Green Wireless Video Sensor Networks using a Low-Power ...Doutora Marília Pascoal Curado Professora...

M A P teleDOCTORAL PROGRAMME IN TELECOMMUNICATIONS

Green Wireless Video Sensor Networks using aLow-Power Control Channel

Filipe Miguel Monteiro da Silva e Sousa

A dissertation submitted in partial fulfilment of

the requirements for the degree of

Doctor of Philosophy (PhD) in

Telecommunications

Supervisor: Manuel Alberto Pereira Ricardo (PhD)Full Professor at Faculdade de Engenharia da Universidade do Porto

Co-Supervisor: Rui Lopes Campos (PhD)Head of the Wireless Networks Area at INESC TEC

January, 2019

c© Filipe Miguel Monteiro da Silva e Sousa: January, 2019

The Jury

President

Doutor Henrique Manuel de Castro Faria SalgadoProfessor Catedrático da Faculdade de Engenharia da Universidade do Porto

Examiners Committee

Doutor Carlos Miguel Tavares de Araújo Cesariny CalafateProfessor Catedrático da Universitat Politècnica de València, Espanha

Doutora Marília Pascoal CuradoProfessora Associada da Universidade de Coimbra

Doutor Adriano Jorge Cardoso MoreiraProfessor Associado da Escola de Engenharia da Universidade do Minho

Doutor José António Ruela Simões FernandesProfessor Aposentado da Faculdade de Engenharia da Universidade do Porto

Doutor Manuel Alberto Pereira RicardoProfessor Catedrático da Faculdade de Engenharia da Universidade do Porto

M A P teleDOCTORAL PROGRAMME IN TELECOMMUNICATIONS

is a joint Doctoral Programme provided by

This work was supported in part by the Project TEC4Growth-Pervasive Intelligence, Enhancers, and Proofs of

Concept with Industrial Impact under Grant NORTE-01-0145-FEDER-000020 and in part by the Project

Symbiotic Technology for Societal Efficiency Gains: Deus ex Machina under Grant

NORTE-01-0145-FEDER-000026, both financed by the North Portugal Regional Operational Programme

(NORTE 2020), through the PORTUGAL 2020 Partnership Agreement, and the European Regional Development

Fund.

To Isabel, Mariana, Inês, and my parents

Abstract

The availability of low cost networked wireless devices and video cameras is enabling

Wireless Video Sensor Networks (WVSNs), which can be used in scenarios such as healthcare,

agriculture, smart cities, intelligent transportation systems, and surveillance. These scenarios

typically require that each node sends a video stream to a server located in the cloud. IEEE

802.11 is considered a suitable technology for transmitting video wirelessly, as it supports

high data rates. However, when using a multi-hop topology to extend IEEE 802.11 coverage,

IEEE 802.11-based WVSNs suffer from three problems: low network performance, throughput

unfairness, and energy inefficiency. In multi-hop networks, the Carrier Sense Multiple Access

– Collision Avoidance (CSMA/CA) leads to low network performance, related to the hidden

node problem. Furthermore, the Wireless Video Sensors (WVS) closer to the gateway tend

to monopolise the medium making the other WVSs starve, causing throughput unfairness.

Energy inefficiency is caused by the low performance of CSMA/CA in multi-hop topologies

since frame collisions imply retransmissions. Moreover, a packet is overheard by several nodes

in the same broadcast domain even when that node is not the destination, thus wasting energy.

Since relay nodes are forced to stay in idle listening most of the time to forward packets from

other nodes, energy is also lost.

In this thesis, we propose a holistic solution for WVSNs, named Green wiReless vidEo

sENsor NEtworks uSing out-of-band Signalling (GREENNESS), to address these problems.

GREENNESS combines a centralised node polling mechanism with out-of-band signalling

over a low power radio. The polling mechanism improves the network performance and

throughput fairness. The use of a low power radio to convey the signalling, namely the node

that shall transmit and the nodes that shall switch OFF their IEEE 802.11 interfaces, saves

energy. By using numerical, simulation, and experimental analysis we show that GREENNESS

can achieve significant energy savings and improve network capacity, packet loss ratio, and

throughput fairness when compared to state-of-the-art CSMA/CA-based WVSN solutions.

GREENNESS can save up to 92 % in the energy consumption, guarantee an improvement

i

ii

in network capacity above 30 % and perfect throughput fairness.

Keywords: Energy-efficiency, Low Power Radio, Network Performance, Out-of-band

Signalling, Wireless Video Sensor Networks.

Resumo

O surgimento de dispositivos de rede sem fios e de câmaras de vídeo de baixo custo

possibilitou o aparecimento das redes de sensores de vídeo sem fios (WVSNs), que podem

ser utilizadas para diferentes aplicações como saúde, agricultura, cidades inteligentes, sistemas

de transporte inteligentes e videovigilância. Nestes cenários, exige-se normalmente que cada

nó envie um fluxo de vídeo para um servidor localizado na nuvem. A tecnologia IEEE 802.11

é considerada adequada para transmissão de vídeo em redes sem fios, já que suporta taxas

de transferência de dados elevadas. No entanto, a utilização de uma topologia de rede com

vários saltos para aumentar a cobertura de uma WVSN basead em IEEE 802.11 encerra três

problemas: baixo desempenho, iniquidade na taxa de transferência de dados e ineficiência

energética. Em redes com múltiplos saltos, o Carrier Sense Multiple Access – Collision

Avoidance (CSMA/CA) leva ao baixo desempenho da rede, relacionado com o problema do

nó escondido. Além disso, os sensores de vídeo sem fios (WVS) mais próximos da gateway

tendem a monopolizar o meio, fazendo com que os outros WVSs não consigam transmitir os

seus dados, causando iniquidade na taxa de transferência de dados. A ineficiência energética

é causada pelo baixo desempenho do CSMA/CA em topologias com múltiplos saltos, dado

que as colisões das tramas implicam a sua retransmissão. A energia também é desperdiçada

quando a transmissão de uma trama é ouvida por vários nós no mesmo domínio de difusão

mesmo quando esse nó não é o destino da trama. Como os nós de retransmissão são forçados

a permanecer ligados grande parte do tempo para encaminhar pacotes de outros nós, existe um

desperdício de energia adicional.

Nesta tese, propomos uma solução holística para WVSNs, denominada Green WiReless

vidEo sENsor NEtworks uSing Out-of-band Signaling (GREENNESS) para endereçar os três

problemas. O GREENNESS combina um mecanismo de polling centralizado com sinalização

fora de banda com recurso a um rádio de baixa potência. O mecanismo de polling melhora

o desempenho da rede e a equidade na taxa de transferência de dados. A utilização de um

rádio de baixa potência para transmitir a sinalização permite controlar os nós da rede que

iii

iv

podem transmitir dados e os que devem desligar as suas interfaces IEEE 802.11, poupando

desta forma energia. Através de análise numérica, simulação e experimentação demonstramos

que o GREENNESS, quando comparado com soluções WVSN do estado da arte baseadas em

CSMA/CA, pode alcançar poupanças significativas de energia, melhorar a capacidade da rede

e a equidade da taxa de transferência de dados. O GREENNESS pode poupar até 92 % do

consumo energético, garantir um aumento da capacidade de rede superior a 30 % e equidade

perfeita na taxa de transferência de dados.

Keywords: Eficiência Energética, Radio de Baixo Consumo, Sinalização Fora de Banda,

Desempenho da Rede, Redes de Sensores de Vídeo Sem Fios.

Acknowledgements

First, I would like to thank Prof. Manuel Ricardo for his support and valuable guidance.

I also would like to thank Rui Campos for supporting me in this endeavour and useful

suggestions that help me to overcome the challenges. I was fortunate for having João Dias

and Filipe Ribeiro’s collaboration and support in the development and testing of the prototype.

Thanks to Fraunhofer Portugal - AICOS for supporting my involvement in the MAP-Tele

doctoral programme and also my close colleagues that encouraged me and motivated to

achieve this objective. I would also like to express my thanks to the contributions from many

individuals, in numerous public presentations and paper reviews that helped to shape my work.

I am grateful to my parents for teaching me to never give up and for their support over the

years. I want to extend my thanks to my close friends and family for their understanding and

relaxing evenings during this journey. Finally, I would like to thank and dedicate this thesis to

my precious treasures and loves of my life Isabel, Mariana, and Inês.

Filipe Sousa

v

“Our greatest weakness lies in giving up. The most certain way to succeed is always to try just

one more time.”

Edison, Thomas A.

vii

Contents

List of Figures xiii

List of Tables xvii

List of Abbreviations xix

1 Introduction 11.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Problem Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.3 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.4 Original Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.5 Publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.5.1 Journals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.5.2 Conferences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.5.3 Workshops and Talks . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

1.6 Document Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2 State-of-the-Art in Energy-Efficient Solutions 92.1 Out-of-Band Control Oriented Solutions . . . . . . . . . . . . . . . . . . . . . 9

2.1.1 Solutions Adopting the Wake-Up Radio Receiver Scheme . . . . . . . 12

2.1.2 Solutions Adopting the Wake-Up Radio Transceiver Scheme . . . . . . 14

2.2 MAC Oriented Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

2.2.1 Contention-Based . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

2.2.2 Hybrid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

2.2.3 Power Saving Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

2.3 Routing Oriented Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

2.3.1 QoS-Based . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30

2.3.2 Swarm Intelligent-Based . . . . . . . . . . . . . . . . . . . . . . . . . 39

ix

x CONTENTS

2.3.3 Network Structure-Based . . . . . . . . . . . . . . . . . . . . . . . . . 42

2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

3 GREENNESS 513.1 GREENNESS Concept and Architecture . . . . . . . . . . . . . . . . . . . . . 51

3.2 Routing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

3.3 WVSN Active Topology Collection Mechanism . . . . . . . . . . . . . . . . . 55

3.4 Node Scheduling Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

3.5 Failure Recovery Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

3.6 Low Power Radio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

3.6.1 Low Power Radio Requirements . . . . . . . . . . . . . . . . . . . . . 62

3.6.2 Candidate Technologies . . . . . . . . . . . . . . . . . . . . . . . . . 64

3.6.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

3.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

4 GREENNESS Evaluation 694.1 Evaluation Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

4.2 Numerical Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

4.2.1 Chain Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

4.2.2 Binary Tree Topology . . . . . . . . . . . . . . . . . . . . . . . . . . 77

4.2.3 Grid Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

4.2.4 Random Network Topologies . . . . . . . . . . . . . . . . . . . . . . 79

4.2.5 Numerical Analysis of Energy Consumption for PACE and GREEN-

NESS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

4.3 Simulation Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

4.4 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

4.4.1 Gateway . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

4.4.2 WVS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88

4.5 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

4.5.1 Numerical and Simulation Results . . . . . . . . . . . . . . . . . . . . 91

4.5.2 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

4.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

4.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

5 Conclusion 1015.1 Work Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101

5.2 Contributions Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

CONTENTS xi

5.3 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

References 105

List of Figures

1.1 Forecast of connected devices with an IP stack [1]. . . . . . . . . . . . . . . . 1

1.2 Reference scenario for WVSNs using Wi-Fi cameras in multi-hop network

topology and streaming video through a gateway to a cloud server. . . . . . . . 3

1.3 Hidden node and exposed node problems in IEEE 802.11-based multi-hop

networks. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

1.4 The GREENNESS concept, with the node scheduling mechanism running over

the LPR control channel illustrated by the arrows in orange. . . . . . . . . . . . 6

2.1 Wireless Video Sensor Architecture with Wake-Up Radio [15]. . . . . . . . . . 10

2.2 Wake-Up Radio communication schemes [14]. . . . . . . . . . . . . . . . . . 11

2.3 Tang et al model of a mobile node [17]. . . . . . . . . . . . . . . . . . . . . . 12

2.4 SleepyCAM power management solution [26]. . . . . . . . . . . . . . . . . . 14

2.5 Mekonnen et al Multi-Tier Architecture [18]. . . . . . . . . . . . . . . . . . . 15

2.6 Data exchange in S-MAC adaptive listen mode [30]. . . . . . . . . . . . . . . 17

2.7 Data exchange in T-MAC with TA [32]. . . . . . . . . . . . . . . . . . . . . . 18

2.8 Data exchange in T-MAC with FRTS [32]. . . . . . . . . . . . . . . . . . . . . 18

2.9 Data gathering in D-MAC [36]. . . . . . . . . . . . . . . . . . . . . . . . . . . 19

2.10 Y-MAC frame format [40]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

2.11 Y-MAC channel hopping mechanism [40]. . . . . . . . . . . . . . . . . . . . . 21

2.12 Z-MAC channel-scheduling algorithm [41]. . . . . . . . . . . . . . . . . . . . 22

2.13 Frame structure of ER-MAC [42]. . . . . . . . . . . . . . . . . . . . . . . . . 23

2.14 Buffer threshold setting in EE-Hybrid MAC based on the hop-count from the

sink [43]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

2.15 M-PSM method to forward packets during a beacon interval [51]. . . . . . . . 27

2.16 MH-PSM method to forward packets during a beacon interval [50]. . . . . . . 28

2.17 EAPSM management components [52]. . . . . . . . . . . . . . . . . . . . . . 28

2.18 OPAMA example with an AP and one Station [53]. . . . . . . . . . . . . . . . 30

2.19 SPEED architecture [59]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32

xiii

xiv LIST OF FIGURES

2.20 Architecture of the real-time scheme for (m,k)-firm streams for WSNs [64]. . . 35

2.21 Architecture diagram for RTLD [66]. . . . . . . . . . . . . . . . . . . . . . . . 36

2.22 WVSN topology for AntSensNet with backbone outlined in black and connect-

ing the cluster heads [74]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

2.23 Greedy forwarding example for TPGF [81]. . . . . . . . . . . . . . . . . . . . 44

2.24 PWDGR example for selecting pair-wise [83]. . . . . . . . . . . . . . . . . . . 45

2.25 Classification of the state-of-the-art energy efficient solutions. . . . . . . . . . 46

3.1 The GREENNESS concept with the node scheduling mechanism running over

the LPR control channel illustrated by the arrows in orange. . . . . . . . . . . . 52

3.2 GREENNESS Architecture. . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

3.3 Registration and Registration Acknowledgement messages used to collect the

WVSN active topology. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

3.4 Typical LPR message format [84]. . . . . . . . . . . . . . . . . . . . . . . . . 64

4.1 Time Sequence Diagram for Chain Topology . . . . . . . . . . . . . . . . . . 76

4.2 Regular network topologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

4.3 Random Network Topology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80

4.4 Network simulation with 30 WVSs randomly positioned in a 500 m x 500 m

area with the gateway identified with GW. . . . . . . . . . . . . . . . . . . . . 84

4.5 Testbed with a gateway and six WVSs using a Raspberry Pi model B boards as

basis for the GREENNESS proof-of-concept prototype. . . . . . . . . . . . . . 86

4.6 The three regular WVSN topologies used to evaluate GREENNESS. . . . . . . 87

4.7 GREENNESS modification to the RDS message. . . . . . . . . . . . . . . . . 88

4.8 FM Tuner and connection with RPi. . . . . . . . . . . . . . . . . . . . . . . . 89

4.9 Energy saving achieved by GREENNESS with respect to PACE for WVSNs

with different sizes and average number of hops. . . . . . . . . . . . . . . . . . 91

4.10 Energy saving exhibited by GREENNESS when varying PLPR and considering

Wi-Fi radios in sleep mode. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

4.11 Impact of different offered network loads and WVSNs sizes in the energy

saving of GREENNESS. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

4.12 Network Capacity of GREENNESS and CSMA/CA for different offered net-

work loads with average number of hops equal to 2. . . . . . . . . . . . . . . . 93

4.13 One-way-delay of GREENNESS and CSMA/CA for different offered network

loads with an average number of hops equal to 2. . . . . . . . . . . . . . . . . 94

4.14 Packet Loss Ratio for GREENNESS and CSMA/CA for different offered

network loads and average number of hops equal to 2. . . . . . . . . . . . . . . 95

LIST OF FIGURES xv

4.15 Energy consumption of GREENNESS and PACE considering testbed, simula-

tion, and numerical evaluations for three scenarios . . . . . . . . . . . . . . . . 96

4.16 Network capacity achieved by GREENNESS and PACE during testbed exper-

iments and simulations for the three scenarios. . . . . . . . . . . . . . . . . . . 96

List of Tables

2.1 Comparison of out-of-band control solutions. . . . . . . . . . . . . . . . . . . 47

2.2 Comparison of MAC oriented solutions. . . . . . . . . . . . . . . . . . . . . . 48

2.3 Comparison of routing oriented solutions. . . . . . . . . . . . . . . . . . . . . 49

3.1 Notations used in the description of the WATCM and NSM mechanisms. . . . . 54

3.2 Candidate Low Power Radios. . . . . . . . . . . . . . . . . . . . . . . . . . . 64

4.1 Notations used in the equations derived in the numerical analysis as well as in

the simulations, and experiments. . . . . . . . . . . . . . . . . . . . . . . . . . 71

