . . . . . . . . . . . . 643.19 Normalized throughput of the network. . . . . . . . . . . . . . . . . . . . . . . . . 643.20 Network lifetime vs No. of sessions. . . . . . . . . . . . . . . . . . . . . . . . . . 653.21 Normalized throughput vs No. of sessions. . . . . . . . . . . . . . . . . . . . . . . 653.22 Analysis of packet delivery. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 653.23 Average delay vs No. of sessions. . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.1 LANETs employed for civilian and military applications. . . . . . . . . . . . . . . 704.2 Architecture of a LANET node. . . . . . . . . . . . . . . . . . . . . . . . . . . . 764.3 Super-slot structure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 814.4 Performance of neighbor discovery vs Ratio of seeking nodes. . . . . . . . . . . . 834.5 Random mode vs Synchronous mode. . . . . . . . . . . . . . . . . . . . . . . . . 834.6 Convergence of neighbor discovery as super-slots increase. . . . . . . . . . . . . . 834.7 Timing diagram of VL-MAC. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 854.8 Throughput comparison between VL-MAC and CSMA/CA. . . . . . . . . . . . . 894.9 Percentage of full-duplex communication established. . . . . . . . . . . . . . . . . 894.10 Packets dropped due to hidden nodes. . . . . . . . . . . . . . . . . . . . . . . . . 89
5.1 Timing diagram. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 975.2 Throughput vs No. of session. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1015.3 Full-Duplex vs No. of session. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1015.4 Throughput vs No. of session. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1025.5 Throughput vs No. of session. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1025.6 Throughput vs Link blockage. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1025.7 Throughput vs Estimation error. . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
v
List of Tables
2.1 Parameters of scenario 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182.2 Parameters of scenario 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192.3 Parameters of scenario 3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202.4 Parameters to baseline end-to-end delay. . . . . . . . . . . . . . . . . . . . . . . . 302.5 Parameters for testbed evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.1 Summary of technology. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403.2 Heterogeneous wireless link parameters. . . . . . . . . . . . . . . . . . . . . . . . 463.3 Neighbor table of node i. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 523.4 Evaluation parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.1 Comparison between LANETs and MANETs. . . . . . . . . . . . . . . . . . . . . 724.2 Summary of MAC protocols for VLC. . . . . . . . . . . . . . . . . . . . . . . . . 804.3 Simulation parameters for neighbor discovery. . . . . . . . . . . . . . . . . . . . . 84
5.1 Parameters of simulation in grid topology. . . . . . . . . . . . . . . . . . . . . . . 100
vi
List of Acronyms
5G 5th Generation
ACK Acknowledgment
ACN Availability Confirmation
AH Aerial HELPER
AP Access Point
API Application Programming Interface
ARQ Automatic Reply Request
ART Availability Request
AODV Ad hoc On-Demand Distance Vector
AXL Adaptive Cross-Layer
BD Block Diagonalization
BE Backoff Exponent
BER Bit Error Rate
BI Beacon Interval
CAP Contention Access Period
CAD Channel Activity Detection
CC Control Channel
CCC Common Control Channel
CCA Clear Channel Assessment
CDMA Code Division Multiple Access
CFP Contention Free Period
vii
CMS Control Micro-Slots
COmBAT CrOss-layer Based testbed with Analysis Tool
CSMA/CA Carrier Sense Multiple Access/Collision Avoidance
CSMA/CD Carrier Sense Multiple Access/Collision Detection
CSK Color-Shift Keying
CTS Clear-to-send
CVS Collaborative Virtual Sensing
DMT Discrete Multi-Tones
D2D Device-to-Device
DC Data Channel
DCN Data Center Network
DCO-OFDM Direct-Current Optical Orthogonal Frequency Division Multiplexing (OFDM)
DoA Direction of Arrival
DoD Department Of Defense
DRS Deadline-based cross-layer Routing and Spectrum allocation
ECC Error-Correction Code
EMI Electromagnetic Interference
ER Emergency Responder
ERC Emergency Response Center
EU End User
FCC Federal Communications Commission
FEC Forward Error Correction
FEMA Federal Emergency Management Agency
FSM Finite State Machine
FSO Free Space Optics
FOV Field Of View
GPS Global Positioning System
viii
GTS Guaranteed Time Slots
HELPER Heterogeneous Efficient Low PowEr Radio
HTL Hop-To-Live
I2V Infrastructure to Vehicle
IM/DD Intensity-Modulation Direct-Detection
IoT Internet of Things
IR Infrared Radiation
ISM Industrial, Scientific and Medical
ISR Intelligence, Surveillance, and Reconnaissance
LAN Local Area Network
LANET Visible Light Ad Hoc Networks
LCD Liquid Crystal Display
LED Light Emitting Diode
Li-Fi Light Fidelity
LOS Line of Sight
LPI/LPD Lower Probability of Intercept/Lower Probability of Detection
LTE Long-Term Evolution
MAC Medium Access Control
MA-DMT Multiple Access Discrete Multi-Tones
MANET Mobile Ad Hoc Network
MHC Minimum Hop Count
MH Mobile HELPER
MUI Multi-User Interference
MU-MIMO Multi-User Multiple-Input Multiple-Output
MU-MISO Multi-User Multiple-Input Single-Output
NAV Network Allocation Vector
NB Number of Backoffs
ix
ND Network Discovery
MC-CDMA Multi-carrier Code Division Multiple Access (CDMA)
NLOS Non-Line-Of-Sight
NRL Naval Research Labs
OAI Optimization Assisting Information
OC Optical Carrier
OCDMA Optical Code-Division Multiple Access
OFDM Orthogonal Frequency Division Multiplexing
OFDMA Orthogonal Frequency Division Multiple Access
OLSR Optimized Link State Routing
OOC Optical Orthogonal Codes
O-OFDMA Optical Orthogonal Frequency Division Multiple Access
O-OFDM-IDMA Optical Orthogonal Frequency Division Multiplexing Interleave Division Multi-ple Access
OOK On-Off Keying
OWC Optical Wireless Communication
OWMAC Optical wireless MAC
PD Photon Detector
PHR PHY Header
PHY Physical
PRO-OFDM Polarity Reversed Optical OFDM
QoS Quality of Service
RA Random Access
RES Reserve Sectors
RF Radio Frequency
ROC Random Optical Codes
RPI Raspberry Pi
RRS Route Reliability Score
x
RSSI Received Signal Strength Indication
SACW Self-Adaptive minimum Contention Window
SDR Software Defined Radio
SH Static HELPER
S-IDLE Synchronous Idle State
SINR Signal-to-Interference-plus-Noise power Ratio
SNR Signal-to-Noise power Ratio
SWaP (Size, Weight, and Power)
TDD Time Division Duplex
TDMA Time Division Multiple Access
THP Tomlinson-Harashima Precoding
TIA Telecommunication Industry Association
TR Transceiving State
UHD Universal Hardware Driver
USRP Universal Software Radio Peripheral
UV Ultraviolet
UVC Ultraviolet Communication
VANET Vehiclular Ad Hoc Network
VHF Very High Frequency
VLC Visible Light Communication
VLN Visible Light Node
V2I Vehicle to Infrastructure
V2V Vehicle to Vehicle
VQ Virtual Queue
VQL Virtual Queue Length
Web App Website Application
WEA Wireless Emergency Alerts
xi
WiFi Wireless Fidelity
WiMAX Worldwide Interoperability for Microwave Access
WSN Wireless Sensor Network
ZF Zero Forcing
xii
Acknowledgments
Being a non-traditional graduate student working full-time, pursuing a doctoral degreewould have been impossible without the immense support and consideration of several people whohave played a crucial role along the way. Firstly, I would like to express my sincere gratitude tomy advisor Dr. Tommaso Melodia for his continuous support of my doctoral study and relatedresearch work. He has been a source of inspiration since the time I joined his group during my M.Sat University at Buffalo. Well-aware of his busy schedule, I am grateful for his patience, motivation,and immense knowledge that has guided me through this process. I have always felt more motivatedand determined after each of our meetings and discussions. I cannot imagine having a better advisorand mentor for my entire graduate studies. Similarly, I would like to express my sincere gratitudeto my supervisor, Mr. Andrew Drozd, who motivated me to pursue graduate school while workingat his company. He has been extremely flexible and accommodating with my requests and needsfor any requirement that I have put forth in this regards. I sincerely thank him for all his support,guidance and inspiration which was a key factor that has helped me achieve this goal.
Besides my advisor, I would like to thank the rest of my committee members; Dr. StefanoBasagni and Dr. Kaushik Roy Chowdhury, for their feedback, comments, and encouragement ateach stage of this process. I want to especially thank them for their time and consideration toaccommodate my request for appointments through their busy schedule even when it was out of theregular schedule.
I would also like to thank my labmates at Wireless Networks and Embedded Systems(WINES) Lab and at ANDRO Computational Solutions for stimulating discussions and workingon projects together. In particular, I would like to thank Sean Furman who was a key member ofmy team at ANDRO. He served as a co-author on a couple of papers that were published duringthis period. I also would like to thank Jessica Griffin and Ramona Smith of our HR department forproctoring my exams for all my remote courses during this period.
I am beyond grateful to my parents and my sister for supporting throughout my life andvalue their immense sacrifices to ensure success in my life. I would not have reached my goal withoutthe constant motivation, support and sacrifice they have made for years at every step of my life.Last but not the least, I would not have started, continued and completed this process without theconstant encouragement and support by my wife, Anu Jagannath, who has been a pillar of supportthroughout this academic pursuit. Being a scientist herself, she has not only been my strength andsupport system but also my colleague and co-author on several of my publications. I will cherish theextended discussions we had at work, in our car, and at home that has led to some interesting ideasand solutions. I am grateful for everything she has done to make this process as easy as possible forme. This would have been unattainable without her reinforcing attitude throughout this process.
xiii
Abstract of the Dissertation
Cross-layer Optimized Networking for Next-Generation 5G Ad Hoc
Networks
by
Jithin Jagannath
Doctor of Philosophy in Electrical Engineering
Northeastern University, August 2019
Dr. Tommaso Melodia, Advisor
The exponential growth of devices that rely on wireless communication to operate hasintroduced significant stress on the limited resources. Additionally, to offset the cost of installingnew infrastructure and to maximize revenue, 5G (5th Generation) network providers are expectedto extend their services beyond traditional cellular communication to support Internet-of-Things(IoT) and machine-to-machine ad hoc communication. To accommodate all these devices over thenext decade, there are two key research directions that need to be adopted; (i) optimizing the use ofavailable resources to meet the application-specific quality of service (QoS) requirements and (ii)develop technology that enables the utilization of unexplored and unlicensed parts of the spectrum tocomplement the current radio frequency (RF) based networks. This work focuses on how cross-layeroptimized algorithms can be the answer to both these requirements for the next-generation of 5Gnetwork.
First, cross-layer optimization is employed to enable both tactical and emergency ad hocnetworks to meet their specific requirements. To this end, a Deadline-based cross-layer Routingand Spectrum allocation (DRS) algorithm is proposed for tactical ad hoc networks to handle theheterogeneous nature of traffic. This work also puts forth a cross-layer architecture that will enableimplementation and evaluation of such cross-layer optimized routing algorithms. The proposedsolution is evaluated on a software-defined radio (SDR) testbed and shown to outperform state-of-the-art routing algorithms. Next, to aid modern emergency response, a low cost Heterogeneous EfficientLow PowEr Radio (HELPER) network is designed and developed to provide complete end-to-endconnectivity for both survivors and first responders. This is realized by designing an energy-aware routing algorithm that aims to maximize network lifetime. The operational feasibility of the
xiv
proposed HELPER network is demonstrated by developing HELPER prototype using Commercial-Off-The-Shelf (COTS) components. Thereafter, extensive quantitative evaluation is performed onthe developed HELPER testbed.
Visible Light Communication (VLC) is envisioned as a major 5G technology that canbe complementary to RF and help mitigate the congestion in the RF spectrum. Visible Light AdHoc Networks (LANET) have the potential to offer capabilities to satisfy growing industrial andmilitary requirements, including low-latency, high bandwidth communication under high networkdensity. The challenges imposed by hidden nodes, deafness and blockage are unique to LANETand influence the network differently from traditional Mobile Ad Hoc Networks (MANET) dueto directionality and Line of Sight (LOS) requirements. Therefore, networking protocols have tobe redesigned with careful consideration of these challenges. These unique challenges demandconsideration of networking problems from the cross-layer perspective. As a significant step inrealizing LANETs, this work first proposes an opportunistic Medium Access Control (MAC) protocoldesigned specifically to mitigate challenges due to deafness, hidden node problem and maximizethe utilization of full-duplex communication. Next, to advance the development of LANETs, adistributed cross-layer routing protocol (VL-ROUTE) that interacts closely with the MAC layerto maximize the throughput of LANET is proposed. In this manner, a LANET with a cross-layeroptimized link and network layer has been successfully designed and evaluated for bolstering thefuture 5G ad hoc networks.
xv
Chapter 1
Introduction
As the relevance of devices relying on wireless communication keeps gaining momentum
in all walks of our modern life, the demand for scarce resources will increase exponentially as the
deployment of 5th Generation (5G) networks progresses. It is estimated that by 2020, over 50 billion
devices will be absorbed into the Internet, generating a global network of “things” of dimensions
never seen before [1]. Given that only a few radio spectrum bands are available to the wireless carriers
[2], battling spectrum congestion while ensuring Quality of Service (QoS) will become the prime
goal in the upcoming decade. Additionally, the reliance on wireless communication has rendered it
an unavoidable component of critical applications including military missions, emergency response,
and medical services. To ensure these applications can sustain the growing number of devices,
novel networking paradigms such as cross-layer optimization need to be exploited to maximize the
utilization of the available spectrum and expand to the unutilized parts of the spectrum.
To this end, in Chapter 2, the Deadline-based cross-layer Routing and Spectrum allocation
(DRS) algorithm is proposed for a tactical ad hoc network. In a tactical ad hoc network, there
exists a constant tension between available resources and the required QoS performance. Nodes
in the network have to deal with severe interference, spectrum crunch, adversarial jamming and
changing network topologies. Additionally, a typical tactical network is required to handle various
traffic classes including regular sampling data, voice, surveillance video, threat alert, among others.
Each of these traffic classes has substantially different QoS based deadline requirements. In these
scenarios, it becomes important to examine the interaction between spectrum management, routing
and session management to develop cross-layer optimized algorithms capable of maximizing the
effective throughput of the network. The proposed solution is extensively evaluated using simulations
and on a Software Defined Radio (SDR) based testbed. The work in this chapter has been presented
1
CHAPTER 1.
at IEEE Conference on Military Communications held on November 2016 at Baltimore, MD, and
at IEEE Global Communications Conference, Washington D.C. on December 2016. The results
and insight obtained from the maturation of this work were also published in IEEE Transactions on
Mobile Computing in 2018.
In the past years, several lives have been devastated by hurricanes, tsunami, floods, earth-
quakes, and other natural disasters. Similar natural and man-made disasters are undesirable but
sometimes unavoidable [3, 4]. In these scenarios, one of the most critical infrastructures affected is
often communication networks [5, 6]. Clearly, wireless communication is an essential component
to maintain connectivity for such alert systems, Emergency Responders (ERs) as well as affected
individuals when traditional infrastructures like cell towers are affected or unavailable. Therefore, in
Chapter 3, an end-to-end solution, Heterogeneous Efficient Low PowEr Radio (HELPER) network is
proposed for emergency ad hoc network to connect survivors and first responders in the aftermath of
disasters. The initial stages of this work were presented at IEEE Consumer Communications and
Networking Conference held at Las Vegas, NV, in January 2019. A comprehensive article discussing
the development of the prototype, experiments, and demonstration of the HELPER network was
published in Ad Hoc Networks (Elsevier) in March 2019.
Visible Light Communication (VLC) has come to the forefront with the advent of modern
Light Emitting Diode (LED) technology that consumes low power and has a short response time. The
unexploited, unregulated spectrum (400 to 800 THz) is a promising candidate to alleviate the Radio
Frequency (RF) spectrum crunch. The exploration of VLC has been limited to various point-to-point
applications including setting up Light Fidelity (Li-Fi) [7] networks using smart lights, among others.
In this context, several topologies such as peer-to-peer, star and broadcast have been considered to
design Medium Access Control (MAC) protocols. This work concentrates on exploiting VLC for ad
hoc networking in military and civilian applications. Visible Light Ad Hoc Networkss (LANETs)
are envisioned to contribute significantly to the upcoming Internet of Things (IoT) revolution in both
indoor and outdoor spaces. Some indoor applications of LANETs include Device-to-Device (D2D)
communication to support IoT technology, enabling indoor positioning system, and aiding the
traditional RF based networks [8, 9]. A simple example is the devices (TV, thermostat among others)
in a smart home forming a LANET. In outdoor scenarios, one of the most promising applications
for LANET is related to vehicular communications. LANETs can also be used for air, ground
and underwater tactical missions such as Intelligence, Surveillance, and Reconnaissance (ISR)
missions entailing deployment of ships, soldiers, and unmanned surface vehicles. LANETs can also
be used in high-security military areas where RF communication is prone to eavesdropping or is
2
CHAPTER 1.
extremely congested. Therefore, as an initial step towards making LANETs a reality, Chapter 4 and
Chapter 5 propose novel MAC and routing protocols that are designed specifically to overcome the
challenges posed by LANETs. Some portion of Chapter 4 was also published in Ad Hoc Networks
(Elsevier) in 2018, introduced the overall concept of LANETs to the community. The MAC and
routing protocols were presented at the International Conference on Computing, Networking and
Communications during March 2018; and IEEE Symposium on a World of Wireless, Mobile, and
Multimedia Networks in June 2019 respectively. Finally, the conclusion of this work is presented in
Chapter 6.
3
Chapter 2
DRS: Deadline Based Routing and
Spectrum Allocation for Tactical Ad Hoc
Network
In a tactical ad hoc network, there exists a constant tension between available resources
and the required QoS performance. Nodes in the network have to deal with severe interference,
spectrum crunch, adversarial jamming and changing network topologies. Additionally, a typical
tactical network as depicted in Fig. 2.1 is required to handle various traffic classes including regular
sampling data, voice, surveillance video, threat alert, among others. Each of these traffic classes has
substantially different QoS based deadline requirements. For example, periodic surveillance data
might have looser deadline constraints when compared to a video or threat alert message. In these
delay-intolerant networks, only packets that arrive at the destination within the specified deadline are
viable and contribute to the overall network throughput. In these scenarios, it becomes important to
examine the interaction between spectrum management, routing and session management to develop
cross-layer control algorithms capable of maximizing the effective throughput of the network. In this
chapter, only the packets that arrive at the destination within the specified deadline is considered in
the computation of effective throughput.
Cognitive radio technology along with various dynamic spectrum access (DSA) techniques
have been proposed to improve the spectrum utilization of the network by enabling opportunistic
access of free spectrum chunks. In previous work [10], an optimization algorithm (ROSA) was
proposed to jointly select route and spectrum such that overall network throughput is maximized. This
4
CHAPTER 2.
Satellite Links
Unmanned Aerial Vehicles (UAVs)Airborne Relays
Vehicular Relays
Dismount Soldiers UsingRifleman Radios
Figure 2.1: Tactical ad hoc network.
algorithm combines the idea of backpressure algorithm [11] with channel-dependent opportunistic
routing. Simulations show that ROSA outperforms the traditional algorithms that use either dynamic
spectrum allocation with fixed route or dynamic routing with fixed spectrum allocation. In this
chapter, ROSA’s weakness has been realized to significantly extended the formulation to examine
the network performance in terms of effective throughput and reliability for multiple sessions with
different deadline constraints. Proposed solutions enable such routing and resource allocation
algorithms to handle heterogeneous QoS based traffic classes efficiently. Accordingly, a distributed
deadline-based optimization algorithm is developed for tactical ad hoc networks. Some of the
challenges in designing a deadline-based algorithm are as follows:
• Each node has to carefully manage multiple sessions to meet the deadline requirements.
For example, sessions with longer backlogs and larger deadlines can be held back while
accelerating shorter backlogged-smaller deadline sessions.
• Adopting an effective resource allocation procedure that would negotiate the access of medium
and choose optimal transmission parameters. In a large network, the spectrum occupancy
varies based on location and time, thus nodes may have to use different parts of the spectrum
in order to route a session in the most effective manner.
• Choosing appropriate routes to meet the needs of each session belonging to different traffic
5
CHAPTER 2.
classes while adapting to broken routes or failed nodes by choosing alternate paths.
• The design should be scalable, reduce communication overhead and yet enable the network to
adjust dynamically to the available resources. Therefore, it is critical to design a distributed
approach that is feasible on a practical network.
Therefore, the overall objective of this chapter is to design and evaluate a distributed
algorithm that utilizes the available resources to determine optimal route, session, and spectrum to
deliver the maximum number of packets to their intended destination within the specified deadline.
The weighted virtual queue (VQ) used in cross-layer optimization ensures proper management of
the sessions. The virtual queue length (VQL) takes into account deadlines associated with each
packet. The joint routing and spectrum allocation aspects of the algorithm provide optimal resource
allocation and enable opportunistic routing. The distributed nature of the proposed algorithm along
with forward progress based routing helps the network to recover from broken routes or failed nodes.
These features are critical in any delay-intolerant applications and will be especially useful in tactical
ad hoc networks where the delayed delivery of critical information in a multihop network can be
fatal.
2.1 Related Work
Dynamic spectrum allocation has been widely investigated with the objective to maxi-
mize spectrum utilization and is mainly divided into centralized [12, 13] and distributed [10, 14]
approaches. While spectrum allocation techniques are designed to improve spectrum utility-based
QoS [15, 16, 17], queue length based backpressure (Q-BP) scheduling algorithm was first proposed
in [11] and was shown to be throughput optimal in terms of achieving network stability under any
feasible load. It is well known that the Q-BP algorithm suffers from high computational complexity
and the last packet problem. To reduce the computational delay for practical implementation, a
greedy maximum scheduling (GMS) algorithm is studied in [18, 19, 20]. This algorithm first chooses
the link l with maximum weight from the set of all links S and eliminates links that interfere with l
from the set S . Next, it again picks the link with maximum weight among the remaining links of
set S and eliminates the link causing interference to it. This process is repeated until all links have
been considered. The trade-off here is the reduced network capacity. In [21], the authors solve a
centralized network throughput maximization problem that uses the backpressure algorithm. The
study also implements the solution on hardware to perform the evaluation. Even though the network
6
CHAPTER 2.
achieves throughput improvement, the network may be prone to last packet problem which is a
crucial hindrance for tactical networks. The last packet problem of Q-BP algorithm arises because of
the assumption that flows have an infinite amount of data packets being injected into the network.
Instead, in practical networks, the flows may be finite with some flows terminating and new flows
emerging. When a finite flow has the last packet in the queue, it may be stagnant for an extended
period of time because of the presence of other queues with a larger backlog. This is referred to as the
last packet problem. It has been shown that in these cases, queue length based schemes may not be
throughput optimal [22]. Accordingly, there has been considerable work on delay-based scheduling
[23, 24, 25, 26, 27, 28, 29] to improve the delay performance of the network and eliminate the last
packet problem.
In [23], the authors use a shadow queuing architecture so that each node maintains only one
queue per neighbor (irrespective of sessions) to reduce the complexity of the queuing structure and
improve the delay performance at the cost of throughput. Each node still has to maintain a separate
shadow queue (a counter) for every flow going through the node. The backpressure algorithm is
executed using the shadow queue counters and these counters are updated according to the optimal
number of shadow packets chosen to be transmitted over each link. The key point here is that the
number of shadow packets is like a permit to transmit on the given link from the real queue but not
associated with the flow of the shadow packet itself. The packet injection rate of the shadow queue is
kept slightly higher than the actual packet injection rate. The rates are designed as follows: if the
packet injection rate of the shadow queue is xt(t), the rate of the real queue is given by βxt(t), where
β is a positive real number smaller than one. Therefore, if the number of real packets in the queue is
less than the number of shadow packets to be transmitted, all the real packets in the corresponding
queue are transmitted. The authors show that the real queue length decreases uniformly at every node
as the value of β decreases, thus leading to lower delays by Little’s law. This decrease in the delay is
accompanied by reduced throughput performance. Maintaining a single queue per neighbor is only
beneficial in scenarios where the number of flows through a node is much greater than the number of
neighbors. Authors propose a self-regulated MaxWeight scheduling algorithm in [30], where each
node estimates the aggregated link rate. They prove that the self-regulated MaxWeight scheduling is
throughput-optimal (i.e. stabilize any traffic that can be stabilized by any other algorithm) when the
traffic flows are associated with fixed routes and the packet arrivals follow some statistical property.
Both [23] and [30] are designed for fixed route scenarios, thus lacking the improvement that could
be achieved by opportunistic routing.
In [24], the authors propose a delay-limiting algorithm to control the burstiness and delays.
7
CHAPTER 2.
They adapt the upper limit for the physical queues to ensure an upper per-hop delay limit at the
expense of throughput. To ensure that nodes in the network remain operational, a lower bound has to
be set on the upper queue limit. If the traffic reduces to a point such that lower bound comes into
play, the delay-limiting approach becomes ineffective. There is also a trade-off between delay and
the degree of multipath and opportunism. As the traffic is spread spatially to utilize multiple routes,
the lower bound on the queue may again render the delay control ineffective.
A cross-layer design is proposed in [25] using VQ structures to provide finite buffer size or
worst-case delay performance. In [26], the authors design a delay-aware joint flow control, routing,
and scheduling algorithm for a multihop network to maximize network utilization. However, due
to their ([26], [25]) centralized nature and high complexity they are not well suited for practical
distributed implementation [31]. In [27], a throughput optimal scheduling algorithm is proposed using
the largest weighted delay first algorithm. The idea is to serve the queue j for which γ jWj(t)r j(t) is
maximal, where Wj( j) is the weighted delay and r j(t) the achievable capacity for link j. Although
this algorithm is an easy and distributed way to achieve throughput optimality, this formulation does
not take into account the dynamic routing possibilities or queuing dynamics of multihop traffic. Since
[27] fails to capture the queuing dynamics of multihop traffic, a new delay metric is defined in [28]
to establish a linear relation between queue length and delay. The authors also propose a greedy
algorithm that is similar to GSM discussed earlier but uses delay differential rather than queue length.
Simulations show that the average queue length of the network is similar in Q-BP and delay-based
backpressure (D-BP) but the tail of the delay distribution is much longer for Q-BP. This implies that
some queues are stagnant over extended periods of time in Q-BP whereas D-BP reduces this problem.
Unlike the proposed algorithm, D-BP is designed for fixed routes and does not consider dynamic
routing.
In [29], a delay-driven MaxWeight scheduler is presented that gets around the last packet
problem and addresses instability of the queue length based algorithms caused by rate variations.
However, it has been shown in [32, 33] that there are other factors that contribute to the inefficiency
of the back-pressure algorithm including, inefficient spatial reuse, failure to opportunistically exploit
better link rates, underutilized link capacity and inefficient routing because of insufficient path
information.
Deadline-based routing has been recently studied in [34, 35] and [36]. In [34], an utility-
based algorithm is proposed for cyclic mobile social networks under the assumption that nodes
follow cyclic mobility, periodically encountering each other with high probability. It is difficult to
extend [34] to tactical ad hoc networks without apriori knowledge of the encounter probability. To
8
CHAPTER 2.
increase the packet delivery ratio, [35] adopts an epidemic based routing algorithm and [36] proposes
a capacity-constrained routing algorithm that decides which packets have to be replicated. The
replication strategies proposed in [35] and [36] to improve the packet delivery ratio may adversely
affect the achievable throughput. The major contributions of the work presented in this chapter can
be outlined as follows,
• A novel deadline-based joint routing and spectrum allocation algorithm is proposed for tactical
ad hoc networks to meet the deadline requirements of multiple sessions. To the best of the
author’s knowledge, this is the first work that combines the interaction of opportunistic routing,
spectrum allocation and deadline constraints to maximize the effective throughput of tactical
ad hoc networks.
• The proposed algorithm is able to adapt to the needs of a dynamic network by managing
multiple sessions with variable QoS. This is accomplished by making an optimal choice about
the session, route, spectrum and power allocation used to maximize the utilization of available
resources.
• A distributed approach is formulated to enable the implementation of the proposed algorithm
in a scalable manner.
• Performance of the proposed algorithm is extensively evaluated under various simulated
scenarios.
• Another major contribution of this chapter is the development of a cross-layer experimental
framework using SDR.
• Finally, to prove the practicality of the proposed algorithm, the deadline-based joint routing and
spectrum allocation algorithm is successfully implemented on this cross-layer SDR testbed.
The rest of the chapter is organized as follows. In Section 2.2, the system model is
described in detail. The design of the deadline-based routing algorithm is discussed in Section
2.3. Next in Section 2.4, a 49 node ad hoc network is simulated to evaluate the performance of the
proposed algorithm. The design and configuration of the cross-layer testbed are described in Section
2.5. The experimental evaluation of the proposed algorithm on a SDR based testbed is discussed in
Section 2.6. Finally, the summary is provided in Section 2.7.
9
CHAPTER 2.
2.2 System Model
Consider a multihop tactical ad hoc network with M primary users and N secondary users
modeled as a directed connectivity graph G(U,E), where U = {u0,u1, ...,uN+M} is a finite set of
wireless transceiver (nodes), and (i, j) ∈ E represents unidirectional wireless link from node ui to
node u j (for simplicity, they are referred to as node i and node j). G is assumed to be link symmetric,
i.e., if (i, j) ∈ E , then ( j, i) ∈ E . The nodes from the subset P U = {u1, ...,uM} are designated as
primary users, and nodes from the subset SU = {uM+1, ...,uM+N} are designated as secondary users.
The secondary network is composed of cognitive nodes capable of adapting to the current spectrum
usage. The primary users hold the license for the specific spectrum bands and have full access to
the spectrum without interference from any other users. In relevant scenarios, the primary user can
also be a non-cooperative node (the adversary). Since the entire spectrum is not always used by
primary users, the aim of the secondary user in a cooperative scenario is to maximize spectrum utility
while ensuring no interference to primary users. Thus, a secondary user has to use the spectrum
holes [10] to maximize the spectrum usage. The secondary network will also allocate resources such
that it maximizes the number of packets delivered at the destination within their respective deadline.
Only packets that reach the destination within the specified deadline contribute towards the effective
throughput computation. The set of neighbors for node i is given by N B i , { j : (i, j) ∈ E}.The secondary users are equipped with cognitive radios capable of scanning the available
spectrum to reconfigure their transceivers on-the-fly. The entire available spectrum is given by BW .
The cognitive transceiver is capable of tuning to a set of contiguous frequency bands [ f , f +∆B],
where ∆B is the bandwidth of the cognitive radio and ∆B < BW . It is assumed that the transmit
power can be varied to exploit any available spectrum opportunity. The spectrum opportunity is
defined as the limited availability of spectrum that might currently be used by nodes (primary or
secondary users) but can be further exploited by adjusting the transmit power such that it does not
violate the Bit Error Rate (BER) constraint of the existing transmission. This work is intended for
any general physical layer but it is assumed that multiple transmissions can occur concurrently on
the same frequency band, e.g., with different spreading codes.
The total spectrum, BW is divided into separate channels, a Common Control Channel
(CCC), and a data channel. All secondary nodes use CCC to share local information for spectrum
negotiation and data channel is used exclusively for data communication. The data channel is divided
into discrete set of carriers { fmin, fmin+1, ..., fmin−1, fmax}, each of bandwidth b and identified by a
unique discrete index. The cognitive radio of the secondary user can tune into a consecutive set of
10
CHAPTER 2.
carriers from [ fmin, fmax]. Let the traffic in the network consist of multiple sessions characterized
by the source-destination pair and the application generating the session. The arrival rates of each
session si ∈ Si at node i is given by λsi (t), and characterized by vector of arrival rates Λ.
2.3 Deadline-Based Routing and Spectrum Allocation
2.3.1 Network Utility Function
Consider that the tactical ad hoc network is assumed to operate over a time slotted channel.
The spectrum utility function is calculated by node i for every time slot t when node i is backlogged
and not already transmitting or receiving packets. Each node i maintains a separate VQ for each
session. Qsi (t) is defined as the VQL formed by packets of session s in node i at time slot t. Unlike
traditional queue length, the VQL gets inflated as time passes to penalize the node for holding packets
whose deadline is approaching. More details about the design of VQL is discussed below. For each
packet qsi ∈ Qs
i (t), that belongs to session s and stored at node i, a set of fields are defined, including,
• L(qsi ) is the length of the packet in bits,
• Tr(qsi ) is the remaining life time of the packet, which is based on the deadline D(qs
i ) assigned
to the packet at the source node,
• Td(qsi ) is the time to the destination as estimated at node i.