4.2 Simulation parameters. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83

4.3 Values of the parameters PLPR, Pidle, and PWiFisleep considered in the numerical

and simulations analysis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

4.4 Values of the measured parameters PLPR and Pidle. . . . . . . . . . . . . . . . . 90

xvii

List of Acronyms

A-MSDU Aggregated MAC Service Data Unit

AC Access Category

ACK Acknowledgement

ACOWMSN Ant Colony Optimization-Based QoS Routing Algorithm

ADV-MAC Advertisement MAC

AGEM Adaptive Greedy-compass Energy-aware Multipath protocol

AM Amplitude Modulation

AntSensNet Ant-based multi-QoS routing metric

AODV Ad-hoc On-Demand Distance Vector

AP Access Point

ARP Address Resolution Protocol

ARQ Automatic Repeat Request

ASAR Ant-based Service-Aware Routing

ATCM Active Topology Creation and Maintenance

ATIM Ad-hoc Traffic Indication Map

ATIM-ACK ATIM-Acknowledgement

B-MAC Berkeley Media Access Control

BLE Bluetooth Low Energy

xix

xx List of Acronyms

CA Collision Avoidance

CAGR Compound Annual Growth Rate

CAQR Correlation-Aware QoS Routing

CBR Constant Bit Rate

CCA Clear Channel Assessment

COTS Commercial Off-The-Shelf

CRC Cyclic Redundancy Check

CSMA Carrier Sense Multiple Access

CSMA/CA Carrier Sense Multiple Access – Collision Avoidance

CTS Clear To Send

CW Contention Window

D-MAC Data–Gathering Medium Access Control

DARA Distributed Aggregate Routing Algorithm

DBP Distance-Based Priority

DCSIS Differential Coding-based Source and Intermediate nodes Selection

DD Directional Diffusion

DDPS Delay-differentiated Packet Scheduling

DGR Directional Geographic Routing

DMA Direct Memory Access

DSR Dynamic Source Routing

DSSS Direct Sequence Spread Spectrum

EA-TPGF Energy-Aware TPGF

EAPSM Energy Aware PSM

EAQoS Energy-Aware QoS

List of Acronyms xxi

EDCA Enhanced Distributed Channel Access

EE-Hybrid MAC Energy Efficient Hybrid MAC

Eo11 Ethernet-over-802.11

EQSR Energy efficient and QoS aware multipath Routing

ER-MAC Emergency-MAC

ESSID Extended Service Set Identification

FCS Frame Check Sequence

FEC Forward Error Correction

FRM Failure Recovery Mechanism

FRTS Future-Request-To-Send

FSK Frequency-Shift Keying

GPIO General Purpose Input Output

GPS Global Positioning System

GPSR Greedy Perimeter Stateless Routing

GREENNESS Green wiReless vidEo sENsor NEtworks uSing out-of-band Signalling

GWR-MAC Generic WUR based MAC protocol

HTSMAC High Throughput Sensor MAC

IAR Improved Adaptive Routing

IBSS Independent Basic Service Set Identifier

IFS Inter-Frame Space

IoT Internet of Things

IP Internet Protocol

IPC Inter-Process Communication

xxii List of Acronyms

ISM Industrial, Scientific and Medical

LEAR Load Based Energy-Aware Multimedia Routing

LECIM Low-Energy Critical Infrastructure Monitoring

LoRaWAN Long Range Wide Area Networking

LPL Low Power Listening

LPR Low Power Radio

LPWAN Low-Power Wide Area Networking

LSI Local Status Indicator

M-IAR Multimedia-Enabled Improved Adaptive Routing

M-PSM Modified PSM

MAC Medium Access Control

MAP Mesh Access Point

MCMP Multi-Constraint Multi-Path

MCRA Multi-Constrained Routing Algorithm

MCU Micro-Controller Unit

MH-PSM Multi-Hop PSM

MLME MAC Sublayer Management Entity

MMSPEED Multipath Multi Stateless Protocol for Real-Time Communication

MSN Maximum Slot Number

MTU Maximum Transmission Unit

NAV Network Allocation Vector

NFL Neighborhood Feedback Loop

NSM Node Scheduling Mechanism

List of Acronyms xxiii

OEDSR Optimized Energy-Delay Subnetwork Routing

OPAMA Optimized Power save Algorithm for continuous Media Applications

OWD One-Way-Delay

PCA Priority Channel Access

PEMuR Energy efficient and perceived QoS aware video routing

PIR Pyroelectric Infrared

PLR Packet Loss Ratio

PRR Packet Reception Rate

PSM Power Saving Mode

PSNR Peak Signal to Noise Ratio

PWDGR Pairwise Directional Geographical Routing

PWM Pulse-Width Modulation

QEMAC QoS-supported Energy-efficient MAC

QMOR QoS aware Multi-sink Opportunistic Routing

QoE Quality of Experience

QoS Quality of Service

RDS Radio Data System

REAR Real-Time and Energy-Aware Routing

RFID Radio-Frequency Identification

RPAR Real-time Power-Aware Routing

RPi Raspberry Pi

RREQ Route REQuest

RTID Radio – Triggered ID

xxiv List of Acronyms

RTLD Real-time routing protocol with load distribution

RTS Request To Send

RTS/CTS Request To Send/Clear To Send

S-MAC Sensor MAC

SAR Sequential Assignment Routing

SFD Start Frame Delimiter

SGF Selective Greedy Forwarding

SHPER Scalable Hierarchical Power Efficient Routing

SI Swarm Intelligence

SNGF Stateless Nondeterministic Geographic Forwarding

SNR Signal-to-Noise Ratio

SoBT Sleep on Beacon Transmission

SPEED Stateless Protocol for Real-Time Communication

SYNC Synchronisation

T-MAC Timeout MAC

TDMA Time Division Multiple Access

TIM Traffic Indication Map

TLS Time To Live

TPGF Two-Phase Geographic Greedy Forwarding

TR Topology Refresh

TxOP Transmit Opportunity

VBR Variable Bit Rate

VHF Very High Frequency

VoIP Voice-over-IP

WATCM WVSN Active Topology Collection Mechanism

WDS Wireless Distribution System

WiFIX Wi-Fi network Infrastructure eXtension

WMN Wireless Mesh Network

WMSN Wireless Multimedia Sensor Network

WSN Wireless Sensor Network

WUR Wake-Up Radio

WVS Wireless Video Sensor

WVSN Wireless Video Sensor Network

Z-MAC Zebra MAC

Chapter 1

Introduction

1.1 Motivation

In 2019, Internet of Things (IoT) devices are expected to surpass the number of mobile

phones and achieve the figure of 18 billion by 2022 [1]. IoT devices are foreseen to increase at

a Compound Annual Growth Rate (CAGR) of 21 %, driven by new use cases such as connected

cars, meters, wearables, industry, and agriculture. Fig. 1.1 presents the growth achieved so far

and the forecast for connected devices with an Internet Protocol (IP) stack until 2022. While the

number of laptops, tablets, mobile phones, and fixed phones connected has stopped growing,

the number of IoT devices using unlicensed, short-range radios such as Wi-Fi and Bluetooth

is growing exponentially and predicted to be around 16 billion by 2022 (bar in light green),

clearly surpassing the other types of devices [1].

Figure 1.1: Forecast of connected devices with an IP stack [1].

On the other hand, the global wireless video surveillance market is growing at a CAGR of

more than 20 %, driven by the shift from analogue to IP cameras, the low prices of cameras, and

1

2 Introduction

the increased security concerns [2]. IP cameras are affordable, flexible, scalable, and easy to

install. Thus, the shift from analogue to digital cameras is facilitated, with many installations

in stadiums, city surveillance projects, and hotels already available. With the emergence of

cloud storage for the video streams generated, the overall costs of surveillance systems have

been reduced, increasing the adoption by small businesses and residential market segments

[2]. Because of political instability and terrorism in many regions around the world, stringent

regulations have been approved to install security systems in public locations such as hospitals,

airports, and railway stations. Notwithstanding, the growth of wireless cameras carries an

energy demand to power-up these devices, either by connecting them directly to the power-grid

or using batteries possibly recharged through renewable energy sources (e.g., solar panels).

The IT-services represent 2 % of all global carbon emissions, the same percentage of

the emissions from the aviation sector [3]. The carbon footprint will grow since, by 2022,

the number of IoT devices will be six times the number of devices in 2016. Based on

these forecasts, governments, organisations, companies, and academia are discussing and

investigating how energy consumption can be reduced. Within this context, a trend on "green

networking" has emerged. Therefore, energy consumption optimisation of IoT devices, even

for sensors without a battery, is important to control the expected growth of carbon emissions.

For self-powered devices, e.g. Wireless Video Sensors (WVSs) running on batteries, the energy

consumption becomes even more critical since it can affect the device regular operation. If the

energy consumption is reduced, the battery can be smaller and the solution cost and carbon

footprint lowered.

The outburst of connected low-cost devices, combined with the availability of affordable

video cameras, is contributing to the emergence of Wireless Video Sensor Networks (WVSNs)

as part of the IoT paradigm [4]. WVSNs enable a range of new applications in fields such as

healthcare, agriculture, smart cities, intelligent transportation systems, and surveillance [5]. In

these scenarios, there is usually the requirement of sending the video streams to a server located

in the cloud [6][7]. The video stream should be transmitted reliably to the cloud with time

constraints and minimal packet loss. Ethernet would be the candidate technology to fulfil these

requirements but, for covering large areas for scenarios such as agriculture and environment

monitoring, becomes expensive and not suitable. IEEE 802.11, also known as Wi-Fi (the two

terms are used interchangeably in this thesis), is a suitable technology for transmitting video

wirelessly, as it is ubiquitous and supports high data rates, especially the new variants IEEE

802.11ac [8] and IEEE 802.11ad [9]. For covering large areas, a single Access Point (AP) is

not sufficient, so cameras need to relay information in a multi-hop topology to assure video

1.2 Problem Statement 3

transmission to the cloud. Although other solutions like multiple APs interconnected using

Wireless Distribution System (WDS) could be used to guarantee network coverage, this would

increase the complexity of the solution. The installation becomes a complex and costly task

since WDS requires a qualified technician to configure the virtual links between APs based

on their radio coverage. Therefore, cameras equipped with an IEEE 802.11 radio and capable

of communicating in multi-hop are envisioned as a suitable solution for these scenarios, as

illustrated in Fig. 1.2.

Figure 1.2: Reference scenario for WVSNs using Wi-Fi cameras in multi-hop networktopology and streaming video through a gateway to a cloud server.

1.2 Problem Statement

The transmission of video from WVSs through the gateway to the cloud has constraints

regarding latency, packet delivery ratio, and throughput. IEEE 802.11-based WVSNs with

multi-hop topologies and formed by WVSs supporting a single radio have three major prob-

lems: low performance, throughput unfairness, and energy inefficiency [10]. In what follows,

we detail each of them.

IEEE 802.11 uses the Carrier Sense Multiple Access – Collision Avoidance (CSMA/CA)

mechanism to control the access to the medium. When used in multi-hop networks, this

mechanism leads to low network performance, affecting latency and packet delivery ratio,

4 Introduction

(a) Hidden node problem. (b) Exposed node problem.

Figure 1.3: Hidden node and exposed node problems in IEEE 802.11-based multi-hopnetworks.

namely due to the hidden node problem [11]. As shown in Fig. 1.3a, when node A transmits

a packet to node B and node C transmits a packet to node D at the same time, since node

B is in the range of C, this generates a collision at node B. Since node C can not sense the

transmission from node A, it can start a new data transmission to node B or D, originating

a collision at node B. The Request To Send/Clear To Send (RTS/CTS) mechanism can solve

the hidden node problem but creates the exposed node problem. In Fig. 1.3b, nodes C and D

start to exchange data using RTS/CTS. Since node B and node E are in the transmission range,

they receive the Request To Send (RTS) and Clear To Send (CTS) messages, respectively, and

enter in the backoff period. During this period, node A and node F can send an RTS message,

but nodes B and E do not reply due to the ongoing data transmission between nodes C and D,

causing the exposed node problem. This affects the network throughput and latency.

The multi-hop nature of a WVSN brings up throughput unfairness since the nodes closer

to the gateway tend to monopolise the medium, making the other nodes to starve [12]. In

a multi-hop topology with every WVS sending video to the cloud, the WVSs closer to the

gateway have the highest throughput, since their own traffic is enough to potentially fill up

their queues, causing the packets from the farther WVSs to be dropped. This leads to the

throughput unfairness problem.

The energy inefficiency problem is a consequence of 1) the collision of packets forcing

their retransmission, 2) the overhearing of packets that are destined to other nodes, and 3)

the idle listening since nodes must actively listen to the channel to receive packets. The

collision of packets is caused by the low performance of CSMA/CA in multi-hop topologies,

as explained above. Collisions imply retransmissions, thus wasting energy. Energy is lost

in multi-hop topologies when a packet is overheard by several nodes in the same broadcast

domain, forcing them to switch to receive mode and decode the packet even when that node is

not the destination. WVS network interfaces must always be ON, even when not transmitting

or receiving any packet, in order to forward packets from other nodes, forcing them to stay in

1.3 Objectives 5

idle listening most of the time.

The transmission of video streams over IEEE 802.11-based WVSNs with multi-hop topolo-

gies brings up challenges. Yet, by minimising collisions and only turning ON Wi-Fi radios

when there is data to transmit/receive, network performance can be improved and energy can

be saved. This is the approach followed in this thesis.

1.3 Objectives

The aim of this thesis is to develop a solution enabling green multi-hop WVSNs. The

reference scenario is illustrated in Fig. 1.2. In this scenario, all the WVSs are equipped with

cameras and send the video streams to the gateway using Wi-Fi, which in turn forwards them

to the cloud server. The main objective is to minimise the energy consumption of WVSNs

when transmitting video from each WVS to the cloud server. To attain the main objective, we

consider the following specific objectives:

• Study related work about increasing energy efficiency in multi-hop scenarios, consider-

ing the problems related to CSMA/CA presented in Section 1.2.

• Design an energy efficient solution that offers time guarantees and a high packet delivery

ratio for wireless video sensing scenarios.

• Evaluate the proposed solution to determine the achievable energy consumption, perfor-

mance, and throughput fairness for comparison with a CSMA/CA-based solution.

1.4 Original Contributions

The main contribution of this thesis is the Green wiReless vidEo sENsor NEtworks

uSing out-of-band Signalling (GREENNESS) solution. Featuring a low power out-of-band

control channel and a traffic-aware Node Scheduling Mechanism (NSM), it enables significant

energy savings while improving network capacity and throughput fairness when compared

to CSMA/CA-based WVSNs. The GREENNESS solution includes the following specific

contributions:

• GREENNESS concept and architecture. GREENNESS combines a node polling

mechanism, such as the one defined in [13], with the use of out-of-band signalling

over a Low Power Radio (LPR) integrated into each WVS node. Fig. 1.4 presents the

6 Introduction

GREENNESS concept, considering a multi-hop WVSN with an LPR installed in each

WVS and the gateway.

LPR range

CWVS#

1

C

C C CC

Wi-Fi Link

Gateway

LPR

WVS#1

WVS#2

WVS#3

WVS#5

WVS#4

WVS#6

mac3

mac4

mac5

mac6

mac2mac

1

LPR Link

Figure 1.4: The GREENNESS concept, with the node scheduling mechanism running over theLPR control channel illustrated by the arrows in orange.

• WVSN Active Topology Collection Mechanism (WATCM). The WATCM mechanism

finds the network topology and computes the optimum polling order for the NSM

mechanism. The polling order is calculated to minimise the number of times the

Wi-Fi radio changes between ON/OFF states, as a higher number of changes affects

the performance. The control messages overhead are also optimised by controlling the

relay nodes and WVSs with a single Poll message.

• Node Scheduling Mechanism (NSM). The NSM mechanism uses the LPR included in

the gateway to schedule the WVS data transmissions and turn the WVS Wi-Fi radios

ON/OFF accordingly. By including a traffic-aware mechanism, energy efficiency is

further improved. When there is no traffic, a WVS is not polled, so the Wi-Fi radio

is kept OFF. The traffic-aware mechanism learns the traffic pattern for Constant Bit Rate

(CBR) flows and changes the WVSs polling order accordingly.

1.5 Publications 7

• Failure Recovery Mechanism (FRM). The FRM mechanism runs when the network

topology changes. When the topology changes or a node is disconnected from the

network, a failure is detected and WVSs are forced to turn ON their Wi-Fi interfaces.

This allows the routing algorithm to run and the WATCM to find the new WVSN

topology.

GREENNESS differs from related work by supporting multi-hop networks and minimising

signalling overhead, without changing the current IP stack. Furthermore, it addresses the low

performance, throughput unfairness, and energy inefficiency that affect WVSNs in a holistic

way.

1.5 Publications

1.5.1 Journals

• F. Sousa, J. Dias, F. Ribeiro, R. Campos, and M. Ricardo, “Green Wireless VideoSensor Networks Using Low Power Out-of-Band Signalling”, IEEE Access, vol. 6,

pp. 30024–30038, Jun. 2018.

• R. Campos, R. Duarte, F. Sousa, M. Ricardo, and J. Ruela, “Network InfrastructureExtension Using 802.1D-based Wireless Mesh Networks”, Wireless Communications

and Mobile Computing, vol. 11, no. 1, pp. 67–89, Jan. 2011.

1.5.2 Conferences

• F. Sousa, J. Dias, F. Ribeiro, R. Campos, and M. Ricardo, “A Traffic-aware Solutionfor Green Wireless Video Sensor Networks”, in Proc. of IFIP/IEEE Wireless Days

2017, Porto, Portugal, Mar. 2017.

• J. Dias, F. Sousa, F. Ribeiro, R. Campos, and M. Ricardo, “Green Wireless VideoSensor Networks using FM Radio System as Control Channel”, in Proc. of WONS

2016, Cortina d’Ampezzo, Italy, Jan. 2016.

• F. Sousa, R. Campos, and M. Ricardo, “Energy-efficient Wireless Multimedia SensorNetworks using FM as a Control Channel”, in Proc. of the 9th IEEE Symposium on

Computers and Communications (ISCC’14), Funchal, Portugal, Jun. 2014.

8 Introduction

1.5.3 Workshops and Talks

• F. Sousa, J. Dias, F. Ribeiro, R. Campos, and M. Ricardo, “A Traffic-aware Solutionfor Green Wireless Video Sensor Networks”, 23o RTCM Seminar, Aveiro, Portugal,

Jul. 2017.

• F. Sousa, R. Campos, and M. Ricardo, “Energy-efficient Wireless Multimedia SensorNetworks using FM as a Control Channel”, in MAP-Tele Workshop, Porto, Portugal,

May 2016.

• F. Sousa, R. Campos, and M. Ricardo, “Energy-efficient Wireless Multimedia SensorNetworks using FM as a Control Channel”, in MAP-Tele Workshop, Aveiro, Portugal,

June 2014.

• F. Sousa, F. Abrantes, and M. Ricardo, ‘Cooperation Between WPAN and WLANNodes For Efficient And Interoperable Communication”, in MAP-Tele Workshop,

Guimarães, Portugal, May 2012.

• F. Sousa, F. Abrantes, and M. Ricardo, ‘Cooperation Between WPAN and WLANNodes For Efficient And Interoperable Communication”, in MAP-Tele Workshop,

Aveiro, Portugal, May 2011.

1.6 Document Structure

The structure of this PhD thesis is as follows. Chapter 2 presents the related work on green

wireless video sensors networks. Chapter 3 describes the GREENNESS solution, namely the

traffic-aware node scheduling mechanism and the candidate wireless technologies for imple-

menting the low-power control channel. Chapter 4 presents the evaluation of the GREENNESS

solution considering numerical, simulations, and experimental analysis. Chapter 5 reviews this

PhD thesis work, draws the main conclusions, recalls the main contributions of the PhD thesis,

and points out the future work.

Chapter 2

State-of-the-Art in Energy-EfficientSolutions

IEEE 802.11-based Wireless Video Sensor Networks (WVSNs) with multi-hop topologies

and formed by Wireless Video Sensors (WVSs) supporting a single radio have three major

problems: low-performance, throughput unfairness, and energy inefficiency. A number of

solutions have been proposed in the literature to address these problems. In the state-of-the-art

survey provided in this chapter, we present energy efficient solutions, offering performance

guarantees for the targeted wireless video sensing scenario.

In this chapter, we classify the state-of-the-art solutions in three types: 1) out-of-band

control oriented; 2) MAC oriented; 3) routing oriented. The out-of-band control oriented

solutions are divided in two sub-categories: solutions using a Wake-Up Radio (WUR)-receiver

and solutions using a WUR-transceiver. The MAC oriented type is further divided into

contention based, hybrid, and Power Saving Mode (PSM) based. The routing oriented type

is further classified according to the following aspects: QoS constraints, “Swarm Intelligence

(SI) based” routing, and network structure.

2.1 Out-of-Band Control Oriented Solutions

In order to improve the energy efficiency in WVSNs, it is necessary to reduce the idle

listening of the Wi-Fi interface. A possible solution to increase the energy efficiency is to use

an out-of-band control channel to reduce the idle listening of Wi-Fi interfaces while keeping

latency low. To achieve this, a WUR is added to the WVS so that it can continuously listen to

the control channel, as shown in Fig. 2.1. When the WUR receives a wake-up signal, the Micro-

9

10 State-of-the-Art in Energy-Efficient Solutions

Figure 2.1: Wireless Video Sensor Architecture with Wake-Up Radio [15].

Controller Unit (MCU) and the main radio are awake. The literature defines three different

schemes to use WURs [14]:

WVS with WUR-receiver the source WVS sends a wake-up signal and the destination WVS

receives it through the WUR-receiver, which wakes-up the Main transceiver of the

destination WVS. Afterwards, an acknowledgement message is sent to the source WVS

and the transmission of data to the destination WVS is started. This is illustrated in

Fig. 2.2a.

WVS with WUR-transceiver where the source WVS sends a wake-up signal through its

WUR-transceiver and the destination WVS receives it through its WUR-transceiver.

The Main transceiver of the destination WVS is woken up, and an acknowledgement

message is sent to the source WVS using the WUR-transceiver. The source WVS can

then transmit the data to the destination WVS using the Main transceiver. This is shown

in Fig. 2.2b.