Based on these parameters, a weight wqsi[L(qs
i ),Td(qsi ),Tr(qs
i )] can be defined for each packet qsi ∈
Qsi (t) as follows,
wqsi(L(qs
i ),Td(qsi ),Tr(qs
i )) =L(qs
i )
max(Tr(qsi ),τ)max(Tr(qs
i )−Td(qsi ),τ)
(2.1)
Examining (2.1), it can be seen that the weight wqsi
assigned to each packet is directly
proportional to L (for simplicity, qsi is removed from these notations) and inversely proportional
to Tr and Td . The τ in (2.1) is a very small value used to avoid negative and infinite weights. The
parameter Tr helps to get rid of the well-known last packet problem since Tr will increase the VQL
as time elapses. This can be interpreted as the holding penalty imposed for packets being stagnant in
the queue for an extended period of time. Since Tr is dependent on the assigned deadline, it helps the
nodes to manage different sessions by pushing critical packets faster even if the actual queue length
is comparatively smaller. Considering just the deadlines alone will not help in cases where there
11
CHAPTER 2.
are two sessions with the same deadline but one is farther away from the destination than the other.
In such cases, Td will ensure that the session farther away from the destination moves through the
network at a faster rate compared to similar sessions closer to the destination. Therefore, Td can be
considered as a variable that either amplifies or diminishes the effect of Tr depending on the time
required to reach the destination. Td also encourages packets to take shorter routes if all other factors
like queue length and spectrum are the same for two different routes. The rationale will become
more evident when the network utility function used for the proposed algorithm is discussed.
Among these three parameters, the exact value of Td is not available at each node and has
to be estimated at each hop. For a centralized network, assuming global knowledge of the network,
Td can be estimated using average queuing delays, transmission rate, propagation delays and using
the knowledge of average delays experienced previously by packets with the same destination.
Estimating Td becomes further challenging in a distributed network where each node is required to
make decisions without global knowledge of the network. One solution is to estimate Td by using
queuing delay experienced by the session in the node itself. This information is used to slightly
overestimate the delay by assuming that the packet has to route through more than one node within
its transmission range itself. Underestimating Td would increase the risk of packets not reaching
the destination within the specified deadline. Therefore, Td is slightly overestimated according to
the characteristics of the network. Since Td is updated at every hop, the estimation error/margin
decreases as the packet moves closer to the destination. This method does not lead to any error
propagation since the value is updated at each hop. A simple way to estimate Td is based on distance
to destination (d), communication range (estimated based on maximum transmit power) of the nodes
deployed (R) and average time spent by the packet during each hop (Th) (estimated based processing
delay, queuing delay, transmission delay and propagation delay). The idea is to assume that a hop is
required every half range of a node and is given by α = R/2. Accordingly, an estimate of how much
time is required to reach the destination can be given as,
Td =d Th
α=
2d Th
R. (2.2)
The value of α can be varied according to the density of the network. Now from the
definition of weights, it can be seen that higher value is assigned to packets with more bits to transmit,
lower Tr and which are farther away from the destination. Accordingly, Virtual Queue Length (VQL)
of a session s in node i is defined as follows,
12
CHAPTER 2.
Qsi (t) = ∑
qsi∈Qs
i (t)wqs
i(L,Td ,Tr). (2.3)
Now, let a(qsi , j, t) = 1 represent a packet qs
i ∈ Qsi (t) is transmitted to node j at time slot t,
and a(qsi , j, t) = 0 otherwise. The routing profile of node i is defined as as
i (t) = [a(qsi , j, t)] j∈∈SU/i
qsi∈Qs
i (t),
and A represents the vector of routing profile asi (t) of all nodes in the network at instant t. The
transmission rate on link (i, j) during time slot t is defined as rsi j(t), and R as the vector of rates.
Then, the VQL of node i can be updated as,
Qsi (t +1) =
[Qs
i (t)+ ∑j∈N /i
∑qs
j∈Qsj(t)
wqsj(L,Td ,Tr)a(qs
j, i, t)− ∑j∈N /i
∑qs
i∈Qsi (t)
wqsi(L,Td ,Tr)a(qs
i , j, t)]+
.
(2.4)
Accordingly, the network link utility function Ui j for link (i, j) ∈ E for session s can be
defined as,
Ui j(asi (t)) =Ci j[Qs
i (t)−Qsj(t)]
+, (2.5)
where [Qsi (t)−Qs
j(t)]+ represents the differential VQL and Ci j is the achievable channel capacity
of the link (i, j) ∈ E at time slot t for a selected frequency ( f ) and the transmission strategy can be
given by,
Ci j( f ,Pi( f )), ∑f∈[ fi, fi+∆ fi ]
b. log2
[1+
Pi( f )PLi j( f )GN j( f )+ I j( f )
](2.6)
In the above equation, Pi( f ) represents the transmit power of node i on the frequency
f , PLi j( f ) is defined as the transmission loss due to path loss (can be computed based on the
chosen path loss model) from i to j, G represents the processing gain, which would be the length
of the spreading code when applicable, N j( f ) is the receiver noise on frequency f and I j( f ) is the
interference experienced by the receiving node j. Assuming a quasi-static channel, i.e. channel
conditions remain constant for the duration in between sensing and transmission of a packet. This
can be achieved with an efficient sensing mechanism and having a dedicated receiver that performs
sensing in parallel to the regular transceiver. As shown in (2.6), the achievable capacity primarily
depends on selected frequency F = [ fi, fi+∆ fi ], power allocation P = [Pi( f )], ∀i ∈ SU, , ∀ f and
the scheduling policy. Therefore, the overall notion of this network utility function is to couple
13
CHAPTER 2.
the constraints of packet deadline to the traditional queue length used in the differential backlog
algorithm. This is then weighted by the dynamic spectrum availability information to provide a joint
routing and spectrum allocation decision. Moreover, algorithms like ROSA [37, 38] does not handle
the QoS requirements of different traffic classes. Since this is essential for improving the reliability
of tactical ad hoc networks, the redefining of the queue length to form the new VQL is where the
proposed algorithm enhances the state-of-the-art.
2.3.2 Distributed Deadline-based Routing and Spectrum Allocation Algorithm
The overall optimization problem is to maximize the utility function discussed in (2.5).
The BER guarantees required for primary and secondary users are denoted as BERP U and BERSU
respectively. Accordingly, the required Signal-to-Interference-plus-Noise power Ratio (SINR)
thresholds required to achieve the target BER for the secondary and primary user can be represented
as SINRthP U and SINRth
P U respectively. Thus, the global objective of the optimization problem is to
find the optimal global vectors R, F, A and P that will maximize the sum of the network utilities,
under the power and BER constraints. The formulation of the optimization problem is as follows,
P1 : Given: G(U,E), PBgt , Qsi , BERSU , BERP U
Find: R, F, P, A
Maximize : ∑i∈SU
∑j∈SU/i
Ui j(asi (t)) (2.7)
subject to :
∑s∈S
rsi j ≤Ci j,∀i ∈ SU, ∀ j ∈ SU (2.8)
SINRk ≥ SINRthP U(BERP U),∀k ∈ P U,∀ f (2.9)
SINRl ≥ SINRthSU(BERSU),∀l ∈ SU,∀ f (2.10)
∑f∈[ fi, fi+∆ fi ]
Pi( f )≤ PBgti ,∀i ∈ SU (2.11)
In the above formulation, the objective is to maximize the network utility of all the active
links. The constraint (2.8) restricts the total amount of traffic in link (i, j) to be lower than or equal to
the physical link capacity. Constraint (2.9) and (2.10) imposes that any transmission by the secondary
user should guarantee the required BER for the active primary users and secondary user respectively.
14
CHAPTER 2.
Finally, PBgti is the instantaneous power available at the cognitive radio. Since solving the overall
optimization problem needs global knowledge of feasible rates and the worst-case complexity of this
centralized problem is exponential, it necessitates the need to design a distributed algorithm that is
scalable for practical implementation.
The resource allocation of the proposed algorithm consists of spectrum and power alloca-
tion. A spectrum opportunity for link (i, j) is a set of contiguous subbands where Oi j( f )≥ 0, when
Oi j( f ) is given by,
Oi j( f ) = Pmaxi ( f )−Pmin
i ( f ), (2.12)
where Pmaxi ( f ) is defined as the maximum power that can be used by the secondary node i on the
frequency f such that it satisfies the BER constraints of primary and secondary users. It is important
to note that Pmaxi ( f ) will be constrained by the maximum transmit power of the wireless radio used in
the network. On the other hand, Pmini ( f ) denotes the minimum power required to reach the required
SINRthSU at the intended secondary receiver. In other words, Pmin
i ( f ) and Pmaxi ( f ) provide the lower
bound and upper bound of transmit power respectively for node i on frequency f . The Pmini ( f ) and
Pmaxi ( f ) values are determined by a node i by gathering spectrum and resource allocation information
from its neighbors. This information is gathered using Collaborative Virtual Sensing (CVS) using the
control packets in the network. The details about resource allocation, CVS and the MAC protocol
employed as it is similar to that in ROSA.
Accordingly, the distributed DRS is proposed to maximize the throughput of a tactical ad
hoc network. In the distributed network, each node makes an adaptive decision to choose optimal
session, next hop, power allocation and spectrum to use during the next time slot based on the
information gathered from the neighbors using CVS. This decision will be different from traditional
ROSA [10] because the network utility defined here is a function of VQL and not the actual queue
lengths. Once a backlogged node senses an idle CCC, it performs the Algorithm 1 to obtain the
optimal resource allocation decision:
1. The proposed algorithm assumes that the location of the intended destination node is known
to the source node. This information is carried by the packet through the intermediate nodes.
Each node selects a feasible set of next hops for each backlogged session j ∈ (us1,u
s2, ...,u
sk),
which are neighbors with a positive advance towards the intended destination.
2. The maximum capacity for each node is calculated by considering all possible spectrum
opportunities. The maximum capacity of each feasible neighbor is used along with the
15
CHAPTER 2.
Algorithm 1 Deadline-based Resource Allocation1: t = 1, ∆ = ∞, Ci j = 0, U∗i j = 02: for si ∈ Si do3: for j ∈ u1,u2, ...uk do4: for fi ∈ [ fmin, ..., fmax−∆ fi ] do5: Calculate Pt
i ( f ) similar to [10]6: Calculate Ctemp as in (2.6)7: if Ctemp >Ci j then8: Ci j =Ctemp
9: [ f ∗i, j,P∗i,j]=[ fi,Pti]
10: end if11: end for12: U s
i j =Ci j ∗ [Qsii −Qsi
j ]13: if U s
i j >U∗i j then14: U∗i j =U s
i j
15: [ f opti ,Popt
i ,sopti , jopt]=[ f ∗i, j,P∗i,j,si, j]
16: end if17: end for18: end for19: Return [ f opt
i ,Popti ,sopt
i , jopt]
corresponding differential VQL to determine the network utility U si j. The optimal decision is
taken such that,
(sopt , jopt) = arg max(U si j). (2.13)
As seen earlier, the network utility function comprises of differential VQL and achievable
capacity. The differential VQL is a function of deadline and estimation of Td . Thus, the
sessions that have smaller deadlines or are further away from the intended destination will be
scheduled more often if the available spectrum for all sessions is comparable. The adaptive
routing will also provide most traffic to VQs that are lightly backlogged.
3. The optimal frequency and power allocation ( f opti ,Popt
i ) correspond to the values that provide
maximum Shannon capacity Ci j over the wireless link (i, jopt), where jopt is the best next hop.
Here, a contention-based MAC protocol is used in the control channel before transmitting
the packet on the selected data channel. In the contention-based MAC protocol, the probability of
accessing the medium is calculated based on the U∗i j. Nodes generate a backoff counter from the
range [0,2CW−1], where CW is the contention window. The CW is a decreasing function of U∗i j. This
will ensure that heavily backlogged VQs with more spectrum resources will have a higher probability
16
CHAPTER 2.
of transmission.
The computational complexity of the DRS algorithm at a node i is directly proportional
to the number of neighbors, number of channels and number of active sessions. Therefore, for a
constant number of channels and sessions in a network the computational complexity for node i is
given as O(|N B i|).
2.4 Performance Evaluation Through Simulation
In this section, the performance of DRS is compared with ROSA in a multihop tactical
ad hoc network. To evaluate DRS, an object-oriented packet-level discrete-event simulator is used,
which implements the features described in the earlier sections of this paper. The metric used for this
evaluation is effective throughput (η) and reliability (ρ) of the network. Effective throughput was
defined based on the number of packets received within the deadline. The reliability is defined as the
ratio of packets received at the destination within the specified deadline with respect to the number
of packets generated at the source node. The evaluation is conducted on a grid topology in a 6000 m
x 6000 m area. The sessions are initiated between disjoint random source-destination pairs and the
packet size of the packets are set at 2500 bytes and the number of packets transmitted per session is
set to 500. The total available spectrum (BW ) is set to be 54 MHz-72 MHz The bandwidth usable by
cognitive radios are restricted to be 2, 4 and 6 MHz. The bandwidth of the common control channel
is set as 2 MHz. Each result was obtained by averaging the values obtained from 50 random seeds
unless specified differently. In all figures except Fig. 2.6, the blue lines represent the performance of
DRS and red lines denote the performance of ROSA.
2.4.1 Scenario 1: Network performance as the number of session increases (All ses-sions started at random time)
In scenario 1, the network performance is evaluated as the number of active sessions in the
network increase. The parameters used during the two experiments for scenario 1 are listed in Table
2.1. The only difference between the two experiments are the deadlines assigned to different sessions.
In experiment 1, all the sessions have a deadline of 2s, which represents a highly constrained network.
Instead, in experiment 2 the odd-numbered sessions have a deadline of 1.5 s and even-numbered
sessions have a deadline of 10 s. Experiment 2 can be considered as a scenario where one session
carries periodic weather monitoring data through the network. These sessions are delay tolerant to an
17
CHAPTER 2.
Table 2.1: Parameters of scenario 1.
Parameter Experiment 1 Experiment 2
Source Rate 2 Mbits/s 2 Mbits/sSession duration 5 s 5 s
Session startrandomly fromt = [0,5] s
randomly fromt = [0,5] s
No. of sessions2, 4, 6, 8, 10, 12
14, 16, 18, 20, 222, 4, 6, 8, 10, 1214, 16, 18, 20, 22
Deadline ofeach session 2 s
Odd session 1.5 sEven session 10 s
Number of sessions2 4 6 8 10 12 14 16 18 20 22
Th
rou
gh
pu
t (M
bit
s/se
c)
0
5
10
15
20
25
ROSA (Exp:1)DRS (Exp:1)ROSA (Exp:2)DRS (Exp:2)
Figure 2.2: Scenario 1: η vs No. of sessions.
Number of sessions2 4 6 8 10 12 14 16 18 20 22
Rel
iab
ility
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
ROSA (Exp:1)DRS (Exp:1)ROSA (Exp:2)DRS (Exp:2)
Figure 2.3: Scenario 1: ρ vs No. of sessions.
extent, hence have a longer deadline. The second type of data can have an extremely small deadline,
consisting of delay-intolerant data like threat detection, incoming missile alert or real-time video
streaming. The proposed algorithm should be able to adapt to the varying requirements of different
sessions and maximize the effective throughput of the network. The session are set to start randomly
any time between start of the simulation (t = 0 s) and session duration (t = 5 s). This ensures that
all sessions are active at some point during the simulation but the number of active sessions will
vary throughout the simulation. The parameters of both the experiments (1 and 2) are listed in Table
2.1. Examining Fig. 2.2 and Fig. 2.3 show that DRS performs much better than ROSA in terms of
reliability and effective throughput in experiment 3 and 4. In these scenarios, traditional backpressure
based algorithm may suffer from the last packet problem. Since DRS is formulated based on VQL
which takes into account the deadlines of each packet in the queue, the penalty for holding packets in
the queue grows as time elapses eliminating the last packet problem.
18
CHAPTER 2.
Rate (Mbits/sec)1 2 3 4 5 6 7 8 9 10
Th
rou
gh
pu
t (M
bit
s/se
c)
0
5
10
15
20
25
30
ROSA (Exp:3)DRS (Exp:3)ROSA (Exp:4)DRS (Exp:4)ROSA (Exp:5)DRS (Exp:5)
Figure 2.4: Scenario 2: η vs No. of sessions.
Rate (Mbits/sec)1 2 3 4 5 6 7 8 9 10
Rel
iab
ility
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
ROSA (Exp:3)DRS (Exp:3)ROSA (Exp:4)DRS (Exp:4)ROSA (Exp:5)DRS (Exp:5)
Figure 2.5: Scenario 2: ρ vs No. of sessions.
2.4.2 Scenario 2: Network performance as the data rate of the sessions increase
Table 2.2: Parameters of scenario 2.
Parameter Exp: 3 Exp: 4 Exp: 5
Source Rate1 to 10Mbits/s
1 to 10Mbits/s
1 to 10Mbits/s
Session duration 5 s 5 s 5 sNo. of sessions 5 5 5Deadline ofeach session 2 s
Odd session 1.5 sEven session 10 s 2 s
In these set of experiments, the network performance is evaluated in a scenario where the
number of sessions are kept constant and the data rate injected at the source node increases from
1 Mbits/s to 10 Mbits/s. Three experiments are conducted in this scenario. Experiments 3 and 4
are similar to the experiments of previous scenario 1, varying only in deadline as shown in Table
2.2. Experiment 5 evaluates the performance of DRS when the packet size is larger (25000 bits).
Figures 2.4 and 2.5 shows that in all three experiments DRS outperforms ROSA. Though the effective
throughput of both ROSA and DRS increased with higher packet size (dotted lines), the reliability
of ROSA decreased more compared to the decrease in DRS. Hence, the difference in performance
between DRS and ROSA increased when larger packets (fewer number of packets per second) were
used in the network.
19
CHAPTER 2.
Table 2.3: Parameters of scenario 3.
Parameter Exp: 6 Exp: 7 Exp: 8
Source Rate 2 Mbits/s 2 Mbits/s 2 Mbits/sSession duration 5 s 5 s 5 s
No. of sessions 5 82, 4, 6, 8, 10,12, 14, 16
Deadlines 2 s 2 s 50 sNo. of seeds 50 50 20
2.4.3 Scenario 3: Examining the effect of different components of DRS
Here, the effect of different components used during the formulation of DRS is evaluated.
Experiment 6 studies how Tr and Tr−Td affect the proposed algorithm individually. Accordingly,
the simulation is run with the parameter shown under experiment 6 in Table 2.3 using two different
weight definitions as shown below,
wTr =L
max(Tr,τ)(2.14)
wTr−Td =L
max(Tr−Td ,τ)(2.15)
Figure 2.6 shows that both the DRS using weights seen in (2.14) and (2.15) perform
considerably better than ROSA but does not maximize the effective throughput like original DRS.
This is because original DRS that uses the weight shown in (2.1) derives the benefits of both weights
((2.14) and (2.15)). Hence, this shows why it is advantageous to have both Tr and Tr−Td in the
denominator of the weight used to calculate VQL. Further, it is interesting to note that in cases where
it is difficult to estimate Td , one can still achieve moderately good performance by using weight
shown in (2.14). Next, the effect of the parameter τ on the effective throughput of the network is
evaluated. Equation (2.1) uses a very small value τ to ensure the correctness of weight, such that
instances with infinite value do not occur.
Figure 2.7 depicts the effective throughput of the network as τ changes while keeping the
number of sessions and source data rates constant. The other parameters of experiment 7 are depicted
in Table 2.3. The result shows that the effective throughput of the network using DRS is consistently
high for values of τ over a range between 10−8 to 0.99. Any value greater than 10−1 takes away the
effect of deadlines and has a degrading effect on effective throughput. As the value of τ moves closer
to 1, the VQL becomes more and more equivalent to traditional queue length. On the other hand,
20
CHAPTER 2.
Number of sessions2 4 6 8 10 12 14 16
Th
rou
gh
pu
t (M
bit
s/se
c)
3
4
5
6
7
8
9
10
DRSDRS(w
Tr-T
d
)
DRS(wT
r
)
ROSA
Figure 2.6: Scenario 3: η vs No. of sessions.
= parameter10-12 10-10 10-8 10-6 10-4 10-2 100
Th
rou
gh
pu
t (M
bit
s/se
c)
0
1
2
3
4
5
6
7
8
9
10
11
DRS (8 Sessions)DRS (6 Sessions)
Figure 2.7: Scenario 3: η vs Parameter τ.
choosing τ to be smaller than 10−8 also affects the algorithm adversely since it bloats the VQL to
an extent where the capacity component of the network utility function becomes insignificant. This
lower bound would depend on the characteristics of the network, specifically, the achievable capacity
determined by the bandwidth of the transceiver. Figure 2.7 also shows that the range of values for τ
outside which the effective throughput of the network starts declining is same for both cases (6 and 8
sessions). This shows that the acceptable value for τ does not change according to the number of
active sessions in the network. Since it has been shown that DRS performs consistently well over a
large range of values of τ, one can choose any value within the acceptable range depending on the
network setup.
Next, the same parameters as in experiment 7 are used to analyze how errors in the
estimation of Td affect the network’s effective throughput. In this case, the number of sessions is set
to eight and ten. Parameter α is varied from R/0.5 to R/5 as shown in Fig. 2.8. When α = R/0.5,
Td is underestimated and when α = R/5, Td is overestimated. As expected, if Td is underestimated,
fewer packets are delivered to the destination within the deadline thereby decreasing the effective
throughput. Meanwhile, overestimation does not impact the throughput negatively because Td is
calculated at each hop and hence error/margin decreases as the packet moves closer to the destination.
Finally, in experiment 8, the performance of DRS is evaluated in a network having sessions
with a very long deadline (50 s). This experiment examines how the network behaves in scenarios
where the deadlines of the sessions are long enough such that packets lost due to the expiration of
deadline are negligible. This experiment evaluates whether there is any loss in throughput while using
DRS as compared to ROSA in the network that is delay tolerant (have extremely long deadlines).
21
CHAPTER 2.
R/0.5 R/1 R/1.5 R/2 R/2.5 R/3 R/3.5 R/4 R/4.5 R/50
1
2
3
4
5
6
7
8
9
10
11
Th
rou
gh
pu
t (M
bit
s/se
c)
DRS (10 Session)DRS (8 Session)
Figure 2.8: Scenario 3: η vs Parameter α.
Number of sessions2 4 6 8 10 12 14 16 18 20 22
Th
rou
gh
pu
t (M
bit
s/se
c)
2
4
6
8
10
12
14
16
18
ROSA (Exp:10)DRS (Exp:10)
Figure 2.9: Scenario 3: η vs No. of sessions(Deadline=50 s).
Figure 2.9 shows that the throughput of both algorithms are equal as the number of sessions in
the network increase. This shows that there is no disadvantage in using DRS over ROSA even in
scenarios where deadlines are insignificant.
2.5 Testbed Implementation
2.5.1 Challenges
In the past decade, cross-layer protocols have been extensively studied and various solutions
have been proposed [39, 40, 41, 42]. Even with these advances in literature, most of the solutions
are evaluated only using simulation tools like MATLAB, ns-3, OPNET among others. The goal
of cross-layer optimization techniques is to utilize the information between different layers and
enable their interaction to jointly optimize objectives including throughput, reliability, delay among
others. This invokes the need for an architecture that enables these interactions and promotes the
design and development of cross-layer optimization algorithms. Lack of such platforms has led to the
growing gap between the number of solutions proposed in literature versus the number of solutions
that are implemented and tested using actual hardware. There are only limited efforts that extend the
implementation of the optimization algorithms to actual hardware and evaluate the performance on a
cross-layer testbed [43, 21, 44, 45, 46, 47]. The major challenge in achieving this implementation is
the lack of a flexible architecture that facilitates the implementation of the cross-layer optimization
algorithm on multi-node networks. This deficiency is being recognized by the community and some
22
CHAPTER 2.
solutions are being proposed to achieve the required flexibility.
GNU Radio is an open-source signal processing software that provides great flexibility
specifically at the physical layer of SDRs. GNU radio comprises of various signal processing and
digital communication blocks and is an excellent tool to control SDR. However, the majority of
the contribution is limited only to the Physical layer. There have also been efforts to relocate some
of the processing functions to a Field-programmable gate array (FPGA)[48] to improve the delay
performance. This makes it difficult to integrate new algorithms for testing and evaluation purpose.
Some other work [49, 50] aims to provide reconfigurable MAC protocols by decomposing the overall
design into core fundamental blocks. In [49], the implementation of these fundamental blocks is
split between PC and FPGA depending on the time-critical nature of the blocks. In [50], the authors
implement an abstract execution machine on a resource-constrained commodity WLAN (wireless
local area network) card. Recently, software-defined network (SDN) using an Open-Flow [51] based
approach has been proposed for evaluating routing protocols. The overall concept of Open-Flow is to
keep the data path on the Open-Flow switch itself while moving the high-level routing decision to
a separate controller (server). The switch performs the packet forwarding based on the flow table
defined by the controller and uses Open-Flow protocol to communicate with each other. The majority
of the work on OpenFlow has been concentrated at the network layer of the protocol stack.
Even with these advancements, a major challenge to transitioning algorithms and protocols
to commercial hardware is the lack of a software-defined testbed with a flexible architecture that
enables easy implementation of cross-layer technologies. These testbeds are essential to corroborate
the results obtained in simulations and evaluate how to refine these algorithms to ensure a successful
transition to relevant hardware. Some of the requirements of such a testbed include a flexible cross-
layer based protocol stack [46], modularity to integrate new algorithms with ease, a framework to
accommodate both centralized and distributed solutions, real-time network performance monitoring
tools, and having the ability to run unsupervised scripted experiments over extended periods of time.
Having such a testbed expedites the design and development process of next-generation wireless
communication technologies destined for a commercial SDR system. The CrOss-layer Based testbed
with Analysis Tool (COmBAT) first introduced in [47] was developed to serve as a software defined
testbed to enable the implementation of cross-layer optimization algorithms. In COmBAT, a Adaptive
Cross-Layer (AXL) communication framework facilitates easy integration of new protocols and
algorithms. The design of AXL is discussed in detail in the next section.
23
CHAPTER 2.
Control Link N1
N3N2
N5
0,2 2,2
1,1
0,0 2,0
MIMO Cable 1,2
1,0
2,10,1
N4
Data link
Ethernet connection
Figure 2.10: Layout of the five-node grid topology.
Register Plane
Data management/storage
Control PlaneMedium access controller
Application Layer
Session ManagerPacket management
Decision PlaneAlgorithms/Decisions
Wireless Data channel & Control channel/ Wired link for NEAT
Physical Layer
ROSA RFA RDA
S1 S2
Thresh
CSMA/CA TDMA FDMA
TEXT Voice
Single Packet or Streaming
Video
Figure 2.11: Adaptive cross-layer(AXL) framework.
2.5.2 Adaptive Cross-Layer (AXL)
The overall AXL framework depicted in Fig. 2.11 consists mainly of the application layer,
session manager, decision plane, control plane, register plane, and the physical layer. Each node in
the network uses the AXL framework in place of a traditional protocol stack. In the implementation,
the AXL framework consists of Python multiprocessing processes which are initialized at node start
up using an AXL daemon. The daemon imports the main modules and properties that are used in the
different layers/planes of the framework as required. The properties include predefined values for
the network such as data timeout duration used by MAC protocol, node IP and MAC addresses and
payload sizes. However, most of the properties are dynamic in nature and they can be reconfigured
on-the-fly based on network optimization strategies or user input. The processes that are started by
the daemon run continuously until shutdown. These processes include the register plane, session
manager, control plane and the physical layer. The decision plane is not a process but a collection of
functions that can be called by the framework when needed. Each plane/layer can share information
with each other by a combination of three methods; direct function calls, shared memory (register
plane) or by overhearing global events (global with respect to the framework, not the entire network)
that can be triggered by any process in the framework. These functionalities allow for a flexible
cross-layer communication between all network protocols.
Application (APP) Layer. The current AXL software package provides data generation
APPs that a user may choose from in order to evaluate the performance of the network. The APPs can
operate in packet streaming mode or packet-by-packet mode. For streaming mode, the source data is
repeated until a user-specified amount of data has been generated. The streaming mode is generally
24
CHAPTER 2.
used in experiments requiring a constant bit rate (CBR) source for a fixed duration of time. The APPs
connect to the AXL daemon via a TCP/IP socket. For each APP that connects, a unique connection
object is created that manages data transfer between the APP and the AXL framework. Each packet
contains the user-generated and QoS parameters. This is where the deadline of the packet can be
defined. The packet is parsed and then sent to the session manager for the next processing stage.
Session Manager. In the AXL framework, the session manager provides the capability of
simultaneous multi-session management. When a packet arrives at the session manager, the session
manager creates a session object based on the packet parameters, which include process ID that is
created by the OS, source and destination IP, data type, any QoS parameters (such as deadline), and
the packet number generated at packet creation. Packets that correspond to existing session objects
are appended to their appropriate session queue. Packets receive their network headers based on
the parameters in the session object mentioned earlier. In the implementation, the session manager
is designed as a multiprocessing first-in, first-out (FIFO) with a user-specified update period. The
update period dictates the timeout for updating packet queues. During each update period, the session
manager stores the current queue length of each session in the register plane and triggers an event
flag which indicates that the transmitter has backlogged session ready for routing decisions.
Decision Plane. As the name suggests, this is the component where all the logical
decision making and algorithm execution takes place. These algorithms pertain to routing algorithms,
spectrum allocation, automatic modulation classification, and other resource allocation decisions.
The complexity of the algorithms can vary from threshold decision to iterative algorithms like
the Expectation and Maximization (EM) algorithms or solvers for convex optimization problems.
Decision algorithms can (i) modify a parameter in a protocol, (ii) trigger switching among different
modes within a protocol, (iii) enable switching among different protocols altogether [46].
In regards to this work, there are currently two routing algorithms (ROSA and DRS), stored
as software modules, available for use (discussed in detail in Section 2.6). This is where all the
calculation for the routing algorithm takes place. Each routing module can be passed as an instance
for the decision plane to use during runtime. The chosen and active routing module is requested
by the session manager to execute the algorithm when a session is backlogged. The results of the
executed algorithm are stored in the register plane for other layers/planes to access. Conversely,
to execute the algorithm, the decision plane obtains the information from the register plane. After
executing the algorithm within the routing module, the decision plane triggers an event flag that
prompts the control plane to schedule a transmission. It should be noted that the interactions between
the decision plane and the other layers/planes in AXL take place via the register plane, through
25
CHAPTER 2.
global events or through direct function calls.
Some DRS specific requirements had to be included in the experimental framework to
ensure successful implementation. Each packet is required to carry the time of generation in its
header. This enables nodes to calculate VQL at each hop. This feature had to be included in the
experimental framework to enable the operation of such a deadline-based cross-layer algorithm.
The ability to track the time instant of generation and arrival of packets at each node including
the destination has been implemented. In a traditional network, the backlog does not change with
time unless a packet is received or transmitted by the node. In the case of DRS, the VQL changes
continuously with time. This was a crucial component necessary to implement DRS-like protocols
that aim to eliminate the last packet problem. Therefore, in this current version of the framework, the
VQL is constantly updated at the decision plane of each node. All of these inclusions have bolstered
the experimental environment/framework to handle algorithms that are required to handle time-based
queue lengths.
Control Plane. The control plane houses the control logic used to access the wireless
medium. The control plane contains the Finite State Machine (FSM) used to implement different
MAC protocols. The chosen MAC protocol defines the exact set of states, events, conditions and
actions required to operate FSM. The control plane can be initialized to use multiple different MAC
protocols depending on the situational awareness gathered from other layers/planes of the stack
as shown in [46]. Each MAC protocol should have its FSM implemented in the control plane as
a separate FSM initialization function. Future developers can take advantage of the baseline FSM
model that is already defined in the control plane by modifying its states and actions as needed by the
protocol. An example of a state transition diagram for a carrier sensing multiple access with collision
avoidance (CSMA/CA) based MAC protocol [10] is given in Fig. 2.12. It is important to point out
that CSMA/CA-based MAC protocol operates only on the CCC and does not restrict concurrently
feasible (in terms of BER constraints) transmission from occurring on the data channel. The state
transition diagram describes the interaction between all possible states, events, and actions for the
receive and transmit paths. As shown in Fig. 2.12, when an event Data_available is set, Send_RT S
(request-to-send) action is taken as the FSM goes from an IDLE state to WAIT CT S (clear-to-send)
state. The next event that the FSM is looking for is either CT S_received or CT S_timeout and the
FSM transitions depending on which event was observed. If CT S was received and CT S_received is
set, Send_DT S (Data Transmission reServation) action is taken as the FSM goes from WAIT CT S
state to SEND DATA state. The rest of the state transition diagram can be interpreted in a similar
manner.
26
CHAPTER 2.
Receive Path
Partial_data or Data_timeout < TH,Send_ACK, clear_state
SENDDATA
WAITACK
IDLE
Data_sent,Clear_state
Data_timeout,Clear_state
Ack_timeout >= Th,Clear_state
WAITCTS
Ack_received or Ack_timeout < Th,Clear_state
Ack_received,Clear_state
Data_available,Send_RTS
CTS_received,Send_DTS
CTS_timeout,Clear_state
WAITDATA
RTS_received,Send_CTS
Data_received,Send_ACK, clear_state
Data_timeout >= TH,Clear_state
Event,Action
Transmit Path
Figure 2.12: FSM of MAC protocol.
The FSM is generally in an IDLE state until the corresponding global AXL events are
flagged to invoke a state transitioning process. These global AXL events are used in cross-layer
communication between different layers/planes and should not be confused with the events used by
the FSM itself. The events in the FSM are strictly defined by the chosen MAC protocol and dictate
the state transitioning process that allows the control plane to manage medium access. The global
events such as SESSION_ROUT ING that transition the FSM from an IDLE state are usually set
in the decision plane after routing decision has been made. Some other examples of such global
events used throughout AXL include SESSION_PROCESSING event which is used by the session
manager to indicate that the node is busy processing a session and START _SENSE event that signals
the PHY layer that it is time to perform spectrum sensing. Therefore, overall it can be stated that the
overall AXL framework follows an event-driven design.
Register Plane. The register plane is essentially a node database used to share information
across layers/planes. Although the register plane does not perform any computation and does not
have any decision-making ability, it is an integral part of the overall cross-layer design. The register
plane can be considered as a central information hub that can be accessed by different layers/planes
of the AXL framework. Data sharing among multiple processes is achieved through Python manager
dictionaries. The global information that needs to be shared among all layers is stored in a manager
dictionary which allows for only one process to read or write information in the register plane at a
time. The main dictionaries that reside in the register plane are a global register dictionary (GRD),
27
CHAPTER 2.
global values dictionary (GVD) and session backlog dictionaries (SBD).
Nodes learn about their environment by overhearing control packets on the CCC. Each
node stores local information in a node dictionary in the GRD. The node dictionary is appended
to every control packet sent on the CCC. The information in the node dictionary is continuously
updated as new information becomes available. Node dictionary information includes IP and MAC
addresses, the node location, local noise plus interference, session packet queue lengths, current
routing algorithm among others. Nodes maintain a copy of their own node dictionary, as well
as a copy of its neighbor’s dictionary in the GRD. The GRD also contains information like the
designated frequencies, possible next hops, and neighbors. The GVD stores the current routing
decision parameters as well as the current state of the FSM. SBD has a list of all local sessions and
their most up to date packet queue lengths. The routing algorithm is able to access this information
stored in the register plane as it optimizes the routing parameters. Other layers/planes can similarly
read or write information in the register plane as needed.
Physical (PHY) Layer. The PHY layer is easily separable from the rest of the framework
as the goal is to allow the integration of different radio front ends and signal processing software.
The PHY layer consists of a hierarchical implementation where the lowest level includes signal
processing software specific libraries such as GNU Radio and a Universal Hardware Driver (UHD)
interface used with the Universal Software Radio Peripheral (USRP) family of products from Ettus.
The PHY hierarchical module consists of functions that are directly accessible by the control layer
and the register plane. This implementation allows for a very simple interface between AXL and a
PHY layer making this design SDR hardware agnostic.
Figure. 2.10 depicts the current configuration of a five-node network that is arranged in the
form of a grid topology. Each node consists of two USRP N210s (one for control link and the other
used as data link) connected to each other via MIMO cable. The SBX and CBX daughterboards
are used with the radios, which cover frequency ranges from 400 MHz to 4.4 GHz and 1.2 GHz to
6 GHz respectively. The receivers (USRP N210s), provide up to 40 MHz of instantaneous analog
bandwidth. The analog-to-digital and digital-to-analog converters on the motherboard use a 100 MHz
master clock and sample at 100 MS/s and 400 MS/s respectively. The on-board Xilinx Spartan
3A-DSP 3400 FPGA performs the required digital interpolation or decimation to provide the required
sampling rate. The host PC interfaces with the USRP using a Gigabit Ethernet (GigE) connection as
shown in Fig. 2.10. The USRPs are controlled using GNU Radio signal processing modules and a
UHD interface. DRS is implemented on the AXL framework and the results are discussed in Section
2.6.
28
CHAPTER 2.
2.6 Testbed Evaluation
N2N1N3
N4N5
Figure 2.13: Five-node USRP testbed.
This section discusses the experimental evaluation of DRS using the AXL framework
implemented on five-nodes USRP testbed shown in Fig. 2.13. In these set of tests, both DRS and
ROSA use the same MAC protocol with FSM as shown in Fig. 2.12. The PHY layer uses Gaussian
minimum shift keying (GMSK) implemented using GNU radio with USRPs as the transceivers of
the AXL node. The USRP is able to operate at frequency, bandwidth and transmit power level as
specified by algorithms (DRS or ROSA) running in the decision plane. This framework provides the
required flexibility to implement and evaluate the two cross-layer optimization algorithm (ROSA
and DRS). For rapid implementation and feasibility analysis, the AXL framework and the routing
algorithms are implemented using Python programming language. The advantage of using Python
is the ease of programming and faster development turnaround time. The drawback is large delays
incurred by the framework [52, 53]. Therefore, it is important to obtain a baseline for the delay
experienced in the network so that one can choose appropriate deadlines for the experimentation
process.
2.6.1 Establishing Baseline
In this section, a baseline is established for the average end-to-end delay (EED) experienced
by the SDR based testbed which is the intended platform for evaluating DRS. To accomplish this,
ROSA is used as the default routing algorithm. Accordingly, the average delay experienced by
packets to traverse from source to destination is calculated as the number of session increases. The
29
CHAPTER 2.
Table 2.4: Parameters to baseline end-to-end delay.
Parameters Baseline Test
Packet size 1250 BytesNumber of packets 3000Number of sessions 1 to 6Source rate 200 kbits/sSource duration 150 sRouting algorithm ROSANo. of seeds 5Maximum transmit power 20 dBm
network parameters used for determining the EED experienced by the current configuration of the
SDR based ad hoc network is listed in Table. 2.4.
Number of sessions1 2 3 4 5 6
En
d-t
o-E
nd
del
ay (
min
ute
s)
0
2
4
6
8
10
12
Average packet delayAverage session delay
Figure 2.14: End-to-End Delay vs Number of Session.
Figure 2.14 depicts the average EED experienced by each packet and the average EED
experienced by the entire session as the number of sessions in the network increase. The EED
experienced by the packet is calculated as the duration between the packet generation at the source
node and packet arrival at the destination node. Similarly, EED experienced by each session represents
the time between the generation of the first packet at the source node and the arrival of the last packet
at the destination node. As expected, the average delay increases in both cases as the number of
sessions increases. Examining these delay values, the impact of the Python-based implementation
of the overall framework can be clearly seen. Nevertheless, this baseline can be used to choose
appropriate deadlines for this network that would enable the performance comparison of the two
30
CHAPTER 2.
cross-layered algorithms (DRS and ROSA). Accordingly, for the next set of tests, 3 min is chosen as
the smaller stringent deadline and 15 min as the larger deadline.
2.6.2 DRS and ROSA
In this section, the effective throughput and reliability of DRS and ROSA are evaluated
on the five-node USRP testbed. In the current implementation, the MAC protocol is able to recover
any loss of packet that occurs due to the channel using retransmission. Therefore, loss of packets
only takes place at the destination when the packets reach after the specified deadlines. In addition to
parameters listed in Table. 2.4, this set of experiments use the parameters in Table 2.5.
Table 2.5: Parameters for testbed evaluation
Param. Test 1 Test 2 Test 3
DeadlinesOdd sess. 3 minEven sess. 15 min
Odd sess. 3 minEven sess. 15 min
60 min
Sessionstart
t = 0s t = [0,2] min t = 0s
τ 10−6 10−6 10−6
Seeds 30 30 30
In the first set of tests, multiple sessions that started at the same time are used to evaluate
the performance of the SDR based network as the number of sessions increased. It is evident from
Fig. 2.15 that for the given network configurations DRS outperforms ROSA in terms of effective
throughput as soon as there are two sessions in the network. The performance trend continues as the
number of session increases achieving up to 17% improvement over ROSA. This is because DRS is
able to manage multiple sessions adaptively to ensure that the effective throughput of the network is
maximized. Similar behavior is also observed in Fig. 2.16 which compares the reliability of DRS
and ROSA as the number of packet increases.
In contrast to the first test, the source nodes are set to choose a random time to start the
session in the second test. This would imply that different sessions will end at different times leading
to the last packet problem in networks using traditional backpressure based algorithm like ROSA.
The effective throughput and reliability of Test 2 are depicted in Fig. 2.17 and Fig. 2.18. As expected
DRS outperforms ROSA in terms of effective throughput (up to 12% improvement) and reliability
(up to 13% improvement) even when the network has finite sessions starting at different times. This
is due to the fact that the use of VQL prevents the network from experiencing the last packet problem.
VQL keeps increasing with time even if the actual queue length does not change.
31
CHAPTER 2.
Number of session1 2 3 4 5 6
Th
rou
gh
pu
t (K
bit
s/s)
160
180
200
220
240
260
280
300
320
ROSADRS
Figure 2.15: Test 1: η vs No. of sessions.
Number of sessions1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
Rel
iab
ility
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
ROSADRS
Figure 2.16: Test 1: ρ vs No. of sessions.
Number of sessions1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
Th
rou
gh
pu
t (K
bit
s/s)
160
180
200
220
240
260
280
300
320
340
360
ROSADRS
Figure 2.17: Test 1: η vs No. of sessions.
Number of sessions1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
Rel
iab
ility
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
ROSADRS
Figure 2.18: Test 1: ρ vs No. of sessions.
In the final test, a very large deadline (60 min) is used for evaluation. The goal was to
evaluate if there is any degradation in performance of DRS compared to ROSA when the deadlines
are large enough to be close to negligible. As shown in Fig. 2.19, there is no significant loss in
performance on using DRS compared to ROSA even in scenarios where the deadlines are long
enough to be insignificant. Overall, these tests follow the same trend as simulations discussed in
Section 2.4. The gains observed with experiments is much smaller than with simulations due to the
smaller network size. Based on the results seen in simulation, it can be predicted that a larger benefit
can be attained on a bigger network deployment. These set of tests provide validity to the proposed
algorithm to be effective in cognitive network that provides the flexibility to adapt according to the
given scenarios.
32
CHAPTER 2.
Number of sessions1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
Th
rou
gh
pu
t (K
bit
s/s)
0
50
100
150
200
250
300
350
400
ROSADRS
Figure 2.19: Test 3: η vs No. of sessions.
2.7 Summary
A novel distributed deadline-based joint routing and spectrum allocation algorithm is
proposed to maximize the effective network throughput. The DRS adapts according to available
resources and is capable of handling sessions with heterogeneous deadline requirements. DRS enables
every node in the network to choose the optimal session, next hop, frequency and transmit power
with an objective to deliver the maximum number of packets to their intended destination before
the specified deadline. Though DRS is designed for tactical ad hoc networks, its application can be
extended to any wireless 5G ad hoc network that handles sessions with heterogeneous QoS based
deadline requirements. Extensive simulations are performed to compare the performance of DRS
with ROSA and showed up to 35 % improvement in effective throughput and up to 26 % improvement
in reliability of the network. Furthermore, the challenges of implementing the proposed algorithm
on a cross-layer framework based software-defined testbed have been successfully overcome. The
experiments conducted on the testbed showed DRS outperforming ROSA in terms of effective
throughput (up to 17%) and reliability (up to 13%). This helped to accomplish the secondary objective
of this chapter which was to validate the utility of the COmBAT framework by implementing novel
cross-layer DRS algorithm.
33
Chapter 3
HELPER: Heterogeneous Efficient Low
Power Radio for Enabling Ad Hoc
Emergency Public Safety Networks
In the past years, several lives have been devastated by hurricanes, tsunami, floods, earth-
quakes, and other natural disasters. Similar natural and man-made disasters are undesirable but
sometimes unavoidable [3, 4]. Even if the disasters may vary in intensity, nature, and duration of
occurrences some of the challenges faced during this period are similar. In these scenarios, one of
the most critical infrastructures affected is often communication networks [5, 6]. Today’s world is
heavily reliant on wireless communication. This is evident from the fact that 99% of the population
is covered by at least 3G network in the United States [54]. Similarly, authorities like Emergency
Response Center (ERC) setup by agencies like Federal Emergency Management Agency (FEMA)
are heavily reliant on wireless communication for information gathering, command, and control.
There are also several disaster alert system [55] that relies on wireless communication to relay the
message. For example, Grillo sensor network (Mexico) is a network of seismic sensors that will
sense and alert local users about the seismic activity. MyShake (U.S) is a mobile application based
solution which leverages the accelerometers of smartphones to detect seismic vibrations and sent
information for analysis to the Berkeley Seismological Laboratory for a final check before alerting
the user. Citizen Flood Detection (U.K) network is based on sensors installed under water bridges
to keep a tab on the water levels using echolocation and update the flood maps while alerting the
connected users over the Internet. Clearly, wireless communication is an essential component to
34
CHAPTER 3.
maintain connectivity for such alert systems, ERs as well as affected individuals when traditional
infrastructures like cell towers are affected or unavailable.
Several steps have been taken to enable wireless communication between ERs in such
situations [56, 57, 58, 59, 60, 61] with an objective to improve interoperability, reliability, and
accessibility. In comparison, there are few solutions designed to connect the affected survivors to the
ERs and the ERCs [62, 63, 64, 65, 66]. Further limited are the solutions that have been implemented
and prototyped to establish feasibility [64, 66, 62]. This aspect of emergency communication is
critical to enable rapid assistance, recovery and ensure the safety of the people in the affected area.
Realizing this gap, this chapter focuses on designing a cross-layer optimized solution (hardware and
software) that can be deployed by civilians (in their households) and ERs (roadside or other locations)
to establish an infrastructure-less network that enables communication between End Users (EUs),
ERs, and ERC during the aftermath of a disaster.
There are several challenges and requirements that have to be considered to enable such
technology that can be accessed by everyone in an emergency scenario. Since there is a high
probability that pre-existing infrastructure like base stations, cables etc. may be partially or completely
damaged during the disaster, the solution proposed for emergency communication must be self-
sustained. The solution should be readily accessible to EU such that there is no learning time or
contingencies for them to be connected to the network. In other words, it should be as simple as
people walking into an airport terminal and connecting to a Wireless Fidelity (WiFi) Internet network
within seconds. There might also be a shortage or absence of electricity during this period leading to
the demand for an energy-efficient solution. Another critical aspect will be the ease of deployment
and cost associated with the technology and the coverage it provides. Since the topology of the
target area may vary from tens to thousands of km2 based on the magnitude of the disaster, the
network must be designed in a distributed manner to ensure scalability. Due to the ad hoc nature
of the network, there could arise network holes which may isolate parts of the network. One way
to mitigate this problem is by deploying dense networks where density is defined as the average
number of neighbors for each node in the network. This can be accomplished by using a physical
layer solution that provides extremely long-range links while maintaining energy efficiency. Finally,
an ideal solution should be portable, low cost and energy-efficient such that large networks can be
deployed and sustained within a short period of time.
In this chapter, a Heterogeneous Efficient Low Power Radio (HELPER) ad hoc network is
developed for enabling emergency wireless communication as shown in Fig. 3.1. The proposed end-
to-end ad hoc networking solution and supporting software is capable of establishing an independent,
35
CHAPTER 3.
low cost, lower power wireless network for off-the-grid users during the aftermath of a disaster. One
of the objectives of this work is to restrict the cost of the proposed device as much as possible such that
each household can have one in their emergency kit and easily set it up when other communication
infrastructures are disrupted. These HELPERs will form a wireless ad hoc network connecting users
among themselves and to a ERC. The goal is not to provide a network with the highest throughput
or minimize delay rather maximize sustained connectivity through energy-efficient operation and
provide key services. These services will include text and voice messages within the neighborhood,
the ability to share resource information (water, food, gas, electricity, and internet) and the ability to
send distress messages to the ERC. On the other hand, the ad hoc network will also be used by the
ERC to remotely monitor the connectivity of the affected area and send alerts regarding imminent
dangers to the connected EU.
Local WiFi network established by HELPER
Aerial Mobile HELPER
First Responder StationFirst Responder Carrying mobile HELPER
Static HELPER in homes and streets
Figure 3.1: HELPER development prototype.
3.1 Related Works
The need for a robust communication system during the recovery period after a disaster is
evident from the previous discussions. Accordingly, several wireless communication technologies
have been developed for public safety [57, 67]. TETRA [59] is a telecommunication standard for the
private mobile digital radio system that provides an interoperability standard for equipment from
multiple vendors. The services provided include voice calls (individual call, group call, broadcast
call) with a data rates from 2.4 kbit/s to 28 kbit/s. TETRA Release 2, known as TETRA Enhanced
Data Service (TEDS) [61] has been deployed in United Kingdom [60] consisting of 3000 base
stations providing national coverage. Similarly, in the United States, Project 25 (or APCO 25) is a
36
CHAPTER 3.
standard setup for public safety communication by Telecommunication Industry Association (TIA)
that ensures interoperability, spectral efficiency and is gaining acceptance worldwide for public
safety application [58]. The protocol supports encrypted communication with a range of few km
with data rates up to a maximum of 9.6 kbit/s. Additionally, there are several instances where
commercial cellular wireless communication systems have been used as an emergency network. For
example, Federal Communications Commission (FCC) in a white paper [68] and the authors of [69]
recommends an approach for public safety broadband communications that leverage the advantages
of Long-Term Evolution (LTE) technologies. All these approaches mentioned above require fixed
infrastructure, which can be significantly degraded or destroyed during the disaster rendering these
services infeasible. An alternative solution that does not require pre-existing on-ground infrastructure
is the use of satellite networks. The satellite network can provide access to mobile or fixed terminals
using various frequency bands including C-Band and Ku Band. While the fixed terminal can achieve
up to 1.5 Mbits/s data rate, the mobile terminals can achieve 256 kbit/s [57]. Another approach to
sustaining communication when on-ground infrastructure is damaged is airborne communication
using avionic communication through helicopters. The traditional avionic communication is in the
Very High Frequency (VHF) band and can be used in three main configurations [70], (i) the system
deployed as an aircraft repeaters, (ii) a base transceiver station on an aircraft or (iii) a complete
system on an aircraft. Avionic communication is not cost-effective but is generally used after a large
natural disaster in a rural area that does not currently have any alternative. Overall, these solutions
are designed for ERs and cannot be cost-effectively extended to enable communication between EUs
and ERs during an emergency. During the aftermath of the earthquake in Haiti, connectivity was
enabled for 100 holding centers for displaced people using Worldwide Interoperability for Microwave
Access (WiMAX) and WiFi [71]. WiMAX was also used during the 2004 tsunami in Indonesia and
after hurricane Katrina in the Gulf Coast in 2005 [57]. This requires setting up a centralized WiMAX
system to provide connectivity to EU. Any such centralized network will limit the scalability of the
network in cases where the affected area is large.
As discussed earlier, it is essential to overcome the reliance on infrastructure and hence
ad hoc networks have been identified as a preferable solution for such scalable networks [72, 73,
62, 74, 65, 75, 76, 77, 66, 63]. The great east Japan earthquake and tsunami motivated authors of
[62] to develop D2D communication capabilities between smartphones to send emergency messages
in areas with affected infrastructure. Accordingly, the authors develop a prototype and conduct
an experimental evaluation in Sendai city which was one among the affected areas. They used
Optimized Link State Routing (OLSR) and epidemic routing protocols to achieve communication
37
CHAPTER 3.
between source and destination. In [78], the authors propose a location-aware wireless mesh network
to assist ER in providing medical support. Zigbee technology that operates in 2.4 GHz band was used
as the physical layer. An emergency and disaster relief system called Critical and Rescue Operations
using Wearable Wireless sensors networks (CROW2) is proposed in [74]. The authors propose
an end-to-end system that employs an Optimized Routing Approach for Critical and Emergency
Networks (ORACE-Net) routing protocol. ORACE-Net accesses every end-to-end link with regards
to its quality (end-to-end link quality estimation) to perform multi-path routing. CROW2 is designed
specifically to offload data from the disaster area to ERC but does not provide any provisions for
the EU to use the network to gather information for themselves. The authors of [65] propose to
use smartphones to form an ad hoc network using a reliable routing mechanism. Reliable Routing
Technique (RRT) [65] uses a broadcast-based routing technique to improve reliability. Broadcasting
every message to determine routes may lead to excessive and non-uniform energy consumption
leading to some devices being excessively drained of energy causing network holes. In contrast,
WIreless DEployable Network System (WIDENS) [75] is a European Project aimed to set up rapidly
deployable emergency services. The system architecture uses a cross-layer interface to provide
enhanced MAC and physical layer interaction. It uses OLSR protocol at the network layer. OLSR is
also used by [76] that aims to provide an efficient broadcast algorithm to reduce network overhead
induced by the control packets. The proposed prototype in [76] uses a hybrid of satellite and WiFi
connectivity to connect the ERC to the affected sites. In addition to considering energy-efficient
routing, the choice of the physical layer will also be essential in determining the network lifetime
and the operational feasibility in energy-constrained scenarios. Therefore, the use of WiFi can be an
ideal choice to connect to the EU but may not be the ideal choice to form ad hoc network that might
have to span over several hundreds of km2. The choice will always be a trade-off between energy
consumption, range and data rate and hence should be made considering the requirement at hand.
Next, some of the emergency ad hoc networking solutions that aim to achieve the required
energy efficiency are discussed. In [77], the author emphasizes on the importance of energy-efficient
operation in emergency conditions and thereby designs Minimum Power Routing (MPR) protocol
that chooses routes that require minimum power using Bellman-Ford algorithm which adapts to
the changing channel conditions (noise and interference). While this may provide optimal energy
efficiency in terms of energy consumed per bit delivered, this may not be the optimal routing strategy
for maximizing the entire network’s lifetime. In other words, this may lead to some nodes being
over-utilized for routing packets in the network depleting these nodes of energy causing network
holes. The authors of [66] propose TeamPhone that uses smartphones to form ad hoc networks
38
CHAPTER 3.
using WiFi. TeamPhone uses opportunistic routing or Ad hoc On-Demand Distance Vector (AODV)
for routing and propose to employ grouping technique along with a wake-up schedule to conserve
energy. This sleeping technique can be adopted by emergency ad hoc network but cannot substitute
an energy-efficient distributed routing algorithm. An interesting framework is proposed in [63]
that enables nodes to harvest energy from an undamaged base station (source) and then act as a
relay to carry the information to an area that does not have direct access to the source. The authors
propose an optimal communication route for networks during an emergency to minimize end-to-end
disconnection and reduce energy consumption while introducing the concept of RF-based energy
harvesting. Clustering is an ideal choice for their framework as the energy harvesting and coordinated
operation is assumed between nodes but in deployments where energy harvesting is infeasible,
clustering may lead to uneven consumption of energy or require frequency re-clustering procedure
that will eventually lead to larger overhead. The above discussion has been summarized in Table 3.1.
Therefore, in this chapter, an emergency ad hoc networking solution, HELPER Network
is developed with an objective to connect EUs of an affected community to each other and the
responding authorities. The significant contributions are summarized below,
• An end-to-end solution is proposed that includes Website Application (Web App) that connects
EU’s mobile devices to the HELPER using WiFi links. The HELPERs then form an energy-
efficient ad hoc network using low power, long-range LoRa links to connect all EUs to each
other and ERC.
• The proposed capabilities include resource information sharing, emergency distress messages
to ERC, imminent danger alert from ERC to all connected EUs and the ability to send text and
voice messages between EUs.
• To accomplish this, a cross-layer protocol stack is designed, implemented and used by each
HELPER to perform optimized energy-aware routing by using information acquired from
different layers.
• Finally, a portable, cost-effective and energy-efficient solution is prototyped to conduct proof-
of-concept demonstration. The six-node network is used to conduct extensive numerical
evaluations of the proposed routing algorithm.
39
CHAPTER 3.
Table 3.1: Summary of technology.
System Users Standard/Band
DesignFocus
Operation/Routing
HardwareEval
TETRA [59] ERTETRA
Release 1 Interoperability Centralized Yes
TEDS [61] ERTETRA
Release 2 Interoperability Centralized Yes
APCO 25 [58] ER Project 25 Interoperability Centralized Yes
Satellite [57] ER C & KuRemote
Connectivity Centralized Yes
Avionic [70] ER VHFRemote
Connectivity Centralized Yes
Haiti [71] EUWiMAX and
WiFi Connectivity Centralized Yes
BRCK [64] EU3G or LTE
WiFi Connectivity Centralized Yes
Gomez et al [69] Both LTEDistributed
LTECentralizedand D2D No
Nishiyama et al [62] EU WiFiSmartphone
RelayOLSR andEpidemic Yes
RRT [65] BothNot
Specified Reliability RRT No
TeamPhone [66] Both WiFiEnergy
EfficiencyAODV and
Opportunistic Yes
WIDENS [75] EREnhanced
802.11Rapid
Deployment OLSR No
Kanchansut et al [76] ERWiFi andSatellite
EfficientBroadcast OLSR Yes
Chandra et al [78] ER ZigbeeEnergy
EfficiencyZigbeeMesh Yes
Ali et al [63] EU LTEEnergy
Harvesting D2D No
3.2 Concept of Operation
3.2.1 Types of HELPERs
As shown in Fig. 3.1, three types of HELPERs are envisioned in the proposed network.
These three HELPERs have the same capabilities in terms of wireless communication and networking
but differ in the context of mobility, size, and survivability (duration of operation).
Static HELPER (SH): These HELPERs are reasonably portable yet considered relatively
static as they are envisioned to be operated in a relatively fixed location (terrace of household,
hospitals, roadside, public buildings, etc.) with abundant sunlight or other energy sources. These
HELPERs have the largest battery and solar panel that supports 24/7 operation. The design goal of
40
CHAPTER 3.
the SH is to survive at least a day or two in the absence of sunlight and to extend for multiple days in
the presence of ample sunlight. The components used to prototype the proposed SH is depicted in
Fig. 3.2. The main board used is a Raspberry Pi (RPI) 3b [79]. The choice was motivated from the
low cost, size, and large open community support for RPI development. Additionally, it is enabled
with WiFi (802.11 b/g/n) and will be set up to operate as an access point for EU. The WiFi link
provides a comparatively lower range of coverage but is an essential choice taking into account the
widespread usage of WiFi by today’s devices (smartphones, tablets, laptops, etc.). This will ensure a
seamless connection from a users point of view due to the abundant familiarity in accessing WiFi.
Adafruit LoRa
Breakout
Arduino
Microcontroller
GPS
Receiver
Raspberry Pi
Zero W
SOLAR PANEL
Raspberry Pi 3
Adafruit RFM95W LoRa
Breakout
Arduino Microcontroller
High Efficiency Buck
ConverterSolar Charge ControllerLatching Power Relay
Relay
GPS Receiver
Figure 3.2: Static HELPER design.
To establish networks covering larger areas, LoRa [80] is chosen to set up low power,
long-range links (2-5 km in urban areas and 15 km in suburban areas [81]) between HELPERs. LoRa
is emerging as a viable communication choice for IoT devices that strive to operate at low power
yet achieve long-range. The long-range of LoRa ensures dense networks because a larger number
of these nodes may be deployed within the communication range. This ensures the availability of
multiple routes to choose from such that energy-based optimization can be performed. A Global
Positioning System (GPS) receiver is also attached to the RPI. This will be used to acquire the
location information to perform geographical routing and to indicate the location of the node when
assistance needs to be dispatched. Lastly, an Arduino microcontroller and power relay can be attached
to the RPI to put the system in a deep sleep mode to conserve energy when multiple devices are
41
CHAPTER 3.
deployed in the same vicinity. Once the RPI has decided to sleep for a given duration of time,
the signal is sent over to the Arduino board which in turn shuts the power relay which disables
the RPI. The Arduino board uses its power management watchdog timers to keep the RPI in a
low energy state for the entire sleep duration. The Arduino will then flip the power relay back on
effectively restarting the RPI. This mechanism will enable energy preservation and cooperation
between multiple HELPERs in the same vicinity.
It has been estimated that the system will require about 1Wh to run at a 25% duty cycle.
Choosing the appropriate solar panel relies on current weather conditions, amount of daily sunlight
and historical weather trends for that location. In a mostly sunny area, a 3 or 4 W solar panel would
be sufficient for 24/7 operation yet some areas may need a 10 to 15 W solar panel. The solar panel
will be attached to a solar controller. This solar controller has a 24 Wh lead-acid battery and a
high-efficiency buck converter to the load. The 24 Wh battery will last an entire day on a full charge.
The more expensive buck converter can be used to supply power to the system because a buck
converter can commonly get up to 90% efficiency, whereas a simple voltage regulator would have a
59% loss of power coming from 12 V down to 5 V. A factor that will affect the portability of the
design is the battery system designed to operate 24/7. Li-Ion Batteries are the ideal choice to ensure
portability but needs a complicated charging and discharging circuit whereas Lead-acid batteries
have a simpler charge-discharge circuit but tend to be on the heavier side. One can make a studied
choice based on the deployment requirements.
Figure 3.3: Envisioned final mobile HELPER design.
42
CHAPTER 3.
Mobile HELPER (MH): Two versions of MHs are envisioned as deployable end-products.
The first version that is indented to go on vehicles will not require a battery as it will draw energy
from the vehicle itself. This version will have the smallest form factor but will have to operate
within the vehicle itself. The second version of the MH will have the same design as the SH with
the exception of eliminating solar panel and using a smaller battery to ensure more portability for
ER and EU. The MH will use lightweight 18650 Li-Ion batteries as a power source with a boost
converter to up the voltage to the required 5 V input of the devices. Li-Ion batteries have a much
higher energy density compared to lead-acid batteries. Since the batteries will be easily swappable,
discharged batteries can be removed to recharge while other charged ones can be used in deployed
nodes. As the batteries do not need to supply a load while being charged, any off-the-shelf charger
can be used. There is a trade space between the battery capabilities with energy storage and max
amperage draw. For the MH, two 3000 mAh 18650 batteries will be used to power the device. Given
the 1 Wh load from the device and two batteries that supply around 22 Wh, the ER devices will be
operable for a total of at least 8-9 hours (22 hours in ideal scenarios considering only 25% duty cycle
and no loss) before needing to be recharged.
Figure 3.4: Aerial HELPER (Erle copter).
Aerial HELPER (AH): When the network is set up, based on accessibility, there might
be parts of the network that is disconnected due to node failure, locally disruptive channel conditions
or uneven distribution of HELPERs during the setup period. These gaps in the network are referred
to as network holes. The goal of an AH is to identify these isolated HELPERs and act as a temporary
sink node that retrieves information. The isolated HELPER can upload the information about the
users currently connected to the given HELPER and this information is carried by the AH to the
ERC. The AH also indicates which locations need more HELPERs to be deployed in order to fill the
43
CHAPTER 3.
network holes. The AH are the most costly and least energy-efficient (taking into consideration the
energy for flight) among the three types of HELPERs but is required in critical scenarios where road
access might be completely cut-off. More about the different deployment scenarios are discussed in
the next section. To meet the needs of the AH, an open style drone (see Fig. 3.4 ) can be used to
allow for the flexibility of programming a completely autonomous drone while minimizing the cost.
The Erle Robotics Drone Kit also has the added benefit that the Erle Brian [82] uses RPI. Therefore,
the development toolchain is the same as the MH and the SH. Therefore, these AHs will use the
battery and RPI that are inherent to the drone itself.
3.2.2 Deployment Scenarios
The deployment scenarios are divided into three major cases based on accessibility and
available resources which are discussed in detail below.
Scenario I (Full Accessibility and resources): In the first scenario, accessibility is not
restricted and the ERs have all the required resources (vehicles, drones and a large number of ERs) to
set up a HELPER network. In this scenario, the SHs can be placed in a strategical manner to ensure
full coverage of the affected area with minimum deployment cost. The placement of the SHs will be
under direct sunlight to enable 24/7 operation. The deployment of SHs can be relatively sparse as
the HELPERs can be arranged optimally to ensure full coverage and extended lifetime. The MHs
and AHs will also supplement this network during the rescue operation. AH will use BEACON
packets (more about BEACON packets is described in Section 3.3) while flying over the affected
areas to determine the HELPERs that might need replacement due to depletion of battery or absence
of sunlight. Overall, there is more control over the deployment of HELPERs and hence easier to
provide full coverage and repair disconnected parts of the network.
Scenario II (Limited Accessibility): In the second scenario, the accessibility is highly
limited during the initial stages. This implies that there will be limited options (few vehicles with
HELPERs during initial stages) to deploy HELPER network. Therefore, a denser deployment of
SHs will be necessary to ensure maximum connectivity and network lifetime. These HELPERs
may be present in the emergency kits of households, or other buildings before the disaster strikes.