WVS with WUR-transceiver only the source WVS sends the wake-up signal and data to

the destination WVS, using the WUR transceiver. The destination WVS receives such

information through the WUR-transceiver and sends an acknowledgement message to

the source WVS. There is no Main transceiver in this case. This is represented in

Fig. 2.2c.

In these three schemes, the WUR-transceiver comprises low-power transmitting and re-

ceiving circuits, while the WUR-receiver includes the receiving circuit only. A WVS with

the WUR-receiver only implies unidirectional communications, while a WVS with the WUR-

transceiver ensures bidirectional communications. The scheme using the WUR-transceiver

2.1 Out-of-Band Control Oriented Solutions 11

only can be used to wake-up a transponder, as proposed in [16]. The proposed solution consists

of a low power radio device which is typically in sleep mode and can be woken up by an event

and moved to the active state. Since our objective is to stream video from several WVSs, a

WUR-transceiver only cannot be adopted because of its low data rate. Thus, the valid schemes

when it comes to the use of out-of-band control channel are WVS with WUR-receiver and

WVS with WUR-transceiver.

(a) WUR-receiver waking up the main transceiver after wake-up signal

(b) WUR-transceiver receiving and transmitting a wake-up signal

(c) WUR-transceiver receiving and transmitting both wake-up signal and data.

Figure 2.2: Wake-Up Radio communication schemes [14].

When the WUR-receiver scheme is used, a central node uses the WUR for sending out-

of-band signalling and for controlling the access to the medium of the WVSs that carry a

WUR-receiver [17]. In the WUR-transceiver scheme, the control is delegated to a WVS which

performs simple tasks of gathering information from the surrounding environment and can also


trigger the access to the medium of WVSs. This scenario is used when the WVSN adopts a

multi-tier architecture [18]. In the following subsections, we refer to state-of-the-art solutions

exploring these two schemes.

2.1.1 Solutions Adopting the Wake-Up Radio Receiver Scheme

Tang et al [17] present a solution for energy and spectrum efficiency by tightly integrating a

low-power WUR with a WLAN module, which is only used for the data transmission/reception.

The WUR is used for carrier sense, contention control, and remote/local wake-up. Fig. 2.3

shows the low power WUR that is kept awake to monitor the channel and the WLAN module

put in sleep mode when the channel is idle. Before sending a packet, the WUR performs carrier

sense and backoff (BO). A packet is sent when the backoff counter reaches 0, and the WLAN

module wakes up (WuR-CS). Furthermore, the Contention Window (CW) is adjusted based on

the length of Inter-Frame Space (IFS) measured by the WUR and the estimation of the number

of contention slots for each transmission (WuR-CSMA). Compared with CSMA, WuR-CSMA

reduces the power consumption by more than 90 % in the saturation case, when the traffic rate

is 5000 packet/s. Nevertheless, WuR-CS and WuR-CSMA were designed for Wi-Fi networks

running in infrastructure mode and not the multi-hop scenarios targeted in this thesis.

Figure 2.3: Tang et al model of a mobile node [17].

RT-Link is a time synchronised real-time sensor networking platform proposed by Rowe

et al [19] that uses an Amplitude Modulation (AM) signal to synchronise the network nodes

globally. This platform employs a Time Division Multiple Access (TDMA) scheme where the

nodes go to sleep and wake up during their transmission time slot. RT-Link was designed for


scenarios that require throughput and latency guarantees, together with energy efficiency. RT-

Link outperforms Berkeley Media Access Control (B-MAC) regarding packet collisions and

end-to-end delay. Nevertheless, RT-Link was not tested for wireless video sensing scenarios

and does not provide throughput fairness.

In [20] a radio-triggered circuit is used to switch the environmental sensors between wake-

up and sleep modes; when a sensor node is in the sleep mode, all its components are shut

down, except the memory, the interrupt handler, and the timer. A radio signal can power-up

the radio-triggered circuit and change nodes’ state to wake-up mode. This solution employs

a multiple-frequency technique by using a Radio – Triggered ID (RTID) to improve the

selectivity of sensors that should be in wake-up mode. The selectivity of the solution is reduced

because it selects more nodes than needed to transmit information and uses multiple radios and

frequencies. The poor selectivity reduces the energy savings. Moreover, RTID does not offer

any Quality of Service (QoS) guarantees.

A working prototype for a WUR is presented by Doom et al [21], which was designed with

standard components and reuses as much as possible the primary radio (CC1000) of the T-node

to operate in the 868 MHz band. The wake-up signals are generated by software that toggles

the transmitting power of the CC1000 radio between the minimum and maximum values. The

receiving circuit is dedicated to filter out interference and retrieve the wake-up signal. Still,

the wake-up radio is not designed with multi-hop topologies in mind and only addresses the

energy inefficiency problem.

The solutions described in [22] and [23], despite being proposed for very different sce-

narios, are also based on the concept of shutting down or entering in low power state when a

sensor or device is in idle state. In [22] a TinyNode 184 [24] is used to carry out-of-band control

information between a terminal and the Access Point (AP), in order to maintain connectivity

and wake up the AP when necessary. Whenever the terminal starts a new session – e.g., Internet

browsing or Voice-over-IP (VoIP) – it sends a beacon through the TinyNode 184 to wake up

the AP. The beacon carries a Time To Live (TLS) field that indicates the amount of time the

AP Wi-Fi interface must be turned ON. Using this scheme, the authors estimate that, for an

average usage pattern of 4 hours of active Internet usage per day, the total real power savings

are 23 %. Nevertheless, this solution was designed for Wi-Fi networks in infrastructure mode

and does not support multi-hop wireless topologies. Furthermore, it only addresses the energy

inefficiency problem. In [23] the authors adopt a similar scheme to increase the battery lifetime

of a PDA-based phone by reducing its idle power consumption. To achieve this, the wireless

network card of the PDA-based phone is shut down when the device is not in use. The device


is powered when an incoming call is received through a MiniBrick, which is a low-power out-

of-band signalling radio added to the PDA-based phone. Although the authors claim that they

double the battery lifetime, the solution was only designed for single-hop scenarios.

2.1.2 Solutions Adopting the Wake-Up Radio Transceiver Scheme

The patent presented in [25] describes an implementation to reduce the battery consumption

of an energy-constrained computing device, by selecting between a low-power radio (low

data rate with a low power consumption) and a high-power radio (high data-rate with high

power consumption) to minimise power consumption while keeping effective wireless data

communications. The dual-radio communications system switches from the low-power radio to

the high-power radio when the user demands high data-rate, the low-power radio is congested,

or the data generated by the low-power radio exceeds a predefined threshold. This patent

addresses the energy efficiency problem for single networks but does not discuss multi-hop

scenarios and the performance and throughput unfairness problems.

Figure 2.4: SleepyCAM power management solution [26].

Mekonnen et al propose the sleepyCAM power management solution [26][27] for wireless

video sensing scenarios, which is illustrated in Fig. 2.4. SleepyCAM uses a Pyroelectric

Infrared (PIR) sensor to detect movement and uses an ATmega1281 to control the power

supply of the camera node composed of a Raspberry Pi (RPi) and a camera module. This

solution evolved to a multi-tier Wireless Sensor Network (WSN) with motion sensors in tier

1, connected through Bluetooth Low Energy (BLE) radios and a Wi-Fi network in tier 2,


Figure 2.5: Mekonnen et al Multi-Tier Architecture [18].

connecting the RPi-Camera nodes [18], as shown in Fig. 2.5. PIR motion sensors, using BLE

radios, activate the streaming from the RPi-Camera nodes to a remote PC when motion is

detected. Although Mekonnen et al proposed solution addresses a video streaming scenario,

it relies on motion to activate the streaming and does not solve the low-performance problem

when movement is detected in the range of all cameras, and multiple streams are sent to the

cloud.

A two-tier strategy for priority based critical event surveillance with wireless multimedia

sensors was proposed by Bhatt et al [28]. The authors developed a two-tier architecture

with audio nodes densely deployed and video nodes sparsely deployed. Audio sensors work

continuously to capture events and send this information to the base station which wakes up the

video nodes in the region of interest. According to the authors, this approach is energy efficient

and has a low deployment cost, but it cannot be applied to the wireless video sensing targeted

scenarios since audio and video nodes coexist in a WVS.

The Generic WUR based MAC protocol (GWR-MAC) was proposed in [15] to improve

energy efficiency by avoiding idle listening. Each node has two radios that are not restricted

to any WUR technology. The wake-up procedure is bidirectional and can be source-initiated

or sink-initiated. In the source-initiated mode, the sensor nodes send a wake-up signal to the

sink node. When the signal is received, the main radio of the sink node is turned ON and

a beacon message is broadcasted to initiate the transmission period for the sensor nodes. In

the sink-initiated mode, the sink node wakes up the other sensor nodes from the sleep mode.

When the sensor nodes receive the wake-up signal, the Main transceiver is turned ON, and

an acknowledgement message is sent to the sink nodes through the WUR. Afterwards, the

sink node sends a beacon message containing information about the following transmission


period to the sensor nodes. Nonetheless, GWR-MAC was not designed for video streaming

scenarios since the energy consumption only decreases significantly when the number of events

transmitted per hour is low [15].

2.2 MAC Oriented Solutions

As mentioned in Chapter 1, most of the problems associated with IEEE 802.11-based

WVSN are caused by the Medium Access Control (MAC) mechanism used. One alternative is

to have sleep periods that are controlled by a duty cycle-based MAC protocol when incoming

transmissions are not detected. Duty cycle based MAC protocols are suitable for applications

with low traffic load since by lowering the duty cycle this will increase the communication

delay. Therefore, this section presents MAC oriented solutions, including contention-based,

hybrid, and power saving solutions, that have been proposed in the state-of-the-art to tackle the

three problems stated in Chapter 1.

2.2.1 Contention-Based

In contention-based MAC protocols, the WVS contend for accessing the media and colli-

sions are avoided through probabilistic coordination.

Sensor MAC (S-MAC) is a contention-based protocol that reduces idle listening by peri-

odically putting nodes into sleep state to save energy [29]. To attain this objective, S-MAC

nodes change between the listen and sleep states in a duty-cycle. In the sleep state, each node

turns OFF its radio and sets a timer to change to the listen state. Since S-MAC requires the

receiver and the sender to be simultaneous in the listening state, they need to be periodically

synchronised to avoid node’s clock drift. Besides, nodes share their own sleep schedules by

broadcasting them to their neighbours. S-MAC adaptive listening was proposed in [30] to

improve the latency in a multi-hop network caused by the periodic sleep state. For a multi-hop

network, neighbouring nodes need to wait for the next listening period to transmit data, which

increases the end-to-end latency. To address this issue, during an adaptive listen period nodes

can overhear a neighbour’s transmission – e.g., Request To Send (RTS) and Clear To Send

(CTS) – to learn its duration and adaptively wake up when the transmission is over. Fig. 2.6

presents a timing diagram with this sequence. When the next-hop node is a neighbour of

the sender, it receives the RTS, or in the case it is a receiver, it receives the CTS. In either

case, the neighbours learn the transmission duration and can schedule the wake-up period

to reduce latency. S-MAC outperforms IEEE 802.11 for light offered traffic. However, for

2.2 MAC Oriented Solutions 17

video transmitting scenarios, S-MAC consumes more energy than IEEE 802.11 because of the

overhead it uses with Synchronisation (SYNC) packets [31].

Figure 2.6: Data exchange in S-MAC adaptive listen mode [30].

Since the listening period duration of S-MAC is fixed, when the duration is too large this

results in a waste of energy; when it is too small, data loss occurs. To overcome the fixed

listening period, Timeout MAC (T-MAC) was proposed. It uses an active period that adapts to

the traffic pattern [32]. Fig. 2.7 shows that during the active time T-MAC sensors send data and

wait for a time TA. When no communication is observed, they return to the low power sleep

mode to minimise the idle listening. The TA has to be long enough so that any neighbour can

hear a potential CTS frame. In multi-hop networks, T-MAC suffers from the early sleeping

problem, caused by a node not hearing the CTS message. Fig. 2.8 exemplifies this problem.

Node A sends an RTS message and node B sends back a CTS message. Although node C

can still hear node’s B CTS message, node D cannot and returns to sleep mode after TA is

over. The Future-Request-To-Send (FRTS) technique was defined to address the problem. The

FRTS message is sent when a CTS message is overheard by node C, as shown in Fig. 2.8.

When node D receives the FRTS message, it keeps active, waiting to receive messages from

node C. The FRTS technique increases the throughput by 75 % [32]. For high traffic loads, both

S-MAC and T-MAC are inefficient since the exchanged messages are concentrated in a short

time. Moreover, in T-MAC the FRTS technique increases the idle listening times and therefore

the energy consumption of nodes.


Figure 2.7: Data exchange in T-MAC with TA [32].

Figure 2.8: Data exchange in T-MAC with FRTS [32].

Another alternative to achieve low power consumption is the B-MAC, which reduces the

duty cycle and idle listening by implementing an adaptive preamble sampling scheme [33]. The

low power consumption is attained by reducing the radio’s duty cycle and idle listening through

a periodic channel sampling named Low Power Listening (LPL). This mechanism is similar to

preamble sampling in Aloha [34] but was designed for different radio characteristics. The

node powers up when it detects an incoming packet and stays awake enough time to process

the packet and return to sleep mode. B-MAC uses noise floor estimation for finding a clear

channel to transmit data but also to find if the channel is active during LPL. To avoid collisions,

B-MAC uses Clear Channel Assessment (CCA) together with backoff mechanisms for reliable

link layer acknowledgements. B-MAC was designed for applications that focus on energy

efficiency and outperforms S-MAC, but it is not fair with respect to packet delivery ratios.

Data–Gathering Medium Access Control (D-MAC) was designed and optimised for data

gathering trees in WSNs, i.e., sequentially wake-up nodes in a multi-hop path to forward data

across different nodes until it reaches the gateway, as shown in Fig. 2.9 [35]. D-MAC divides


the node’s schedule in three modes: receiving, sending, and sleeping. In Fig. 2.9 the tree leaf

node starts in receiving mode, but the neighbour does not have data to send. Afterwards, the

leaf node switches to sending mode and sends a packet to the upstream neighbour, which was

already in receiving mode and sends back an Acknowledgement (ACK) message. The leaf node

can now turn OFF its radio and change to sleeping mode. The node can request more slots if

it needs to transmit more data. The receiving and sending periods are fixed and identical,

being enough to transmit and receive one packet. When the depth of the tree is known, the

nodes can set their wake schedule ahead from the gateway, and periodically move between the

three states. D-MAC decreases the latency and energy consumption for tree-based multi-hop

topologies and can adapt the duty cycle to the traffic variation. Nevertheless, D-MAC does

not consider node fairness, and interference between nodes in the same depth level is handled

through further protocol overhead.

Figure 2.9: Data gathering in D-MAC [36].

2.2.2 Hybrid

A hybrid MAC protocol is the combination of a TDMA with Carrier Sense Multiple Access

(CSMA). TDMA has the advantage of being suitable for high traffic scenarios since it avoids

collisions, but it requires global synchronisation, it does not adapt to topology changes, and

hardly discovers interference amongst neighbour nodes. On the other hand, CSMA provides

the best result for low traffic scenarios, but it experiences problems with the hidden terminal,

which can cause more packet collisions. During the data period, only the intended receivers

are awake, and the other nodes are in a low power sleep mode.


Advertisement MAC (ADV-MAC) is a hybrid MAC protocol similar to S-MAC and T-

MAC, but it uses an advertisement message for intended receivers to minimise the energy lost

in idle listening while keeping throughput and latency [37]. ADV-MAC is composed of four

different periods: synchronisation, advertisement, data, and sleep. The synchronisation period

works as S-MAC and T-MAC. Next, there is a fixed advertisement period that the nodes use

to transmit advertisement packets containing an ID of the intended receivers. Nodes that do

not participate in the data transmission switch to the sleep period. The ADV-MAC energy

consumption is 30 % less when compared with T-MAC and 41 % when compared with S-MAC

with a 20 % duty cycle while improving throughput and latency [37]. Nonetheless, ADV-MAC

has two significant disadvantages [31]: advertisement packets are not confirmed, and in the case

of collisions the intended receiver will be in sleep mode while the transmitting node is awake,

thus wasting energy; when many nodes assign slots in the advertisement period, the data period

is not enough to assure data transmission for all nodes. Therefore, ADV-MAC cannot guarantee

packet loss ratio and throughput fairness for a wireless video sensing scenario.

X-MAC is an asynchronous MAC protocol that uses short preambles, embedding the target

ID of the receiver, instead of one long preamble, as in B-MAC [38]. When the receiver node

wakes up, it checks the ID on the preamble packet. If it is not the intended recipient, the node

returns to sleep, continuing its duty cycle. Otherwise, it sends an ACK message and remains

awake to receive the data packet. Since X-MAC avoids the synchronisation overhead, it is

more energy efficient than S-MAC. Moreover, experiments demonstrate that X-MAC achieves

significant gains over LPL implementations when it comes to energy consumption, latency,

and throughput. Furthermore, the performance gains of X-MAC continually increase with the

density of the network. In [39] X-MAC was improved with a collision avoidance algorithm

named X-MAC/CA. When combining X-MAC with a Collision Avoidance (CA) mechanism,

the transmissions are randomised in the overcrowded network, thus reducing the probability of

collisions. X-MAC/CA improves the throughput in 30 % when compared to X-MAC [39], but

cannot offer packet loss ratio and throughput fairness guarantees.

Y-MAC combines TDMA and CSMA protocols with light-weight hopping mechanism to

achieve both high performance and energy efficiency under high-traffic conditions [40]. Since

Y-MAC is a TDMA-based protocol, the frame is divided into fixed-length time slots, with each

frame composed of broadcast and unicast periods, as shown in Fig. 2.10. The number of time

slots in each frame can be increased to allocate exclusive time slots to more nodes, but latency

shall also increase due to the extended length of the frame period. Nevertheless, Y-MAC

proposes the use of multiple channels, as an alternative to increasing the number of possible


time slots. Since multiple channels are used, sensor nodes require time synchronisation. A

simple synchronisation technique is included to synchronise the upcoming timer events of

nodes by adjusting the expiration times of these events. After a predefined time, if nodes

do not receive any control message, they are considered detached from the network and are

moved to sleep mode. Under light traffic conditions, this scheme is energy efficient because

nodes access the medium only during the broadcast and unicast receive time slots. For high

traffic levels, nodes have to wait for a unicast time slot, and messages have to wait in a queue,

thus increasing latency.

Figure 2.10: Y-MAC frame format [40].

Figure 2.11: Y-MAC channel hopping mechanism [40].

Y-MAC overcomes this problem by implementing a light-weight channel hopping mecha-

nism that uses multiple channels to reduce latency. This mechanism is presented in Fig. 2.11,

assuming that four channels are available and f1 is the base channel. When a node receives

a message in the f1 channel, it hops to the next channel using a hopping sequence generation

algorithm to receive the next message. If a node has a pending message destined for the same


receiver it also hops to the same channel and competes to access the medium in the contention

window. The contention winner is penalised by limiting the range of its back-off timer value

for the next transmission. This way per node fairness is guaranteed. In Y-MAC the overhearing

problem is reduced since receive time slots are allocated to nodes; thus, Y-MAC maintains a

low energy consumption while achieving a high delivery rate of bursty messages under high

traffic conditions. Nonetheless, Y-MAC cannot guarantee a packet loss ratio and throughput

fairness for offered loads typical in wireless video sensing scenarios.

Figure 2.12: Z-MAC channel-scheduling algorithm [41].