Additionally, a large number of supplementary SHs can be deployed via air. The role of AH will
also be critical in these scenarios to determine network holes (areas without coverage or isolated
HELPERs). When isolated HELPERs are determined, AH will act as a temporary sink and will fly
back to the ERC with this information. This will enable ERs to have access to survivor information
44
CHAPTER 3.
in isolated areas and prepare rescue efforts. The AHs will also be able to plug the detected network
holes by promoting ERC to deploy HELPERs to provide complete network connectivity.
Scenario III (Limited Accessibility and resources): In the third scenario, the assumption
is the lack of access and resources. There is no availability of the costlier AH. Since the proposed
SH and MH is highly cost-effective, multiple HELPERs can be deployed in a dense manner such
that maximum area is covered for connectivity. The dense network will operate in an ad hoc manner
bolstered by the proposed routing algorithm that aims to maximize the network lifetime.
3.3 HELPER Design and Implementation
This section discussed the overall HELPER framework consisting of the communication
protocol stack, novel routing protocol and discuss the corresponding packet structure, and packet
handling while determining the necessary interactions between different layers to enable a cross-
layered approach to optimize the network performance.
3.4 HELPER’s Cross-Layer Protocol Stack
The significance of cross-layer optimization in wireless communication has been widely
studied [83, 84, 85, 45, 86, 87] across various domains and optimization problems. Identifying the
advantages of cross-layer optimization there has been some work recently to develop cross-layer
platforms to facilitate these technologies [88, 46, 89]. Figure 3.5 depicts the design concept of a
HELPER and how the design is currently implemented in a modular manner on the selected platform.
The design considerations for each of these layers are discussed in detail in the upcoming sections.
Physical Layer
As discussed earlier, HELPER is enabled using two wireless technologies WiFi (802.11
b/g/n) and LoRa which gives it the heterogeneous nature of operations. The prominent reason behind
using both these well established wireless technologies are as follows, (i) WiFi is ubiquitous in
today’s devices and this will ensure seamless access for EUs, (ii) LoRa is becoming a prominent
communication technology enabling IoT devices that requires low power, long-range wireless links
and (iii) it is extremely cost-effective to use off-the-shelf physical layer to ensure low (Size, Weight,
and Power) (SWaP). The features of the physical layer are shown in Table. 3.2. Though both WiFi
and LoRa are used as wireless technologies to enable HELPER, only LoRa can be considered as the
45
CHAPTER 3.
Cross-Layer Controller
ApplicationLayer
TransportLayer
NetworkLayer
Data LinkLayer
HELPER PROTOCOL STACK
PhysicalLayer
Web/Mobile Application
MOBILE DEVICES / PC
Responder Dashboard
HELPER NODE/ RASPBERRY PI
Service Layer
Network Layer
Medium Access Control Layer
LoRa Physical Layer
DESIGN CONCEPT
IMPLEMENTATION
WiFi Ethernet
Application Server
Figure 3.5: HELPER’s cross-layer protocol stack design and implementation.
physical layer from the point of view of the ad hoc HELPER network. WiFi can be considered as
the interface between the application layer and the service layer of the HELPER’s protocol stack as
shown in Fig. 3.5.Table 3.2: Heterogeneous wireless link parameters.
Features WiFi (802.11 b/g/n) LoRa
Frequency Range 2.4 GHz 915 MHzBandwidth 20 MHz−40 MHz 7.8 kHz−500 kHzTransmission Range Medium HighPHY techniques DSSS, OFDM, MIMO-OFDM CSS, FSK [90]
The HELPER network stack interfaces to the LoRa radio module using the Radio Head and
BCM2835 C++ libraries. This interface is implemented in an Application Programming Interface
(API) that is utilized by the MAC layer to send and receive packets over-the-air. The API also
provides functionalities for reading and writing radio parameters.
Data-Link Layer
One of the primary function of the data-link layer is negotiating the medium access. In
the proposed HELPER network, similar to the discussion in the physical layer, HELPER has two
levels of medium access, (i) local WiFi links between HELPER and devices (phone, laptop, and
tablets) of users and (ii) LoRa links between different HELPERs that form the ad hoc network. The
standard off-the-shelf MAC protocol employed by WiFi (IEEE 802.11) is used to allow multiple
46
CHAPTER 3.
users access to HELPERs within the local area. A similar Carrier Sense Multiple Access/Collision
Avoidance (CSMA/CA) based MAC protocol has also been implemented to setup multihop ad hoc
network using LoRa with the intention to utilize the Channel Activity Detection (CAD) offered as
a hardware feature on the RF95 LoRa. CAD is a valuable tool since LoRa uses spread spectrum
transmissions. The spread spectrum is known to operate at low signal to noise ratio making traditional
approaches like power detection with Received Signal Strength Indication (RSSI) unreliable. CAD
helps to detect if there is ongoing transmission in the channel chosen within two symbols according to
the RF95 hardware documentation. This feature can be leveraged to implement the MAC protocol for
the multihop LoRa based network. Since WiFi and LoRa operate on different parts of the Industrial,
Scientific and Medical (ISM) band (as shown in Table. 3.2) they do not interfere with each other’s
operation.
IDLE WAIT CTS WAIT ACKWAIT DATA
RX RTSSEND CTS
Unicast Data AvailableSEND RTS
CTS Timeout
RX CTSSEND DATA
ACK RXOr
ACK Timeout
DATA RXSEND ACK
Or DATA
TimeoutEVENT ACTION
Timer ExpiresBROADCAST
BEACON
Broadcast Data Available
BROADCAST DATA
Figure 3.6: FSM of the MAC protocol.
Therefore, the data-link layer shown in Fig. 3.5 contains the control logic used by
HELPERs to negotiate access to the wireless medium. It houses the FSM used to implement
the CSMA/CA like MAC protocol used by the HELPER network. As seen in Fig. 3.6, the traditional
RTS (request-to-send), CTS (clear-to-send) handshake is used before transmitting a unicast data
packet. The successful reception is followed by the receiver transmitting ACK (Acknowledgement)
packet. In addition to this, a BEACON packet is broadcasted periodically by a HELPER that has
not transmitted any control packet for a pre-determined duration. Each of these control packets
(RTS, CTS, BEACON) carry information including, instantaneous backlogged queue length, resid-
ual energy, location and the observed goodput per neighbor, which is referred to as Optimization
Assisting Information (OAI). In this manner, each HELPER gathers OAI from its neighbor and uses
47
CHAPTER 3.
this updated information to perform optimized energy-efficient routing (which will be discussed in
detail in upcoming sections). Therefore, in implementation, the MAC layer continuously monitors
the physical layer receive queue for inbound messages and handles them according to the current
state of the FSM. All OAI received from the control packets are used to update the inputs to perform
cross-layer optimization. Once the network layer has performed the required optimization and chosen
the optimal next hop, the MAC layer negotiates the medium and forwards the data packets.
3.4.0.1 Network Layer
The network layer is responsible for packet queuing and routing. As shown in Fig. 3.5,
the network layer interfaces to the service layer and data-link layer. When packets are received
from the service layer or data-link layer, the network layer encapsulates/decapsulates network layer
fields as needed and places packets in the appropriate queue. The network layer maintains two
transmit queues: one for priority traffic and a second for best-effort traffic. Each packet is sorted
into one of these queues depending on application message type and other fields in the header. The
energy-efficient routing algorithm is used for routing unicast packets which ensure maximum network
lifetime. Every broadcast packet contains a Hop-To-Live (HTL). Packets with HTL greater than zero
are broadcasted by the receiving HELPER. In addition, the network layer uses shared memory to
manage neighbor lists and OAI information in order to perform optimized cross-layer routing. The
network layer also has access to HELPER’s current GPS location via libgps and stores it in its local
OAI data which is then shared with neighboring HELPERs.
A critical aspect of the proposed HELPER network is its energy efficiency. Since the
majority of energy consumption is attributed to the transmission of packets, routing becomes a
significant aspect of the design. Accordingly, the utility function takes into account energy efficiency,
goodput and a measure of congestion (using differential backlog) to formulate an optimization
problem with the objective to maximize network lifetime while maintaining reliable connectivity.
Since the goal is to deploy a scalable network, a distributed version of the optimized routing
algorithm is formulated such that each node can make its own routing decision based on the limited
OAI gathered from its neighborhood.
System Model: To design the routing algorithm, the most constrained scenario (scenario
III of Section 3.2.2) which has restricted access and minimal resources is considered. Accordingly,
consider a dense multihop wireless ad hoc network comprising of several N HELPERs (which is
referred to as nodes in this section) modeled as a directed connectivity graph G(Nnet ,E), where
48
CHAPTER 3.
Nnet = {H1,H2...,HN} is a finite set of wireless transceiver (nodes), and L(i, j) ∈ E represents
unidirectional wireless link from node Hi to node H j (for simplicity, they are also referred to as node
i and node j). G is assumed to be link symmetric, i.e., if L(i, j) ∈ E, then L( j, i) ∈ E. Each node is
assumed to have the transmission range R based on the chosen transmit power Pt . As seen before,
all the nodes are equipped with GPS and therefore the location (longitude/latitude) coordinates are
known. The knowledge of node locations is important for a geographical/position-based routing
algorithms proposed in this work. In Fig. 3.7, the nodes within the transmission radius of i will
constitute its neighbors. Let us denote the set of neighboring nodes of node i as NBi = { j,k} and
the sink (ERC) node as s. The location of s can be predefined in every node or as in this case, this
information is flooded at the time of network setup. This formulation considers packets that have to
be transmitted from node i to sink s but this can be extended to any source-destination pair.
j
k
Sinki
𝒅𝒋𝒔
𝒅𝒊𝒔
𝒅𝒌𝒔
Figure 3.7: Network diagram.
The distance between any two nodes i and j is represented by di j. If a node j exist within
the transmission range of node i, there exists a link L(i, j), i.e., a wireless communication link L(i, j)
exists when di j ≤ R. The power consumed over L(i, j) or the power required by the source node (i)
to transmit to a neighboring node ( j) is denoted by Pi j. The initial and residual battery energy at node
i can be denoted as Ei0 and Ei
r respectively. Every node maintains a queue that holds the outbound
packets. Let qi represent the instantaneous number of packets retained in the queue of node i, also
called the queue backlog. The transmission bit rate and BER over L(i, j) are denoted by Ri jb and ei j
b
respectively.
Routing Algorithm: The proposed diStributed Energy-Efficient bacKpressure (SEEK)
49
CHAPTER 3.
routing algorithm utilizes the geographic information of nodes, differential queue backlog, residual
battery energy and transmission power levels to compute the optimal next hop. In this section, a
formal derivation of the utility function (Ui j is presented with respect to L(i, j)) and formulate the
network optimization problem.
The utility function considers the following parameters associated with potential next-hop;
(i) proximity to sink, (ii) differential queue backlog, (iii) residual battery energy, (iv) power required
to transmit over the link and (v) the corresponding link throughput. This information is gathered from
traditional control packets like RTS, CTS and BEACON packets. As discussed earlier, these packets
will contain updated OAI and a PROBE field. In a realistic scenario, the real-time Signal-to-Noise
power Ratio (SNR) is unknown to the device. Therefore, measures derived from SNR estimated
based on a radio propagation model might not be a suitable guideline for signifying transmission
reliability over a link L(i, j). Therefore, a PROBE field is proposed in the control packets to perform
link probing [91, 92]. The PROBE field would contain a bit sequence known by the nodes in the
network. Upon receipt of the control packet, each node will compute effective throughput (goodput)
measure in bits per second (bps). The term goodput is used to signify the effective number of
bits successfully received. For example, once node i receives a control packet from node j, it will
compute the corresponding goodput measure (Gi j) with respect to L(i, j) and transmission strategy,
Ti j (which includes choice of Ri jb and Pi j). The energy efficiency of a given link can be expressed as
a ratio between goodput and transmission power as [93],
ηi j =Ri j
b
(1− ei j
b
)Pi j
=Gi j
Pi j(3.1)
where ηi j gives the measure of number of bits successfully transmitted over L(i, j) per Joule of
transmission energy. Another key factor that needs to be considered in routing is the differential
queue backlog (∆Qi j = qi−q j) with respect to the source node (i) and next-hop ( j) [11, 94, 95]. The
queue backlog at the destination node is considered to be zero. Considering the queue backlog is
necessary to mitigate congestion in the network and traditional backpressure algorithms has been
shown to be throughput optimal [11]. Since achieving maximum throughput is not the sole objective
of HELPER network, differential backlog is just one parameter in the utility function. The effective
progress made by a packet can be represented as dis− d js. Choosing nodes that provide larger
progress implies fewer hops to the sink node which in turn could lead to smaller energy consumption.
Finally, to ensure uniform depletion of energy per node, one must consider E jr of potential next hops
[96]. Therefore, utility function is defined as follows,
50
CHAPTER 3.
Ui j = ηi j
(max [∆Qi j,0]
qi
)(dis−d js
dis
)(E
jr
Ej0
),∀ j ∈ NBi
ηi j aims to improve the energy efficiency of the network. It is also interesting to note that the
maximum value of Ui j = ηi j when each of the three normalized terms is 1. This implies that each
of the other terms penalizes the utility function based on the instantaneous value. For example, a
small differential backlog (qi−q j < qi) will dampen the value of Ui j. Both dis− d js and Ejr will
have similar effects on Ui j.
The objective of the network is to maximize the summation of Ui j for all possible links
L(i, j) in order to maximize the overall energy efficiency of the network. This, in turn, will ensure
reliable communication while maximizing the network lifetime (which is defined as the time when
the first node in the network depletes its energy leading to a network hole). The optimization problem
is subject to residual battery energy, queue backlog, bit error rate, and capacity constraints. This is
formulated as Problem P1 shown below,
P1 : Given: G(Nnet ,E), G, Er, Q
Find: NH∗,T∗
Maximize : ∑i∈Nnet
∑j∈NBi
Ui j (3.2)
subject to :
Ri jb ≤Ci j, ∀i ∈ Nnet , ∀ j ∈ NBi (3.3)
ei jb > ei j
b∗, ∀i ∈ Nnet , ∀ j ∈ NBi (3.4)
Eir > 0, ∀i ∈ Nnet (3.5)
qi ≥ 0, ∀i ∈ Nnet (3.6)
Where the objective is to find the set of next-hop and transmission strategy for all nodes in
the network which can be represented as NH∗ = [NH∗i ] and T∗ = [T∗i j] respectively, ∀i ∈ N j ∈ NBi.
In the above optimization problem P1, G= [Gi j], Er = [Eir] and Q= [qi], ∀i ∈Nnet , ∀ j ∈NBi denote
the set of goodput measure, residual battery energy and queue backlogs respectively. The constraint
3.3 restricts the total amount of data rate in link (i, j) to be lower than or equal to the physical link
capacity. Constraints 3.4 impose that any transmission should guarantee the required BER. Finally,
51
CHAPTER 3.
constraints 3.5 and 3.6 ensure the residual energy and queue backlog of each node will not have
negative values. It can be seen that for solving the above optimization problem, nodes would require
global knowledge of the network. Since the centralized optimization method is not a scalable solution,
it motivates the need for a scalable distributed solution. SEEK is proposed to operate in a distributed
fashion and enable each node to find the next-hop based on the local information available to them.
Each node with a packet to transmit chooses an optimal next-hop and transmission parameters such
that it maximizes its own local utility function. The probability of channel access will be controlled
by utility-based random backoff. This can be considered as a divide-and-conquer approach to solving
the optimization problem in a distributed manner. Accordingly, every source node (i) will aim
to maximize the utility function Ui j and select the optimal next-hop and transmission strategy as
follows,
[ j∗,T∗i j] = argmaxj
Ui j,∀ j ∈ NBi (3.7)
Each node will maintain a neighbor table with node parameters of its neighbors and will
update the table as needed based on information from the control packets. Considering the scenario
in Fig. 3.7, the source node i will listen to control packets and maintain a neighbor table as in table
3.3.Table 3.3: Neighbor table of node i.
Node ID Distance toDestination
QueueBacklog
ResidualBattery Energy Goodput
j d js q j Ejr Gi j
k dks qk Ekr Gik
The ERC (sink) collects and disseminates vital information like availability of resources,
drop-off locations, emergency updates for EU among others. ERC needs to strategically flood this
information in the network to enable all HELPERs to obtain the updated information. This implies
the requirement to implement one of the energy-efficient flooding technique that has been widely
studied in literature [97, 98].
Service Layer
The Service Layer provides a common interface between HELPER applications and the
lower layers of the protocol stack. This layer communicates to HELPER applications using local
sockets and the Network Layer via direct function calls. Messages received from applications are
translated from HELPER Send format to HELPER Packet format (shown in Fig. 3.8) and are passed
52
CHAPTER 3.
HELPER SEND
Message Type
Destination
Time To Live
Payload
Reserve
HELPER RECEIVE
Message Type
Source
Time To Live
Source Location
Payload
HELPER PACKET
Message Type
Source
Destination
Time To Live
Packet ID
Sequence ID
Sequence Total
Beacon Data
MAC Msg Type
MAC Source
MAC Destination
Payload
Figure 3.8: Packet formats.
to the network layer. Messages received from the network layer are translated from HELPER Packet
structure to HELPER Receive format and are then passed to the application. In the implementation,
the Service Layer uses an MQTT messaging socket to communicate with the Web Application and
messages are encoded using JSON. The Paho MQTT CPP and Rapid JSON libraries are used to
implement the messaging to the application. The implementation is such that more application
message types and message handling can be added in the future to expand the capabilities of the
HELPER network.
3.4.0.2 Application Server
Users can join the HELPER network by connecting to a HELPER node via a WiFi or LAN
connection. The HELPER nodes are configured to act as a WiFi access point, allowing users to
connect their smartphones, laptops, tablets, etc. to the network. A wired connection to a HELPER
node is also possible and is utilized by the ERC. Each node runs a web server that hosts the web
applications. Once connected to a HELPER node, a user can launch a web application. Currently,
two web applications are developed, one of EU and the other for ERC. These are discussed in detail
in Section 3.4.3 and Section 3.4.4.
The web server interfaces to the HELPER network stack to send and receive data across
the network. This interface is implemented using MQTT sockets. The web server and service layer
both connect to the MQTT broker and setup two publish-subscribe channels. One channel is for
53
CHAPTER 3.
sending messages from the web server to the HELPER stack and the other is for sending messages
from the HELPER stack to the web server. The data sent on these MQTT sockets are in the Helper
Send and Helper Receive formats of Fig. 3.8 and are encoded with JSON.
3.4.1 HELPER Packet Handling
The HELPER network consists mainly of two kinds of HELPERs; (i) HELPERs that are
deployed in households, hospitals, and other building that residents (survivors of disaster) connect to
and usually have limited power supply and (ii) HELPER that forms ERC and usually has an unlimited
power supply. Accordingly, messages can be classified as EU messages and ERC messages. Each
message types used by HELPER is described below,
3.4.1.1 End-User Messages
Emergency HELP Messages: A HELP message is used to indicate that an individual is in
need of immediate assistance. This is similar to or a substitute for a 9-1-1 call when cell phone and
other services are disrupted or inaccessible. All HELP messages are handled by HELPER in two
ways. First, the HELP message is send destined for the ERC with maximum HTL. Additionally, at
the service layers, these HELP messages are also broadcasted with a predefined HTL (currently set
to 2). The intention of broadcasting the HELP message with HTL= 2 is to find a first responder who
may be in the vicinity of the individual in distress to provide faster response rather than waiting for
ERC to react. The HTL is limited to avoid excessive energy consumption and mitigate problems of
congestion. Overall, the HELP message will enable users to alert authorities of their location, need,
and situation when all other communication infrastructures are down.
Local Messages: Every user connected to a HELPER is able to chat with each other using
Local Chat messages. These messages are exchanged using WiFi itself and do not have to use the
LoRa on Physical Layer. These links can achieve high data rates and in future support video chatting
as long as HELPER is plugged in and does not have energy constraints. Therefore, everyone in
the range of a single HELPER can use local messaging to remain connected to each other. These
messages are handled by the service layers itself and are not passed to the lower layers of the
HELPER protocol stack.
Neighborhood Messages: A neighborhood message is a chat message that is transmitted
to all immediate neighboring HELPERs. In the network layer, this is a broadcast message with an
HTL of 1. This message is intended to enable communication between the community in the close
54
CHAPTER 3.
neighborhood (within 2 km radius). These messages will be used by the community members to
help each other and mark themselves safe even if they are not connected to the same HELPERs. The
design is flexible enough such that the broadcast can be extended beyond immediate neighbors by
setting appropriate HTL.
Resource Messages: A resource message is sent by a user to indicate that a resource (food,
water, gas, medicine, internet, etc.) is available in proximity to the local HELPER. This message
is transmitted to the ERC with an HTL set to maximum. The Responder Station aggregates these
messages, approves them and transmits periodic HELPER resource update messages to let all the
HELPERs in the network get the updated Map.
3.4.1.2 Emergency Response Center Messages
Network Discovery Message: A Network Discovery (ND) message is used at network
initialization. The ERC broadcasts this ND message to its immediate neighbors. All HELPERs
that receive this message use an efficient flooding technique to broadcast ND messages to other
HELPERs. As the deployed HELPERs receive a ND, they reply with a HELPER Update Message
containing their information (Node ID, location, etc.). These are unicast messages to the ERC with a
maximum HTL. In this manner, the ERC performs network discovery to map all the nodes that are
actively deployed in an affected area.
HELPER ALERT Message: Similar to the Wireless Emergency Alerts (WEA) that is used
over the cellular network, the ALERT message is intended to inform every connected user about
an imminent threat like high winds, rising water level, flash flood, fires, etc. This message is also
distributed using an efficient flooding technique. Every user connected to a HELPER sees a message
labeled from ERC and hence are aware of the steps to take to remain safe during the upcoming
situation. This will be highly beneficial in situations where the cellular network is not operable due
to infrastructural damage.
Resource Update Message: Once the ERC receives a resource message from EU connected
to any of the HELPERs in the network, it has to first approve the resource update. Upon approval, the
ERC floods the resource update message to the entire network using an efficient flooding technique.
In this manner, every HELPER in the network receives updated resource information for users to
access.
55
CHAPTER 3.
3.4.2 Applications
As discussed, to provide a complete end-to-end solution, two applications have been
developed; one for EU to connect to the HELPER network using their mobile devices; and the second
for ERC to remotely monitor the network and provide assistance and alerts to the EU. In this section,
the functionalities that have been enabled through these applications are described.
3.4.3 End-User Application
Location of HELPER
user is connected to
Resource Update Options
Chat Tabs to filter various
Types HELPER Messages
Neighboring HELPERs
Location of Emergency
Response Center (ERC)
Login Details
User choose to send messages
locally, to neighborhood or to ERC
alerting emergency situation
HELPER Turns red on all the
maps when someone connected
to it needs help
Figure 3.9: Website application.
Every user connected to a HELPER via WiFi will be prompted to access the services of
HELPER network by logging in to the Web App shown in Fig. 3.9. As you can see, the login page
consists of the location of the HELPER (marked using a black marker) that the user is currently
connected. The final version of the App will also have a short message describing the network and
utilities to encourage people to use the HELPER network. Next, the features availed by the Web App
for the connected EU to use are discussed below.
Text and Voice Messaging: The primary goal of the HELPER Network is to keep individu-
als in the affected community connected. Therefore, an intuitive chatbox is developed for EUs to
interact with and help each other and ERC. As shown in Fig. 3.9, there are three kinds of messages
that can be sent/received by a user (as indicated by the 3 tabs or options in the pop-up menu).
i The local messages are exchanged between users connected to the same HELPER. These
messages go directly over WiFi and do not need to interact with the LoRa physical layer. These
56
CHAPTER 3.
links can achieve high throughput since it is not bottle-necked by the lower data-rates of LoRa. In
the future, video call and higher throughput applications can be enabled based on the availability
of energy in the affected area.
ii Next, the message sent using Neighbor tab are broadcasted to h-hop neighbors (where h is
predetermined by the operator). The choice of h would be a trade-off between energy consumption
and range of connectivity. In this case, a message sent by a user to the neighbors is received by
all users connected to all the HELPERs (black and blue markers) within h hops from the source
node.
iii Finally, and most importantly, the Emergency tab is used to send distress messages directly to the
ERC to seek help during distress. These messages will be carried over a multi-hop path to the
ERC and inform the ERC of the location where help is required. This serves as an alternative
to 9-1-1 calls when the degradation of infrastructure renders traditional 9-1-1 calls infeasible.
Similarly, ERC can broadcast ALERT messages so that each user is alerted to situations like high
winds, rising water level, flash floods, etc. The All tab displays all the above messages in one
place.
Live Map Updates: A regional map with live updates on the availability of resources
like gas stations, operational hospitals, food and water gas station, internet access, electricity, etc
are accessible to the connected users. The ERC will collect information about the availability of
resources using HELPERs deployed in hospitals, stores, gas stations, households, etc. Periodically,
this information is flooded by the ERC in the ad hoc network to update the map at each HELPER.
The periodicity of this flooding can be controlled by the ERC based on the update information and
status of the network. This information sharing is accomplished as follows,
• A connected user (ER or EU) who has information about available resources to share with rest
of the users, drags and drops the corresponding resource on the known location on their local
map.
• This action triggers a packet that is directed towards the nearest ERC. This packet is routed
using the proposed SEEK algorithm towards ERC.
• Upon reception of the packet, a message containing the information about the type of resource
and its location shows up on the ERC Application. Once the operator verifies this information,
it is flooded to the rest of the HELPER network. The method of verification will be controlled
57
CHAPTER 3.
by the agencies. This can be based on trusted nodes, the number of similar requests or physical
verification using an on-field ER.
It can be argued that the above three-step process may incur a delay in disseminating
information as compared to the information being flooded by the source HELPER itself without
going through the ERC. While this may be true, authorizing any node to update resource information
may lead to the propagation of misinformation, duplicate information, and overall larger energy
consumption.
3.4.4 Emergency Response Center Dashboard
Network Discovery &
Resource Update Button
Emergency HELP message
from user
Location of HELPER sending
user’s distress message
Resource update from
user
HELPERs in the network
Live resources updates
Figure 3.10: ERC application.
As shown in Fig. 3.5, at the ERC, a HELPER is connected to a PC using an Ethernet cable.
ERC Dashboard with some critical features have been developed with the following capabilities,
Remote Monitoring: A ND phase can be initiated by using the Network Discovery Message
button on the ERC Dashboard. Accordingly, the ERC broadcasts ND packet to its immediate
neighbors. All HELPERs that receive this message use an efficient flooding technique to broadcast
ND packet to other HELPERs. As HELPERs receive a ND packet, they reply with a HELPER
Update packet. The HELPERs deployed in the field use this unicast HELPER Update packet during
ND phase to reply to the ND packet with their information (Node ID, location, etc.). The operator
can perform this remote monitoring intermittently to ensure all the HELPERs in the network are
active. If some of the HELPERs do not show up during these intermittent monitoring phases, the
operator will be aware of the lack of connectivity in those areas and can deploy more nodes or take
other corrective actions to keep the network fully connected.
58
CHAPTER 3.
Critical Alert Message: Similar to the Wireless Emergency Alerts (WEA) that is used
over the cellular network, the ALERT message is intended to inform every connected user about an
imminent threat like high winds, rising water level, flash flood, fires, etc. This message will also be
distributed using an efficient flooding technique. Every user connected to a HELPER sees a message
labeled from ERC and hence are aware of the steps to take to remain safe during the upcoming
situation. This will be highly beneficial in situations where cellular network is not operable due to
damage and a large number of individuals need to be informed about imminent dangers.
3.5 Evaluation
Figure 3.11: HELPER protoype.
This section discusses the HELPER prototype that was developed to establish proof of
concept and conduct an experimental evaluation of the proposed solution on the HELPER testbed.
3.5.1 Operational Proof-of-Concept
To establish feasibility, the focus was on developing the prototype for MH as shown in
Fig. 3.11. The decision to prototype the MH stems from its portability, cost and because it can be
used to demonstrate the envisioned operation of the entire HELPER network. This concept can be
easily extended when more SH and/or AH are added to the HELPER network. The proof-of-concept
demonstrations conducted by us were recorded using mobile screen recorders and compiled in the
form of a video [99]. In this section, the parts of the video are used to discuss the experiments and
functionalities it intended to display. Accordingly, six HELPERs (see Fig. 3.11) have been developed,
59
CHAPTER 3.
Figure 3.12: 6-node HELPER network.
five of which are deployed with given location value as shown in Fig. 3.12 (white markers) for EU to
connect using their mobile devices. The sixth one is connected to a PC and acts as the ERC which is
indicated as a building in Fig. 3.12. There were four EUs connected to the network through three of
the deployed HELPERs. It can be seen that two users (Jithin and Nick) are connected to the same
HELPER. The Web App with the chatbox corresponding to each connected users are displayed at
the edges of Fig. 3.12.
First, the operation of the local messaging using the HELPER where two users were
connected is assessed. Several text messages were exchanged between Nick and Jithin, as one
can see in Fig. 3.13 connected to the same HELPER. As mentioned earlier, these messages are
exchanged over WiFi and do not need to use LoRa. Since these are local messages, the other two
users (Andrew and Anu) connected to their respective HELPERs will not receive these messages.
Next, Jithin switches his chat option from local to neighbor which implies all users connected to
HELPERs within h hop will receive his messages. In this deployment h = 1, which implies one-hop
neighbors will receive chat messages. Accordingly, the text message sent by Jithin is received
by the other three users even if they are not connected to the same HELPER as Jithin. This part
of the demonstration proves how HELPER can be used to keep community members connected
by exchanging text messages with each other. The EU is completely abstracted from the ad hoc
networking operation that happens in the background. To the EU, they are just sending messages to
60
CHAPTER 3.
two different groups, one local group, and other more extended community groups.
Figure 3.13: Local text messaging. Figure 3.14: Neighbor text messaging.
As mentioned before, ERC has the ability to flood the HELPER network with ALERT
messages to inform connected EU of imminent dangers. In this case, the ERC sends an ALERT
message regarding the high winds. In Fig. 3.15, it can be seen that three users connected to different
HELPERs have received the “HIGH WIND" alert message transmitted by ERC. The packet was
dropped for one user since a transport layer is not currently implemented that would ensure end-to-
end reliability. During this period of demonstration, multiple users have shared the availability of
resources like a gas station, water, food, etc. These packets are first sent directly to ERC and upon
approval, the information is flooded in the network. Accordingly, the connected users are able to see
the location of the resource on their local map in their Web App as shown in Fig. 3.16.
Figure 3.15: ERC’s ALERT. Figure 3.16: Distress message.
The final part of the demonstration was to evaluate how distress messages can be sent
directly to the ERC. In this case, Jithin realized Nick needs medical attention and uses the “help"
option to send a message. This message is sent directly to ERC and nearby HELPERS simultaneously
just in case there is ER or others in the vicinity who can provide assistance as compared to the
61
CHAPTER 3.
ERC itself. Accordingly, the connected users see the HELPER which Jithin is connected to turn red
indicating distress at that location. This will enable community members to reach out to the nearest
ER and provide the required assistance. This location information is also available at the ERC which
instantly dispatch help to the given location. Overall, this service acts as the replacement for 9-1-1
calls when traditional infrastructures like cell towers or the internet are unavailable. This technology
is envisioned to enable low cost, efficient public safety system. Additional demonstration [100] is
provided from the point of view of an ERC.
3.5.2 Testbed Evaluation
The previous section established the feasibility of the proposed HELPER network. Here, an
extensive evaluation of the underlying SEEK algorithm and analyze various aspects of its operation
in a unicast setting is performed. To accomplish this, the six HELPER prototypes are set up in a
grid topology as shown in Fig. 3.17. The proposed algorithm is compared against the shortest-path
routing algorithm. This shortest-path routing is implemented using a greedy geographical forwarding
technique. In this algorithm, nodes that have a packet to forward elects the node closest to the
destination as the next hop. This can also be seen as a greedy distributed version of MPR used in [77]
discussed earlier in Section 3.1 with the assumption that paths with the smallest number of hops may
indeed be the path with minimum energy consumption. Both protocols have similar complexity. In
other words, all the possible next-hops are considered by both. The greedy algorithm calculates the
forward progress of each next-hop and SEEK calculates the utility function for each next-hop. Both
then choose the neighbor providing the highest value. In terms of complexity for a given number of
transmission strategies, the complexity of both the algorithms is O(|NB|).
C
A
B
D
E
F
Figure 3.17: 6-node HELPER networkfor quantitative evaluation.
Table 3.4: Evaluation parameters.
Parameters Values
Payload Size 200 BytesPacket interval 100 msPhysical Layer LoRaBandwidth 125 kHzEini 25 JDuration 120 min
The parameters used in the evaluation are depicted in Table 3.4. As discussed earlier,
62
CHAPTER 3.
LoRa consumes extremely low power. This means that for a realistic battery to drain completely,
the evaluation may have to be run over multiple days. To save time and yet without loss of rigor,
a virtual energy level is used to evaluate the HELPER network so that one can see the network’s
behavior in experiments lasting less than 120 min. Each node is assumed to start at a total energy of
25 J and is depleted as each packet (control or data) is transmitted.
In the first experiment, F is set as the destination (would represent ERC in a real-life
scenario) and HELPER A and B are the source nodes. As shown in Table 3.4, packets are generated at
the source node at a constant rate and it has to choose appropriate routes to reach the destination. The
first metric evaluated is the minimum residual energy (Eminr ) among all HELPERs in the network. In
other words, at any given time instant t, the residual energy value of the HELPER that has consumed
the highest energy is plotted. The second metric under evaluation is the normalized throughput of the
network calculated with respect to observed point-to-point link throughput (T hl) and can be referred
to as,
T hnet =T hnet
T hl(3.8)
First, let’s look at the initial 14 minutes of the experiments. As you can see in Fig. 3.18,
the Eminr in both cases are the same since SEEK operates similarly to the greedy algorithm in this
stage even after gathering information from immediate neighbors. This is because at the beginning
most of the possible next hops have similar parameters including backlog length and residual energy.
Additionally, it can be seen from Fig. 3.19 that during the same period, the greedy algorithm seems
to marginally outperform SEEK. This can be attributed to the overhead involved in SEEK to compute
the optimal next hop from the gathered information. This marginal superiority is short-lived as SEEK
starts learning about the environment and begins to exploit spatial diversity to choose multiple paths
to the destination. This provides HELPER network with two advantages, (i) the energy consumption
is evenly spread between nodes and (ii) higher throughput is achieved. Accordingly, from Fig. 3.19,
it is evident that the death of the first node in the aggressive greedy algorithm happens much earlier
than the death of the first node in SEEK. This provides a proof-of-concept that SEEK can be applied
to maximize the network lifetime in a distributed manner.
Next, to extend the experiments further, the performance of SEEK is evaluated while
increasing the number of sessions in the network to 4. This is to evaluate if SEEK can adapt to
multiple traffic partners in the network which is expected behavior in a large distributed network.
These sessions include A→ F , B→C, C→ E and F → A and are chosen to ensure no source in a
session has it’s destination via a direct link (i.e. destination is not the source’s immediate neighbor).
63
CHAPTER 3.
Time (min)0 20 40 60 80 100 120
Min
imu
m r
esid
ual
en
erg
y (J
ou
les)
0
5
10
15
20
25
Greedy Geographic RoutingSEEK
First greedy nodedies
First node of SEEKdies
SEEK starts to share traffic load tomaximize network lifetime
Figure 3.18: Maximum energy consumed by anode.
Time (min)0 20 40 60 80 100 120
No
rmal
ized
th
rou
gh
pu
t
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
Greedy Geographic RoutingSEEK
First greedy node dies
First node of SEEK dies
28% Improvement
Figure 3.19: Normalized throughput of thenetwork.
Each source in the session is set up to generate packets at a constant rate as mentioned in Table 3.4.
Both residual energy of each node and packets received are constantly monitored. First, the network
lifetime is analyzed which is defined as the duration of operation until the first node in the network
dies. This is important for such emergency networks as the death of a node would imply unconnected
users. Figure 3.20 shows how SEEK outperforms the greedy algorithm regardless of the number of
sessions. The experiments show an improvement of up to 53% in terms of network lifetime. One
interesting finding is that the network lifetime seemed to increase with the increase in sessions which
might be counter-intuitive at first sight. Further evaluation using Fig. 3.21 will reveal that the small
network is saturated even with two sessions in the network as portrayed by the throughput decline.
This implies that more collision may occur at the MAC layer leading to a larger backoff and lower
throughput as the number of sessions increase. In a saturated network, the overall throughput even
while operating for a longer period of time is better for SEEK compared to the greedy algorithm. To
further substantiate the importance of network lifetime, the percentage increase in packets delivered
by SEEK as compared to the greedy algorithm is depicted in Fig 3.22. This keeps increasing as the
number of sessions in the network grows which can be related to delivering critical information from
survivors to the ERC during the aftermath of the disaster.
Finally, the average delay per packets is evaluated as the number of sessions in the network
increases. To accomplish this, each session is set to transmit 100 packets while keeping the rest of
the setting similar to the earlier experiment. As expected, the delay per packet of both the schemes
increases as the number of sessions in the network increases due to congestion. The more critical
64
CHAPTER 3.
observation from Fig. 3.23 is that the delay incurred by packets serviced using SEEK is up to 40%
less than greedy algorithm especially when the traffic increases (3 sessions). This is because SEEK
is able to use multiple paths to distribute traffic spatially among nodes to reduce congestion at the
bottleneck nodes. This is further substantiated by the fact that the advantage in terms of lower delay
diminishes as the network saturates (4 sessions) since all the nodes are involved in either case (greedy
and SEEK) leaving no extra nodes for SEEK to distribute traffic load.
Number of sessions1 2 3 4
Net
wo
rk li
feti
me
(min
)
0
20
40
60
80
100
120
SEEKGreedy Geographic Routing
53% improvement
Figure 3.20: Network lifetime vs No. ofsessions.
Number of sessions1 2 3 4
No
rmal
ized
th
rou
gh
pu
t
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
SEEKGreedy Geographic Routing
Figure 3.21: Normalized throughput vs No. ofsessions.
Number of sessions1 2 3 4
Per
cen
tag
e o
f in
crea
se in
pac
kets
del
iver
ed b
y S
EE
K
0
10
20
30
40
50
60
70
80
90
Figure 3.22: Analysis of packet delivery.
Number of sessions1 2 3 4
Ave
rag
e D
elay
(s)
0
5
10
15
20
25
30
35
SEEKGreedy Geographic Routing
40% decrease
Figure 3.23: Average delay vs No. of sessions.
Overall, the experiments showed how nodes in SEEK share information among each other
using the control packets which is then used to perform cross-layer optimization to choose optimal
65
CHAPTER 3.
routes that ensure all nodes share the load of the traffic to maximize the network lifetime. The
improvement in the performance may be more significant on a larger network consisting of hundreds
of nodes. Here, HELPER has been prototyped and a small yet effective testbed has been set up with
a limited number of nodes to perform extensive testing. The results provide proof-of-concept that the
proposed HELPER network can be deployed in the near future to enable off-the-grid connectivity.
3.6 Summary
In this chapter, a complete end-to-end solution to enhance and enable public safety commu-
nication systems has been proposed, prototyped and demonstrated to establish the proof-of-concept.
The proposed HELPER uses heterogeneous wireless communication techniques; (i) WiFi which en-
ables EU to connect to the HELPER like any WiFi access point thereby ensuring easy and widespread
adoption, and (ii) LoRa, that provides extremely low power, long-range wireless link to implement
the ad hoc operation. The HELPER network is used to set up a completely self-sustained network
that does not require the support of any traditional communication infrastructure like cell towers or
satellite. The HELPER network is designed to serve a dual purpose; (i) enable affected individuals to
stay connected and maintain situational awareness, and (ii) equip authorities to remotely monitor the
situation, provide assistance and warnings in an efficient manner.
The proposed solution provides connected EU with live map updates to share the location
of known resources. It enables text messages between community members and equips EU with an
alternative to traditional 9-1-1 like emergency calls. Similarly, it provides ERC with the capability
to monitor the network connectivity, manage resource sharing information and send out ALERT
messages to connected users. Additionally, numerical evaluations using HELPER testbed showed
up to 53% improvement in network lifetime and up to 28% improvement in network throughput as
compared to a greedy scheme that routes using shortest path. All these demonstrated capabilities will
enhance the state-of-the-art public safety response system.
66
Chapter 4
VL-MAC: Opportunistic MAC Protocol
for Visible Light ad Hoc Network
The proliferation of wireless entities including IoT devices, multimedia devices among
others is causing significant growth in demand for bandwidth and spectrum resources. While new
portions of the RF electromagnetic spectrum are being made available and are increasingly leveraged
to meet this demand, RF communications inevitably suffer from problems including spectrum crunch,
co-channel interference, vulnerability to eavesdroppers, among others [101, 102] in the new 5G era.
Moreover, RF-based communications are not always permitted because of the potentially dangerous
effect of Electromagnetic Interference (EMI), which occurs when an external device generates
radiations that affect electrical circuits through electromagnetic induction, electrostatic coupling, or
conduction. For example, cellular and WiFi emissions are prohibited in airplanes during takeoff and
landing because electromagnetic radiations can interfere with onboard radios and radars; electronic
equipment can emit unintentional signals that allow eavesdroppers to reconstruct processed data at a
distance by means of directional antennas and wideband receivers.
Optical communications have attracted significant attention as a valid alternative over
legacy RF-based wireless communications. Optical communications are classified into two main
categories, fiber-based and Optical Wireless Communications (OWCs). Fiber-based systems are
frequently employed in the backbone network cabling because of their robustness, reliability, and
high-rate in delivering large amounts of data. OWCs are rapidly growing in popularity as an emerging
and promising wireless technology capable of high-speed data transfer over short distances [103]
[104]. An optical wireless-based system relies on optical radiation to deliver information in free
67
CHAPTER 4.
space, with wavelengths included in the Infrared Radiation (IR), visible-light, and Ultraviolet (UV)
bands. In the last decades, OWCs have been deployed in medium to long communication distance
environments, e.g., OWC has been applied for inter-chip connection as short-range transmission
while VLC found applications in medium-range indoor wireless access. Moreover, inter-building
connections can be established using IR communications whereas ultraviolet communications have
been recently adopted in outdoor non-line-of-sight scenarios and specifically for ad hoc and Wireless
Sensor Networks (WSNs). Recently, satellite communications and deep-space applications based on
OWC have been demonstrated, especially for military applications [105]. In particular, the recent
rapid increase in the use of LEDs for lightning has paved the way for the development of new
communication systems based on leveraging visible light as a communication medium. That is,
LEDs can act as illumination devices as well as information transmitters at the same time, thus
delivering data by digitally modulating the emitted light beam intensity at a very fast rate [106].
The exploration of VLC has been limited to various point-to-point applications including
setting up Li-Fi [7] networks using smart lights, among others. In this context, several topologies
such as peer-to-peer, star and broadcast have been considered to design networking protocols. In
this work, the emphasis is on enabling the use of VLC for ad hoc networking in military and civilian
applications. LANETs is expected to contribute considerably to the upcoming IoT revolution in
both indoor and outdoor spaces and this chapter explores how this can be achieved. The major
contributions of this chapter can be summarized as follows,
• The vision of using LANET for military and civilian applications requiring short-range,
low-latency, high-data rate links have been discussed in great details and the crucial role of
cross-layer technology in realizing it has been established.
• Accordingly, this process has been initiated by developing a mechanism to perform neighbor
discovery for LANETs taking into account the challenges introduced by directionality.
• A novel multi-utility based opportunistic MAC protocol, VL-MAC is designed to maximize
the throughput of LANET improving the probability of establishing links and promoting
the percentage of full-duplex communication in the network. The proposed MAC protocol
overcomes deafness, blockage and hidden node problem.
• Extensive simulations are performed to demonstrate improvements achieved by the proposed
VL-MAC protocol.
68
CHAPTER 4.
The rest of this chapter is organized as follows. In Section 4.1, the concept of LANET has
been defined and the envisioned applications have been discussed in detail. Section 4.1 also provides
a high-level comparison between LANETs and traditional Mobile Ad Hoc Networks (MANETs)
and discusses the major design challenges that need to be overcome. As a first step in the designing
LANET to support upcoming 5G networks, this chapter provides the first MAC protocol developed
for LANETs. To accomplish this, some of the recent efforts to design MAC protocol for VLC and
their shortcoming are discussed in Section 4.2. A neighbor discovery scheme for LANETs has
been presented in Section 4.3. Next, the detailed design of the proposed MAC protocol for LANET,
VL-MAC is provided in Section 4.4. Thereafter, the effectiveness of the proposed solution is proven
using extensive simulation in Section 4.5. Finally, Section 4.6 provides the summary of the chapter.
4.1 LANET: Visible-Light Ad Hoc Networks
LANETs refer to an infrastructure-less mobile ad hoc network where Visible Light Nodes
(VLNs) are wirelessly connected using multi-hop visible light links, capable of configuring their
protocol stacks in a cross-layer, online and software-defined manner, adapting to various networking
environments and demands.
4.1.1 Envisioned Applications
LANETs have a great potential for enabling a rich set of new civilian and military applica-
tions, as illustrated in Fig. 4.1, ranging from low-latency high-bandwidth indoor communications and
outdoor intelligent transportation networking, to highly secure Lower Probability of Intercept/Lower
Probability of Detection (LPI/LPD) operations under high network density and jamming conditions,
among others. Some examples of these applications are discussed below.
Intelligent Transport Systems. One of the most promising outdoor applications of
LANETs is for ad hoc vehicular communications [107] [108], including Vehicle to Infrastructure
(V2I), Infrastructure to Vehicle (I2V) and Vehicle to Vehicle (V2V) communications. LANETs can be
employed to design intelligent transport systems with better road safety. For V2V, a communication
link can be established using head and tail lights or photo-diodes and image sensors at the receiver
side, while for V2I the urban infrastructures (e.g., traffic lights, street lights) can be utilized for
transmitting useful information related to current circulation of traffic including vehicle safety, traffic
information broadcast and accident signaling. Additionally, in Vehiclular Ad Hoc Network (VANET),
69
CHAPTER 4.
Figure 4.1: LANETs employed for civilian and military applications.
the network topology is highly dynamic and often large-scale. This makes realizing visible-light
VANETs more challenging because of the limited FoV, and the relatively short transmission ranges
[109]. Moreover, different from legacy RF-VANETs, the quality of visible links can be significantly
degraded by weather conditions, including fog and rain, among others.
Internet of Things. The vision of IoT anticipates that large amounts of mobile embedded
devices and/or low-cost resource-constrained sensors will communicate with each other via the
Internet. To allow networking among a massive number of devices, the communication system
must be ubiquitous, low-cost, and bandwidth and energy-efficient. Infrastructure-less LANETs are a
promising choice for communication in the IoT because of its inherent advantages as discussed in
Section 4.1.2, e.g., orders of magnitude available bandwidth, reusing ubiquitously existing lighting
infrastructure, low-cost front-end devices, among others. Therefore, LANETs can easily enable
a wide range of IoT services, such as localization, smart home, smart city, air/land/navy defense,
70
CHAPTER 4.
among others.
D2D Communications. D2D communications are rapidly emerging in recent years [110]
and is expected to be one of the key services provided by 5G service providers to offset the cost of
deploying additional infrastructures. Beyond the crowded RF spectrum, LANETs are a promising
candidate to support D2D communications. VLC-D2D applications [111] can use LEDs and Photon
Detectors (PDs) or Liquid Crystal Display (LCD) screens and camera sensors. The ubiquitous pres-
ence of LCD screens and surveillance cameras in urban environments creates numerous opportunities
for practical D2D applications since information can, for example, be encoded in display screens
while camera sensors can record and decode data using image processing techniques [112].
Indoor Positioning. Recently proposed VLC-based indoor localization schemes have
shown improved performance, in terms of accuracy, given the higher density of LEDs as compared
to Wi-Fi access points [113]. To set up a light-weight indoor positioning network, LANET-enabled
sensors can be organized to form an ad hoc network with a tree-like structure (i.e., having a sensor
connected to a Local Area Network (LAN) as the root node) and a simplified protocol stack only
providing basic data transfer and routing functionalities that can be run on devices with limited
resources.
RF-Suppressed Applications. LANETs can provide a reliable and accurate solution
for data transmission in scenarios where RF communications are suppressed or prohibited, like
hospital and climbing/landing airplanes. For example, wireless technology is applied in hospitals for
updating information related to patient records, collecting data in a real-time way from handheld
patient devices, detecting changes in a patient’s condition, and also for observing medical images
via medical equipment (e.g. ultrasound). There, security and safety are essential to maintain the
confidentiality of patient records and to ensure that only authorized personnel has access to the data
being transferred wirelessly while limiting the interference to those interference sensitive medical
devices like EMI.
Military Applications.
In the last decades, the most common optical/visible light communication for military
applications employ IR short-range transmissions [114]. In recent years, the emerging of VLC has
shown promising advancements making possible the extensive deployment of VLC for military
communication strategies [115]. The use of VLC is turned out to be beneficial in the tactical field
with enhanced network capacity and better resistance against adversary jamming, and the research
is focused in this direction by military organizations and defense companies. Novel and advanced
visible light-based military applications include personal area networks, warfighter-to-warfighter
71
CHAPTER 4.
communication, vehicular networks, underwater networks, and space applications including inter-
satellite and deep-space links. For example, in underwater, autonomous vehicles will be able to
self-organize in a LANET to exchange high-data rate traffic via visible light carriers as a high-rate
short-range alternative to acoustics; in ground, marine soldiers can self-organize in a LANET in case
of RF interference and be connected to command; finally, in air/space LANETs, nanosats can be
connected to a satellite station via VLC and be relay-assisted by other nanosats when in proximity in
a delay-tolerant ad hoc network.
4.1.2 LANETs vs Traditional MANETs
Similar to traditional RF-based MANETs, LANETs also have the ability to self-organize,
self-heal, and self-configure. Because of the unique characteristics of visible light compared to RF
signals, in LANETs visible light point-to-point links require mutual alignment of transmitters and
receivers given the directivity of light signal propagation, which is not easy to obtain with mobile
nodes; communication links in LANETs can be easily interrupted by intermittent blockage since
light does not propagate through opaque materials. Table 4.1 summarizes the differences between
LANETs and MANETs, in terms of critical aspects including transmitter and receiver, network
capacity, channel modeling, efficiency, and security, among others.
Transmitter and Receiver. In MANETs, the front-end components of each node are
typically antenna-based, operating at high frequency. In contrast, simple LED luminaires and
PDs or imaging sensors are typically adopted as transmitters and receivers in LANET. They are
relatively simple and inexpensive devices that operate in the baseband and do not require frequency
or sophisticated algorithms for the correction of RF impairments, e.g., phase noise and IQ imbalance
Property MANET LANET
Power Consumption Medium LowBandwidth Regulated, Limited Unlimited (400nm∼ 700nm)Infrastructure Access Point Illumination/Signaling LEDEMI Yes NoSecurity Reduced HigherMobility High ReducedLine of Sight Not required Strictly requiredTechnology Mature Early stageCoverage - Range Medium - Long Narrow - Short
Table 4.1: Comparison between LANETs and MANETs.
72
CHAPTER 4.
[116]. As a consequence, SWaP (size, weight, and power) and cost of front-end components involved
in LANET systems are often lower than equivalent MANET systems.
Spectrum Regulation. The visible light spectrum is mostly unused for delivering infor-
mation, which implies potential high throughput and an opportunity to alleviate spectrum congestion,
particularly evident in the ISM band. The bandwidth available in the visible light portion of the
electromagnetic spectrum is considerably larger than the radio frequency bandwidth, which ranges
from 3 kHz to 300 GHz. The availability of this mostly unused portion of the spectrum provides
the opportunity to achieve high data rates through low-cost multi-user broadband communication
systems. VLC solutions could be complementary to traditional RF systems and alleviate the spectrum
congestion that especially impacts the ISM band.
Network Capacity. In MANETs, all the nodes usually operate in a shared wireless
channel with a single radio at each node, where the number of channels, the operating frequency, and
maximum transmit power are stringently regulated [117], and consequently, the network capacity is
unavoidably limited and affected by co-located networks. LANETs, instead, can rely on a substantial
portion of the unlicensed and currently unregulated spectrum as described above, which have the
potential to make significant capacity available for networked operations.
Spatial Reuse. Visible light cannot pass through opaque objects, thus resulting in low
penetration. Moreover, in contrast to omnidirectional RF communications, because of predefined
limited Field Of View (FOV) of LEDs, visible light links are typically directional. This provides
a higher degree of spatial reuse with respect to omnidirectional transmissions typically used in RF.
For example, since light cannot propagate outside of a closed room, there is no interference from
VLC signals in adjacent rooms. Because of this unique characteristic of VLC, most existing MAC
and network layer MANET protocols cannot be directly applied to LANETs and hence need to be
redesigned, including neighbor discovery and route selection, among others.
Security. Since they operate in dynamic distributed infrastructure-less configurations
without centralized control, MANETs are vulnerable to various kinds of attacks, ranging from
passive attacks such as eavesdropping to active attack such as jamming [118]. Differently, in
LANETs, the inherent security property that stems from the spatial confinement (low penetration and
restricted FOVs) of light beams, will enable secure communications since jammers or eavesdroppers
can be easily spotted than in legacy RF communication.
Costs. As discussed above, LANETs are more cost-efficient than MANETs because
of much simpler front-end devices (e.g., LEDs, PDs) compared to RF solutions for transmitting,
sampling and data processing. Moreover, nodes in MANETs are usually battery-powered to enable
73
CHAPTER 4.
communications in the absence of a fixed infrastructure. The sensing unit, the digital processing
unit and the radio transceiver unit are the main consumers of the battery energy, and therefore
more sophisticated energy-efficient algorithms, e.g., energy-efficient MAC or routing schemes [119]
[120], are needed, which are however challenging in such resource-limited and infrastructure-less
MANETs. Differently, LEDs used as transmitters in LANETs highlight themselves by high energy
efficiency, longevity, and environment-friendly factor enabled by recent tremendous advances in
LED technologies [116]. Moreover, VLC manifests its low-power baseband processing property,
which further results in low-cost LED devices compared to high-frequency passband RF front-end
antennas.
4.1.3 Main Design Challenges
VLC has found many applications in short-, medium-, as well as long-range communi-
cations in the last decade. These include inter-chip connections, indoor wireless access, as well as
satellite and deep-space applications, among others [116, 105]. However, while there has been a sig-
nificant advancement in understanding efficient physical layer design for visible-light point-to-point
links, the core problem of developing efficient networking technology specialized for visible-light
networks is substantially unaddressed. One of the main challenges is that VLC relies on optical
radiations to deliver information in free space through a substantial portion of the unregulated spec-
trum between 400 and 800 THz, with corresponding wavelengths in the IR, visible light, and UV
bands [116]. This makes VLC substantially different from RF-based communications in terms of
communication range, transmission alignment and shadowing effect, ambient light interference and
receiver noise, and VLC ad hoc networking, among others.
Short Communication Range. Because of the limited propagation range of short-
wavelength signals, the transmission range of VLC is relatively short (typically a few meters),
compared to tens of meters for WiFi [121, 122]. When increasing the link distance, for a given
desired level of reliability the achievable data rate decays sharply, thus limiting the number of
applications where VLC high data rate transmissions can be employed.
Transmission Alignment and Shadowing Effect. Because of the low penetration of light,
while visible light signals in adjacent rooms do not interfere with each other, this also presents several
limitations. First, the transmitter and the receiver must be aligned to each other, especially for Line
of Sight (LOS) short distance communications with small FOVs, and this is challenging especially if
LANET nodes are moving [123]. Second, VLC link quality can be significantly degraded because of
74
CHAPTER 4.
shadowing effects caused by obstructing objects, e.g., mobile human bodies [124].
Ambient Light Interference and Receiver Noise. Noise and interference in VLC are
mainly caused by exposure of the receiver to direct sunlight and by the presence of other sources of
illumination (i.e. other LED sources, fluorescent and bulb lamps) [125] [126] that cause shot noise
and consequently decrease the SNR. In turn, the receiver can be affected by thermal noise caused by
the pre-amplification chain.
Lack of Well-established Channel Models. Factors that affect the performance of visible
light links include free space loss, absorption, scattering, scintillation noise induced by atmospheric
turbulence and alignment between transmitters and receivers, among others [127]. Different from RF,
channel modeling for visible light links is still largely based on preliminary empirical measurements,
especially for outdoor Non-Line-Of-Sight (NLOS) environments [128, 129]. The applicability of
existing theoretical channel models in the design of LANETs still needs to be verified and tested in
different transmission media [130].
VLC Ad Hoc Networking. Existing work on VLC mostly focuses on increasing the
data rate for a single VLC link using advanced modulation schemes [131, 132, 133, 134, 135,
136]. However, VLC ad hoc networking with a large number of densely co-located VLC links
(i.e., LANETs) is still substantially unexplored because of the unique characteristics of VLC,
including intense modulation/direct detection (IM/DD) channel model, FOV based directionality, low-
penetration, among others. To the best of the author’s knowledge, there are no existing architectures
and protocols designed specifically for LANETs..
As mentioned earlier, this work focuses on exploiting VLC for ad hoc networking in
military and civilian applications that have been discussed in detail in the previous section. As an
initial step towards making LANETs a reality, this chapter provides a design of a LANET-specific
MAC that mitigates specific challenges put forth by LANETs.
A key distinguishing feature of VLC is directionality. While it enables better spatial
re-use, directionality is the direct reason for some of the major challenges experienced in LANETs.
The classical challenges like hidden node problem are amplified by transmission directionality,
since the control packets such as Clear-to-send (CTS) transmitted by a receiver may not be received
by nodes because of limited FOV. When a receiver is oriented towards a certain spatial sector and is
therefore unable to receive from all the remaining sectors, it is referred to as deafness. Thus, a node
may try to initiate communication with its neighbor who is experiencing deafness with respect to the
node, leading to additional delays during the contention phase. Another unique challenge of LANET
is the sudden communication discontinuity which may happen during the contention phase; thus,
75
CHAPTER 4.
trying to access one particular neighbor may not be the most efficient way to forward packets. This
problem is referred to as blockage. Some challenges of LANET are similar to the ones experienced
by directional RF networks. The list of instantaneous neighboring nodes may change depending on
the FOV. Unlike typical RF transceiver systems equipped with a single antenna to transmit or receive,
VLC devices are usually equipped with a LED for transmission and PD for reception making these
devices inherently capable of full-duplex communication. Therefore, network protocols designed
for LANETs should be able to take advantage of full-duplex links to improve network throughput.
The unique characteristics of VLC impose the need for cross-layer design as shown in Fig.
4.2 to address these challenges [9]. Here, the process is initiated by designing a novel opportunistic
MAC protocol that optimizes the throughput of LANETs using a divide-and-conquer approach aimed
at achieving the following objectives:
• Maximize the probability of establishing a link in a given direction while overcoming chal-
lenges caused by the hidden node problem, deafness, and blockage;
• Improve the probability of full-duplex communication;
• Maximize the amount of spatial re-use.
Cross-Layer Controller
ApplicationLayer
TransportLayer
NetworkLayer
Data LinkLayer
LANET PROTOCOL STACK
Firmware
Custom Logic
Interp DACDUCLED
Driver
Decim ADCDDC TIA
Signal Processing Chain
LANET HARDWARE
LED
PDPhysical
Layer
Figure 4.2: Architecture of a LANET node.
4.2 Related Works
The study of MAC solutions for VLC is still in its infancy and even more limited is the
attention given to MAC in the context of ad hoc networks. Unfortunately, the few existing MAC
schemes designed for point-to-point VLC are not easily extendable to LANETs as they do not
76
CHAPTER 4.
consider some unique challenges and opportunities related to VLC. Some of the existing MAC
protocols are discussed below.
CSMA-based Channel Access [137, 138, 139]. In [137], the authors propose a full-
duplex MAC protocol with Self-Adaptive minimum Contention Window (SACW) that delivers
higher throughput from the central node to the terminal nodes in a star topology. The proposed
algorithm still uses the basic slotted CSMA/CA mechanism as in [140] with adaptive contention
window. The objective of SACW MAC is to allow the central node to monitor the data traffic to
increase the probability of full-duplex operation. The authors of [138] also propose a high-speed
full-duplex MAC protocol based on Carrier Sense Multiple Access/Collision Detection (CSMA/CD)
by considering a start topology with Access Point (AP) at the center and multiple terminal nodes
trying to communicate with the AP. Another example of VLC using CSMA/CA is in [139], which
uses LED to transmit and receive to reduce hardware cost and size. This work uses LED charged in
reverse bias to receive the incoming light.
Cooperative MAC [141]. A cooperative MAC protocol is proposed in [141] to reduce
latency and for on-demand error correction. The sender and receiver will initiate a cooperative
mechanism to find relay nodes when the direct link does not provide the required bandwidth to meet
the QoS requirement. Once the cooperative mode is initiated, the sender broadcasts a RelayRequest.
Nodes within range save the sender’s identification number. Next, the destination broadcasts a
RelayRequest. Nodes that receive both RelayRequests will broadcast its information to sender and
destination if the node decides to be a relay. The relay overhears the sender’s packets and saves them
till an Acknowledgment (ACK) is received from the destination. If the ACK is not received, the relay
transmits the saved packets to the destination.
Orthogonal Frequency-division Multiple Access (OFDMA) [142, 143, 144, 145]. Re-
cently, the OFDM used in the Physical (PHY) layer of VLC has been extended to enable multi-user
access through Orthogonal Frequency Division Multiple Access (OFDMA). In [142], authors com-
pare the BER performance, receiver complexity and power efficiency of two multicarrier-based
multiple access schemes namely, Optical Orthogonal Frequency Division Multiplexing Interleave
Division Multiple Access (O-OFDM-IDMA) and Optical Orthogonal Frequency Division Multiple
Access (O-OFDMA). The authors of [143] evaluate a self-organizing interference management pro-
tocol implemented inside an aircraft cabin. The goal of the work is to allocate time-frequency slots
(referred to as chunks) for transmitting data in an Intensity-Modulation Direct-Detection (IM/DD)-
based OFDMA-Time Division Duplex (TDD) systems. Another OFDMA technique for indoor
VLC cellular networks is analyzed in [144] using Direct-Current Optical OFDM (DCO-OFDM)
77
CHAPTER 4.
as multi-user access scheme. In [145], the authors propose a heuristic subcarrier reuse and power
redistribution algorithm to improve the BER performance of conventional Multiple Access Discrete
Multi-Tones (MA-DMT) used for VLC.
Code Division Multiple Access (CDMA) [146, 147, 148, 149, 150, 151]. There have
been several contributions aimed at employing CDMA in VLC. A system using Multi-carrier
CDMA (MC-CDMA) along with OFDM platform is proposed in [146]. The proposed design uses
Polarity Reversed Optical OFDM (PRO-OFDM) to overcome the inherent light-dimming problem
associated with using CDMA with visible light. In this design, a unipolar signal is either added
or subtracted to the minimum or maximum current respectively in the LED’s linear current range
to provide various levels of dimming. In [147], the authors discuss how Gold sequences and
Wash-Hadamard sequences can be adapted for VLC. Optical Orthogonal Codes (OOC) [148]
comprising of sequences of 0s and 1s have also been explored as a prime candidate to establish
Optical Code-Division Multiple Access (OCDMA) for visible light communication. Since as the
number of users increases in the system, it becomes challenging to generate OOC for each user,
Random Optical Codes (ROC) has been proposed as an alternative, even though they do not provide
optimal performance [149, 150]. There have also been efforts to combine Color-Shift Keying (CSK)
modulation and OCDMA to enable simultaneous transmission to multiple users [151].
QoS-Based MAC. In [152], the authors propose a QoS based slot allocation to enhance
the broadcasting MAC of IEEE 802.15.7 standard. They use a super frame structure similar to the
standard. When a new channel wants to join the AP, it sends a traffic request to the access point
along with its QoS parameters (data rate, maximum burst traffic, delay requirements, and buffer
capacity). Optical wireless MAC (OWMAC) [153] is a Time Division Multiple Access (TDMA)
based approach aimed to avoid collision, retransmission, and overhead due to control packets. In
OWMAC, each node reserves a time slot and advertises the reservation using a beacon packet.
OWMAC also employs Error-Correction Code (ECC) in their ACK to ensure that retransmission are
reduced to corrupted ACK packets. This protocol is designed to handle start like topologies.
MU-MIMO [154, 155, 156, 157, 158, 159, 160]. An alternative method uses multiple
LED arrays as transmitters to serve multiple users simultaneously [154, 155]. In contrast to the RF
counterpart, the VLC signal is inherently non-negative leading to the necessity of modifying the
design of the Zero Forcing (ZF) precoding matrix. In [155], a ZF precoder is chosen in the form
of the specific generalized inverse of the channel matrix known as the pseudo-inverse. The authors
of [154] recognize that the pseudo-inverse may not be the optimal precoder. Accordingly, they
design an optimal ZF precoding matrix for both the max-min fairness and the sum-rate maximization
78
CHAPTER 4.
problems. Block Diagonalization (BD) algorithm [156] has also been used to design the precoding for
Multi-User Multiple-Input Multiple-Output (MU-MIMO) VLC system [157] to eliminate Multi-User
Interference (MUI) and its performance has been evaluated in [158]. Finally, Tomlinson-Harashima
Precoding (THP) [159] has been utilized in [160] to achieve better BER performance compared to
the block diagonalization algorithm in VLC systems.