Like in Y-MAC, Zebra MAC (Z-MAC) combines the strengths of CSMA and TDMA

mechanisms, by adapting itself to the level of contention in the network [41]. Z-MAC was

designed to behave like CSMA under low contention and switch to TDMA under high con-

tention. Furthermore, by combining both mechanisms, Z-MAC becomes better than a stand-

alone TDMA with respect to the robustness to topology changes, time synchronisation failures,


time-varying channel conditions, and slot assignment failures. Z-MAC reuses channels by

adopting a Distributed RAND [41], an efficient, scalable channel-scheduling algorithm, which

allocates slots for all the nodes in the network. A slot is periodically assigned to a node, and

each node can reuse its slot. Two-hop neighbours of a node can get the same slot since DRAND

allows any two nodes beyond their two-hop neighbourhoods to own the same slot. Before

allocating time slots to nodes, Z-MAC runs a neighbour discovery protocol, which broadcasts

a ping to its one-hop neighbours. Using this protocol, each node collects the information from

its one-hop neighbours, which constitutes the two-hop neighbours. This two-hop neighbour

list is used as input to the DRAND algorithm. After allocating a time slot, each node needs to

decide the time frame of the node, i.e., the period a node can use the time slot for transmission.

To calculate the time frame each node propagates the Maximum Slot Number (MSN) to its

neighbours. Fig. 2.12 shows on the top the network topology with the slot numbers assigned

to each node; the numbers in parenthesis are the MSN. The bottom part of Fig. 2.12 shows the

slots schedule for all nodes. The dark slots represent "empty" slots; the shaded slots represent

time slot that can be used to transmit. Besides the neighbour discovery, slots assignment, and

local framing, Z-MAC requires local synchronisation since this protocol only involves one-

hop and two-hop neighbours. After the setup phase, the nodes forward the frame size, the slot

number to two-hop neighbours, and maintain synchronisation. Z-MAC is more efficient at high

contention levels, showing 40 % higher fairness index than B-MAC. It is energy inefficient for

low contention levels and does not offer packet loss ratio and throughput guarantees.

Figure 2.13: Frame structure of ER-MAC [42].

Emergency-MAC (ER-MAC) was designed for fire monitoring in building scenarios but

can also be useful for other WSN emergency applications [42]. For high volumes of traffic,

ER-MAC allows contention in TDMA time slots, trading energy efficiency for higher packet

delivery ratio and lower latency. Furthermore, ER-MAC was designed with two priority queues

to distinguish high priority packets for emergency scenarios, from low priority data, i.e., non-

critical data. When the high priority queue is empty, the non-critical data is sent from the low

priority queue. Nodes enter a low power sleep mode when there is no data to be transmitted.


The ER-MAC frame structure, as shown in Fig. 2.13, is composed of contention-free slots

with ts duration and a contention period with tc duration. Similarly to Z-MAC, a contention-

free slot is assigned to more than one node within the two-hop neighbourhood. In case of

emergency, each contention-free slot is further divided into sub-slots (t0, t1, t2, t3). The

contention period at the end of the frame is used to add new nodes. In [42], simulation

results prove that ER-MAC has higher packet delivery ratio, lower latency, and lower power

consumption when compared with Z-MAC. ER-MAC consumes more energy in normal mode

than in emergency mode and does not guarantee a packet loss ratio and throughput fairness.

Figure 2.14: Buffer threshold setting in EE-Hybrid MAC based on the hop-count from the sink[43].

For industrial monitoring scenarios, Pandeeswaran et al [44] developed Energy Efficient

Hybrid MAC (EE-Hybrid MAC), that aims to be a low latency MAC protocol, assuring high

packet delivery ratio and energy efficiency. Like Z-MAC, EE-Hybrid MAC is a hybrid protocol

combining the best features of TDMA and CSMA depending on the traffic pattern. The EE-

Hybrid MAC was designed based on Z-MAC, but it includes a priority region and changes the

buffer memory level depending on the node distance to the sink node. When a node from a

high priority region sends data to its neighbours, it identifies the data as high priority, switches

to TDMA mode, and allocates the first slot to that node. If more than one node is in the

high priority region, adjacent slots are allocated by neighbours. To assure energy efficiency,

the total memory to store information in the node before forwarding it to neighbours can also

be adjusted, as presented in Fig. 2.14. The buffer threshold Qthreshold is set up based on the

hop-count from the sink to the node. Nodes closer to the sink will get a higher threshold;

the threshold diminishes with the hop-count. Nodes are continuously in sleep mode. They

only move to active mode if the number of arriving packets exceeds the buffer threshold [43].

EE-Hybrid MAC provides better results for packet delivery ratio, and insignificant energy


consumption reduction when compared to S-MAC, Z-MAC, and T-MAC. Nevertheless, in the

end-to-end delay results, T-MAC outperforms EE-Hybrid MAC, and EE-Hybrid MAC does

not offer throughput fairness.

In [45] the authors have proposed the High Throughput Sensor MAC (HTSMAC) for

surveillance applications. The streaming of images is triggered when a sensor node detects

high temperature, thus indicating a possible fire hazard in the zone. HTSMAC improves S-

MAC and makes the protocol switch between two operation modes: normal mode – using S-

MAC for WSN nodes to sense temperature, humidity, and luminance; image mode – using

RIPPLE, sensor nodes power ON the camera and send images to the sink. The SYNC

packets of S-MAC are used to switch between normal and image modes. In our targeted

scenarios, the cameras are continuously transmitting video, so there is no need to switch

between modes. Moreover, Ripple protocol was designed for multi-hop network multimedia

applications without considering power efficiency.

QoS-supported Energy-efficient MAC (QEMAC) improves the throughput fairness and

energy-efficiency of the standard IEEE 802.11e for scenarios of video surveillance, locali-

sation, telemedicine, and industrial processes [46]. While the QoS support is based on an

improved version of IEEE 802.11e, energy-efficiency is achieved by employing a dynamic duty

cycling sleep/awake mechanism. QEMAC provides rate fairness by assuring that no internal

collisions occur by assigning the Transmit Opportunity (TxOP) to each Access Category (AC).

The sleep/awake mechanism is based on the Request To Send/Clear To Send (RTS/CTS)

exchange, which can happen multiple times in a single TxOP. The QEMAC’s main drawback

is the support of single-hop networks only and the periodic wake up of nodes to receive RTS

frames.

In [13] PACE was proposed as an evolution to Wi-Fi network Infrastructure eXtension

(WiFIX). WiFIX is a simple and efficient tree-based routing solution overlaid on the 802.11

MAC. It configures an active tree topology rooted at the gateway and uses 802.1D bridges and

their simple learning mechanism for frame forwarding. Still, it also suffers from performance

inefficiency and throughput unfairness due to the use of the Carrier Sense Multiple Access –

Collision Avoidance (CSMA/CA). PACE enables coordinated access among nodes to prevent

collisions, without requiring explicit synchronisation and fixed packet size. PACE assumes that

a logical tree topology, rooted at the gateway, is configured over the physical network using

WiFIX. In PACE, the gateway controls the access to the medium by limiting transmissions

to a single node at each time. Thus, each node can transmit a packet in each network-wide

transmission round, ensuring performance and throughput fairness. The packet received by the


destination node is implicitly used as a token that grants the permission to send a packet. When

the gateway receives the packet coming from the destination node, the same process is repeated

with another destination node, until all the nodes have had the opportunity to transmit one

packet and the first destination node can be authorised to send again. Since control packets can

degrade performance [13], PACE exchanges packets in both directions, whenever possible, by

using a flag in the poll message to indicate whether the gateway has a packet to be transmitted

to a node. An explicit control packet is sent by the gateway when there is no data to transmit.

Additional signalling is embedded in data packets to minimise the control overhead. In multi-

hop networks, PACE default is to send one packet per poll to avoid intra-flow interference

and performance degradation by relay nodes. PACE outperforms CSMA/CA-based solutions

for near-saturation or saturated Wireless Mesh Networks (WMNs) regarding goodput, delay,

and fairness [13]. In non-saturated WMNs, PACE’s delay can be higher than in CSMA/CA-

based networks since sending an explicit poll control message has a high cost. Nevertheless,

PACE was not designed to be energy efficient and wastes resources when polling signals are

not embedded in data frames.

2.2.3 Power Saving Mode

IEEE 802.11 standard proposed an amendment [47], which introduces PSM to increase the

lifetime of IEEE 802.11 stations running on battery, such as smartphones. PSM was firstly

designed for single hop networks running in infrastructure mode, so it performs poorly for ad-

hoc mode, especially in multi-hop networks [48][49]. PSM increases the packet delay when

a data frame is forwarded across multi-hop networks since nodes on subsequent hops stay

in the doze state until a traffic announcement is received. On each hop, the frame waits for

the beacon interval, before being forwarded, and for a high number of hop-count the end-to-

end delay increases, affecting time-sensitive applications. Moreover, nodes are forced to stay

awake to respond to probe requests from nodes that are scanning the medium for joining the

network. For instance, in an IEEE 802.11 ad-hoc network with two nodes, at least, one node

remains awake, limiting the sleep time to 50 %. Therefore, PSM is not suitable for low-energy,

low-latency multi-hop WVSNs. Many solutions described in [50] allow increasing the energy

efficiency by keeping a node in Deep Sleep mode when it is not involved in data transmissions.

Next, we present solutions/schemes that use modified versions of PSM or PSM combined with

other energy-efficient mechanisms.

If using IEEE 802.11 PSM all nodes in a multi-hop WVSN need to be time synchronised

and periodically wake-up at the beginning of each beacon interval to exchange the Ad-hoc


Traffic Indication Map (ATIM) frame. When there is data to be transmitted nodes remain

awake. Even if there is no data to be transmitted, nodes are forced to be awake during the

ATIM window, therefore energy is wasted. Furthermore, a frame travels one hop during a

beacon interval, which causes a high end-to-end delay and energy consumption.

Modified PSM (M-PSM) proposes a variation to PSM, allowing a frame to traverse several

hops from the source to the sink within a beacon interval [51]. M-PSM slices the ATIM window

into smaller contention slots. When a node has data to transmit, it first sends an empty packet

during the contention slot and all receiving nodes will be awake during the ATIM window.

Using this dynamic ATIM method, when there is less traffic in the WVSN all nodes go to

sleep after the contention slot, not staying awake for the entire ATIM window. To enable a

frame to travel more than one hop in a beacon interval, an history-based prediction method was

proposed. Fig. 2.15 shows how this method works. When node A needs to transmit a frame

to node C, through node B, it first transmits an ATIM packet to node B. Node B replies to

node A with an ATIM-Acknowledgement (ATIM-ACK), which is listened by node C. Node C

transmits to node B an ACK informing that data should be forward. Using this method during

a single beacon interval a frame is transmitted from node A to node C, through node B. This

method saves energy and reduces the number of beacon intervals used in a multi-hop network

to transmit data and also the packet latency. Nonetheless, M-PSM cannot guarantee a packet

loss ratio and throughput fairness.

Figure 2.15: M-PSM method to forward packets during a beacon interval [51].

Multi-Hop PSM (MH-PSM) addresses the same issues raised by M-PSM. MH-PSM for-

wards frames in a multi-hop network minimising the number of beacon intervals used [50].

To achieve this goal, the ATIM frame is used to indicate the destination node in the multi-hop

network. As shown in Fig. 2.16, when node A needs to send data to node D, through nodes

B and C, it advertises an ATIM frame with the MAC address of node D inside. During the

ATIM window, the advertisement is forwarded until it reaches node D. Since all nodes on the

path are in the awake state, the data frame will be forwarded end-to-end in one beacon interval.


Figure 2.16: MH-PSM method to forward packets during a beacon interval [50].

Moreover, MH-PSM proposes a mechanism named Sleep on Beacon Transmission (SoBT)

to increase the doze time when nodes are in idle mode. As mentioned, in the ad-hoc mode

the node transmitting the beacon should remain awake until the end of the beacon interval, in

order to assure that the other nodes can discover the Independent Basic Service Set Identifier

(IBSS) to which that station belongs to. SoBT mechanism, instead of keeping the node awake

during the entire beacon interval, wakes up the node periodically by transmitting intra-beacons.

Simulation results demonstrate that MH-PSM improves end-to-end delay and doze time when

compared with PSM in lightly-loaded multi-hop networks.

Energy Threshold

Identify Transmission Mode

Energy Calculation

ComparatorGet

Get

Get

PSM SchedulerSet Mode

Active

Deep sleep

Light sleep

Change Mode

Figure 2.17: EAPSM management components [52].

Energy Aware PSM (EAPSM) was designed to improve the low packet delivery ratio

caused by PSM [52]. EAPSM is composed of three major components, as shown in Fig. 2.17:

energy consumption calculator, PSM scheduler, and transmission mode identifier. The PSM

scheduler compares the remaining energy, calculated by the energy consumption calculator


module, with the node’s energy threshold required by the node to be involved in a transmission,

reception, or relay. Furthermore, the nodes involved in the transmission are given by the

transmission mode identifier module. If the required energy to transmit/receive/relay a packet

by the node is lower than the node’s remaining energy the PSM scheduler changes the node

from light sleep to active mode; otherwise, the node remains in light sleep mode. When the

node’s remaining energy reaches 0, the PSM changes the state of the node to deep sleep mode.

EAPSM enhances QoS by increasing the packet delivery ratio when compared to PSM, but for

high traffic loads the packet loss ratio increases together with throughput unfairness.

Optimized Power save Algorithm for continuous Media Applications (OPAMA) improves

PSM energy efficiency for continuous media applications (e.g., video, voice) by combining data

aggregation techniques with application specific requirements [53]. OPAMA uses an algorithm

that is based on the user feedback about the required video quality to save energy while keeping

the desired application performance. For example, the user can opt to diminish the video

quality to save energy when the device battery is low. This scenario is achieved by managing the

AP buffer and frame aggregation according to the required application performance, allowing

a station to enter the doze state. Fig. 2.18 presents a simple example of the OPAMA algorithm

operation. AP-1 informs in Beacon-2 that it has data to be transmitted using the Traffic

Indication Map (TIM) field in the beacon. The OPAMA algorithm running in STA-1 enters

in doze state since it decides to wait for more data. When STA-1 receives Beacon-3, it sends a

PS-Poll message to the AP asking for the data, and AP-1 sends the aggregated frames Frame-

1, Frame-2, Frame-3 and Frame-4 using the Aggregated MAC Service Data Unit (A-MSDU)

scheme into Frame-A1. The same happens to the remaining frames and after STA-1 enters

the doze state. OPAMA is capable of enhancing energy efficiency without compromising the

Quality of Experience (QoE), achieving energy savings up to 63 % for high tolerance delay

scenarios when using IEEE 802.11g default Maximum Transmission Unit (MTU). Although

OPAMA was built for a video streaming scenarios, it does not support multi-hop networks.

E-Mili is an interesting mechanism in which nodes downclock their Wi-Fi network inter-

face cards during the idle listening periods, thus reducing the energy consumption by around

44 % for 92 % of users in real-world wireless networks [54]. However, this solution requires a

hardware modification to the standard Wi-Fi cards and to the existing MAC layer protocol.

The µNAP solution was proposed in [55] to limit nodes overhearing by adopting local

energy saving mechanism which uses micro-sleeps inherent to the CSMA operation in 802.11

networks. Moreover, the authors performed a characterisation of timing and consumption of

a Commercial Off-The-Shelf (COTS) wireless card when it switches between idle mode and


sleep mode. µNAP uses the time the channel will be busy, carried on each MAC frame to

update the Network Allocation Vector (NAV), to estimate the amount of time a station can be

at sleep when the ongoing transmission involves another station. This mechanism reduces in

57 % the time spent by each node in overhearing, which causes an energy saving of 16 %.

Figure 2.18: OPAMA example with an AP and one Station [53].

2.3 Routing Oriented Solutions

Energy-efficient routing techniques are also proposed in the state-of-the-art literature. In

[56] a review on routing protocols for Wireless Multimedia Sensor Networks (WMSNs) is

presented. The authors divide the routing protocols in three major categories: QoS-based,

"swarm intelligence-based", and network structure-based.

2.3.1 QoS-Based

QoS-based routing protocols focus on guaranteeing qualitative and quantitative perfor-

mance parameters for WVSNs, besides assuring energy efficiency for battery constrained

devices. The typical QoS parameters considered in these routing protocols are reliability,

latency, bandwidth, and jitter. The QoS-based routing protocols are further divided into

latency-constrained or multi-constrained. Latency-constrained routing protocols guarantee a

specific delay for multimedia applications while multi-constrained consider other parameters

such as packet delivery ratio, bandwidth, and nodes interference.

2.3 Routing Oriented Solutions 31

Ling et al present a delay-constrained routing protocol in [57], which guarantees a delay

upper bound for real-time applications and minimises the energy consumption by selecting the

least energy-consumption path, therefore extending the lifetime of the WVSNs. This routing

protocol uses a global link-state routing algorithm to collect information about the network

topology, the queues of the nodes, and the residual energy for each node. Firstly, the algorithm

removes nodes from the graph whose residual energy is below a certain threshold. Secondly,

the Dijkstra’s algorithm is run to find the shortest and more energy-efficient routes. Thirdly,

all routes are checked against a delay upper bound constraint. When the delay is higher than

the defined threshold, a new route is computed using the Dijkstra’s algorithm with delay as

the metric. By following these steps, a least energy-consumption tree, meeting the delay

upper bound, is constructed. The use of a link-state routing algorithm has some drawbacks.

It increases the memory and bandwidth requirements due to the information that needs to be

exchanged. Moreover, it is only suitable for small WVSNs since it is hard to store and maintain

state information for large WVSNs.

Energy-Aware QoS (EAQoS) computes optimal paths to the gateway considering the

residual energy of the nodes and the packet loss ratio for each link while meeting an end-

to-end delay requirement [58]. EAQoS was designed to handle real-time and non-real-time

traffic by using a queuing model. Moreover, a cost function combining the distance between

the nodes, the residual energy of the nodes, the expected time for the current energy level to

be depleted, and packet loss ratio is computed for each link. The EAQoS algorithm selects

a set of least cost paths using Dijkstra’s algorithm and finds which paths meet the bandwidth

ratio r to maximise throughput for different paths. The bandwidth ratio r is initially set in each

link by the gateway to reserve bandwidth for real-time and non-real-time traffic. The algorithm

generates different r-values for different paths, and the maximum of them is selected to be

disseminated by the gateway to the whole network. The r-value satisfies the end-to-end delay

requirement for all paths. Although EAQoS considers QoS and energy efficiency requirements,

it requires the knowledge about the residual energy of the nodes and the packet loss ratio for

the whole network links.

The Stateless Protocol for Real-Time Communication (SPEED) was designed for sensor

networks and real-time communication services. It uses Stateless Nondeterministic Geographic

Forwarding (SNGF) for routing packets to keep the desired delivery speed and reducing

network congestion by dropping packets. Therefore, SPEED assures end-to-end soft real-

time communications [59]. SNGF is responsible for selecting the candidate next hop that can

assure the desired delivery speed, as shown in Fig. 2.19. The modules named Neighborhood


Feedback Loop (NFL) and Backpressure Rerouting are intended for reducing or diverting

traffic when congestion is detected, by providing a list of candidates to the SNGF. Geographic

based routing is achieved by exchanging location information of neighbours using the beacon

exchange module, while the delay estimation module computes the round trip single-hop

delay after receiving the ACK to determine whether or not congestion occurs. The last mile

process is used to support three types of real-time communication services: real-time unicast,

real-time area-multicast, and real-time area-anycast. Since SPEED forwards traffic using

nondeterministic geographic information that is kept in one localised database, the routing

overhead is minimised. Moreover, for scenarios where the resources of the nodes are scarce,

SPEED is a highly efficient and scalable protocol. Nevertheless, SPEED uses a per-flow

reservation, which is not scalable and performance degrades when unexpected congestion

occurs.