MAC protocols [137, 138, 142, 143, 144, 145] that are designed for centralized operation
in a star topology are not easily extensible to LANETs. Cooperative operations like in [141] can be
employed in LANETs but cannot be the primary MAC protocol used to negotiate reliable medium
access. Techniques based on CDMA or MU-MIMO are suitable for centralized networks as it may
be complex to negotiate different codes for each link in a distributed network. Similarly, QoS-based
techniques can be used to improve a stable MAC protocol that has been primarily designed to
overcome inherent problems of LANETs such as deafness, blockage and hidden node problem.
4.3 Neighbor discovery
Consider a multihop LANET with N static VLNs modeled as a directed connectivity graph
G(U,E), where U = {u0,u1, ...,uN} is a finite set of VLN of the graph, and (i, j) ∈ E represents
a feasible unidirectional wireless link from node ui to node u j (for simplicity, they are referred
to as node i and node j) representing neighboring relationships, i.e., there is a feasible link if the
nodes are close enough. In LANET, each node consists of LED luminaires and PDs adopted as
transmitters and receivers, respectively. Since the transmissions are directional, the directions to
which the FOV of each node can be set to are represented by Ns equal sectors s ∈ S . The FOVs of
typical LEDs and PDs can vary from ±10◦ to ±60◦ [161, 162], e.g. Vishay TSHG8200, OSRAM
LCW W5SM Golden Dragon and Vishay PD TESP5700. Here, for the sake of simplicity, but without
loss of generality, the FOV for both LED and PD is chosen to be ±22.5◦, leading to eight sectors.
This can be easily extended according to the FOV of the hardware available on specific VLN. It
is also assumed that a VLN is capable of directing its FOV to all the Ns sectors when required for
transmission and reception. This is possible with multiple LEDs and PDs, that can be used depending
on which sector the nodes want to access (only one sector of a node is activated at any given time).
In non-ideal scenarios, interference mitigation techniques [163] can be employed to reduce the
interference between sectors. The neighbors are grouped into sectors based on their location which
can be provided when the network is deployed or learned by exchanging of control packets. Thus,
the superset of neighbors for node i consists of the set of neighbors in each sector represented as
79
CHAPTER 4.
Table 4.2: Summary of MAC protocols for VLC.
MAC Protocol Medium AccessMethod
Topology/Operation Modes Other Comments
IEEE 802.15.7 [140] CSMA/CAPeer-to-peer, star,broadcast Standardization for VLC
SACW MAC [137] CSMA/CA Star Full-duplexLin et al [138] CSMA/CD Star Full-duplexSchmid et al [139] CSMA/CA Peer-to-peer LED-to-LEDCooperative MAC [141] CSMA/CA Peer-to-peer Cooperative relay
Broadcasting MAC [152] TDMA BroadcastFrame synchronization andsupports QoS
OWMAC [153] TDMAStar, with unicast,broadcast, & multicast 84 Mb/s data rates
Dang et al [142] OFDMA StarComparison of O-OFDMA& O-OFDM-IDMA
Ghimire et al [143] OFDMA-TDD StarSelf-organizinginterference management
Chen et al [144] DCO-OFDMIndoor downlinktransmission
Spectral efficiency of5.9 bits/s/Hz
Bykhovsky et al [145] DMT StarInterference-constrainedsubcarrier reuse
Shoreh et al [146]MC-CDMA withPRO-OFDM Star
Handles dimming usingPRO-OFDM
He et al [147] OCDMA with OOC Peer-to-peer, starBipolar-to-Unipolarencoding and decoding
Gonzalez et al [149] OCDMA with ROC Peer-to-peer, starSpecific design of OOC,higher complexity
Chen et al [151] OCDMA with CSK Peer-to-peer, starMobile phone cameraused as receiver
Yu et al [155] MU-MISO Cooperative broadcast ZF algorithm usinggeneralized inverse
Pham et al [154] MU-MISO Cooperative broadcast ZF algorithm usingoptimal precoding
MU-MIMO (BD) [157] MU-MIMO StarPrecoding using BDalgorithm
MU-MIMO (THP) [160] MU-MIMO StarPrecoding using THPalgorithm
N B i ∈ {N B i1,N B i
2, ...,N B iNs}, where N B i
s , { j : (i, j) ∈ E} is the neighbor of node i in sector s.
Let the traffic in the network consist of multiple sessions q = 1,2, ...,Q, characterized by
the source-destination pairs. In this chapter, the feasible next hop for a session is defined as any
neighbor that is closer to the destination and is termed as forward progress. In this context, each
session q in node i belongs to one or more sector queue sets q ∈ Q si (q can be a component of more
than one sector queue sets) such that the sector contains neighbors that ensure forward progress for
packets in a queue. This information will be used by the VLN while choosing an optimal sector to
80
CHAPTER 4.
forward packets. The arrival rate of each session q ∈ Q si at node i is given by λi
q(t), and characterized
through the vector of arrival rates Λ. The VLNs in the network are assumed to be synchronized with
each other using techniques like GPS based clock synchronization. The time spent listening to each
sector is called sector duration (tsec) and this forms a sector slot as shown in Fig. 4.3. The sector
slot is further divided into multiple Control Micro-Slots (CMS). Control packets are transmitted
only at the beginning of a CMS. The duration of a CMS is set such that transmission of a control
packet can be completed in one CMS. A set of Ns sector slots forms a super-slot. VLNs have two
operational states; Synchronous Idle State (S-IDLE) and Transceiving State (TR). In S-IDLE, nodes
sequentially listen in each sector following a fixed pattern. In this way, a VLN that has to transmit in
a given sector knows the appropriate sector slot when the idle neighbors (in the given sector) will be
listening, thus mitigating the effect of deafness. The channels used by the LANET are divided into
CCC and Data Channel (DC) using, for example, orthogonal CDMA codes.
Super-slot
Sector slot NsSector slot 1
Control micro-slot (CMS)
. . .
Figure 4.3: Super-slot structure.
Neighbor discovery is critical for any ad hoc network. Due to the unique characteristics
of VLC, a neighbor discovery mechanism has to be designed specifically for LANETs. Each VLN
needs to know the neighbors that correspond to each sector. Unlike some RF directional network,
LANETs do not have the option to operate in omni-directional mode when required. This is true even
when LEDs and PDs are available in each sector since they may not have a dedicated receive and
transmit circuitry (due to cost) corresponding to each LED/PD. During the neighbor discovery phase,
it is challenging for nodes to communicate with each other due to deafness. This also invokes the
need to cooperatively share information among nodes to enable faster neighbor discovery. Therefore,
synchronization and cooperation among neighbors are used to overcome these challenges and perform
neighbor discovery.
This section describes how LANETs perform neighbor discovery using two mechanisms
namely; synchronized and random. Regardless of the mechanism followed, there is a common
message exchange procedure that takes place during the neighbor discovery phase. The nodes in the
network can be divided into two kinds of nodes; seeking nodes and responding nodes. The seeking
nodes initiate the neighbor discovery process during the first half of the sector slot using Hello
packets and responding nodes reply to the seeking nodes using Hello ACK packets during the second
81
CHAPTER 4.
half of the sector slot. Consider the situation where nodes A and B are in the neighbor discovery
phase. Let A be the seeking node and B be the responding node. Accordingly, A activates its FOV
to a sector and chooses a random backoff within the first half of the sector slot and broadcasts a
Hello packet in a CMS. The Hello packet consists of the Node ID, location, and A’s neighbor list and
their corresponding location. When B receives the Hello packet, it adds A to its neighbor list for the
corresponding sector. B also uses the Node IDs and location of A’s neighbors to evaluate if any of
A’s neighbor has the potential of being B’s neighbor based on their location. Accordingly, B will
add them as its tentative neighbors for the corresponding sector estimated according to the node’s
location. The procedure is repeated as B transmits Hello ACK packet to A in the second half of the
sector duration. In this way, nodes collaboratively help each other to discover the neighbors. The
tentative neighbors are confirmed in the neighbor list once it receives any control packets from the
tentative neighbors. It is important to note that a seeking node may receive a Hello packet during the
backoff period from other seeking nodes. In such cases, the seeking node uses the received Hello
packet and responds with Hello ACK packet like a responding node. The two schemes to perform
neighbor discovery are discussed below.
Random Scheme. In this scheme, the responding nodes are called static nodes as they
randomly pick one sector and remains in the same sector during the neighbor discovery phase. In
contrast, the seeking nodes randomly pick a sector to start from and then sequentially activates
the FOV in a counter-clockwise direction such that it covers all the eight sectors one-by-one. The
performance of this scheme is plotted in Fig. 4.4 (dotted lines) assuming that the sector duration is
large enough for nodes to exchange neighbor information without collision using random backoff.
The parameters used for simulation are tabulated in Table 4.3. The percentage of neighbors discovered
(including tentative neighbors) after one super-slot (eight sector slots in this case) is used to evaluate
the performance of neighbor discovery scheme. It can be observed from Fig. 4.4 (dotted lines)
that as the ratio of seeking nodes (γ) in the network increases, the performance of the random
neighbor discovery scheme increases. This implies that the best approach is to have no static nodes
(γ = 1) during the random scheme of neighbor discovery. As expected, the performance of neighbor
discovery also increases with the increase in density of nodes in the network. This can be seen from
the increased performance as the number of nodes increases from 25 to 200. The advantage of this
scheme is that it does not require synchronization.
Synchronous Scheme. In contrast to the earlier scheme, here there are no static nodes.
Instead, the responding nodes synchronously activate FOV in different directions such that it covers
all eight sectors sequentially, listening to one sector at a time. These nodes are referred to as
82
CHAPTER 4.
Ratio of seeking nodes0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Per
cen
tag
e o
f n
eig
hb
ors
dis
cove
red
0
20
40
60
80
100
Random: N=25Sync: N=25Random: N=100Sync: N=100Random: N=200Sync: N=200
Figure 4.4: Performance of neighbor discoveryvs Ratio of seeking nodes.
Number of nodes1 2 3 4 5 6
Per
cen
tag
e o
f n
eig
hb
ors
dis
cove
red
0
20
40
60
80
100
Random discoverySynchronous discovery
Figure 4.5: Random mode vs Synchronous mode.
Number of super-slots1 2 3 4 5 6 7 8 9 10
Per
cen
tag
e o
f n
eig
hb
ors
dis
cove
red
0.4
0.5
0.6
0.7
0.8
0.9
1
Random: N=15Sync: N=15Random: N=50Sync: N=50
Figure 4.6: Convergence of neighbor discovery as super-slots increase.
synchronous (sync) nodes. The remaining nodes are still called seeking nodes but they do not start
from a random sector. They are also synchronized but activates its FOV such that they face the
intended synchronous node during each sector slot. For example, when the sync nodes are facing
sector 1, all the seeking nodes will be turned to sector 5. Therefore, if the pattern of sector access
of the synchronous nodes follows sectors 1→ 2→ 3→ 4→ 5→ 6→ 7→ 8; the corresponding
sector access pattern of seeking nodes will be 5→ 6→ 7→ 8→ 1→ 2→ 3→ 4. In this way, the
seeking nodes know which sector to turn to during each sector slot, thus increasing the chance of
discovering its neighbors. The solid lines in Fig. 4.4 represents how the synchronous neighbor
discovery performs as γ varies from 0.1 to 1. As you can see, the performance is maximum when
there is an approximately equal number of sync nodes and seeking nodes represented by the peak
83
CHAPTER 4.
Table 4.3: Simulation parameters for neighbor discovery.
Network Parameters Values
Number of nodes 15,25,50,100,150,200Area of the network 25 m x 25 mFOV 45◦
Range of VLN 5 mNumber of super-slots 1 and 10Number of seeds 200
near the middle of the curve. It is also worth noting that the synchronous scheme performs better
than random scheme especially when the density of the network is lower. To visualize this fact better,
the highest values of percentage neighbors discovered by both schemes for different network size are
plotted in Fig. 4.5.
The convergence of the two neighbor discovery schemes is evaluated in Fig. 4.6. In this
simulation, the neighbor discovery phase is extended to 10 super-slots to evaluate how quickly all the
neighbors in the network can be discovered in a relatively sparse network with N = 15 and N = 50. At
the beginning of each sector, nodes randomly choose to become seeking nodes or responding nodes
based on the given γ value. The value of γ is set corresponding to the best performance achieved for
each scheme (γ = 1 for the random scheme and γ = 0.4 for the synchronous scheme). The solid line
represents the synchronous scheme and the dotted line represents the random scheme. It is evident
that the synchronous scheme manages to discover all the neighbors faster than the random scheme
since the seeking nodes know when neighboring nodes will be facing a given sector. As expected, it
can also be seen that neighbor discovery is faster in a denser network (N = 50) compared to a sparser
network (N = 15) regardless of the scheme.
Overall, the synchronous scheme performs better in terms of percentage neighbors discov-
ered within one super-slot and has a faster rate of convergence. The random scheme does not require
any synchronization and eventually converges albeit at a slower rate. Depending on the application,
the user could choose from the two schemes to perform the neighbor discovery for LANETs.
4.4 Design of VL-MAC
Unlike RF based MANETs, the accessibility of neighbors changes drastically with time
due to deafness and blockage and therefore the traditional approach of choosing a neighbor to forward
a packet before negotiating medium access may not be the most efficient approach. The concept of
84
CHAPTER 4.
opportunistic link establishment has been introduced wherein a VLN uses a utility function to identify
the optimal sector that maximizes the probability of establishing a link. The VLN then broadcasts a
control packet in the chosen sector for two purposes; to negotiate the access of the medium and find
the optimal next hop. The neighboring VLNs that receive the broadcast respond based on a second
utility function and provide parameters for efficient transmission. This step further contributes to
maximizing the network throughput by favoring the establishment of full-duplex links. Thereafter, the
channel is reserved to complete the data transmission.
D(ACP2)
C(INI2)
B(INI1)
A(ACP1)
A
R
T
A
R
T
A
R
T
A
C
N
A
R
T
A
R
T
A
C
N
A
R
T
A
C
N
R
E
S
R
E
S
A
C
N
ACN is Ignored since C is deferring
ART Transmissions
C
I
F
S
R
E
S
A
C
KPACKET TRAIN
A
C
K
EXPLOITING FULL DUPLEX WHEN POSSIBLE
A
C
K
A, B, C, D = nodes
= random backoff
= deferring access
A
C
N DEFERRED and switches to S-IDLE
DEFERRED and switches to S-IDLE
A
C
K
PACKET TRAIN
ACN & RES Transmissions
Sector Duration
In Control Channel In Data Channel
PACKET TRAIN / BUSY TONE
Figure 4.7: Timing diagram of VL-MAC.
The timing diagram of the mechanism followed by VL-MAC is depicted in Fig. 4.7. Since
VL-MAC is designed to encourage full-duplex communication, the use of terms transmitter and
receiver becomes confusing. Therefore, hereafter, the terms transmitter and receiver will be replaced
by initiator and acceptor depending on which node initiates communication. Consider four nodes A,
B, C and D, among which B and C are the initiators with packets to be transmitted and A and D are
prospective acceptors in S-IDLE. Once a node has packets to transmit, it has to choose a sector to
transmit such that it maximizes the initiator’s utility function (U iini(s)). This is a joint function of
backlog and the achievable forward progress through the chosen sector. To better understand U iini(s),
consider a node i with intended destination k, and let j be the possible next hop. The Uini(s) for node
i is given by,
85
CHAPTER 4.
U iini(s) = ∑
q∈Q is
∑j∈N B i
s
biq di j, ∀ j : dik−d jk > 0 (4.1)
where,
di j =dik−d jk
di j, (4.2)
biq is the backlog length of session q ∈ Q s
i at node i, Q is is set of all sessions with packets that can
be forwarded through sector s. This term ensures that heavily backlogged sessions result in higher
utility function. The distance between nodes i and k is denoted as dik (each node stores the last known
location of the neighbors acquired from control packets). In (4.2), dik−d jk represents the function
that evaluates forward progress achieved by choosing j as the next hop. The denominator (di j) can be
considered as the cost of achieving the forward progress. Greater transmission distance implies more
resources (power) may have to be utilized to reach the neighbor. This also implies that a larger area
will be under the interference range of the transmitter. Therefore, this parameter helps to be more
conservative by providing nodes at a smaller angular distance preference over the others that have
the same dik−d jk values. The summation over all feasible neighbors (that provide forward progress)
ensures that the utility function increases proportionally to the number of feasible neighbors in the
given sector, which in turn increases the probability of finding an available next hop. This is a critical
differentiating feature of the proposed MAC protocol since it introduces the concept of opportunistic
link establishment in contrast to traditional methods where a forwarding node is chosen before the
negotiation for channel access begins. This mitigates the inaccessibility caused due to deafness or
blockage. Accordingly, the sector that maximizes the utility function for i is chosen as the optimal
sector s∗ and can be represented as,
s∗i = arg maxs∈S (U iini(s)) (4.3)
Accordingly, B and C choose the sector corresponding to their maximum Uini(s). In
this example, assume that both choose the same sector. Nodes B and C choose a random backoff
depending on their Uini and broadcast an Availability Request (ART) packet if the channel is idle
within the ART transmission period of the sector duration. The ART consists of the information
regarding the source node (initiator) such as node ID, location, backlog length of all sessions
considered for the given sector and channel state. As shown in Fig. 4.7, both A and D listen to
control packet during the corresponding sector duration. On reception of ARTs, A and D will switch
86
CHAPTER 4.
to TR and calculate their respective acceptor’s utility function, U jacp(i), using information from all
the ARTs received during the sector duration. The U jacp(i) for any initiator-acceptor pair i and j can
be computed as follows,
U jacp(i) = δi j(q∗i )Ci jdi j +δ ji(q∗j)C jid ji (4.4)
where q∗i is the session selected for transmission from i to j such that it maximizes the differential
backlog. The maximum differential backlog between nodes i and j is given by,
δi j(q∗i ) = arg maxq∈Q is[bi
q−b jq] (4.5)
It can be seen that (4.4) includes the product of maximum differential backlog, channel
capacity and forward progress from both directions. This is because a VLN uses hardware that
inherently supports full-duplex communication. Therefore, the initiator-acceptor pair that can
achieve higher combined throughput using duplex communication gets access to the channel thereby
improving the overall throughput of the network. Assuming LOS transmission, the Shannon capacity
can be computed as,
Ci j = B log2
(1+
(PtLP)2
σ2shot +σ2
thermal
)(4.6)
where σsho and σthermal are the standard deviations of shot noise and thermal noise respectively, Pt is
the transmit power and LP is the path loss value for a Lambertian LED source and is given as [164],
LP =(m+1)Ar
2πd2 cos(α)cosm(β) (4.7)
where m is the order of Lambertian emission, Ar is the receiver’s aperture area, α is the incident
angle and β is the irradiation angle and d is the distance between the initiator and acceptor. Using
(4.6) and (4.7), the acceptor can calculate the transmit powers required to satisfy the SNR threshold
for the full-duplex communication.
According to the above discussion, A and D choose the initiator (B or C) that they want
to provide access to. The acceptors also select the initiator’s session and acceptor’s session for
full-duplex communication such that it maximizes their respective U jacp(i) and the transmission
power to be used by the initiator-acceptor pair to ensure the required BER is achieved as shown
87
CHAPTER 4.
below,
(i∗, p∗i , p∗j ,q∗i ,q∗j) = arg max(U j
acp(i)). (4.8)
This is the second critical step taken by the MAC protocol to maximize the network throughput by
choosing initiator-acceptor pairs favoring opportunities for establishing full-duplex communication.
These chosen parameters are encapsulated in a Availability Confirmation (ACN) packet and trans-
mitted by the acceptors to the chosen initiators. In this case, A transmits a ACN to B after a random
backoff which is dependent on Uacp. The ACN contains information that is used by the initiator to set
the transmission parameters (modulation, power, and channel if applicable). In this example, the
ACN from A is received by intended node B and overheard by C. Accordingly, B transmits Reserve
Sectors (RES) packet to reserve the time required to complete the transmission. Node C learns that it
was not chosen for transmission by overhearing the ACN, and hence defers access and returns to the
S-IDLE. Similarly, D overhears the RES packet and returns to S-IDLE.
After the three-way handshake, nodes A and B perform full-duplex data transmission as
shown in Fig. 4.7. Each data packet is followed by an ACK packet from the respective receiver.
After the completion of the full-duplex transmission, both the nodes return to the S-IDLE. In cases
where there is no opportunity for full-duplex communication (acceptor does not have any session to
be transmitted to the initiator), a busy tone is transmitted by the acceptor. This ensures that other
nodes sense the channel to be busy from both directions of the initiator-acceptor pair and reduces the
hidden node problem. All these factors collectively mitigate the effects of deafness, blockage and
hidden node problem while favoring the establishment of full-duplex links thereby maximizing the
throughput of the network.
4.5 Performance Evaluation
To evaluate the performance of the proposed MAC protocol, a packet-level simulator is
implemented that operates at the data-link layer. To abstract the evaluation from the effects of the
physical layer, the simulator only considers packet loss caused due to collisions. This can be easily
extended to include any modulation and coding scheme at the physical layer and will show a similar
trend in performance at the data-link layer. The simulator is used to compare the performance of
VL-MAC and a CSMA/CA based MAC. To ensure a fair comparison, VLNs are synchronized in both
cases. The network consists of 50 VLNs with transmit range of 5 m deployed at random locations
within a 25 m x 25 m area on the same plane. The size of control and data packets were 20 Bytes
88
CHAPTER 4.
and 2500 Bytes respectively and the data rate was set to 10 Mbps. Each session in the network is
characterized by the source node and a random point on the topology indicating the destination. The
destination is used to determine the direction of forward progress in case of VL-MAC and to choose
next hop for the session in case of the CSMA/CA scheme. To this end, the neighbor that provides
the most forward progress is chosen as the next hop for a given session. The CSMA/CA based MAC
protocol can also establish full-duplex links if the intended acceptor has a session to be forwarded to
the initiator. This is a single hop simulation so a packet being successfully forwarded to a suitable
next hop in the specified direction contributes towards the overall network throughput.
Number of sessions0 20 40 60 80 100 120 140
No
rmal
ized
Net
wo
rk T
hro
ug
hp
ut
0
5
10
15
20
25
30
35
VL-MACCSMA/CA
Figure 4.8: Throughput comparison betweenVL-MAC and CSMA/CA.
Number of sessions0 20 40 60 80 100 120 140
Per
cen
tag
e o
f F
ull-
Du
ple
x lin
ks e
stab
lish
ed
10-2
10-1
100
101
102
3.63
8.8212.98
16.4920.77 22.08 23.88 26.11 27.36
0.05
0.31
0.761.19 1.43
1.78 1.91 2.09 2.15
VL-MACCSMA/CA
Figure 4.9: Percentage of full-duplex communica-tion established.
Number of sessions4 20 36 52 68 84 100 116 132
Pac
ket
colli
tio
ns
du
e to
hid
den
no
des
0
500
1000
1500
2000
CSMA/CA
Figure 4.10: Packets dropped due to hidden nodes.
The throughput normalized to link rate is computed to evaluate the percentage of full-
duplex links established and track collisions due to the hidden node as the number of session increases
89
CHAPTER 4.
from 4 to 140 with each session containing 500 packets. The results were computed over 100 seeds.
The comparison of normalized throughput between VL-MAC and CSMA/CA is depicted in Fig. 4.8.
The difference in performance grows as the number of sessions in the network increases. As discussed
earlier, the opportunistic approach of VL-MAC along with the carefully designed utility function
(U iini(s)) increases the probability of establishing a link to forward the packets. Similarly, the second
stage of negotiation using U jacp(i) favors the full-duplex links which can be further corroborated
by Fig. 4.9. Since U jacp(i) utilizes both differential backlog and capacity, it aims to maximize the
throughput achieved by initiator-acceptor pair thereby maximizing the throughput of the overall
network. Next, in Fig. 4.10, the total number of packets dropped due to collision from the hidden
node problem is depicted. It can be seen that the number of packets dropped in CSMA/CA increases
with increase in the number of sessions. Throughout the evaluations, no collision due to hidden nodes
was observed for the proposed VL-MAC. This is because VL-MAC employs separate CC and DC
and uses either busy tone or full-duplex communication links (always keeping both directions busy).
All these factors jointly provide up to 61% improvement in the throughput achieved by LANET
employing proposed VL-MAC. These experiments establish how the performance of LANETs can
be significantly improved by taking into consideration the unique properties of VLC while designing
the communication protocols.
4.6 Summary
The proposed protocol, VL-MAC, implements a three-way handshake procedure to negoti-
ate access to the medium. VL-MAC optimizes the throughput of the network by dividing the complex
problem into a step-by-step utility-based opportunistic negotiation to establish links. Results show
how VL-MAC significantly mitigate deafness, blockage and hidden node problems to drastically
improve the percentage of full-duplex links established, essentially leading to a 61% improvement in
network throughput over CSMA/CA.
90
Chapter 5
VL-ROUTE: A Cross-Layer Routing
Protocol for Visible Light Ad Hoc
Network
5.1 Motivation
The key applications and challenges in the context of enabling several indoor [165, 166]
and outdoor applications [167] have been discussed in detail in the previous chapter. It is evident
that significant work is required at the network layer of LANET to overcome some of the challenges
specific like deafness (due to directionality) and blockage (due to the nature of propagation) that
induces highly volatile route conditions. In this chapter, some of these specific challenges affecting
different types of routing techniques are discussed and a novel cross-layer routing protocol, VL-
ROUTE is designed with an objective to enable reliable routing in LANETs.
Proactive Routing. Each node in the network maintains routing information for the entire
network in proactive (table-driven) routing protocol. This approach usually ensures lower end-to-end
delays at the expense of larger overhead to maintain routes. Usually, in a traditional network with
omnidirectional antennas, the nodes may use broadcast messages regularly to learn route changes. In
a directional network, this becomes challenging and time-intensive due to deafness and the need to
exchange messages in every sector. This problem is further aggravated in LANETs due to the limited
route lifetime due to varying link connectivity. Thus, there is a constant need to update routes but at
the same time, it is extremely challenging and expensive to learn changes in the network. All these
91
CHAPTER 5.
factors render it extremely difficult to maintain updated routing tables for the entire network.
Reactive Routing. In contrast to proactive routing protocols, a source node discovers a
route when it has to transmit a packet in a reactive routing protocol. This eliminates the need to
maintain routing tables at every node and hence reducing the overhead and power consumption but
may lead to higher delays. In the context of LANETs, it is difficult to discover all possible routes due
to the narrow FOV and without an adequate neighbor discovery scheme that overcomes blocking.
Similarly, there is no guarantee that the route still exists once the route is discovered and the source
starts transmitting packets towards the destination. Therefore, a route that theoretically provides the
highest throughput but lacks alternate paths that might help route recovery in case of link failure may
not be an ideal choice for LANETs.
It has been established how traditional design considerations may not be directly applicable
to LANETs. Reliability of route and the opportunity to change the routing decision quickly will
be the critical features distinguishing a routing protocol from traditional approaches. Previously,
cross-layer network optimization has been explored in RF networks [86, 168, 45, 87] but is especially
crucial for LANETs to combat the volatile nature of links [169, 170, 107]. Therefore, considering
the above challenges, the features essential for a routing protocol for LANETs can be summarized as
follows,
• A cross-layer approach is required to ensure collaboration with the MAC layer to mitigate the
degradation that is caused due to deafness and blockage and to maximize the probability of
establishing full-duplex links.
• Due to the highly dynamic nature of LANETs, an opportunistic routing protocol that uses a
distributed algorithm to determine the optimal hops at each intermediate node in the multihop
network is required.
• Reliability of nodes and the routes they can provide for each session should be considered as a
key metric while making routing decisions. The absolute channel condition itself may not be
the best indicator of successful routes.
The primary focus in VLC has been to enable point-to-point communication with the goal
of improving link data rates. Network layer protocols are usually derived from traditional methods to
act as a facilitator of VLC application [134]. In [171], authors propose a novel platform aimed at
distributed multihop visible light communication that has 360 degree coverage and is compatible with
experimental boards such as Arduino, Beaglebone, Raspberry Pi. They identify the open problem
92
CHAPTER 5.
in developing multihop solution but do not propose a solution. Authors propose a greedy routing
algorithm to support a fully wireless Data Center Network (DCN) across racks in [166]. The greedy
algorithm chose next-hop such that it has the shortest distance to the destination. Their objective
was to use VLC to eliminate hierarchical switches and inter-rack cables, and thus reducing hardware
investment, as well as maintenance cost. They were successful in showing an effective application of
VLC but the routing protocol does not consider any LANET specific challenges.
There has been recent effort to explore cooperative relaying [172, 173, 165] mainly for
linear and triangular topologies in an indoor environment. Cooperative relaying can be used to
enhance the link reliability and extend coverage but cannot be directly extended to a multihop
network. The necessity of a cross-layered approach has been identified by the community and there
has been some effort in this direction. In [107], authors show that improved end-to-end delivery ratio
can be achieved by using multipath routing that accounts for the intermittent blockage problem of
VLC links in vehicular visible light communication (V2LC) networks.
In [174], the author proposes a hexagonal cylindrical design to provide omnidirectional
access to directional VLC network. Each face in the hexagonal design has IR transmitters and PD
receivers to provide omnidirectional access. The author designs methodology to avoid the sudden
blockage by finding alternate paths to the intended destination. Accordingly, when the base station
(intended for the ceiling) or user device (intended for the desk) looses connection it first checks if
a connection can be established using any other faces. If that fails, source checks if a previously
known route exists and sends validate packet if it does. If such route does not exist, the source sends
reactive Route Discover Packet with preset forward depth count looking for rendezvous node which
has the path to the destination node. If in a given period of time (associated with forward depth
count) there is no response from any node, they consider that there is no such rendezvous node.
While the proposed solution aims to mitigate the effects of blockage, the constant disconnect and
route discovery may deteriorate the overall network throughput.
To best of the author’s knowledge, this is the first work that addresses the challenges in
enabling distributed and dynamic routing algorithm specifically for LANETs. Accordingly, the major
contributions of this chapter are as follows,
• A cross-layer routing algorithm that interacts closely with the MAC layer to ensure oppor-
tunistic packet forwarding thereby mitigating effects of deafness, blockage and hidden node
problem as been designed for LANET.
• Due to the volatile nature of routes in LANET, the requirement of highly dynamic routing
93
CHAPTER 5.
technique has been established and hence a routing algorithm is designed where hop-by-hop
decision making is employed to ensure adaptability.
• The reliability score is formulated such that it enables complete decentralized operation of the
network with the objective to maximize the expected throughput of the network.
• Extensive simulations are conducted to analyze the behavior of the proposed routing algorithm
in the various operating environment and compared against two other routing algorithms.
5.2 Design of VL-ROUTE
Most routing algorithms in RF based ad hoc network are designed to optimize network
parameters like throughput, delay, energy consumption etc. These cannot be the the sole metric in
LANET due to its unique challenges discussed earlier. The link state in LANETs are highly dynamic
and can be interrupted due to blockage or deafness in addition to channel conditions itself. Therefore,
route reliability becomes a key metric for consideration while designing routing algorithm for LANET.
The reliability of a route can be defined as the probability of successfully delivering a packet from i
to the desired sink k on first attempt. It is given as follows,
pr(i : k) = ∏(i, j)∈Lr
p(i, j) (5.1)
where Lr is the set of all links (i, j) in route r, p(i, j) is the probability that packet is successfully
forwarded from i to j in the first attempt. The value of p(i, j) depends on packet error probability
(pe), probability of blockage (pbi j), and probability of i winning the contention to establish link with
j in the first attempt (pacsi j ). Therefore, p(i, j) can be represented as follows
p(i, j) = (1− pe).pacsi j .(1− pb
i j) (5.2)
where pacsi j denotes the probability that node i can negotiate access to node j which in turn depends
on number of nodes (M) in set N B js where s is the sector to which i belongs. Assuming worst case
scenario where every node has a packet to transmit and each node chooses a random backoff value
between the range (0,CW −1) where CW is the contention window size, it can be represented as,
pacsi j = p0(1− p0)
M−1 (5.3)
94
CHAPTER 5.
such that p0 is the probability that a VLN transmits in a timeslot and is given by [175],
p0 =2
CW +1(5.4)
Using (5.1), the probability of delivering packets from i to k over at least one of the possible
routes can be given by,
p(i : k) = 1− ∏r∈R i
k
[1− pr(i : k)] (5.5)
where R ik is the set of all possible routes from node i to sink k (only routes in which each hop makes
some forward progress towards k is considered). Therefore, the expected throughput of a session q
can be defined as follows,
E[T (q)] = p(i : k).T (q) (5.6)
where T (q) is the maximum achievable throughput for a session q from node i to k. The overall
objective of the proposed routing algorithm for LANET is defined as shown below
Maximize : ∑q∈Q
E[T (q)] (5.7)
subject to :
Link capacity constraint (5.8)
Maximum queue size constraint (5.9)
Power budget constraint (5.10)
Computing p(i : k) requires global knowledge of the network in order to consider all
possible routes and link probabilities from node i to k. Therefore, in this paper, to enable distributed
operation, Route Reliability Score (RRS) is defined based on approximation of p(i : k) for each node
i to indicate a measure of expected success in reaching the sink k through i. Each node will use the
RRS of its immediate neighbors to determine its routing strategy at each hop. The implications of
RRS of node i with respect to sink k is explained in detail as follows,
Γki = β
ki .
1−mins∈S
∏j∈N B i
s/hki <hk
j
[1− p(i, j).Γk
j
] (5.11)
95
CHAPTER 5.
where,
βi(s) =bmax−bi
kbmax
(5.12)
where bik is the total number of packets destined for k backlogged at i and bmax represents the
maximum buffer size. βi(s) is meant to penalize the score of i w.r.t particular sink k when the node
is heavily backlogged with packets intended for k. The absence of such penalization may lead all
traffic to follow same route regardless of congestion. The use of βi(s) aims to ensure a balance
between reliability and congestion. Next, hki denotes the Minimum Hop Count (MHC) from i to k. It
can be seen in the formulation of (5.11) that only neighbors with lower MHC to the respective sink
k contributed towards the RRS. Since VLN can operate on only one sector at a time, this should
be reflected in the calculation of RRS. Therefore, the final value of RRS corresponds to the sector
that provides the highest value among the all sectors in S . Looking closely at the formulation of
RRS, one realizes that this distributed estimation still reflects the overall structure of (5.5) along with
the addition of βi(s). More importantly, it can be seen from the above definition of RRS that every
node can calculate its score with respect to each sink using information gathered from immediate
neighbors. This critically ensures a scalable and distributed algorithm.
Initially, each VLN sets its RRS to zero and MHC to infinity for each sink in the network.
Thereafter, VLNs listen to neighbor’s control packets to compute RRS and MHC. The first set of RRS
is calculated by VLNs within one hop from the sinks from the overheard control packets transmitted
by the sink. In this case, for a node x one hop away from sink k, RRS is simply Γkx = p(x,k) and
updates its MHC to sink k as hkx = 1. Subsequently, the computed RRS is appended to the control
packets that it transmits. In this manner, whenever a node i receives an updated Γkj from neighbor
j (that provides forward progress w.r.t k), i updates Γki and hk
i . In this manner, each node keeps
updating its RRS and MHC w.r.t each known sink in the network.
As discussed earlier, it is understood that a cross-layer approach is required to combat the
unique challenges posed by LANETs. The data-link layer and network layer need to coordinate with
each other to optimize the network’s performance. Therefore, a cross-layered routing algorithm is
designed such that it embeds the RRS into an opportunistic MAC protocol designed specifically for
LANETs [176]. To this end, VL-MAC discussed in the previous chapter is significantly extended to
interact with the network layer and include optimized routing decisions while negotiating the access
of the medium.
The timing diagram of the mechanism followed by VL-MAC is depicted again in Fig.
5.1. Consider four nodes A, B, C and D, among which B and C are the initiators with packets to be
96
CHAPTER 5.
D(RX2)
C(TX2)
B(TX1)
A(RX1)
A
R
T
A
R
T
A
R
T
A
C
N
A
R
T
A
R
T
A
C
N
A
R
T
A
C
N
R
E
S
R
E
S
A
C
N
ACN is Ignored since C is deferring
ART Transmissions
C
I
F
S
R
E
S
A
C
KPACKET TRAIN
A
C
K
EXPLOITING FULL DUPLEX WHEN POSSIBLE
A
C
K
A, B, C, D = nodes
= random backoff
= deferring access
A
C
N DEFERRED and switches to S-IDLE
DEFERRED and switches to S-IDLE
A
C
K
PACKET TRAIN
ACN & RES Transmissions
Sector Duration
In Control Channel In Data Channel
PACKET TRAIN / BUSY TONE
Figure 5.1: Timing diagram.
transmitted and A and D are prospective acceptors in S-IDLE. Once a node has packets to transmit,
it has to choose a sector to transmit such that it maximizes the initiator’s utility function (U iini(s)).
Consider a node i with packets intended for sink k, and let j be the possible next-hop. The initiator’s
utility function for node i is given by,
U iini(s) = ∑
q∈Q is
∑j∈N B i
s
bi(q) di j Γkj, ∀ j : dik−d jk > 0 (5.13)
where,
Γkj =
Γkj
max j∈N B i [Γkj]
and di j =dik−d jk
di j, (5.14)
where bi(q) is the backlog length of session q∈Q si at node i, Q i
s is the set of all sessions with packets
that can be forwarded through sector s. The measure of forward progress is provided by di j [176].
The normalized RRS, Γkj provides higher utility to sector with neighbors that provide more reliable
routes. It is critical to understand why Γkj is used over just Γk
j. Investigating (5.11) closely, it can be
seen that Γkj generally decreases as the number of hops to sink increases. Therefore, using Γk
j instead
of Γkj would give unfair advantages to session whose destination is closer to i. The goal of Γk
j in (5.13)
is to ensure that neighbor that provides relatively higher RRS w.r.t a given sink contributes to a larger
utility value. The summation over all feasible neighbors ensures that the utility function increases
97
CHAPTER 5.
proportionally to the number of feasible neighbors in the given sector, which in turn increases the
probability of finding an available next-hop mitigating the effect of deafness.
The goal of this utility function is to introduce the concept of opportunistic link establish-
ment in contrast to traditional methods where a forwarding node is chosen before the negotiation
for channel access begins. This mitigates the inaccessibility caused due to blockage or deafness.
Accordingly, i chooses the optimal sector s∗ that maximizes its utility function and can be represented
as,
s∗i = arg maxs∈S(U i
ini(s))
(5.15)
Accordingly, in this example, lets assume B and C choose the same sector which corre-
sponds to their maximum Uini(s). Nodes B and C choose a random backoff depending on their Uini
and broadcast an ART packet if the channel is idle within the ART transmission period of the sector
duration. The ART consists of the information regarding the source node (initiator) such as node ID,
location, RRS, backlog length of all sessions considered for the given sector and channel state. As
shown in Fig. 4.7, both A and D listen to control packet during the corresponding sector duration.
On reception of ARTs, A and D will switch to TR and calculate their respective acceptor’s utility
function, U jacp(i), using information from all the ARTs received during the sector duration. The
U jacp(i) for any initiator-acceptor pair i and j can be computed as follows,
U jacp(i) = [ηi j(q∗i )Ci j]+
[η ji(q∗j)C ji
](5.16)
where q∗i is the session selected for transmission from i to j such that it maximizes the weighted
differential backlog given as follows,
ηi j(q∗i ) = arg maxq∈Q is
[Γk
j di j(bi(q)−b j(q)
)](5.17)
It can be seen that (5.16) includes the product of maximum weighted differential backlog and channel
capacity (Ci j) in both directions as defined in [176]. This implies the initiator-acceptor pair that can
achieve higher combined throughput using full-duplex communication gets access to the channel
thereby improving the overall throughput of the network. It is important to note that in contrast
to [176], the utility functions include the RRS that governs the routing decision and hence will
implicitly lead to reliable routes and higher throughput.
According to the above discussion, A and D choose the initiator (B or C) that they want to
provide access. The acceptors also select the initiator’s session and acceptor’s session for full-duplex
98
CHAPTER 5.
communication such that it maximizes their respective U jacp(i) as shown below,
(i∗,q∗i ,q∗j) = arg max
(U j
acp(i)). (5.18)
This is the second critical step taken by the cross-layer routing protocol to maximize the expected
network throughput by choosing initiator-acceptor pairs favoring opportunities for establishing full-
duplex communication. These chosen parameters are encapsulated in a ACN packet and transmitted
by the acceptors to the chosen initiators. In this particular example, after a Uacp dependent random
backoff, A transmits an ACN to B. The ACN contains information that is used by the initiator to set
the transmission parameters (modulation, power, and channel if applicable). Accordingly, B receives
the ACN from A and C overhears this ACN intended for B. Next, B transmits RES packet to reserve
time required to complete the transmission. Node C learns that it was not chosen for transmission by
overhearing the ACN, and hence defers access and returns to the S-IDLE. Similarly, D overhears the
RES packet and returns to S-IDLE.
After this three-way handshake, nodes A and B perform full-duplex data transmission as
depicted in Fig. 4.7. The respective receivers transmit the ACK packet after the reception of data
packet. After the completion of the full-duplex transmission, both the nodes return to the S-IDLE.
In cases where there is no opportunity for full-duplex communication (acceptor does not have any
session to be transmitted to the initiator), a busy tone is transmitted by the acceptor. This is to ensure
that other nodes sense the channel to be busy from both directions of the initiator-acceptor pair and
reduce the hidden node. All these factors collectively mitigate the effects of deafness, blockage and
hidden node problem while favoring the establishment of full-duplex links. These factors along with
the carefully designed RRS based route selection strategy maximizes the expected throughput of the
network.
5.3 Performance Evaluation
To evaluate the performance of VL-ROUTE, a packet-level simulator that operates at the
network layer but interacts closely with the data link layer has been implemented. The simulator only
considers packet loss caused due to collisions and channel condition (i.e. based on the packet error
rate of each link). This framework can be easily extended to include any modulation and coding
scheme at the physical layer and will show a similar trend in performance at the network layer. The
simulator is used to compare the performance of the proposed VL-ROUTE with a greedy routing
99
CHAPTER 5.
that employs CSMA/CA based MAC (for simplicity this is referred to as GR-CSMA). The greedy
routing is similar to [166], a VLN that has a packet to transmit choose a neighbor that is closest to
the intended sink to forward the packet. To ensure a fair comparison, VLNs are synchronized in both
cases and perform full-duplex communication links whenever possible. The network consists of
100 VLNs with a transmission range of 4 m and each session in the network is characterized by the
source node and an indented sink. The size of control and data packets were 20 Bytes and 2500 Bytes
respectively and the data rate was set to 10 Mbps.
Table 5.1: Parameters of simulation in grid topology.
Parameters Values
Size 25 m × 25 mMean packet error rate 0.2Number of sessions 2 to 20Total Packets per session 200Number of Sinks 5
Grid Topology. VL-ROUTE is first evaluated in a fully connected 10 × 10 grid network
using the parameters shown in Table 5.1. To perform a rigorous evaluation, in addition to GR-CSMA,
VL-ROUTE is compared to VL-MAC. Though VL-MAC was designed in order to facilitate cross-
layered operation, it is important to recognize that VL-MAC by itself cannot serve as a stand-alone
routing algorithm. This is because VL-MAC by itself does not have the complete mechanism to
determine existing routes to the sink. Whereas, in VL-ROUTE, the presence of a neighbor j with a
non zero Γkj indicates the presence of route to the sink k. But here, to perform a thorough evaluation
of the proposed VL-ROUTE, a method has been formulated to compare it to VL-MAC. To this end,
it has been assumed that all the VLNs using VL-MAC know the location of the sinks and hence uses
VL-MAC in each hop to perform a geographical routing to the sink and ensures forward progress in
each hop.
As discussed earlier, blockage is one of the most critical challenges of LANET. To simulate
this behavior, 25% of the links are randomly set to have a 90% chance of blockage and the remaining
75% of the links to have 5% chance of blockage. In real-life scenarios, this would be the difference
between a busy walkway or corridor that has a high probability of blockage versus most parts of the
building that might receive considerably less foot traffic or obstruction. Later, the impact of varying
blockage levels on VL-ROUTE has been evaluated in more detail. Both pe and pbi j can be estimated
by monitoring the previous activity on the given link (pe will depend significantly on modulations
and coding used by the physical layer) with some estimation error associated with it. In this first
100
CHAPTER 5.
Number of sessions2 4 6 8 10 12 14 16 18 20
No
rmal
ized
Th
rou
gh
pu
t
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
VL-ROUTEGreedy Routing + CSMA/CAGeographic Routing + VL-MAC
100% improvement
21% improvement
Figure 5.2: Throughput vs No. of session.
Number of sessions2 4 6 8 10 12 14 16 18 20
Per
cen
tag
e o
f d
up
lex
links
10-3
10-2
10-1
100
VL-ROUTEGreedy Routing + CSMA/CAGeographic Routing + VL-MAC
4.7 times higher
Figure 5.3: Full-Duplex vs No. of session.
simulation, the estimation error has been set to 5%.
First, the throughput normalized to link rate is computed as the number of sessions in the
network increases. As seen in Fig. 5.2, VL-ROUTE takes into account the unique characteristics
of LANET and outperforms traditional approach employed by GR-CSMA by up to 100%. This
improvement in network throughput can be attributed to three main reasons; (i) consideration of
route reliability in a distributed manner encourage packets to select route that provides the least
resistance (caused by blockage or unfavorable channel conditions) to intended sink (ii) the cross-layer
interaction with the link layer that provides opportunistic link establishment mitigates the effects
due to deafness, blockage and hidden node and (iii) combining RRS with the remaining factors of
VL-MAC maximizes the probability of establishing full-duplex links while choosing optimal routes.
The influence of the novel design of VL-ROUTE can be further substantiated by comparing the
performance of VL-ROUTE to the network that used geographic routing (which assumes the location
of sink is known) with VL-MAC. This improvement in performance can be primarily attributed to
route choices since the ratio of full-duplex links are similar for both VL-MAC and VL-ROUTE
(although both are much higher with compared to GR-CSMA).
Random Topology. Since the grid network is uniform deployment and has a uniform
neighborhood, a random topology has been simulated to evaluate the performance of VL-ROUTE
in a non-uniform deployment. Therefore, Fig. 5.4 shows that even in a random deployment the
proposed algorithm outperforms GR-CSMA and VL-MAC because not only is it able to identify all
the feasible hops can also direct traffic in such a way that it maximizes the expected throughput of the
network. VL-ROUTE achieves up to 124% improvement in throughput with respect to GR-CSMA
and achieves 15% improvement w.r.t VL-MAC. Figure 5.5 depicts the number of packets that were
101
CHAPTER 5.
Number of sessions2 4 6 8 10 12 14 16 18 20
No
rmal
ized
Th
rou
gh
pu
t
0
0.2
0.4
0.6
0.8
1
1.2
VL-ROUTEGreedy Routing + CSMA/CAGeographic routing + VL-MAC
124% improvement
15%
Figure 5.4: Throughput vs No. of session.
Number of sessions2 4 6 8 10 12 14 16 18 20
Nu
mb
er o
f p
acke
ts d
eliv
ered
to
sin
k
0
500
1000
1500
2000
2500
3000
3500
VL-ROUTEGeographic routing + VL-MACGreedy Routing + CSMA/CA
14 %
98 %
Figure 5.5: Throughput vs No. of session.
Percentage of nodes with 90% probability of blockage0 10 20 30 40 50 60
No
rmal
ized
Th
rou
gh
pu
t
0
0.2
0.4
0.6
0.8
1
1.2
VL-ROUTEGreedy Routing + CSMA/CA
69% improvement
114%
Figure 5.6: Throughput vs Link blockage.
Percentage of estimation error5 10 15 20 25 30 35 40
No
rmal
ized
Th
rou
gh
pu
t
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
VL-ROUTE (No. of session=10)
12 % decline
Figure 5.7: Throughput vs Estimation error.
successfully delivered to the sink by each routing algorithm. Since the random topology does not
guarantee a fully connected network there could be several dead-end paths. The construction of RRS
score ensure the maximum delivery of packets while the opportunistic MAC protocol itself perform
reasonably well. Overall, VL-ROUTE delivers up 98% and 14% more packets than GR-CSMA and
VL-MAC respectively.
Blockage Analysis. Next, to study how VL-ROUTE is effected due to various levels of
blockage that might be encountered in real-life implementation, varying levels of blockage scenarios
have been simulated. To this end, the same parameters as in the random topology are used while
varying the percentage of nodes experiencing 90% blockage from 0 to 60%. All other nodes in each
test point will be set to experience 5% chance of blockage. The number of sessions in the network
is set to 5. As expected, the throughput decreases as the percentage of nodes experiencing severe
blockage increases but in all the scenarios VL-ROUTE outperforms GR-CSMA. The improvement
102
CHAPTER 5.
is minimum (69%) at the lowest level of blockage and increases up to 114% for higher levels of
blockage. This proves how VL-ROUTE can adapt to any level of blockage and provide optimal
performance in any given scenario while operating in a distributed manner.
Estimation Error Analysis. Since the algorithm depends on the accuracy of estimation
of pe and pb, the effects of change in estimation error on VL-ROUTE has been studied here. It is
important to analyze this effect to provide clarity to the reader on how the implementation accuracy
affects the performance. Accordingly, the estimation error of both pe and pb has been varied from
5% to 40% and evaluate the performance by setting the number of sessions to 10. It can be seen from
Fig. 5.7 that there is an obvious decrease in performance with the increasing error. The degradation
in performance is not drastic (about 12%) and the resulting normalized throughput is comparable to
one achieved by VL-MAC (see Fig. 5.2). Therefore, having a highly accurate estimation mechanism
is advantageous but some error (up to 10%) is acceptable which makes VL-ROUTE a feasible choice
for actual deployment.
5.4 Summary
Enabling LANET for several indoor and outdoor applications require significant effort at
the network layer. In this chapter, this problem has been tackled and a cross-layer routing protocol,
VL-ROUTE has been designed specifically to mitigate challenges like blockage, deafness, and hidden
node to ensure the inherent full-duplex capability of VLC is completely utilized. Realizing that route
reliability is a significant factor to be considered in LANETs in contrast to other ad hoc networks,
a RRS is formulated. This enables each node in the network to estimate the reliability of the route
through a neighbor for a given sink. The RRS is then embedded in a well designed opportunistic
MAC protocol to exploit the interaction between the network and data-link layer. The measure of
reliability along with the cross-layer opportunistic link establishing mechanism provides up to 124%
improvement in throughput over GR-CSMA. The effectiveness of the formulated RRS is evident
when VL-ROUTE outperforms VL-MAC with geographic routing by 21%. Additionally, there is
improvement also in the percentage of full-duplex links and the number of packets delivered to the
sink.
103
Chapter 6
Conclusion
The objective of this work was to design, develop and evaluate cross-layer optimized
networking solutions for next-generation 5G ad hoc networks to meet the growing needs of the
community. The two main directions that will be adopted by the next-generation wireless ad hoc
networks to manage the proliferation of devices will include, (i) designing optimized networking
algorithm to maximize the utilization of the constrained resources and (ii) use cross-layer optimization
to exploit unlicensed and underutilized parts of the spectrum like visible light by overcoming its
unique challenges. This proposal looks at algorithms designed to tackle these problems across
commercial and military applications.
First, in Chapter 2, a cross-layer deadline-based routing and spectrum allocation algorithm
designed to enable tactical networks to handle different classes of data adeptly is discussed. This
work also proposes a cross-layer architecture which is then used to implement and evaluate DRS on a
SDR-based testbed. DRS was evaluated via simulations and testbed evaluations against state-of-the-
art ROSA protocol which has been shown to outperform traditional approaches. DRS outperformed
ROSA by 35 % in terms of effective throughput and by up to 26 % in terms of reliability. This
implies the critical messages for a tactical deployment will be delivered with higher reliability in a
network that employs DRS.
Next, in Chapter 3, HELPER network, an end-to-end solution is proposed to enable modern
infrastructure-less emergency ad hoc network. The overarching goal of the proposed solution is to
keep authorities and survivors connected by easily deploying an energy-efficient ad hoc network.
The HELPER network includes services like text and voice messages, live map updates, ability to
send distress messages to authorities. HELPER network can also be used by authorities to remotely
monitor the connectivity of the affected area, alert users of imminent dangers and share resource
104
CHAPTER 6.
information. The effectiveness of the proposed solution with respect to network lifetime has been
demonstrated on hardware testbed. The results showed the proposed cross-layer, energy-aware
SEEK algorithm providing 54 % improvement in network lifetime and 28 % improvement in terms
of network throughput when compared to the greedy scheme that uses shortest path routing. The
demonstration using the prototype was able to successfully establish the direct application of the
solution in emergency scenarios.
The second major objective of this work was to design communication protocols that will
enable utilization of unlicensed visible light spectrum which has been substantially underutilized.
To address this, in Chapter 4, the work begins by motivating the applications of LANETs in various
indoor and outdoor scenarios. Thereafter, the distinction between the LANETs and traditional
MANETs have been illustrated to describe some of the advantages put forth by this new domain
of wireless communication. This leads to the discussion of significant design challenges that have
to be overcome in order to make LANET a reality. Therefore, this work takes significant strides in
this direction by designing MAC and routing protocols for enabling the deployment of LANETs to
bolster low range high data rate applications of the 5G ad hoc networks.
The MAC protocol, VL-MAC, introduces the novel concept of opportunistic link assign-
ment such that nodes opportunistically select the intended forwarding node during the process of
negotiating channel access. This is different from traditional MAC protocols where the intended next-
hop is known before the negotiation of channel access takes place. To accomplish this, a utility-based
three-way handshake MAC protocol has been proposed along with the necessary synchronization
mechanism. The proposed solutions showed up to 61% increase in throughput and the percentage of
full-duplex links established increased from 2% to 27% when compared to CSMA/CA. Thereafter,
in Chapter 5, this was extended to design a cross-layer routing protocol which uses RRS to ensure the
nodes choose reliable routes. Accordingly, VL-ROUTE provides significant improvement (124%) in
throughput and reliability (98%) when compared to approaches derived from traditional RF networks.
These results clearly illustrated the need for LANET-specific design of the protocol at various layers
of the protocol stack.
Overall, this work has established the need to go beyond the stringent layered architecture
of communication systems and has shown the gains that can be obtained by considering the cross-
layer optimized design. These techniques will be beneficial to mitigate ever-growing resource scarcity
and explore the underutilized parts of the spectrum.
105
Bibliography
[1] Cisco Systems, “Cisco Visual Networking Index: Global Mobile Data Traffic Forecast Update,
2016-2021 White Paper,” http://tinyurl.com/zzo6766, 2017.
[2] Federal Communications Commission [2016], “Spectrum Crunch,” https://www.fcc.gov/
general/spectrum-crunch.
[3] “FEMA: Disaster List,” https://www.fema.gov/disasters, accessed: April. 27, 2018.
[4] “The deadliest natural disasters of 2017,” https://www.telegraph.co.uk/education/stem-
awards/design/deadliest-natural-disasters/, accessed: April. 27, 2018.
[5] T. Gomes and J. Tapolcai and C. Esposito and D. Hutchison and F. Kuipers and J. Rak and A.
de Sousa and A. Iossifides and R. Travanca and J. Andre and L. Jorge and L. Martins and P. O.
Ugalde and A. Pasic and D. Pezaros and S. Jouet and S. Secci and M. Tornatore, “A survey
of strategies for communication networks to protect against large-scale natural disasters,” in
8th International Workshop on Resilient Networks Design and Modeling (RNDM), September
2016, pp. 11–22.
[6] “When Communications Infrastructure Fails During a Disaster,”
https://www.drj.com/articles/online-exclusive/when-communications-infrastructure-
fails-during-a-disaster.html, accessed: April. 27, 2018.
[7] D. Tsonev, S. Videv, and H. Haas, “Light Fidelity (Li-Fi): Towards All-Optical Networking,”
Proc. of SPIE, vol. 9007, pp. 900 702–10, 2013.
[8] A. Jovicic, L. Junyi, and T. Richardson, “Visible light communication: opportunities, chal-
lenges and the path to market,” IEEE Communications Magazine, vol. 51, no. 12, pp. 26–32,
December 2013.
106
BIBLIOGRAPHY
[9] N. Cen, J. Jagannath, S. Moretti, Z. Guan, and T. Melodia, “LANET: Visible-Light Ad Hoc
Networks,” Ad Hoc Networks, vol. 84, pp. 107 – 123, 2019.
[10] L. Ding, T. Melodia, S. Batalama, J. Matyjas, and M. Medley, “Cross-layer Routing and
Dynamic Spectrum Allocation in Cognitive Radio Ad Hoc Networks,” IEEE Transactions on
Vehicular Technology, vol. 59, pp. 1969–1979, May 2010.
[11] L. Tassiulas and A. Ephremides, “Stability properties of constrained queueing systems and
scheduling policies for maximum throughput in multihop radio networks,” in Proc. of IEEE
Conference on Decision and Control (CDC), Honolulu, HI, December 1990.
[12] V. Brik, E. Rozner, S. Banerjee, and P. Bahl, “DSAP: a protocol for coordinated spectrum
access,” in Proc. of IEEE Intl. Symp. on New Frontiers in Dynamic Spectrum Access Networks
(DySPAN), Baltimore, MD, November 2005.
[13] M. Bkassiny, S. Jayaweera, Y. Li, and K. Avery, “Optimal and Low-Complexity Algorithms for
Dynamic Spectrum Access in Centralized Cognitive Radio Networks with Fading Channels,”
in Proc. of IEEE Vehicular Technology Conference (VTC), Budapest, Hungary, May 2011.
[14] Y. Yuan, P. Bahl, R. Chandra, T. Moscibroda, and Y. Wu, “Allocating Dynamic Time-spectrum
Blocks in Cognitive Radio Networks,” in Proc. of ACM Intl. Symp. on Mobile Ad Hoc
Networking and Computing (MobiHoc), Montreal, Canada, September 2007.
[15] R. Doost-Mohammady, M. Y. Naderi, and K. R. Chowdhury, “Spectrum Allocation and QoS
Provisioning Framework for Cognitive Radio With Heterogeneous Service Classes,” IEEE
Transactions on Wireless Communications, vol. 13, no. 7, pp. 3938–3950, July 2014.
[16] L. Luo and S. Roy, “Analysis of dynamic spectrum access with heterogeneous networks:
Benefits of channel packing scheme,” in Proc. of IEEE Global Communications Conference
(GLOBECOM), November 2009.
[17] A. Alshamrani, X. S. Shen, and L. L. Xie, “Qos provisioning for heterogeneous services in
cooperative cognitive radio networks,” IEEE Journal on Selected Areas in Communications,
vol. 29, no. 4, pp. 819–830, April 2011.
[18] A. Dimakis and J. Walrand, “Sufficient conditions for stability of longest-queue-first schedul-
ing: second-order properties using fluid limits,” Transactions of Advances in Applied Proba-
bility, vol. 38, no. 2, pp. 505–521, June 2006.
107
BIBLIOGRAPHY
[19] C. Joo, X. Lin, and N. Shroff, “Understanding the Capacity Region of the Greedy Maximal
Scheduling Algorithm in Multi-Hop Wireless Networks,” in Proc. of IEEE Conference on
Computer Communications (INFOCOM), Phoenix, AZ, April 2008.
[20] M. Leconte, J. Ni, and R. Srikant, “Improved Bounds on the Throughput Efficiency of Greedy
Maximal Scheduling in Wireless Networks,” IEEE/ACM Transactions on Networking, vol. 19,
no. 3, pp. 709–720, June 2011.
[21] R. Laufer, T. Salonidis, H. Lundgren, and P. L. Guyadec, “A Cross-Layer Backpressure
Architecture for Wireless Multihop Networks,” IEEE/ACM Transactions on Networking,
vol. 22, no. 2, pp. 363–376, April 2014.
[22] P. van de Ven, S. Borst, and S. Shneer, “Instability of MaxWeight Scheduling Algorithms,” in
Proc. of IEEE Conference on Computer Communications (INFOCOM), Rio de Janeiro, Brazil,
April 2009.
[23] L. Bui, R. Srikant, and A. Stolyar, “A Novel Architecture for Reduction of Delay and
Queueing Structure Complexity in the Back-Pressure Algorithm,” IEEE/ACM Transactions
on Networking, vol. 19, no. 6, pp. 1597–1609, December 2011.
[24] M. Kurth and J.-P. Redlich, “Delay properties of opportunistic back-pressure routing in CSMA-
based wireless mesh networks,” in Proc. of IEEE Intl. Symp. on Personal, Indoor and Mobile
Radio Communications Workshops (PIMRC), Istanbul, Turkey, September 2010.
[25] D. Xue and E. Ekici, “Delay-Guaranteed Cross-Layer Scheduling in Multihop Wireless
Networks,” IEEE/ACM Transactions on Networking, vol. 21, no. 6, pp. 1696–1707, December
2013.
[26] H. Xiong, R. Li, A. Eryilmaz, and E. Ekici, “Delay-Aware Cross-Layer Design for Net-
work Utility Maximization in Multi-Hop Networks,” IEEE Journal on Selected Areas in
Communications, vol. 29, no. 5, pp. 951–959, May 2011.
[27] M. Andrews, K. Kumaran, K. Ramanan, A. Stolyar, P. Whiting, and R. Vijayakumar, “Provid-
ing quality of service over a shared wireless link,” IEEE Communications Magazine, vol. 39,
no. 2, pp. 150–154, February 2001.
108
BIBLIOGRAPHY
[28] B. Ji, C. Joo, and N. Shroff, “Delay-Based Back-Pressure Scheduling in Multihop Wireless
Networks,” IEEE/ACM Transactions on Networking, vol. 21, no. 5, pp. 1539–1552, October
2013.
[29] B. Sadiq and G. de Veciana, “Throughput optimality of delay-driven MaxWeight scheduler for
a wireless system with flow dynamics,” in Proc. of Annual Allerton Conf. on Communication,
Control, and Computing, Monticello, IL, September 2009.
[30] S. Liu, E. Ekici, and L. Ying, “Scheduling in Multihop Wireless Networks Without Back-
Pressure,” IEEE/ACM Transactions on Networking, vol. 22, no. 5, pp. 1477–1488, October
2014.
[31] D. Xue, R. Murawski, and E. Ekici, “Capacity Achieving Distributed Scheduling With Finite
Buffers,” IEEE/ACM Transactions on Networking, vol. 23, no. 2, pp. 519–532, April 2015.
[32] B. Ji, C. Joo, and N. Shroff, “Exploring the inefficiency and instability of Back-Pressure
algorithms,” in Proc. of IEEE Conference on Computer Communications (INFOCOM), Turin,
Italy, April 2013.
[33] P. Van de Ven, S. Borst, and L. Ying, “Spatial inefficiency of MaxWeight scheduling,” in
Proc. of Intl. Symp. on Modeling and Optimization in Mobile, Ad Hoc and Wireless Networks
(WiOpt), Tsukuba Science City, Japan, May 2011.
[34] M. Xiao, J. Wu, H. Huang, L. Huang, and W. Yang, “Deadline-sensitive opportunistic utility-
based routing in cyclic mobile social networks,” in Proc. of IEEE Intl. Conf. on Sensing,
Communication, and Networking (SECON), Seattle, WA, June 2015.
[35] C. Lee and K. I. Kim, “A Deadline Aware DTN Approach Based on Epidemic Routing,” in
Proc. of IEEE Intl. Symp. on Network Computing and Applications (NCA), Cambridge, MA,
August 2014.
[36] Q. Fu, B. Krishnamachari, and L. Zhang, “DAWN: A density adaptive routing for deadline-
based data collection in vehicular delay tolerant networks,” Tsinghua Science and Technology,
vol. 18, no. 3, pp. 230–241, June 2013.
[37] K. R. Chowdhury and M. D. Felice, “SEARCH: A routing protocol for mobile cognitive radio
ad-Hoc networks,” in Proc. of IEEE Sarnoff Symposium (SARNOFF), March 2009.
109
BIBLIOGRAPHY
[38] L. Ding, K. Gao, T. Melodia, S. Batalama, D. Pados, and J. Matyjas, “All-spectrum Cog-
nitive Networking through Jointly Optimal Distributed Channelization and Routing,” IEEE
Transactions on Wireless Communications, vol. 12, no. 11, pp. 5394–5405, November 2013.
[39] V. Srivastava and M. Motani, “Cross-layer design: a survey and the road ahead,” IEEE
Communications Magazine, vol. 43, no. 12, pp. 112–119, December 2005.
[40] L. D. Mendes and J. J. Rodrigues, “A survey on cross-layer solutions for wireless sensor
networks,” Journal of Network and Computer Applications, vol. 34, no. 2, pp. 523 – 534,
2011.
[41] B. Fu, Y. Xiao, H. J. Deng, and H. Zeng, “A survey of cross-layer designs in wireless networks,”
IEEE Communications Surveys Tutorials, vol. 16, no. 1, pp. 110–126, Firtquarter 2014.
[42] I. Al-Anbagi, M. Erol-Kantarci, and H. T. Mouftah, “A survey on cross-layer quality-of-service
approaches in wsns for delay and reliability-aware applications,” IEEE Communications
Surveys Tutorials, vol. 18, no. 1, pp. 525–552, Firstquarter 2016.
[43] R. Subramanian, B. Drozdenko, E. Doyle, R. Ahmed, M. Leeser, and K. R. Chowdhury, “High-
Level System Design of IEEE 802.11b Standard-Compliant Link Layer for MATLAB-Based
SDR,” IEEE Access Journal, April 2016.