Figure 2.19: SPEED architecture [59].

The Real-time Power-Aware Routing (RPAR) protocol can attain application end-to-end

delay requirements by dynamically changing the transmission power together with other

routing metrics [60]. RPAR is composed of four modules: velocity assignment policy,

neighbourhood manager, delay estimator, and forwarding policy. The velocity assignment

policy maps a packet’s deadline to the number of hops crossed by the packet divided by the

end-to-end packet delay, i.e., velocity. The neighbour manager maintains a table with all the

forwarding choices and corresponding power level. For each forwarding choice, the delay

estimator evaluates the one-hop delay, and the forwarding policy compares it with the required

velocity. When no alternative is found in the neighbour table that meets the required velocity,

the neighbourhood manager finds a new choice through power adaptation and neighbour

discovery. Although RPAR is related to SPEED, RPAR is better in handling congestion

and employs a novel neighbourhood management mechanism, which is more efficient than


the beacon exchange mechanism of SPEED. Furthermore, since RPAR specifies the required

application deadline, it controls the trade-off between energy consumption and end-to-end

delay. RPAR does not offer packet loss ratio and throughput fairness guarantees.

The Optimized Energy-Delay Subnetwork Routing (OEDSR) protocol was developed for

WSNs with a cluster-based architecture [61]. In OEDSR, nodes are in sleep or idle mode and

awake when an event is detected. Afterwards, nodes near the event become active and form

a cluster-based sub-network. Before selecting the cluster heads, a temporary head is elected

based on the remaining energy of the nodes. This interim head selects the required number

of cluster heads to route information from sensor nodes to the base station, broadcasts this

decision to all sensor nodes, and becomes a regular sensor node. The cluster heads need to

find the optimum path to send the detected event using relay nodes between the cluster heads

until it reaches the base station. The selection of relay nodes between cluster heads is based

on a link cost factor that depends on the energy available in the node, delay between any two

nodes, and hop distance to the base station. OEDSR outperforms Ad-hoc On-Demand Distance

Vector (AODV) and Dynamic Source Routing (DSR) regarding energy consumption and end-

to-end delay results [61]. Since minimising OEDSR limits the number of nodes that access

the channel, the number of collisions is minimised. Nonetheless, OEDSR cannot guarantee

throughput fairness for a wireless video sensing scenario.

Adaptive Greedy-compass Energy-aware Multipath protocol (AGEM) was proposed for

WVSNs and uses the information of the positions of the nodes for packet forwarding decisions

[62]. The decisions to forward a packet are made online at each node but using a smart greedy

forwarding, based on adaptive compass policy to limit the number of neighbours to be selected.

The following parameters are considered for forwarding a packet: neighbour remaining energy,

neighbour hop-count to the sink, the distance between the current node and its neighbours,

and previous history of forwarded packets from the same stream. When the packet cannot be

forwarded to any neighbour, it changes to the Walking Back Forward mode to find an alternative

path. Simulation results performed in [62] demonstrate that AGEM maximises the network

lifetime since data streams are routed using different paths and guarantees a low transmission

delay and packet loss ratio. Moreover, AGEM outperforms Greedy Perimeter Stateless Routing

(GPSR) regarding energy efficiency for network sizes above 50 nodes [62]. Although AGEM

presents a low packet loss ratio, it was evaluated only with one multimedia flow and does not

offer throughput guarantees.

The Energy efficient and QoS aware multipath Routing (EQSR) protocol was designed

for multimedia applications to differentiate delay sensitive from delay tolerant traffic while


maximising network lifetime [63]. To distinguish real-time from non-real-time traffic, a

queuing model is employed by EQSR, while an increase of network lifetime is achieved by

balancing the energy consumption across multiple nodes by spreading traffic through multiple

disjoint paths. During the initialisation phase, nodes advertise HELLO messages with the

following parameters: hop-count from the originator node, the residual energy of the node,

the status of the buffer of the node, and quality of the links between the current node and its

neighbours. Based on the collected information, each node computes a link cost function for

each neighbour, which includes the following factors: remaining energy, available buffer, and

Signal-to-Noise Ratio (SNR). The primary path discovery phase is used by the sink to send

Route REQuest (RREQ) messages towards the source to select next-hop based on the highest

link cost function value. The alternative paths discovery phase is used to find disjoint paths

from the primary path discovery phase. To avoid paths sharing the same node, each node is

limited to accept only one RREQ message. From the discovery phase, only one set of paths

is selected based on a defined data delivery bound. Before sending data, the source segments

and encodes each data segment using an XOR-based Forward Error Correction (FEC) coding

algorithm to increase reliability. Data is distributed across the selected paths using a weighted

traffic allocation strategy. Moreover, to avoid flooding of keepalive messages to update the

cost function metrics, EQSR includes the metrics above in the data messages, i.e., residual

energy, remaining buffer size, and link quality. The simulation results show that EQSR attains

lower average delay, higher energy saving and packet delivery than Multi-Constraint Multi-

Path (MCMP) protocol [63]. Nonetheless, EQSR was tested with one flow and throughput

fairness cannot be guaranteed.

Li et al proposed in [64] a real-time fault-tolerant mechanism to fulfil a hard QoS require-

ment regarding reliability and timeliness for multimedia applications. A real-time message

stream can have an (m,k)-firm guarantee requirement when m out of k consecutive messages

meet the pre-defined QoS parameter values. Furthermore, Distance-Based Priority (DBP)

was developed to maintain a state for each stream about the number of messages that meet

and miss the QoS requirements. The real-time scheme for (m,k)-firm streams uses a Local

Status Indicator (LSI) so that intermediate nodes share their local condition and announce

network faults. Fig. 2.20 presents the five modules that compose the solution: 1) beacon

exchange, 2) delay estimation, 3) calculation of LSI, 4) orphan node backpressure, and 5)

routing mechanism. The beacon exchange module is used to share and collect information

about the location of the nodes and their residual energy. Three types of beacons exist to

implement different functionalities: DBP stream beacon, single hop delay estimation beacon,


and orphan node removal beacon. The DBP beacons are regularly transmitted by the sink

node to the source node to inform intermediate nodes about the stream DBP value, used by

the routing algorithm, and to assist in the fault recovery mechanism. For measuring the local

transmission conditions the single hop delay estimation beacon is used, while the orphan node

removal beacon is adopted to address the void region problem. The delay estimation module

employs the information from single hop beacons to estimate the delay, which is used by

the calculation of LSI module. Using the information of the orphan node removal beacons

a mechanism runs in the orphan node backpressure module to prevent the void occurrence.

Finally, the routing mechanism is responsible for making the forwarding decisions and for

maintaining a neighbour table with the following entries: neighbour ID, neighbour energy level,

estimated delay computed by the estimation delay module, expiration time or standard Round-

Trip Time, (m,k)-firm requirement for each stream, packet deadline, and DBP value obtained

from the respective beacon. The optimum routing decisions are made based on the LSI, packet

deadline, and remaining power. Furthermore, this scheme implements different fault recovery

mechanisms to address congestion, link failure, and void problems. The simulation results [64]

prove that this scheme outperforms SPEED by consuming less energy and meeting timeliness

and QoS.

Figure 2.20: Architecture of the real-time scheme for (m,k)-firm streams for WSNs [64].

The Distributed Aggregate Routing Algorithm (DARA) was designed for WSNs, sup-

porting both time-critical and non-time-critical applications with the objective of assuring


reliability, delay guarantees, while being energy efficient [65]. Besides, DARA is a multi-

sink, multi-path, and location-aware protocol. This protocol defines three routing metrics: 1)

expected sojourn time of the packet, 2) progress speed towards the destination, and 3) residual

energy. The expected sojourn time of a packet is the difference between the insertion time of the

packet in the interface queue and the time it is successfully transmitted. The packet’s sojourn

time includes the queuing delay, back-off timeout, contention period, and retransmissions due

to error or collision. Progress speed is calculated as the difference between the minimum hop

distance from the current node to the destination, and any downstream node minimum hop

distance towards the destination, divided by the link-delay between the two nodes. The third

routing metric is the residual energy of the node. In DARA, before forwarding a packet to

the destination, each node finds the downstream node by calculating the maximum aggregated

weight of the three combined routing metrics. Moreover, DARA implements 1) a Selective

Greedy Forwarding (SGF) algorithm that together with transmission power control reduces

collision drops, and 2) a Delay-differentiated Packet Scheduling (DDPS), which reduces packet

drop due to deadline failure. This is achieved by maintaining different queues for time-critical

and non-time-critical packets. DDPS also assures a lower end-to-end delay because time-

critical packets are assigned higher priority at the MAC in two levels: inter-node and intra-node.

Nodes collect network information using beacon messages which are dynamically broadcasted

by DARA. The source sends duplicate packets to ensure communication reliability. When com-

pared with Multipath Multi Stateless Protocol for Real-Time Communication (MMSPEED)

and MCMP, DARA saves energy by using a transmission power control to minimise collisions

and reduce retransmissions [65]. Moreover, DARA improves reliability and delay guarantees

for time-critical and non-time-critical applications. Nevertheless, DARA increases the routing

overhead since it duplicates the number of transmitted packets and does not efficiently solve

the void problem.

Figure 2.21: Architecture diagram for RTLD [66].


The Real-time routing protocol with load distribution (RTLD) was proposed for large

WSNs to guarantee high packet throughput and prolong the lifetime of the network with

minimal packet overhead [66]. RTLD adopts a novel communication mechanism named

"geodirection-cast forwarding”, which forwards the packet through multiple paths to the

destination, in order to increase the packet delivery ratio. To achieve this goal, it combines

geocast with directional forwarding. The Packet Reception Rate (PRR), remaining energy,

and one-hop packet velocity are used by the routing protocol to balance the load of real-

time traffic and compute the optimum forwarding node. Fig. 2.21 shows the four main

modules that compose RTLD: 1) neighbourhood management, 2) location management, 3)

power management, and 4) routing management. The neighbourhood management runs a

discovery protocol to find the forwarding candidate nodes and stores this information in a

neighbour table. RTLD assumes that all sensor nodes are fixed and the sink node in at

the origin (0,0). The location management module determines the location of each sensor

based on the signal strength received from 3 neighbours. The power management module

finds the state of the transceiver and adjusts the transmission power of the sensor node to

minimise the energy wasted in idle listening. The routing management computes the optimal

forwarding paths and adopts two forwarding methods: unicast and geodirectional-cast, which

forwards packets in the direction of the destination quadrant. Additionally, it includes a routing

problem handler mechanism to recover from network holes, caused by node failures over the

lifetime of the network. The path recovery can use two different methods: fast recovery

using power adaptation and slow recovery using feedback control packet. The simulation

results demonstrate that RTLD outperforms SPEED regarding packet delivery ratio and energy

consumption. Nevertheless, RTLD was not tested for video streaming scenarios and depends

on the reliability of the node location mechanism.

QoS aware Multi-sink Opportunistic Routing (QMOR) was designed for WVSNs to assure

1) the timely delivery and reliability requirements of real-time multimedia streams and 2)

minimise the energy consumption of the sensor nodes [67]. While many other WVSN

routing protocols are designed for delivering video streams to a single sink, QMOR supports

multiple sinks, thus contributing to improve network performance [67]. Moreover, QMOR

adopts an opportunistic routing scheme to enhance communication performance by using the

broadcast nature of IEEE 802.11. Each node keeps a forwarding list of neighbour nodes

that can route the broadcast packet. The list is prioritised to meet the QoS requirements

of the multimedia application. Furthermore, to reduce redundant information, which results

from highly correlated video streams, a differential coding scheme for optimal node selection


is included. To minimise the energy consumption and reduce data redundancy, QMOR

selects the source and intermediate nodes that belong to the same correlation group (a set

multimedia sensor node that observes the same object within its field-of-view). This is

addressed by the Differential Coding-based Source and Intermediate nodes Selection (DCSIS)

algorithm. Simulations were conducted to study the performance of QMOR that led to the

conclusion that QMOR outperforms Correlation-Aware QoS Routing (CAQR), by improving

video transmission quality and minimising energy consumption. However, for a multi-sink

network scenario, QMOR consumes more energy, but provides a higher number of decodable

video frames per units of energy consumption.

The Multi-Constrained Routing Algorithm (MCRA) was designed to assure end-to-end

delay and packet loss guarantees for multimedia communication while being energy efficient

[68]. MCRA adopts a logical coordinate system that estimates the position of the nodes

based on their hop-count information, which helps to reduce message overhead. Based on

the position of the nodes, MCRA inhibits message transmissions and retransmissions due to

collisions. MCRA uses query-flooding and query-driven data driven models because of its

resilience and reliability. To query an event, MCRA uses the query event named interest. The

sink starts by sending interest messages to all its neighbours. Upon receiving this message,

the current node calculates the packet drop ratio, residual energy, and end-to-end energy. Two

fields in the interest message are updated with the packet loss and end-to-end information.

This intermediate node checks if it does not belong to the list of interest nodes, if the packet

drop ratio and end-to-end delay are below a threshold, and if residual energy is above the

threshold. If all these conditions are verified the interest message is flooded to all its neighbours

(excluding the upstream node); otherwise, the message is discarded. These steps are repeated

until the interest message arrives at the source node. In case the source node receives multiple

interest messages, which travelled through different paths, it selects the interest with the lowest

hop-count. Moreover, MCRA can optionally adopt service differentiation by using different

forwarding priority levels for real-time and best-effort data. Simulation results prove that

MCRA outperforms Direct Diffusion and SPEED regarding the end-to-end delay, packet loss

ratio, and average energy consumption [68]. Moreover, MCRA reduces routing overhead and

minimises packet collisions by adopting a logical coordinate system. Still, MCRA cannot

guarantee packet loss ratio and throughput fairness.


2.3.2 Swarm Intelligent-Based

SI-based routing protocols incorporate the collective behaviour of biological species (e.g.,

bees and ant colonies), leading to self-organised and decentralised protocols. This behaviour

is emulated by software agents that are part of the network routing protocol, which is designed

to be adaptive, scalable, fault tolerant, fast, and autonomous.

Multimedia-Enabled Improved Adaptive Routing (M-IAR) is a flat multi-hop routing

protocol based on Improved Adaptive Routing (IAR) that includes the end-to-end delay and

jitter as QoS parameters to enable the support for real-time multimedia traffic [69]. This routing

protocol starts by sending a forward ant, which starts by calculating the probability of choosing

a node as the next-hop considering the distance to the sink and to the sender node. Moreover,

based on this information the neighbour with the highest probability value is selected, and the

routing table is updated. Each forward ant transports a set of global parameters: 1) visited

nodes and their neighbours; 2) corresponding probability values; 3) total number of visited

hops; and 4) distance for each link. Intermediate nodes follow the same procedure until the

forward ant reaches the sink node, where it is destroyed. A backward ant is sent through the

same visited path by the forward ant to update the probability values and update the routing

table. M-IAR results demonstrate that it can handle multimedia traffic and is optimised for a

bounded end-to-end delay and jitter. Moreover, M-IAR finds in 98 % of the cases the shortest

path with only three route discovery attempts. Although M-IAR finds the shortest path and

consumes less energy, it can degrade when a hole is created in the network and the overhead

increases exponentially when the network load is high.

Ant-based Service-Aware Routing (ASAR) was designed for WVSNs considering three

types of services: query-driven, data-driven, and stream-driven. Since it runs in all the cluster

heads [70], it can also be classified as a hierarchical routing protocol. For each service, this

algorithm searches three different routing paths towards the sink by using a positive feedback

mechanism, adopted in ant-based algorithms. The paths are selected based on a routing metric

which combines four parameters: 1) available bandwidth, 2) queuing delay, 3) packet loss ratio,

and 4) energy consumption for sending data in the link. Since the pheromone value is quantified

in the sink, the frequency for sending reverse ants is reduced and the algorithm convergence

time also diminishes. The simulation results demonstrate for the stream-driven service that

ASAR outperforms the Dijkstra’s algorithm regarding delay and consumes less energy than

Directional Diffusion (DD) routing algorithm. Nevertheless, the packet loss ratio is higher for

ASAR than for DD and Dijkstra, and network performance can be affected by the bottleneck

problem of the hierarchical model and the selection of optimum path due to congestion [71].


The Ant Colony Optimization-Based QoS Routing Algorithm (ACOWMSN) was proposed

for transmitting video and imaging data in WVSNs based QoS and energy efficient routing

model [72]. Like in ASAR, ACOWMSN uses forward ants to collect QoS parameters, and

backward ants carry back to the source the QoS information and update the nodes pheromone

and routing table. The QoS parameters carried by forward ants are the following: minimum

residual energy of visited nodes, cumulative queue delay, packet loss ratio, and available

memory of each visited node. The next forwarding node of the forward ant is selected using

a probability function that combines the pheromone value with the residual energy of the

nodes. ACOWMSN uses the routing discovery phase to send forward ants towards the sink

either using unicast or broadcast when the current node does not have information about the

destination. The backward ants perform the route confirmation. From the simulation results,

the authors conclude that ACOWMSN increases network lifetime, ensuring a lower delay and

high packet delivery ratio when compared to AODV and M-IAR. Notwithstanding, for large

WVSNs, the algorithm will converge very slowly since it is required to collect and exchange

the QoS parameters.

Ant-based multi-QoS routing metric (AntSensNet) is based on T-ANT [73] and was

designed for WVSNs to be energy efficient and meet the QoS requirements of the applications

[74]. Moreover, AntSensNet includes a packet scheduling mechanism to minimise video

distortion during transmission through multiple paths. Like T-ANT the clustering operation

is divided into the cluster phase and the steady phase. In the cluster phase, the cluster heads are

elected, and the rest of the nodes join the cluster. To avoid maintenance of many state variables,

AntSensNet uses agents, named cluster ants, to control the election process. During the election

process, a node with a cluster ant becomes a cluster head while the remaining nodes join the

cluster. HELLO packets are periodically broadcasted with the node’s ID, clustering pheromone

value, and nodes state. Fig. 2.22 presents a typical topology constructed with AntSensNet.

The cluster radius parameter is configurable and determines the minimum distance between

two cluster heads. In the steady phase, data transmission occurs between the sources and sink.

AntSensNet updates the routing tables using three type of ants: forward ant, backward ant,

maintenance ant. When a route to the sink node does not exist, forward ants are generated

and carry information about the traversed routing paths. The found routes are evaluated in the

sink node when the forward ant arrives. When the route meets the application requirements,

a backward ant is created and forwarded through the reverse path, updating the pheromone

values. A maintenance ant is generated in case of congestion or a lost link problem to update

the routing table. AntSensNet includes a traffic classifier on each cluster head to differentiate


packets according to the assigned class and priority. From the simulation results, AntSensNet

performs better than AODV regarding packet delivery ratio and end-to-end delay. Moreover,

when comparing the achieved Peak Signal to Noise Ratio (PSNR) or video quality, AntSensNet

attains a higher value than Two-Phase Geographic Greedy Forwarding (TPGF) and ASAR due

to the adopted distortion reduction mechanism.

Figure 2.22: WVSN topology for AntSensNet with backbone outlined in black and connectingthe cluster heads [74].