[44] G. Sklivanitis, E. Demirors, A. Gannon, S. N. Batalama, D. A. Pados, and S. N. Batalama,
“All-spectrum cognitive channelization around narrowband and wideband primary stations,” in
Proc. of IEEE Global Communications Conference (GLOBECOM), San Diego, CA, December
2015.
[45] A. L. Drozd, T. Arcuri, J. Jagannath, D. A. Pados, T. Melodia, E. Demirors, and G. Sklivanitis,
“Network Throughput Improvement in Cognitive Networks by Joint Optimization of Spectrum
Allocation and Cross-layer Routing,” in Proc. of NATO Symp. on Cognitive Radio and Future
Network (IST-123), The Hague, The Netherlands, May 2014.
[46] E. Demirors, G. Sklivanitis, T. Melodia, and S. N. Batalama, “RcUBe: Real-Time Reconfig-
urable Radio Framework with Self-Optimization Capabilitites,” in Proc. of IEEE Intl. Conf.
on Sensing, Communication, and Networking (SECON), Seattle, WA, June 2015.
[47] J. Jagannath, H. Saarinen, W. Timothy, J. O’Brien, S. Furman, T. Melodia, and A. Drozd,
“COmBAT: Cross-layer Based Testbed with Analysis Tool Implemented Using Software
110
BIBLIOGRAPHY
Defined Radios,” in Proc. of IEEE Conference on Military Communications (MILCOM),
Baltimore, MD, USA, November 2016.
[48] M. C. Ng, K. E. Fleming, M. Vutukuru, S. Gross, Arvind, and H. Balakrishnan, “Airblue: A
System for Cross-layer Wireless Protocol Development,” in Proc. of the ACM/IEEE Sympo-
sium on Architectures for Networking and Communications Systems (ANCS), 2010.
[49] G. Nychis, T. Hottelier, Z. Yang, S. Seshan, and P. Steenkiste, “Enabling MAC Protocol Imple-
mentations on Software-defined Radios,” in Proc. of the USENIX Symposium on Networked
Systems Design and Implementation, 2009.
[50] I. Tinnirello, G. Bianchi, P. Gallo, D. Garlisi, F. Giuliano, and F. Gringoli, “Wireless MAC pro-
cessors: Programming MAC protocols on commodity Hardware,” in Proc. of IEEE Conference
on Computer Communications (INFOCOM), March 2012.
[51] N. McKeown, T. Anderson, H. Balakrishnan, G. Parulkar, L. Peterson, J. Rexford, S. Shenker,
and J. Turner, “Openflow: Enabling innovation in campus networks,” ACM Transaction
SIGCOMM Computer Communication Review, vol. 38, no. 2, pp. 69–74, March 2008.
[52] G. Nychis, T. Hottelier, Z. Yang, S. Seshan, and P. Steenkiste, “Enabling MAC Protocol
Implementations on Software-defined Radios,” in Proc. USENIX Symp. on Networked Systems
Design and Implementation (NSDI), April 2009.
[53] K. R. Chowdhury and T. Melodia, “Platforms and testbeds for experimental evaluation of
cognitive ad hoc networks,” IEEE Communications Magazine, vol. 48, no. 9, pp. 96–104,
September 2010.
[54] “Proportion of population covered by a mobile network, by technology,” https://sdg.data.gov/9-
c-1/, accessed: April. 27, 2018.
[55] P. P. Ray, M. Mukherjee, and L. Shu, “Internet of Things for Disaster Management: State-of-
the-Art and Prospects,” IEEE Access, vol. 5, pp. 18 818–18 835, 2017.
[56] O. A. Dobre, A. Abdi, Y. Bar-Ness, and W. Su, “Survey of automatic modulation classification
techniques: classical approaches and new trends.” IET Communications, vol. 1, no. 2, pp.
137–156, 2007.
111
BIBLIOGRAPHY
[57] G. Baldini, S. Karanasios, D. Allen, and F. Vergari, “Survey of Wireless Communication
Technologies for Public Safety,” IEEE Communications Surveys Tutorials, vol. 16, no. 2, pp.
619–641, Second 2014.
[58] “Project 25 Technology Interest Group (PTIG),” http://www.project25.org/, accessed: April.
27, 2018.
[59] ETSI EN 300 392-1 V1.4.1 (2009-01), “Terrestrial Trunked Radio (TETRA); Voice plus Data
(V+D); Part 1: General network design,” 2009.
[60] D. Rogers, “Interoperability: ensuring joined-up communication, 2007,”
https://www.geoconnexion.com/articles/interoperability-ensuring-joined-up-
communication/, accessed: April. 27, 2018.
[61] ETSI TR 102 580 V1.1.1, “Terrestrial Trunk Radio (TETRA); Release 2; Designer’s Guide;
TETRA High-Speed Data (HSD); TETRA Enhanced Data Service (TEDS),” 2007.
[62] H. Nishiyama, M. Ito, and N. Kato, “Relay-by-smartphone: realizing multihop device-to-
device communications,” IEEE Communications Magazine, vol. 52, no. 4, pp. 56–65, April
2014.
[63] K. Ali, H. X. Nguyen, Q. T. Vien, P. Shah, and Z. Chu, “Disaster Management Using D2D
Communication With Power Transfer and Clustering Techniques,” IEEE Access, vol. 6, pp.
14 643–14 654, 2018.
[64] “BRCK,” https://www.brck.com/connectivity/, accessed: April. 27, 2018.
[65] V. G. Menon, J. P. Pathrose, and J. Priya, “Ensuring Reliable Communication in Disaster
Recovery Operations with Reliable Routing Technique,” Mobile Information Systems, 2016.
[66] Z. Lu, G. Cao, and T. L. Porta, “TeamPhone: Networking SmartPhones for Disaster Recovery,”
IEEE Transactions on Mobile Computing, vol. 16, no. 12, pp. 3554–3567, December 2017.
[67] K. Ali, H. X. Nguyen, Q. T. Vien, P. Shah, and Z. Chu, “Disaster Management Using D2D
Communication With Power Transfer and Clustering Techniques,” IEEE Access, vol. 6, pp.
14 643–14 654, 2018.
[68] FCC White Paper, “The Public Safety Nationwide Interoperable Broadband Network, A New
Model for Capacity, Performance and Cost,” June 2010.
112
BIBLIOGRAPHY
[69] K. Gomez, L. Goratti, T. Rasheed, and L. Reynaud, “Enabling disaster-resilient 4G mobile
communication networks,” IEEE Communications Magazine, vol. 52, no. 12, pp. 66–73,
December 2014.
[70] J. Deaton, M. Schmitt, S. Cherry, and C. Papke, “Analyzing Options for Airborne Emergency
Wireless Communications,” Idaho National Laboratory, March 2008.
[71] “WiMAX in Haiti,” http://www.dailywireless.org/2010/02/12/wimax-in-haiti/, accessed: April.
27, 2018.
[72] A. Yarali, B. Ahsant, and S. Rahman, “Wireless Mesh Networking: A Key Solution for
Emergency & Rural Applications,” in Proc. of Second International Conference on Advances
in Mesh Networks, June 2009.
[73] S. S. Anjum, R. M. Noor, and M. H. Anisi, “Survey on MANET Based Communication
Scenarios for Search and Rescue Operations,” in Proc. of 5th International Conference on IT
Convergence and Security (ICITCS), August 2015.
[74] D. Ben Arbia, M. M. Alam, A. Kadri, E. Ben Hamida, and R. Attia, “Enhanced IoT-Based
End-To-End Emergency and Disaster Relief System,” Journal of Sensor and Actuator
Networks, vol. 6, no. 3, 2017. [Online]. Available: http://www.mdpi.com/2224-2708/6/3/19
[75] H. Aïache, V. Conan, G. Guibé, J. Leguay, C. Le Martret, J. M. Barcelo, L. Cerda, J. Garcia,
R. Knopp, N. Nikaein, X. Gonzalez, A. Zeini, O. Apilo, A. Boukalov, J. Karvo, H. Koskinen,
L. R. Bergonzi, J. C. Diaz, J. Meessen, C. Blondia, P. Decleyn, E. Van de Velde, and
M. Voorhaen, “WIDENS: Advanced Wireless Ad-Hoc Networks for Public Safety,” in Proc.
of IST Summit, 14th Mobile & Wireless Communication Summit, June 2005.
[76] K. Kanchanasut, A. Tunpan, M. Awal, T. Wongsaardsakul, D.Das, and Y.Tsuchimoto, “Build-
ing A Long-distance Multimedia Wireless Mesh Network for Collaborative Disaster Emer-
gency Responses,” Internet Education and Research Laboratory, Asian Institute of Technology,
2007.
[77] M. W. Subbarao, “Dynamic Power-Conscious Routing for MANETs: An Initial Approach,”
Journal of Research of the National Institute of Standards and Technology, vol. 104(6), pp.
587–593, 1999.
113
BIBLIOGRAPHY
[78] A. K. Chandra-Sekaran, A. Nwokafor, P. Johansson, K. D. Mueller-Glaser, and I. Krueger,
“ZigBee Sensor Network for Patient Localization and Air Temperature Monitoring During
Emergency Response to Crisis,” in Proc. of Second International Conference on Sensor
Technologies and Applications (SENSORCOMM), August 2008.
[79] “Raspberry Pi 3 Model B,” https://www.raspberrypi.org/products/raspberry-pi-3-model-b/,
accessed: April. 27, 2018.
[80] “LoRa (RF95),” https://www.digikey.com/catalog/en/partgroup/rfm95w-868s2/50906, ac-
cessed: April. 27, 2018.
[81] F. Adelantado, X. Vilajosana, P. Tuset-Peiro, B. Martinez, J. Melia-Segui, and T. Watteyne,
“Understanding the limits of lorawan,” IEEE Communications Magazine, vol. 55, no. 9, pp.
34–40, 2017.
[82] “Erle Robotics,” https://erlerobotics.com/blog/erle-copter/, accessed: April. 27, 2018.
[83] D. Pompili, M. C. Vuran, and T. Melodia, “Cross-layer Design in Wireless Sensor Networks,”
in Sensor Network and Configuration: Fundamentals, Techniques, Platforms, and Experiments.
Germany: Springer-Verlag, October 2006.
[84] R. Mehta and D. K. Lobiyal, “Energy efficient cross-layer design in MANETs,” in Proc. of
4th International Conference on Signal Processing and Integrated Networks (SPIN), February
2017, pp. 448–453.
[85] S. Colonnese, F. Cuomo, T. Melodia, and I. Rubin, “A Cross-Layer Bandwidth Allocation
Scheme for HTTP-Based Video Streaming in LTE Cellular Networks,” IEEE Communications
Letters, vol. 21, no. 2, pp. 386–389, 2017.
[86] J. Jagannath, T. Melodia, and A. Drozd, “DRS: Distributed Deadline-Based Joint Routing and
Spectrum Allocation for Tactical Ad-hoc Networks,” in Proc. of IEEE Global Communications
Conference (GLOBECOM), Washington D.C., USA, December 2016.
[87] J. Jagannath, S. Furman, A. Drozd, and T. Melodia, “Design and Experimental Evaluation
of a Cross-Layer Deadline-Based Joint Routing and Spectrum Allocation Algorithm,” IEEE
Transactions on Mobile Computing, pp. 1–1, 2018.
114
BIBLIOGRAPHY
[88] Q. Jiang, Y. Zhou, J. Wang, Y. Wang, and X. Wang, “A lightweight cross-layer cooperative
testbed for evaluation of connected vehicles,” in Proc. of 11th International Conference on
Mobile Ad-hoc and Sensor Networks (MSN), December 2015, pp. 194–201.
[89] J. Jagannath, H. Saarinen, T. Woods, J. O’Brien, S. Furman, A. Drozd, and T. Melodia,
“COmBAT: Cross-layer Based Testbed with Analysis Tool Implemented Using Software
Defined Radios,” in Proc. of IEEE Conf. on Military Communications (MILCOM), Baltimore,
MD, USA, November 2016.
[90] “Semtech LoRa Documentation,” https://www.semtech.com/uploads/documents/an1200.22.pdf,
accessed: April. 27, 2018.
[91] T. Nadeem, S. Banerjee, A. Misra, and A. K. Agrawala, Energy-Efficient Reliable Paths for
On-Demand Routing Protocols. Boston, MA: Springer US, 2005, pp. 485–496.
[92] S. Dawans, S. Duquennoy, and O. Bonaventure, “On link estimation in dense RPL deploy-
ments,” in Proc. of 37th Annual IEEE Conference on Local Computer Networks - Workshops,
Clearwater, FL, USA, October 2012.
[93] X. Chen, Y. Xu, and A. Liu, “Cross layer design for optimizing transmission reliability, energy
efficiency, and lifetime in body sensor networks,” Sensors, vol. 17, no. 4, p. 900, 2017.
[94] S. Liu, E. Ekici, and L. Ying, “Scheduling in multihop wireless networks without back-
pressure,” IEEE/ACM Transactions on Networking, vol. 22, no. 5, pp. 1477–1488, October
2014.
[95] J. Jagannath, T. Melodia, and A. Drozd, “DRS: Distributed Deadline-Based Joint Routing and
Spectrum Allocation for Tactical Ad-hoc Networks,” in Proc. of IEEE Global Communications
Conference (GLOBECOM), Washington DC, USA, December 2016.
[96] K. Hao, Z. Jin, H. Shen, and Y. Wang, “An efficient and reliable geographic routing protocol
based on partial network coding for underwater sensor networks,” Sensors, vol. 15, no. 6, pp.
12 720–12 735, 2015.
[97] H. Song, Z. Xu, and X. Wang, “Energy-efficient flooding with minimum latency for low-
duty-cycle WSNs,” in Proc. of 10th International Conference on Sensing Technology (ICST),
November 2016, pp. 1–6.
115
BIBLIOGRAPHY
[98] S. Ahn, Y. Lim, and H. Yu, “Energy-Efficient Flooding Mechanisms for the Wireless Sensor
Networks,” in Proc. of International Conference on Information Networking, Jan 2008, pp.
1–5.
[99] “Demonstration Video: HELPER Network - 6 Node Demonstration (User Perspective),”
https://youtu.be/wfXkuAtFlfQ, accessed: April. 27, 2018.
[100] “HELPER Network - 6 Node Demonstration (ERC Perspective),” https://youtu.be/TtWm3oX-
4z8, accessed: April. 27, 2018.
[101] A. Goldsmith, Wireless Communications. UK: Cambridge University Press, 2005.
[102] E. Hossain, M. Rasti, and L. B. Le, Radio Resource Management in Wireless Networks: An
Engineering Approach. UK: Cambridge University Press, 2017.
[103] A. Mahdy and J. S. Deogun, “Wireless optical communications: a survey,” in Proc. of IEEE
Wireless Communications and Networking Conference (WCNC), vol. 4, March 2004, pp.
2399–2404.
[104] K. Thyagarajan and A. Ghatak, Fiber Optic Communication Systems. Wiley-IEEE Press,
2007, pp. 100–124. [Online]. Available: http://ieeexplore.ieee.org/xpl/articleDetails.jsp?
arnumber=5224674
[105] H. Hemmati, A. Biswas, and I. Djordjevic, “Deep-space optical communications: Future
perspectives and applications,” Proceedings of the IEEE, vol. 99, no. 11, pp. 2020–2039,
November 2011.
[106] M. Kendall and M. Scholand, “Energy savings potential of solid state lighting in general
lighting applications,” US Department of Energy, Washington, DC, 2001.
[107] C. B. Liu, B. Sadeghi, and E. W. Knightly, “Enabling Vehicular Visible Light Communication
(V2LC) Networks,” in Proc. of the ACM Intl. Workshop on Vehicular Inter-networking,
September 2011.
[108] D. M. S and A. Berqia, “Li-Fi the Future of Vehicular Ad hoc Networks,” Transactions on
Networks and Communications, vol. 3, no. 3, 2015.
116
BIBLIOGRAPHY
[109] A. M. Cailean and M. Dimian, “Current Challenges for Visible Light Communications Usage
in Vehicle Applications: A Survey,” IEEE Communications Surveys Tutorials, vol. 19, no. 4,
pp. 2681–2703, Fourthquarter 2017.
[110] H. Yao, Z. Yang, H. Jiang, and L. Ma, “A Scheme of Ad-hoc-Based D2D Communication in
Cellular Networks,” Ad Hoc & Sensor Wireless Networks, vol. 32, pp. 115–130, 2016.
[111] H. Zhang, W. Ding, J. Song, and Z. Han, “A Hierarchical Game Approach for Visible
Light Communication and D2D Heterogeneous Network,” in IEEE Global Communications
Conference (GLOBECOM), Washington DC, USA, December 2016.
[112] W. Du, J. C. Liando, and M. Li, “SoftLight: Adaptive Visible Light Communication over
Screen-Camera Links,” in Proc. of IEEE International Conference on Computer Communica-
tions (INFOCOM), San Francisco, CA, April 2016.
[113] T. Li, C. An, Z. Tian, A. T. Campbell, and X. Zhou, “Human Sensing Using Visible Light
Communication,” in Proc. of MobiCom, Paris, France, September 2015.
[114] D. L. Begley, “Free-space laser communications: a historical perspective,” in The 15th Annual
Meeting of the IEEE Lasers and Electro-Optics Society, vol. 2, 2002, pp. 391–392.
[115] S. Ucar, S. Coleri Ergen, O. Ozkasap, D. Tsonev, and H. Burchardt, “SecVLC: Secure Visible
Light Communication for Military Vehicular Networks,” in Proceedings of the 14th ACM
International Symposium on Mobility Management and Wireless Access (MobiWac), Malta,
Malta, November 2016.
[116] A. Jovicic, J. Li, and T. Richardson, “Visible Light Communication: Opportunities, Challenges
and the Path to Market,” IEEE Communications Magazine, vol. 51, no. 12, pp. 26–32,
December 2013.
[117] S. Misra, S. C. Misra, and I. Woungang, Guide to Wireless Mesh Networks. Springer, 2009.
[118] B. Wu, J. Chen, J. Wu, and M. Cardei, A Survey of Attacks and Countermeasures in Mobile
Ad Hoc Networks, ser. Signals and Communication Technology. Springer US, 2007.
[119] I. Chlamtac, M. Conti, and J. J.-N. Liu, “Mobile ad hoc Networking: Imperatives and
Challenges,” Ad Hoc Networks, vol. 1, no. 1, pp. 13 – 64, 2003.
117
BIBLIOGRAPHY
[120] Y. Zhang, L. T. Yang, and J. Chen, RFID and Sensor Networks: Architectures, Protocols,
Security, and Integrations, 1st ed. Boca Raton, FL, USA: CRC Press, Inc., 2009.
[121] D. Karunatilaka, F. Zafar, V. Kalavally, and R. Parthiban, “LED Based Indoor Visible Light
Communications: State of the Art,” IEEE Communications Surveys Tutorials, vol. 17, no. 3,
pp. 1649–1678, 3rd Quarter, 2015.
[122] P. H. Pathak, X. Feng, P. Hu, and P. Mohapatra, “Visible Light Communication, Networking,
and Sensing: A Survey, Potential and Challenges,” IEEE Communications Surveys Tutorials,
vol. 17, no. 4, pp. 2047–2077, Fourthquarter 2015.
[123] J. Zhang, X. Zhang, and G. Wu, “Dancing with Light: Predictive In-frame Rate Selection for
Visible Light Networks,” in Proc. IEEE Conf. on Computer Communications (INFOCOM),
Hong Kong S.A.R., PRC, April 2015.
[124] T. Li, C. An, Z. Tian, A. T. Campbell, and X. Zhou, “Human Sensing Using Visible Light
Communication,” in Proceedings of the 21st Annual International Conference on Mobile
Computing and Networking (MobiCom), Paris, France, September 2015.
[125] S. Verma, A. Shandilya, and A. Singh, “A model for Reducing the Effect of Ambient Light
Source in VLC System,” in Proceedings of IEEE International Advance Computing Conference
(IACC), Gurgaon, India, February 2014.
[126] S. Yin and O. Gnawali, “Towards Embedded Visible Light Communication Robust to Dynamic
Ambient Light,” in Proceedings of IEEE Global Communications Conference (GLOBECOM),
Washington DC, USA, December 2016.
[127] H. Joshi, R. Green, and M. Leeson, “Channel Models for Optical Wireless Communication
Systems,” in Proc. IEEE International conf. on Transparent Optical Networks (ICTON),
Azores, June 2009.
[128] S. Rajagopal, R. Roberts, and S.-K. Lim, “IEEE 802.15.7 Visible Light Communication:
Modulation Schemes and Dimming support,” IEEE Communications Magazine, vol. 50, no. 3,
pp. 72–82, March 2012.
[129] Z. Ghassemlooy, W. Popoola, and S. Rajbhandari, Optical Wireless Communications: System
and Channel Modelling with MATLAB. Boca Raton, FL, USA: CRC Press, Inc., 2012.
118
BIBLIOGRAPHY
[130] F. Miramirkhani and M. Uysal, “Channel Modeling and Characterization for Visible Light
Communications,” IEEE Photonics Journal, vol. 7, no. 6, pp. 1–16, December 2015.
[131] S. Park, D. Jung, H. Shin, Y. Hyun, K. Lee, and Y. Oh, “Information Broadcasting System
based on Visible Light Signboard,” Proc. of Wireless Optical Communication, pp. 311–313,
2007.
[132] J. Vucic, C. Kottke, S. Nerreter, K. Habel, A. Buttner, K.-D. Langer, and J. Walewski, “125
Mbit/s over 5 m Wireless Distance by Use of OOK-Modulated Phosphorescent white LEDs,”
in Proce. IEEE European Conference on Optical Communication, Vienna, Austria, September
2009.
[133] ——, “230 Mbit/s via a Wireless Visible-Light Link based on OOK Modulation of Phosphores-
cent White LEDs,” in Proc. IEEE International Conference on Optical Fiber Communication,
collocated National Fiber Optic Engineers (OFC/NFOEC), San Diego, CA, USA, March
2010.
[134] Q. Wang, D. Giustiniano, and D. Puccinelli, “OpenVLC: Software-defined Visible Light
Embedded Networks,” in Proceedings of the 1st ACM MobiCom Workshop on Visible Light
Communication Systems, ser. VLCS ’14. New York, NY, USA: ACM, 2014, pp. 15–20.
[Online]. Available: http://doi.acm.org/10.1145/2643164.2643167
[135] C. Gavrincea, J. Baranda, and P. Henarejos, “Rapid Prototyping of Standard-compliant Visible
Light Communications System,” IEEE Communications Magazine, vol. 52, no. 7, pp. 80–87,
July 2014.
[136] Y. Qiao, H. Haas, and K. Edward, “Demo: A Software-defined Visible Light Communications
System with WARP,” in Proc. ACM MobiCom Workshop on Visible Light Communication
Systems (VLCS), Maui, Hawaii, September, 2014.
[137] Z. Wang, Y. Liu, Y. Lin, and S. Huang, “Full-duplex MAC protocol based on adaptive
contention window for visible light communication,” IEEE/OSA Journal of Optical Communi-
cations and Networking, vol. 7, no. 3, pp. 164–171, March 2015.
[138] L. KIX and K. Hirohashi, “High-Speed Full-Duplex Multiaccess System for LED-Based
Wireless Communications using Visible Light,” in Proc. of Intl. Symp. on Optical Eng. and
Photonic Tech. (OEPT), July 2009.
119
BIBLIOGRAPHY
[139] S. Schmid, G. Corbellini, S. Mangold, and T. R. Gross, “LED-to-LED Visible Light Commu-
nication Networks,” in Proc. of the Fourteenth ACM Intl. Symp. on Mobile Ad Hoc Networking
and Computing (MobiHoc), July 2013.
[140] “IEEE 802.15.7 Standard for Local and Metropolitan Area Networks–Part 15.7: Short-Range
Wireless Optical Communication Using Visible Light,” IEEE Std 802.15.7-2011, pp. 1–309,
September 2011.
[141] N.-T. Le, S. Choi, and Y. M. Jang, “Cooperative MAC protocol for LED-ID systems,” in Proc.
of Intl. Conf. on ICT Convergence (ICTC), Seoul, Korea, September 2011.
[142] J. Dang and Z. Zhang, “Comparison of optical OFDM-IDMA and optical OFDMA for uplink
visible light communications,” in Proc. of Intl. Conf. on Wireless Communications & Signal
Processing (WCSP), Huangshan China, October 2012.
[143] B. Ghimire and H. Haas, “Resource allocation in optical wireless networks,” in Proc. of IEEE
Intl. Symp. on Personal Indoor and Mobile Radio Communications (PIMRC), Toronto, Canada,
September 2011.
[144] C. Chen, M. Ijaz, D. Tsonev, and H. Haas, “Analysis of downlink transmission in DCO-
OFDM-based optical attocell networks,” in Proc. of IEEE Global Communications Conference
(GLOBECOM), Austin, TX, December 2014.
[145] D. Bykhovsky and S. Arnon, “Multiple Access Resource Allocation in Visible Light Com-
munication Systems,” Journal of Lightwave Technology, vol. 32, no. 8, pp. 1594–1600, April
2014.
[146] M. H. Shoreh, A. Fallahpour, and J. A. Salehi, “Design concepts and performance analysis of
multicarrier CDMA for indoor visible light communications,” IEEE/OSA Journal of Optical
Communications and Networking, vol. 7, no. 6, pp. 554–562, June 2015.
[147] C. He, L. liang Yang, P. Xiao, and M. A. Imran, “DS-CDMA assisted visible light communi-
cations systems,” in Proc. of IEEE 20th Intl. Workshop on Computer Aided Modelling and
Design of Communication Links and Networks (CAMAD), Guildford, UK, September 2015.
[148] J. A. Salehi, “Emerging OCDMA communication systems and data networks,” Journal of
Optical Networking, vol. 6, no. 9, pp. 1138–1178, Sep 2007.
120
BIBLIOGRAPHY
[149] J. A. Martin-Gonzalez, E. Poves, and F. J. Lopez-Hernandez, “Random optical codes used in
optical networks,” IET Communications, vol. 3, no. 8, pp. 1392–1401, 2009.
[150] M. F. Guerra-Medina, O. Gonzalez, B. Rojas-Guillama, J. A. Martin-Gonzalez, F. Delgado,
and J. Rabadan, “Ethernet-OCDMA system for multi-user visible light communications,”
Electronics Letters, vol. 48, no. 4, pp. 227–228, February 2012.
[151] S. H. Chen and C. W. Chow, “Color-Shift Keying and Code-Division Multiple-Access Trans-
mission for RGB-LED Visible Light Communications Using Mobile Phone Camera,” IEEE
Photonics Journal, vol. 6, no. 6, pp. 1–6, August 2014.
[152] N.-T. Le and Y. M. Jang, “Broadcasting MAC protocol for IEEE 802.15.7 visible light
communication,” in Proc. of Intl. Conf. on Ubiquitous and Future Networks (ICUFN), Da
Nang, Vietnam, July 2013.
[153] O. Bouchet, P. Porcon, M. Wolf, L. Grobe, J. Walewski, S. Nerreter, K. Langer, L. Fernàndez,
J. Vucic, T. Kamalakis, G. Ntogari, and E. Gueutier, “Visible-light communication system
enabling 73 Mb/s data streaming,” in Proc. of IEEE GLOBECOM Workshops (GC Wkshps),
Miami, FL, USA, December 2010.
[154] T. V. Pham and A. T. Pham, “Max-Min Fairness and Sum-Rate Maximization of MU-VLC
Local Networks,” in Proc. of IEEE Globecom Workshops (GC Wkshps), San Diego, CA, USA,
December 2015.
[155] Z. Yu, R. J. Baxley, and G. T. Zhou, “Multi-user MISO broadcasting for indoor visible light
communication,” in Proc. of IEEE Intl. Conf. on Acoustics, Speech and Signal Processing
(ICASSP), Vancouver, Canada, May 2013.
[156] Q. H. Spencer, A. L. Swindlehurst, and M. Haardt, “Zero-forcing methods for downlink spatial
multiplexing in multiuser MIMO channels,” IEEE Transactions on Signal Processing, vol. 52,
no. 2, pp. 461–471, February 2004.
[157] J. Chen, Y. Hong, Z. Wang, and C. Yu, “Precoded visible light communications,” in Proc. of
Intl. Conf. on Information, Communications and Signal Processing (ICICS), Tainan, Taiwan,
December 2013.
121
BIBLIOGRAPHY
[158] Y. Hong, J. Chen, Z. Wang, and C. Yu, “Performance of a Precoding MIMO System for
Decentralized Multiuser Indoor Visible Light Communications,” IEEE Photonics Journal,
vol. 5, no. 4, pp. 7 800 211–7 800 211, August 2013.
[159] V. Stankovic, M. Haardt, and M. Fuchs, “Combination of block diagonalization and THP
transmit filtering for downlink beamforming in multi-user MIMO systems,” in Proc. of
European Conference on Wireless Technology, Amsterdam, The Netherlands, October 2004.
[160] J. Chen, N. Ma, Y. Hong, and C. Yu, “On the performance of MU-MIMO indoor visible
light communication system based on THP algorithm,” in Proc. of IEEE/CIC International
Conference on Communications in China (ICCC), Shenzhen, China, October 2014.
[161] “LED light specification,” http://www.theledlight.com/technical1.html.
[162] S. Dimitrov and H. Haas, “Principles of LED Light Communications.” Cambridge, United
Kingdom: Cambridge University Press, 2015.
[163] S.-Y. Jung, D.-H. Kwon, S.-H. Yang, and S.-K. Han, “Inter-cell interference mitigation in
multi-cellular visible light communications,” Opt. Express, vol. 24, no. 8, pp. 8512–8526,
April 2016.
[164] P. H. Pathak, X. Feng, P. Hu, and P. Mohapatra, “Visible light communication, networking,
and sensing: A survey, potential and challenges,” IEEE Communications Surveys & Tutorials,
vol. 17, no. 4, pp. 2047–2077, 2015.
[165] O. Narmanlioglu, R. C. Kizilirmak, F. Miramirkhani, and M. Uysal, “Cooperative visible light
communications with full-duplex relaying,” IEEE Photonics Journal, vol. 9, no. 3, pp. 1–11,
June 2017.
[166] Y. Qin, D. Guo, G. Cheng, D. Zhang, and L. Luo, “Wswdc: Vlc enabled wireless small-world
data centers,” in 2017 IEEE 23rd International Conference on Parallel and Distributed Systems
(ICPADS), December 2017.
[167] B. BÃ c©chadergue, L. Chassagne, and H. Guan, “Suitability of visible light communication
for platooning applications: An experimental study,” in Proc. of Global LIFI Congress (GLC),
February 2018.
122
BIBLIOGRAPHY
[168] D. Pompili, M. C. Vuran, and T. Melodia, “Cross-layer Design in Wireless Sensor Networks,”
in Sensor Network and Configuration: Fundamentals, Techniques, Platforms, and Experiments.
Germany: Springer-Verlag, October 2006.
[169] M. Bilgi and M. Yuksel, “Capacity Scaling in Free-Space-Optical Mobile Ad Hoc Networks,”
Ad Hoc Networks (Elsevier), vol. 12, pp. 150–164, January 2014.
[170] X. Li, R. Zhang, J. Wang, and L. Hanzo, “Cell-Centric and User-Centric Multi-User Scheduling
in Visible Light Communication aided Networks,” in Proc. International Conference on
Communicaitons (ICC), London, UK, June 2015.
[171] L. Klaver and M. Zuniga, “Shine: A step towards distributed multi-hop visible light communi-
cation,” in 2015 IEEE 12th International Conference on Mobile Ad Hoc and Sensor Systems
(MASS), October 2015.
[172] H. Yang and A. Pandharipande, “Full-duplex relay VLC in LED lighting linear system
topology,” Proc. of 39th Annual Conference of the IEEE Industrial Electronics Society
(IECON), 2013.
[173] ——, “Full-duplex relay VLC in LED lighting triangular system topology,” Proc. of Interna-
tional Symposium on Communications, Control and Signal Processing (ISCCSP), 2014.
[174] Z. Wu, “Free Space Optical Networking With Visible Light: A Multi-hop Multi-Access
Solution,” Ph.D. Dissertation, College of Engineering, Boston University, Boston, MA, 2012.
[175] G. Bianchi, L. Fratta, and M. Oliveri, “Performance evaluation and enhancement of the
CSMA/CA MAC protocol for 802.11 wireless LANs,” in Seventh IEEE International Sympo-
sium on Personal, Indoor and Mobile Radio Communications (PIMRC), October 1996.
[176] J. Jagannath and T. Melodia, “An Opportunistic Medium Access Control Protocol for Visible
Light Ad Hoc Networks,” in Proc. of International Conference on Computing, Networking
and Communications (ICNC), Maui, Hawaii, USA, March 2018.
123