ICACR is a QoS-aware, hierarchical, and ant-based routing protocol designed for WVSNs

[75]. This routing protocol integrates IPACR, a 2D plain-based routing protocol, which

optimises the distribution of pheromones to diminish the algorithm convergence rate. While

standard ant colony routing algorithms set the same initial pheromone value for every link

between nodes and their neighbours, IPACR stores information about adjacent nodes and

ensures one feasible path. The optimum path is selected based on the first HELLO packet,

arriving at the node. Hence, the initial pheromone values of neighbours are increased according

to the arrival of HELLO packets. To support large WMSNs deployments, the authors proposed

ICACR, which uses a divide-and-conquer methodology to assure scalability [75]. Since the

network is divided into clusters and each can be considered a sub-network, ICACR protocol

first searches the optimum local paths inside each cluster, which are connected through the


cluster heads until the sink node. With IPACR, it is 4 times more likely to find an optimal path

than with standard ant colony routing protocols. Moreover, ICACR outperforms IPACR and

Dijkstra regarding of network lifetime and QoS metrics for large WVSNs and video quality,

PSNR increases when using multiple paths. Nonetheless, when the source node is located near

the sink node, ICACR performance decreases.

2.3.3 Network Structure-Based

Routing can also be performed based on the network structure. In flat routing protocol,

sensor nodes transfer data directly to the gateway or base station and use data aggregation to

reduce redundant information, thus saving energy. For hierarchical routing protocols, sensor

nodes are grouped into clusters and nodes with higher processing capacity are elected as

cluster heads. Data is routed through cluster heads until it reaches the gateway or base station.

Hierarchical routing protocols are suitable for large WVSNs. Location-based routing protocols

use nodes location information to route packets efficiently. The location of the sensor nodes

is obtained from Global Positioning System (GPS) or using other mechanisms [76], which

minimises the routing overhead and makes the routing protocol fully distributed and stateless.

The Real-Time and Energy-Aware Routing (REAR) protocol was proposed for transmitting

video sensing data in a WVSN using metadata to establish multiple routing paths, and selecting

the path which consumes less energy and enables the best delay [77]. Metadata is used to

minimise the amount of repeated data exchange, thus saving network bandwidth and energy.

REAR uses an advanced Dijkstra’s algorithm and a cost function to find the optimum routing

path. The cost function uses the following parameters: 1) distance between nodes, 2) remaining

energy of next hop, and 3) the queue length of the next hop. To offer delay guarantees

for real-time traffic, REAR includes a classified queue model in each node to distinguish

between real-time and non-real-time traffic. From the simulation results, REAR increases the

network lifetime when compared to Sequential Assignment Routing (SAR) because it reduces

the repeated transmission and energy consumption, by using metadata for route discovery.

However, REAR does not take into consideration the bandwidth and reliability requirements

of the application.

Load Based Energy-Aware Multimedia Routing (LEAR) is based on AODV and is designed

to be a reactive, self-organising, and energy efficient protocol for WVSNs [78]. As soon as an

event occurs, the multimedia sensor closest to the event broadcasts a RREQ message to all

its neighbours. If none of the nodes is the final destination, the nodes re-broadcast the RREQ

message. Each node selects the best route based on a β selection factor that is carried on the


RREQ message. The β selection factor considers the available energy of each node and the

number of active routes. When the RREQ message arrives at the destination, an RREQ message

is sent back to the source, and the node with the maximum β selection factor is selected to

participate in the route. When the energy of the node reaches 25 % of its total energy, the node

is labelled as “Swap Node” and only transmits packets in certain critical conditions. By using

this mechanism, LEAR avoids network holes caused by the node’s battery energy depletion.

Since LEAR considers the total number of active routes in the β selection factor, disjoint

paths are first selected to maintain network load balance and increase the network lifetime.

Nevertheless, LEAR does not offer reliability, delay, and bandwidth control for multimedia

applications.

Energy efficient and perceived QoS aware video routing (PEMuR) is a hierarchical routing

protocol based on Scalable Hierarchical Power Efficient Routing (SHPER), aiming to be energy

efficient and attain high QoS for multimedia based applications [79]. In PEMuR, nodes are

grouped into clusters, which elect a cluster head. The cluster heads closer to the base station

are called upper-level cluster heads and can communicate directly with the base station. The

cluster heads located outside the vicinity of the base station are the lower-level cluster heads

and need to perform multi-hop routing to send video streams to the base station. Using this

scheme the network size can still increase, not affecting the network performance. In the

initialisation phase of the algorithm, the base station defines a TDMA schedule, so that all

nodes advertise themselves and the distance between them is calculated. Moreover, random

upper and lower-level cluster heads are elected by the base station, and the remaining nodes

are associated with the cluster heads with the maximum signal strength. After the initialisation

phase, nodes enter a steady phase and elect as cluster heads the ones with higher residual

energy. When the available bandwidth is limited, a sensor node can drop video packets to

reduce the retransmission rate. By using a video distortion model [79], less significant packets

are dropped to minimise video distortion. PEMuR outperforms TEEN in retarding the time of

node depletion, thus increasing network lifetime. Moreover, PEMuR keeps high-level PSNR

when compared to TEEN, but PEMuR requires the base station for the initialisation phase,

therefore it is not a fully distributed protocol.

Directional Geographic Routing (DGR) is a location-based protocol that was proposed for

streaming real-time video over bandwidth and energy constrained WSN [80]. In DGR any

node can send video packets to the sink using disjoint paths, and FEC coding is adopted to

assure reliable data delivery. Furthermore, a video stream is divided into multiple sub-streams

and transfer is simultaneous through the multiple disjoint paths. Using this mechanism, DGR


assures network load balancing, bandwidth aggregation, and fast packet delivery. The authors

proved through simulation that DGR compared with GPSR provides lower packet end-to-end

delay and better PSNR, and increases network lifetime. Still, DGR does not perform well when

several video sensor nodes transmit simultaneously to a sink since the selection of multiple

paths would interfere with each other.

Figure 2.23: Greedy forwarding example for TPGF [81].

TPGF considers the requirements for the transmission of real-time multimedia through

WVSNs [81]. To assure a low end-to-end delay in communications, TPGF selects the shortest

path or near-shortest path when it needs to avoid holes. Furthermore, this routing algorithm

can find more disjoint routing paths, allowing multi-path transmission of multimedia streams.

It can also bypass holes, by avoiding the usage of face routing (right/left and rules) and the

identification of hole/boundary nodes. When forwarding traffic, TPGF selects the next hop

with less hop-count to the base station even in case the next hop is further than itself (i.e., has

a higher hop-count), as shown in Fig. 2.23. A locally based optimisation method is used to

remove circular paths and optimise the routing path. TPGF differs from other location-based

protocols by not adopting face routing protocols. When compared with GPSR, TPGF provides

paths with fewer average hop-count for different network sizes. Nevertheless, this routing

protocol requires the knowledge of the network which is not scalable. Since it uses the shortest

path, network lifetime will be reduced because nodes battery will be depleted faster.

Energy-Aware TPGF (EA-TPGF) is an enhanced version of TPGF, considering the residual

energy of the nodes when computing the routing paths [82]. Since TPGF only uses the distance

metric to find the optimum path to the base station, some nodes can be overloaded with

several transmissions, which decreases the network lifetime. EA-TPGF proposes a score that

combines distance between the sensor and the sink with the current residual energy of the node.


Simulation results demonstrate that EA-TPGF increases nodes residual energy, thus extending

the network lifetime, especially when compared with TPGF. Nonetheless, EA-TPGF augments

the average communication delay.

Figure 2.24: PWDGR example for selecting pair-wise [83].

Pairwise Directional Geographical Routing (PWDGR) is based on DGR and addresses

the energy bottleneck problem [83]. As depicted in Fig. 2.24, PWDGR uses pair-wise nodes

around the sink, using all the nodes around the sink and prolonging network lifetime and not

only a few nodes as in previous solutions. The algorithm starts by finding the pair-wise node

around the sink, in this case three hops from the sink, to work as the destination node. The

selection of this pair-wise node depends on the angle of the source sending path and the number

of hops to the sink. Secondly, around the source, a set of cooperative nodes are selected to

avoid a fixed choice of nodes by each path, which would lead to a decrease of the network

lifetime. PWDGR routing scheme is divided into 3 stages: 1) source node collects video data

and creates several segments, including a list of cooperative nodes in the packet header, and

packets are broadcasted to the neighbour nodes; 2) neighbour nodes check whether they are in

the cooperative list and in that case the coordinates for the next hop are computed; and 3) source

cooperative node is selected based on the distance between the ideal point and the neighbour

node or the cumulative numbers to be selected for cooperative node. PWDGR outperforms

DGR and can increase by 70 % the network lifetime. However, the delay is increased by 8 %

when compared with DGR. When the sink is located on the edge of the network, PWDGR

results are similar to DGR.


2.4 Summary

Video streaming from nodes to a gateway in an IEEE 802.11 single-radio, multi-hop topol-

ogy leads to three problems related to CSMA/CA: low performance, throughput unfairness, and

energy inefficiency. This chapter has presented a set of solutions that address these problems.

We have classified the state-of-the-art solutions in 3 types: 1) solutions using out-of-band

control; 2) solutions using MAC oriented approaches; 3) routing oriented solutions. The

classification considered is summarised in Fig. 2.25. In what follows, a comparison of each

type of solutions and their relationship with the requirements settled for wireless video sensing

in Section 1.3 is presented. Each table classifies the reviewed solutions based on 5 criteria:

multi-hop support, energy, delay, packet loss, and fairness. These criteria were considered

taking into account the aim of this thesis, namely the development of a solution for green

multi-hop WVSNs. A check in the column means that the solution complies with the criterion.

SI-Based

Classification

RoutingMACOut-of-Band Control

WUR-Receiver WUR-Transceiver QoSBased

NetworkStructure

HybridContentionBased

Tang et al. [17]Rowe et al. [19]RTID [20]Doorn et al. [21]Haratcherev et al. [22]Shih et al. [23]

Bahl et al. [25]sleepyCAM [18]Bhatt et al. [28]GWR-MAC [15]

Ling et al. [57]EAQoS [58]SPEED [59]RPAR [60],OEDSR [61]AGEM [62]EQSR [63]Li et al. [64]DARA [65]RTLD [66]QMOR [67]MCRA [68]

M-IAR [69]ASAR [70]ACOWMSN [72]AntSensNet [74]ICACR [75]

REAR [77]LEAR [78]PEMuR [79]DGR [80]TPGF [81]EA-TPGF [82]PWDGR [83]

S-MAC [29]T-MAC [32]B-MAC [33]D-MAC [35]

ADV-MAC [37]X-MAC [38]Y-MAC [40]Z-MAC [41]ER-MAC [42]EE-Hybrid [44]HTSMAC [45]QEMAC [46]PACE [13]

M-PSM [51]MH-PSM [50]EAPSM [52]OPAMA [53]E-Mili [54]µNAP [55]

Power SavingMode

Figure 2.25: Classification of the state-of-the-art energy efficient solutions.

Table 2.1 presents a summary of the criteria that are fulfilled by out-of-band control

oriented solutions. Although all the solutions can save energy, some do not support multi-hop

network topologies, and only a few demonstrate that they can offer QoS guarantees. Moreover,

these solutions were not prepared for a scenario where all nodes transmit a video stream to the

gateway with throughput fairness.

MAC oriented solutions are summarised and compared in Table 2.2. Since these solutions

minimise frame collisions, they offer energy savings, excluding PACE which only offers QoS

guarantees. Furthermore, solutions [53][54][55] that minimise the idle listening period were

2.4 Summary 47

Table 2.1: Comparison of out-of-band control solutions.

SolutionsMulti-Hop

Energy DelayPacketLoss

Fair-ness

Tang et al [17] ! ! !

Rowe et al [19] ! ! ! !

RTID [20] ! !

Doorn et al [21] ! !

Haratcherev et al [22] !

Shih et al [23] !

Bahl et al [25] !

sleepyCAM [18] ! !

Bhatt et al [28] ! !

GWR-MAC [15] ! ! !

not designed for multi-hop topologies. None of the reviewed MAC oriented solutions can

guarantee the required QoS parameters (delay, packet loss ratio, and throughput fairness).

Table 2.3 shows a comparison of the routing oriented solutions. The routing solutions

focus on prolonging the network lifetime of a multi-hop network and minimising congestion

in links, so energy inefficiency related to collisions is minimised. Also, QoS guarantees are

offered by all routing solutions, assuring quality for video transmission from a source to a sink.

Nevertheless, energy is lost since nodes are kept ON to forward information.

None of the out-of-band solutions can support multi-hop network topologies and offer QoS

guarantees. The MAC oriented solutions that support the video streaming scenarios have poor

energy savings. The compared routing solutions can extend the network lifetime, but energy is

wasted since nodes are kept on to forward packets. Also, from all the solutions discussed, none

of them can guarantee throughput fairness per node in a scenario with all WVSs transmitting

to a sink located in the cloud.


Table 2.2: Comparison of MAC oriented solutions.

SolutionsMulti-Hop


Fair-ness

S-MAC [29] ! !

T-MAC [32] ! !

B-MAC [33] ! ! !

D-MAC [35] ! !

ADV-MAC [37] ! ! !

X-MAC [38] ! ! !

Y-MAC [40] ! ! !

Z-MAC [41] ! ! !

ER-MAC [42] ! ! !

EE-Hybrid [44] ! ! !

HTSMAC [45] ! ! !

QEMAC [46] ! !

PACE [13] ! ! ! !

M-PSM [51] ! ! !

MH-PSM [50] ! ! !

EAPSM [52] ! !

OPAMA [53] ! !

E-Mili [54] !

µNAP [55] !

2.4 Summary 49

Table 2.3: Comparison of routing oriented solutions.

SolutionsMulti-Hop


Fair-ness

Ling et al [57] ! ! !

EAQoS [58] ! ! !

SPEED [59] ! ! !

RPAR [60] ! ! !

OEDSR [61] ! ! ! !

AGEM [62] ! ! ! !

EQSR [63] ! ! ! !

Li et al [64] ! ! !

DARA [65] ! ! ! !

RTLD [66] ! ! ! !

QMOR [67] ! ! ! !

MCRA [68] ! ! ! !

M-IAR [69] ! ! !

ASAR [70] ! ! ! !

ACOWMSN [72] ! ! ! !

AntSensNet [74] ! ! ! !

ICACR [75] ! ! ! !

REAR [77] ! ! !

LEAR [78] ! !

PEMuR [79] ! ! ! !

DGR [80] ! ! ! !

TPGF [81] ! ! ! !

EA-TPGF [82] ! ! ! !

PWDGR [83] ! ! !

Chapter 3

GREENNESS

In Chapter 2, we reviewed a set of solutions that address the three major problems of multi-

hop Wireless Video Sensor Networks (WVSNs) using either out-of-band control, MAC-based,

or routing oriented approaches. This chapter describes our proposed solution, named Green

wiReless vidEo sENsor NEtworks uSing out-of-band Signalling (GREENNESS).

GREENNESS addresses the three problems by combining a centralised multi-hop node

polling mechanism with out-of-band control over a low power radio. We start by presenting

the GREENNESS concept and architecture, followed by the description of its companion

mechanisms: WVSN Active Topology Collection, Node Scheduling, and Failure Recovery.

Finally, we discuss the requirements for the out-of-band control channel and describe possible

Low Power Radio (LPR) candidates.

3.1 GREENNESS Concept and Architecture

In a video monitoring scenario, the majority of the traffic is exchanged between the

Wireless Video Sensors (WVSs) and the cloud server through a gateway; the traffic between

any pair of WVSs is assumed to be null. For this scenario, Wi-Fi network Infrastructure

eXtension (WiFIX) [12] has been demonstrated to be a suitable solution for Wireless Mesh

Networks (WMNs). WMNs and WVSNs are fundamentally similar. The differences have

mostly to do with the type of nodes forming the network – WVSNs are formed by wireless

video sensors instead of Mesh Access Points (MAPs). WiFIX does not include a mechanism

to coordinate the access amongst nodes to prevent collisions. PACE has been proposed to

address this problem [13]. Nevertheless, PACE was not designed to be energy efficient, so we

51

52 GREENNESS

LPR range

CWVS#

1

C

C C CC

Wi-Fi Link

Gateway

LPR

WVS#1

WVS#2

WVS#3

WVS#5

WVS#4

WVS#6

mac3

mac4

mac5

mac6

mac2mac

1

LPR Link

Figure 3.1: The GREENNESS concept with the node scheduling mechanism running over theLPR control channel illustrated by the arrows in orange.

developed GREENNESS and its companion algorithms, WVSN Active Topology Collection

Mechanism (WATCM) and Node Scheduling Mechanism (NSM). GREENNESS is inspired by

PACE, which already addresses the low network capacity and throughput unfairness problems

[13].

The GREENNESS novelty lies in a centralised NSM that runs in the gateway, together

with a LPR integrated into each WVS node. The NSM, by using an associated protocol,

enables/disables WVS transmissions, and turns the WVS Wi-Fi radios ON/OFF accordingly.

The LPR allows establishing an energy-efficient out-of-band control channel between the

gateway and the WVS nodes. Fig. 3.1 illustrates the GREENNESS concept.

The GREENNESS architecture is presented in Fig. 3.2. During the network bootstrap, the

Routing module is responsible for creating the routes between each WVS and the gateway.

Typically, the Routing module will configure a logical tree rooted at the gateway, hereafter

called the active tree topology. Also, through the Routing module, the gateway can route

packets to each WVS and each WVS can route packets to the gateway. GREENNESS is

agnostic to the actual routing protocol used. Herein, we have assumed WiFIX [12] is used;

WiFIX is described in Section 3.2. The WATCM runs in the gateway and in each WVS

3.1 GREENNESS Concept and Architecture 53

WVS

802.11

FR

M

NSM

WATCM

Routing

LPR

Gateway

Poll

Registration ACK

Registration

DATA802.11

FR

M

NSM

WATCM

Routing

LPR

Figure 3.2: GREENNESS Architecture.

to discover the network topology after the Routing module created the routes between these

nodes. Each WVS registers itself in the gateway using the IEEE 802.11 interface. The WATCM

module running in the gateway computes the network topology, giving as output the polling

vector, which is an ordered list of WVSs that will be polled. When the WATCM finishes, the

NSM starts sending poll messages through the LPR and turns the WVS Wi-Fi radios ON/OFF

accordingly. In case a failure is detected, the gateway and the WVSs run the Failure Recovery

Mechanism (FRM).

In the following sections, we describe in detail each of the GREENNESS modules, in-

cluding the WATCM, NSM, and FRM mechanisms as well as the Routing and LPR modules.

Table 3.1 provides the notations used in the description of the WATCM and NSM mechanisms.

54 GREENNESS

Table 3.1: Notations used in the description of the WATCM and NSM mechanisms.

Notation Description

S[k] Medium Access Control (MAC) address of WVS k

H[k] Hop count of WVS k to the gateway

PM[k] MAC address of the parent of WVS k

P[k] nodeId of the parent of WVS k

l Branch of the WVSN active topology

N Number of nodes forming the WVSN (WVSs plus the gateway)

D Depth of the active tree topology

Nleaves Number of leaf nodes in the active tree

Bl nodeIds associated to branch l

T [u] Last nodeId parent included in vector Bl and stored in position u

L[l] Last nodeId inserted in vector Bl and stored in position l

hmaxval Highest hop count found in vector H

V [x] nodeId of the xth node to be polled, x ∈ 1, ...,N−1MPoll Number of Poll messages sent by the gateway

Mwarm−up Number of Poll messages used by the mechanism to learn WVS traffic

pattern

TGWtimeout Gateway polling timeout value

TWV Stimeout WVS polling timeout value

O[i][nodeId] Returns 0 or 1 depending on the nodeId and polling interval i. The

pattern is stored during the warm-up period using WVS’s queue

information

Q[nodeId] Queue status for a given WVS with the nodeId

QS Queue size threshold used to trigger the next Poll message

R[s] Snooped nodeId in each Registration Acknowledgement message and

stored in position s by each relay WVS

Rn Number of different Registration messages received

3.2 Routing

WiFIX is a simple and efficient tree-based routing solution overlaid on the IEEE 802.11

MAC. It configures an active tree topology rooted at the gateway. For that purpose, WiFIX

3.3 WVSN Active Topology Collection Mechanism 55

defines a single-message protocol that enables the self-organisation of a WVSN and reuses con-

cepts such as IEEE 802.1D bridges and their simple learning mechanism for frame forwarding.

In WiFIX, the active topology is a tree rooted at the WVSN gateway. WiFIX supports multi-hop

forwarding by using IEEE 802.1D bridges and a new encapsulation method, named Ethernet-

over-802.11 (Eo11), for the creation of virtual links between neighbour nodes. The virtual

links are created and managed by the Active Topology Creation and Maintenance (ATCM) and

form a tree topology rooted at the gateway. The ATCM periodically sends a Topology Refresh

(TR) message at the gateway, which is forwarded by all nodes. The TR message includes the

number of hops to the gateway, the parent address, and the original address of the frame. Each

node selects the parent node with the lower hop count to the gateway. When more than one

parent node has the same hop count, the parent node with the lower MAC address is selected.

Thus, the TR message is used to announce the gateway address and notify a parent node that

was selected by the child node. The virtual links are seen as Ethernet links by the IEEE 802.1D

bridges, so packets are forwarded between the virtual links up to the gateway. Since WiFIX

runs on top of the IEEE 802.11 interface and processes all packets, a scheduling mechanism

can be developed on top of this interface.

3.3 WVSN Active Topology Collection Mechanism

WATCM is responsible for registering the WVSs, assigning a shorter identifier to each

WVS, named nodeId, and computing the polling vector. The registration process is used to

collect information about the network topology. In each Registration Acknowledgement, a

nodeId is assigned to a WVS. The usage of a shorter identifier than the MAC address minimises

the size of the Poll message. By collecting the network information topology, the optimum

polling vector can be calculated. Since the time required to switch a Wi-Fi radio ON or OFF

can affect performance, it is required to minimise the number of times the Wi-Fi radio changes

between ON/OFF states. WATCM uses the collected network topology information to find

the branches and leaves of the WVSNs and construct the optimum polling vector then used

by the NSM. The WATCM starts with the registration phase. Each WVS registers itself in

the gateway by sending a Registration message with its MAC address and the MAC address

of its parent in the logical tree rooted at the gateway. WATCM assumes each WVS learns its

parent in the active topology through the routing protocol used. This set of messages allows the

gateway to compute the WVSN active topology. For each new Registration message received,

the gateway generates a nodeId. The nodeId is an integer with an initial value equal to 1.

56 GREENNESS

Each time a new Registration message is received at the gateway the nodeId is incremented.

Therefore, a unique nodeId is guaranteed for each WVS. The nodeId will be used by the NSM

running at the gateway to address each WVS through the LPR. The gateway then sends a

Registration Acknowledgement message to the source WVS with the nodeId. These messages

are sent through Wi-Fi. In order to minimise the number of bytes sent through the LPR in

the next phases, all nodes in the path between the current WVS and the gateway snoop the

message and get the nodeId; this way when the gateway signals a WVS to turn ON its Wi-Fi

radio all the nodes in the path will also turn ON their Wi-Fi radio, allowing the message to be

relayed all the way up to the gateway. To avoid flooding of Address Resolution Protocol (ARP)

messages, the Registration message can include the WVS Internet Protocol (IP) address and

MAC address; the gateway stores in its ARP table the WVS MAC address and corresponding

IP address. For the same reason, the Registration Acknowledgement can include the streaming

server MAC address; when the Registration Acknowledgement message reaches the WVS, it

also stores the streaming server MAC address and corresponding IP address in its ARP table.

Fig. 3.3 shows the format of the Registration and Registration Acknowledgement messages.

Ethernet HeaderGateway

MAC Address

6 bytes 6 bytes26 bytes

WVSMAC Address

WVS IP Address(Optional)

4 bytes

WVS ParentMAC Address

6 bytes

Hop Count

2 bytes

(a) Registration message.

Ethernet HeaderNode registration

MAC Address Node

Id

6 bytes 1 byte 6 bytes26 bytes

StreamingServer

MAC Address(Optional)

StreamingServer

IP Address (Optional)

4 bytes

(b) Registration Acknowledgement message.

Figure 3.3: Registration and Registration Acknowledgement messages used to collect theWVSN active topology.

The main purpose of the registration procedure is to let the gateway compute the WVSN

active topology and construct the polling vector, which is an ordered list of nodeIds that will

be used by the NSM. Every time a new Registration message arrives at the gateway a set

of local vectors are updated. Vector S contains the MAC addresses of nodes received in the

Registration messages, S[k] is the MAC address of the node having the nodeId k, where k ∈1, ...,N−1 and N is the number of nodes in the WVSN, including the gateway. In Fig. 3.1,

N = 7 and S[1] = mac1. Vector H stores the hop count of each WVS to the gateway, with H[k]

3.3 WVSN Active Topology Collection Mechanism 57

representing the hop count of WVS with nodeId k (in Fig. 3.1, H[3] = 2). The Registration

message includes the hop count which is incremented in each WVS. Vector PM stores the MAC

addresses of parents, PM[k] representing the parent of WVS k (in Fig. 3.1, PM[3] = mac1); PM[k]

is an auxiliary vector. After receiving all Registration messages, the vector P, representing the

parent nodes, is created by searching each PM[ j] in S[k] and making P[ j] = k, j ∈ 1, ...,N−1(in Fig. 3.1, P[3] = 1). The first objective of the algorithm is to find the set of WVSs that

compose each branch of the active tree topology. For each branch of the tree, the gateway

creates the vector Bl which contains the list of nodeIds that belong to branch l. Bl includes the

nodeIds of the WVS nodes that belong to the branch l ∈ 1, ...,Nleaves, where Nleaves is the

number of leaf nodes in the active tree topology (in Fig. 3.1, Nleaves = 4). Bl[d] is the nodeId of

the WVS from branch l at position d, d ∈ 1, ...,D, where D is the depth of the active tree (in

Fig. 3.1, D = 2). The first position of the vector includes the nodeId of the leaf node; the other

positions include the sequence of WVSs up to the gateway. The algorithm starts by finding

in H the elements with hop count, hmaxval , i.e., the leaf nodes in the active topology. The tree

network topology in Fig. 3.1 is a balanced binary tree, so all leaves have the same hop count

and hmaxval = 2. Next, all nodeIds from H that have a hop count equal to hmaxval are added

to Bl[0]; one Bl vector is created for each nodeId found in H. Then, Bl[1] includes the parent

nodeId, which can be obtained by looking up in P; this process is repeated for Bl[2], Bl[3],

..., Bl[Nleaves], and stops when the parent nodeId is the gateway. For the network topology in

Fig. 3.1, four vectors are created with the following values: B1 = [3,1], B2 = [4,1], B3 = [5,2],

and B4 = [6,2]. Subsequently, the algorithm has to identify the missing branches with hop

count lower than hmaxval . It starts by finding the nodeIds that have (hmaxval−1) hop count and

were not yet added to Bl . If a new branch is found, l is incremented, and the first element and

its parent (in case the WVS parent is not the gateway) are added to Bl . The new Bl is created

using the same recursive process explained above.

Finally, the vector V can be obtained. Vector V represents the polling order of the WVSs.

It can be obtained by concatenating the vectors Bl . The repeated nodeIds are removed since

different branches can have the same node included in their list. For instance, in Fig. 3.1,

WV S#1 is common to leaf nodes WV S#3 and WV S#4. Vector V is used by the NSM to poll

the WVSs (in Fig. 3.1, V = [3,1,4,5,2,6]). The WATCM is formally described in Alg. 1.

The NSM is controlled by the gateway and is responsible for scheduling the WVS data

transmission and turning the WVS Wi-Fi radios ON/OFF accordingly. The scheduling is based

on the polling order computed by the WATCM. Furthermore, it is designed to be traffic-aware

to improve energy-efficiency; it does not poll a WVS when there is no traffic, so the Wi-Fi

58 GREENNESS

radio of the WVS is kept OFF.

Algorithm 1 WVSN Active Topology Collection

Input: Rn, Registration messages dataOutput: V [x]

1: N← Rn +12: S[0]← GW MAC address3: for k ∈ [1..Rn] do4: S[k]←WVS source MAC address5: H[k]← hop count6: PM [k]←WVS parent MAC address7: end for8: for j ∈ [1..(N−1)] do9: for i ∈ [0..(N−1)] do

10: if PM [ j] = S[i] then11: P[ j]← i12: end if13: end for14: end for15: l← 016: d← 017: u← 018: hmaxval ← max(H[k])19: for k ∈ [1..sizeo f (S[k])−1] do20: if H[k] = hmaxval then21: if d 6= 0 then22: l← l +123: d← 024: end if25: Bl [d]← k26: d← d +127: Bl [d]← P[k]28: T [u]← P[k]29: L[l]← P[k]30: u← u+131: d← d +132: end if33: end for34: for i ∈ [1..hmaxval ] do

35: for j ∈ [0..l] do36: if P[L[ j]] 6= 0 AND L[ j] /∈ T then37: B j[sizeo f (B j)]← P[L[ j]]38: L[ j]← P[L[ j]]39: if L[ j] /∈ T [u] then40: T [u]← P[L[ j]]41: u← u+142: end if43: end if44: end for45: for all k ∈ [1..sizeo f (S[k])−1] do46: if H[k] = (hmaxval − i) and k /∈ T then47: l← l +148: d← 049: Bl [d]← k50: if P[k] 6= 0 then51: d← d +152: Bl [d]← P[k]53: L[l]← P[k]54: if L[l] /∈ T then55: T [u]← L[l]56: u← u+157: end if58: end if59: end if60: end for61: end for62: x← 063: for j ∈ [0..l] do64: for i ∈ [0..sizeo f (B j)−1] do65: if B j[i] /∈V then66: V [x]← B j[i]67: x← x+168: end if69: end for70: end for

3.4 Node Scheduling Mechanism

As explained in Section 3.3, for each Registration message received the gateway generates

a new nodeId that is used to address the corresponding WVS through the LPR. Alg. 2 and

Alg. 3 formally define the NSM that runs over the LPR installed in the gateway and in each

WVS node, respectively. It works as follows. Initially, each node keeps the Wi-Fi radio

switched ON. After successfully registering in the gateway, all WVS nodes switch OFF their

Wi-Fi radios. Then, for each element found in vector V , the gateway sends a Poll message

3.4 Node Scheduling Mechanism 59

Algorithm 2 Node scheduling algorithm run-ning in the WVSN gateway

Input: V [x], packet from nodeId, data available, queuesize, QS, Mwarm−upOutput: Poll message with nodeId and hasData

1: i← 02: MPoll ← 03: while True do4: Increment MPoll5: for all nodeId ∈V do6: if MPoll =< Mwarm−up then7: if data available for nodeId then8: hasData← True9: end if

10: send Poll message with nodeId and hasData11: if hasData=True then12: send packet to WVS13: end if14: set TGWTout15: while packet from nodeId not received OR

TGWTout > 0 do16: if packet from nodeId is received then17: set Q[nodeId]18: end if19: Decrement TGWTout20: end while21: end if22: if TGWtimeout < 0 OR Q[nodeId] = 0 then23: O[MPoll ][nodeId]← 024: else25: O[MPoll ][nodeId]← 126: end if27: if MPoll > Mwarm−up then28: if O[i][nodeId] = 1 OR Q[nodeId] => 1 then29: send Poll message with nodeId30: if hasData=True then31: send packet to WVS32: end if33: set TGWTout34: while packet from nodeId not received

OR TGWTout > 0 do35: if packet AND queue size > QS then36: O[i+1][nodeId]← 137: else38: O[i+1][nodeId]← 039: end if40: Decrement TGWTout41: end while42: end if43: i← i+144: if i => Mwarm−up then45: i← 046: end if47: end if48: end for49: end while

Algorithm 3 Node scheduling algorithm run-ning in the WVS upon receiving Poll message

Input: Poll message, queue size, data available, nodeId,hasData, myNodeIdOutput: turn ON/OFF radio, send packet

1: for all Poll messages received do2: Increment MPoll3: if nodeId = myNodeId then4: turn Wi-Fi radio ON5: if hasData=True then6: set TWV STout7: while packet from gateway not received OR

TWV STout > 0 do8: Decrement TWV STout9: end while

10: end if11: if MPoll <= Mwarm−up then12: if data available then13: send packet and include queue size14: else15: send a packet with queue size16: end if17: else18: while New Poll message is not received do19: if data available then20: send packet and include queue size21: end if22: end while23: end if24: else25: if nodeId ∈ R then26: turn Wi-Fi radio ON and forward packets27: else28: turn Wi-Fi radio OFF29: end if30: end if31: end for

through the LPR containing the nodeId of the WVS that should turn the Wi-Fi radio ON and

60 GREENNESS

a bit which is set to 1 when the gateway has data to be transmitted to the WVS. Each WVS

verifies whether its nodeId is included in the Poll message. Next, the WVS whose nodeId

matches the one included in the Poll message checks whether the gateway has data to transmit

by checking the bit in the Poll message. When the gateway has data to transmit, a timeout

TWV STout is configured, and the WVS waits for a packet from the gateway. After receiving it or

TWV STout has expired, the WVS can send its own packet to the gateway. During the registration

phase of the WVSN Active Topology Collection Mechanism, each WVS stores in vector R

the nodeId that was assigned to a child node by snooping the Registration Acknowledgment

message (in Fig. 3.1, R = [3,4] in node WV S#1). Every relay WVS that finds its nodeId in

R switches its Wi-Fi radio ON. In the gateway, a timeout TGWTout is configured to assure that

in case of failure another WVS is scheduled to transmit. The gateway checks if data from

nodeId has been received or TGWTout has expired; while these conditions are not met TGWTout

is decremented. If a packet from nodeId is received or TGWTout expires, the gateway polls the

next element in V . When the WVSs do not find their nodeId in the Poll message or in R, the

Wi-Fi radio is switched OFF. This process is repeated until all nodeIds have been polled by the

gateway and the polling cycle has been completed. After that, the first node is scheduled again,

and a new polling cycle is initiated.

In a video monitoring scenario, the WVS bit rate will depend on the camera video quality

and resolution. So, for low video quality, the WVS duty cycle would be short. This may mean

that a WVS receiving a Poll message may not have packets to transmit to the cloud server. To

further improve energy-efficiency, the NSM is designed to be traffic-aware. During the period

of time the gateway knows there will be no traffic from a given WVS, the WVS is not polled,

and the Wi-Fi radio is kept OFF.

In Alg. 2, after a warm-up period, the gateway will only send a Poll message according

to the WVS traffic pattern. During the first set of Poll messages, Mwarm−up, Alg. 2 stores

in O[ ][ ], for each nodeId and Poll messageId, the current queue status (0 - empty; 1 -

otherwise); even if the WVS does not have a packet to transmit, a packet with this information

is sent to the gateway. During the warm-up period, Mwarm−up, the gateway stores 1 or 0 in

O[messageId][nodeId] according to the queue status received from each nodeId. After the

warm-up period, Mwarm−up, the gateway will replay the traffic pattern initially learned for each

nodeId. The WVSs will no longer be forced to send a packet, but the queue size is still included

in the packets sent to the gateway. When the gateway receives a packet from the nodeId whose

length is higher than the QS threshold, it forces a change in the state of next Poll message by

setting the array O[MPoll + 1][nodeId] to 1. This enables the possibility to change the polling

3.5 Failure Recovery Mechanism 61

order when the rate is modified. For Constant Bit Rate (CBR) video streams, this estimation

can be easily performed and enables further energy savings.

3.5 Failure Recovery Mechanism

GREENNESS was designed considering that the network topology remains stable most of

the time. Nevertheless, it supports a Failure Recovery Mechanism (FRM) to deal with WVSN

topology changes. The FRM runs at the gateway and is triggered when a timeout, TGWTout ,

occurs more than three times for a given nodeId; this means that the gateway did not receive

any packet from the WVS with nodeId in 3 consecutive polling intervals, and the topology has

changed or the node was disconnected from the network. Upon this event, the gateway issues a

Poll message through the LPR with the nodeId set to 0. This will force all the WVSs to turn ON

their Wi-Fi interfaces and start the WiFIX ATCM mechanism together with in-band polling,

i.e., in this phase the NSM runs over the Wi-Fi interface, not the LPR, forcing the sensors re-

registration. By returning to WiFIX and in-band polling operation, GREENNESS can rebuild

the network topology while the in-band polling minimises interruption of video streams and

performance is maintained. The WiFIX ATCM mechanism at the gateway broadcasts periodic

TR messages. Upon receiving these messages, the WVSs choose the best upstream neighbour.

Then, the WVSs send a Registration message, and the gateway runs the WATCM again to

discover the new network topology. Once the network is reconfigured, GREENNESS returns

to its regular operation using out-of-band control. Besides, when it is required to add a new

WVSs or change the location of the node, the network manager can manually trigger the FRM

so that the new node is added to the topology and registered in the gateway. GREENNESS was

not designed considering mobile WVSs. Nevertheless, the solution could evolve to support

scenarios when WVSs change position. Upon detecting the WVSN topology change triggered

by the mobile WVS, the gateway could periodically send a Poll message with the nodeId set

to 0 in order to force the update of the network topology.

3.6 Low Power Radio

In GREENNESS, the LPR plays an important role not only for reducing the energy con-

sumption of the WVSN but also as a mean to control the access to the medium. In this section

we define the requirements that need to be fulfilled by the LPR, present a set of candidate

62 GREENNESS

low power technologies, and discuss their suitableness to be used as the GREENNESS LPR

module.

3.6.1 Low Power Radio Requirements

The LPR technology needs to fulfil the following set of requirements:

1. Maximum Range: the radio coverage shall be large enough to enable communications

between the gateway and each WVS forming the Wi-Fi-based multi-hop network;

2. Power Consumption: the power consumption shall be lower than the power consump-

tion of a Wi-Fi radio in idle mode;

3. Maximum Payload Length: the payload length should be at least 9 bits, 8 bits for the

nodeId and 1 bit to indicate whether the gateway has traffic to be delivered;

4. Minimum Inter-frame Interval: the inter-frame interval should be similar to the one

achieved by the in-band control using Wi-Fi in [13] to assure similar performance results;

5. Communication Direction: unidirectional communications between the gateway and

each WVS shall be guaranteed – bidirectionality is optional.

The radio coverage will depend on the transmission power but also on the receiver sensitiv-

ity. Under ideal conditions the Friis propagation model can be used to get the power received:

Prx = PtxGtxGrx

(c

4πDr f0

)2

(3.1)

Ptx is the transmission power, while Gtx and Grx represent the gain of the antennas of the

transmitter and receiver, respectively. Furthermore, the received power depends on the distance

Dr between the transmitter and the receiver, the selected frequency f0, and c the speed of light

in vacuum. Since we are aiming to cover the range of Wi-Fi multi-hop networks, the LPR

range should at least be higher than 400 m, which is the typical Wi-Fi range. Nevertheless,

we can use LPR transceivers with radio range of 100 m, for instance, considering that we can

adopt a multi-hop LPR network, i.e., some WVSs can relay information from neighbour LPRs.

Using Eq. (3.1), we can calculate the received power when Dr is equal to 100 m. Assuming

the LPRs is operating at 2.4 GHz, the transmission power Ptx is 0 dBm, and the gains of the

antennas Gtx and Grx are 3 dBi, the received power Prx is −74 dBm. So, when selecting a

receiver, the sensitivity should be lower than −74 dBm. Signal-to-Noise Ratio (SNR) values

3.6 Low Power Radio 63

for common digital modulations and Noise Figures for custom equipment may add about 10 dB

to the receiver sensitivity. However, the usage of techniques such as Forward Error Correction

(FEC) and Repetitions may enable the system to work with zero or even negative SNR (in

dBs) at the cost of increasing latency. In this case, the receiver sensitivity requirement may be

relaxed.

The power consumption of the LPR transceivers should be bounded by the power con-

sumption of a Wi-Fi radio in idle mode to assure that even by adding a second radio we can

still save energy. By switching OFF or moving the Wi-Fi card to sleep state, we can save energy

that can be used to power up a second radio for controlling the access to the medium by the

WVSs. Moreover, if the radio coverage needs to be extended, the transmission power may have

to be increased. This will increase the power consumption of the transmitter. However, only

unidirectional communications are required. As such, only the transmission power of the LPR

running at the gateway is relevant. The transmission power Ptx has no influence on the power

consumption of the WVSs. They only need to receive control messages from the gateway.

The LPR transceiver must be able to address the WVS that needs to wake-up, in order

to avoid the selection of several WVSs, which increases the nodes overhearing time. The

authors of [84] proposed the message format shown in Fig. 3.4 that can be used to fulfil this

requirement. The message includes a header with a wake-up preamble and a Start Frame

Delimiter (SFD) to respectively synchronise the transmitter and receiver, and indicate the

beginning of information. The message also includes an address field that is used to indicate

the nodeID of the destination WVS. The size of this field is 2 Bytes [84], but in our case,

8 bits is enough for a WVSN with 255 nodes. Additionally, the message may contain data

or extra instructions. In our case, this is used to indicate whether the gateway has data to be

transmitted to the WVSs. The message ends with the Frame Check Sequence (FCS) that uses

a Cyclic Redundancy Check (CRC) to provide a high degree of error detection. In our case,

the message can be limited to 8 bit for the nodeId and 1 bit to indicate whether the gateway has

traffic to be delivered.

The energy consumption of an LPR depends on the amount of time required to deliver the

data. This time depends on the transmission time, that relies on data rate supported by the radio

link and frame size, but also on the amount of time that is required to generate the message.

For the latter, we have defined an inter-frame interval as the minimum interval between Poll

messages. The inter-frame interval affects the WVSN performance since a long interval will

increase the Poll message interval and the latency of packets sent from each WVS to the

64 GREENNESS

Figure 3.4: Typical LPR message format [84].

gateway. In order to maintain the performance of PACE, the minimum inter-frame interval

value should be similar to the one used in Wi-Fi [13].

Although GREENNESS requires only unidirectional communications between the gateway

and each WVS, Automatic Repeat Request (ARQ) based recovery mechanisms for control

messages could be implemented if a bidirectional channel is available. In that case, a WVSs

could also request access to the medium using the control channel and possibly change the

polling order.

Finally, the LPR should be cost-effective in order not to increase the overall cost of the

WVS. By selecting off-the-shelf LPRs, we can reduce the overall cost and development time

when compared to designing a customised solution [84].

3.6.2 Candidate Technologies

Taking into account the requirements defined in Section 3.6.1, we have identified the set of

candidate LPRs shown in Table 3.2. In order to select this set of candidates, we have conducted

an exhaustive review of wake-up radios [84][85][86][87]. In the following sections, we briefly

describe each LPR candidate and characterise it with respect to the requirements identified

above.

Table 3.2: Candidate Low Power Radios.

Standard Max. Range Power con-sumption

Max. PayloadLength

Min. Inter-frameInterval

FM-RDS 100 km 49.2 mW 104 bit 87.58 ms802.15.4 100 m 30.4 mW 127 B 3 ms802.15.4g 1 km 57.0 mW 2,047 B 21 ms802.15.4k 5 km 52.8 mW 2,047 B 25 msBLE 100 m 44.2 mW 27 B 7 msDASH7 5 km 55.5 mW 256 B 16 ms


FM Radio Data System (FM-RDS)

FM Radio Data System (FM-RDS) [88] has been initially considered in [10], motivated by

its high radio range (from 50 m to 100 km) and low power consumption (49.2 mW). Although

an RDS group of 104 bits will be enough to carry the nodeId and 1 bit to indicate whether

the gateway has traffic to be delivered, there is a limitation related to the use of the RDS

control channel to run the GREENNESS NSM. The standard states that the RDS bitrate must

be precisely 1187.5bit/s± 0.125bit/s [88]. Since the standard also specifies RDS groups of

104 bits and the transceivers require the generation of an entire RDS group, it takes 104bit1187.5bit/s ≈

87.58ms to send a polling request. This results in a packet rate of ≈ 11.4 packets/s, which

requires the transmission of more than one packet per polling interval to maintain the same

performance. By allowing the transmission of more than one packet the jitter and delay of the

video transmission are increased. The jitter and delay are affected because the time interval

between consecutive polls to a given node becomes large. Moreover, FM-RDS only allows

unidirectional communications and relies on message repetition to recover from errors.

IEEE 802.15.4

IEEE 802.15.4 radios using the 868/915/2450 MHz bands are an alternative solution. The

Atmel AT86RF212 transceivers, for example, are optimised for low power communications

and support up to 250 kbit/s. Although the radio range of IEEE 802.15.4 is similar to the radio

range of IEEE 802.11, IEEE 802.15.4 supports mesh topologies. When the transceiver is ON,

the power consumption is only 30.36 mW [86]. The minimum inter-frame interval is 3 ms, for

a payload of 10 Bytes in a single-hop network [89], which is similar to the one achieved in

[13] with Wi-Fi. Moreover, IEEE 802.15.4 allows bidirectional communications, which can

enable the implementation of recovery mechanisms when a polling message is not received,

for example.

IEEE 802.15.4g

The IEEE 802.15.4g standard was defined by the Smart Utility Networks (SUN) Task

Group as an amendment to 802.15.4. It enables the deployment of large-scale process control

applications, such as the utility smart grid network capable of supporting large, geographi-

cally dispersed networks with minimal infrastructure. The transceiver CC1200 from Texas

Instruments, for example, has a radio range up to 1 km, enabling scenarios of sparse WVS

nodes across a large geographic area. This transceiver can operate at 1250 kbit/s with a typical

66 GREENNESS

power consumption of 57 mW for receiving data in low-power mode. As in IEEE 802.15.4, the

minimum inter-frame interval meets the requirements and also allows bidirectional communi-

cations.

IEEE 802.15.4k

For Low-Energy Critical Infrastructure Monitoring (LECIM) application the Task Group

(TG4k) proposed the standard IEEE 802.15.4k for operating in the Industrial, Scientific and

Medical (ISM) bands (Sub-GHz and 2.4 GHz). This standard adopts Direct Sequence Spread

Spectrum (DSSS) and Frequency-Shift Keying (FSK) as two new PHY layers to increase

the range to 5 km [87]. The MAC layer specification has been amended to support Carrier

Sense Multiple Access – Collision Avoidance (CSMA/CA), and CSMA/CA and ALOHA with

Priority Channel Access (PCA). The addition of PCA enables traffic prioritisation, providing

service differentiation for critical applications like emergency services. A star topology

network deployment to monitor air quality was presented in [90], proving that the IEEE

802.15.4k technology supports radio ranges up to 3 km. The ML7404 transceiver from Rohm

semiconductor [91] reports a power consumption of 52.8 mW with an average delay of 25 ms,

considering 8 packet retransmissions tagged as an emergency [92].

Bluetooth Low Energy (BLE)

Bluetooth Low Energy (BLE) – operating in the 2.4 GHz frequency band – is another

candidate LPR. BLE has radio range lower than Wi-Fi, but the standard includes mesh

networking capabilities. In [93] a multi-hop latency test was performed in a network with

6 BLE nodes; for a payload of 8 Bytes, the attained round-trip time was 15 ms and 100 ms

respectively for 1 and 6 hops. Thus, we estimate a minimum inter-frame interval of 7.5 ms

for a single-hop network, reaching 50 ms for networks with 6 hops. This calculation assumes

that the uplink and downlink paths in BLE are symmetric. The nRF51822, for example, is a

System-on-Chip that implements BLE. The maximum power consumption is 44.2 mW [86].

Furthermore, BLE also supports bidirectional communications.

DASH7

The standard ISO/IEC 18000-7 [94] was created to define the air interface for the active

Radio-Frequency Identification (RFID), but the DASH7 Alliance developed a full stack to

provide mid-range connectivity for sensors and actuators. This technology was named DASH7


and uses a narrow band modulation, operating in the Sub-GHz bands. It enables radio ranges

up to 5 km. DASH7 includes a MAC protocol with a mechanism to periodically check for

possible downlink transmissions, which reduces latency in downlink communications to 5 ms

for reduced duty cycles [95]. The periodic monitoring of the communication channel increases

the power consumption of the transceiver, but the value announced in the datasheet of SH1030

– Ultra low power DASH7 Arduino Shield Modem [96] is in the same range of the other

technologies, around 44.5 mW.

3.6.3 Discussion

There are several candidate LPR technologies available. FM-RDS was already demon-

strated and tested in [11], which allowed us to prove that it could be used despite the long

inter-frame interval imposed. The radio coverage of FM-RDS also imposes restrictions since

the 100 km range is only possible for licensed broadcasters; for unlicensed broadcasters, a

range of 50 m can be achieved using the so-called Low Power FM. We believe that the

use of IEEE 802.15.4g or IEEE 802.15.4k can bring advantages such as higher range (for

802.15.4k, 5 km range) and bidirectional communications, allowing the implementation of

recovery mechanisms for control messages or more advanced scheduling mechanisms, for

instance, the possibility of changing the polling order. Moreover, by having a higher bit

rate and a lower inter-frame interval (lower than 25 ms), we can improve the overall latency

of the network because the Poll of each WVS can be performed faster. Long Range Wide

Area Networking (LoRaWAN) is an example of the most adopted Low-Power Wide Area

Networking (LPWAN), but with a duty cycle of 1 %, it is not an LPR candidate. The

same happens for WEIGHTLESS standards [97], and INGENU [98], which provide different

features, radio range, and power consumption but the downlink latency is 8 s [99] and 2 s [100]

respectively. Since the inter-frame interval is too high, it is not possible to have real-time

control of the Wi-Fi radios. 3GPP NB-IoT [101] can also be considered, but it is operator-

based and requires a data plan, which increases the cost of the solution. In the future, we

can also consider IEEE 802.11ah (Wi-Fi HaLow), since this new standard promises higher

ranges and lower power consumptions while providing a low inter-frame interval. It was not

included in our study because the chipsets along with their actual power consumption will

become available only in 2019.

68 GREENNESS

3.7 Summary

For a wireless video sensing scenario where WVSs transmit to a sink located in the cloud,

the state-of-the-art solutions cannot address the three problems described in Chapter 1. In this

chapter, we presented GREENNESS, a holistic solution for multi-hop WVSNs. GREENNESS

aims at reducing energy consumption and improving performance and throughput fairness of

multi-hop WVSNs when compared to state-of-the-art IEEE 802.11-based WVSNs. GREEN-

NESS is composed of five modules: 1) Routing module, 2) WATCM, 3) NSM, 4) FRM, and 5)

LPR. Each of these modules was described in this chapter.

The Routing module maintains the routes between the WVSs and the gateway. Although

GREENNESS can run on top of any state-of-the-art routing solution, WiFIX was selected since

it is a suitable solution for tree-based WMNs [12].

The WATCM mechanism is responsible for registering the WVSs in the gateway and

computing the polling vector. The registration process is used to collect information about

the network topology. After knowing the network topology and computing the polling vector,

the WATCM mechanism triggers the NSM mechanism.

The NSM mechanism uses the polling vector calculated by WATCM mechanism to sched-

ule the WVS data transmission and turn the WVS Wi-Fi radios ON/OFF accordingly. NSM

mechanism is designed to be traffic-aware and only polls a WVS when there is traffic to

transmit from that node. NSM mechanism learns the CBR pattern in order to keep the WVS

Wi-Fi radio OFF when there are no packets to transmit, improving energy-efficiency.

The FRM detects failures and rebuilds the network topology, for instance, when a WVS is

removed from the WVSN. FRM is triggered when a timeout happens in 3 consecutive polling

intervals. The gateway starts the recovery mode by sending a Poll message through the LPR

with the nodeId set to 0. The WVSs turn ON their Wi-Fi interfaces, and start WiFIX ATCM,

forcing the re-registration of the sensors.

The LPR module is used to transport the out-of-band control signal, which is employed to

poll each WVS. The requirements for the LPRs are discussed as LPR is essential for reducing

the energy consumption of the WVSN and to control the access to the Wi-Fi medium. Based

on these requirements, a set of LPR candidates were studied and discussed.

Chapter 4

GREENNESS Evaluation

GREENNESS was evaluated using numerical analysis, simulations, and experimentation

with a proof-of-concept prototype. This chapter first presents the evaluation methodology used.

Then, we analyse the energy consumption of a WVSN numerically when GREENNESS is used

and compare it with the state-of-the-art solutions. Finally, we describe the ns-3 simulations and

experimental evaluation carried out also considering traffic aspects.

4.1 Evaluation Methodology

The evaluation methodology was devised to verify whether GREENNESS can actually

save energy while improving delay, packet loss ratio, throughput, and throughput fairness.

A comparison with the state-of-the-art CSMA/CA-based WVSN solutions was considered.

To achieve this objective, we have proceeded as follows: 1) estimated the GREENNESS

and PACE energy consumption for regular and random network topologies numerically; 2)

assessed by means of ns-3 simulations the energy saving for random network topologies and

compared it with the numerical results for validating the numerical analysis; 3) characterised

the impact of different LPR power consumptions and the switching of Wi-Fi radios to sleep

mode on GREENNESS energy savings; 4) used simulations to study the energy savings for

different offered traffic loads and compared performance of GREENNESS with CSMA/CA-

based WVSN solutions; and 5) deployed a testbed to measure energy consumption and

achieved performance using a proof-of-concept prototype.

The numerical estimation of GREENNESS and PACE energy consumption for random

network topologies was performed to evaluate the energy saving that can be attained using

GREENNESS. PACE and GREENNESS energy consumption numerical analysis was first

69

70 GREENNESS Evaluation

studied in [10] for regular topologies, and then extended in [102] for random topologies.

Although PACE was not designed to be energy efficient, by controlling the access to the

medium, it avoids packet collisions and is indeed more efficient than CSMA/CA-based WVSN

solutions, as shown in [13]. Using this analysis, the energy savings were estimated for a

scenario where one packet is transmitted from every single WVS to the gateway.

GREENNESS was implemented in the network simulator 3 (ns-3) in order to further

analyse the energy savings that could be attained and to compare them with the numerical

results. Moreover, the implementation in ns-3 allowed studying the performance regarding the

delay, throughput, packet loss ratio, and throughput fairness. The simulations were performed

with 8,874 WVSNs with random wireless multi-hop networks topologies and composed of

10, 20, and 30 WVSs. The 8,874 WVSNs were used to analyse the energy consumption for

12 different average number of hops from 2 to 4.2, assuring at least 50 simulations for each

average number of hops. Afterwards, an analysis of ns-3 simulation results was performed and

the numerical analysis validated.

Upon validation, the numerical analysis was also used to study the impact of LPR power

consumption in the energy saving of GREENNESS. This analysis assessed the energy saving

achieved by GREENNESS in case we select one of the candidate LPRs discussed in Chapter 3.

Moreover, we evaluated the scenario when Wi-Fi radios are not switched OFF since in practice

this can cause energy transient effects or delays as the WVSs need to re-associate to a neighbour

WVS. Instead, the Wi-Fi radios of WVSs were considered to be switched to sleep mode and

so the Wi-Fi sleep power consumption was also considered in the numerical analysis and its

impact on GREENNESS energy savings.

The energy saving of the traffic-aware feature of the NSM mechanism was also evaluated

for different offered network loads using ns-3 simulations. Afterwards, GREENNESS perfor-

mance was assessed and compared with CSMA/CA-based WVSN solutions. Using the same

WVSN topologies, the delay, throughput, packet loss ratio, and Jain’s Index – which measures

the level of fairness – were compared between GREENNESS and CSMA/CA for different

offered network loads.

Finally, a prototype was developed for a wireless video sensing scenario to demonstrate

that GREENNESS reduces the energy consumption and keeps the performance and throughput

fairness. The prototype was developed to create a testbed to measure the energy consumption

of GREENNESS and validate the simulation results. Table 4.1 provides the notations used in

the numerical analysis, simulations, and experiments.

4.1 Evaluation Methodology 71

Table 4.1: Notations used in the equations derived in the numerical analysis as well as in thesimulations, and experiments.

Notation Description

E Energy consumption of the WVSN

Etx Energy consumption of the WVSN for transmitting frames

Erx Energy consumption of the WVSN for receiving frames

Eidle Energy consumption of the WVSN in idle mode

Eoverhear Energy consumption of the WVSN for receiving frames destinated to other

nodes

ttotal Total time spent by all WVSs in transmitting, receiving, idle mode, and

overhearing frames

tPACE Total time spent by all WVSs in transmitting, receiving, idle mode, and

overhearing frames when running PACE

tGREENNESS Total time spent by all WVSs in transmitting, receiving, idle mode, and

overhearing frames when running GREENNESS

N Number of nodes forming the WVSN (WVSs plus the gateway)

D Depth of the tree defining the active topology of the WVSN

T Time for a node to transmit a data frame to its neighbour and receive the IEEE

802.11 ACK

Nleaves Number of leaf nodes in the active tree topology

Fcor Correction factor to cancel the duplicate transmissions of WVSs relays

C Number of WVSs on one side of the network grid topology

h j Number of hops for branch j

B j WVSs associated with branch j

xi j Returns 1 when the WVS, i, belongs to branch j or 0 otherwise

EPACE Energy consumption of the WVSN when running PACE

EGREENNESS Energy consumption of the WVSN when running GREENNESS

ELPR Energy consumption of the LPRs of the WVSN

PWiFisleep Power consumption of the Wi-Fi radio in sleep mode

PLPRrx Power consumption of the LPRs of the WVSN when receiving data

tLPRrx Total time the LPRs of the WVSN are receiving data

PLPRtx Power consumption of the LPRs of the WVSN when transmitting data

tLPRtx Total time the LPRs of the WVSN are transmitting data

Esaving Energy saving ratio achieved by GREENNESS when compared with PACE


For the evaluation of GREENNESS we used the following figures of merit: energy saving,

network capacity, delay, packet loss ratio, and throughput fairness. These metrics are defined

below.

Energy Saving

To calculate the amount of energy that is saved by Solution 1 with respect to Solution 2,

Eq. (4.1) can be used, where E1 is the energy spent by Solution 1 and E2 is the energy spent by

Solution 2.

Esaving =(

1− E1

E2

)×100 (%) (4.1)

Network Capacity

The performance of a network can be measured by calculating the throughput of each

flow, which gives the rate of successful packets delivered over the Wi-Fi channel. Eq. (4.2)

calculates the throughput for flow i using the number of bytes received in the gateway, divided

by the duration of the flow, which is the difference between the last and first timestamp of the

received packets. Since in the wireless video sensing several flows will coexist in the WVSN,

the throughput of all flows in the network is summed to calculate the network capacity. Eq. (4.3)

calculates the rate received by the gateway for flow i, and sums it for all flows.

T Hi =RxBytesi×8

LastRxTimei−FirstRxTimei(4.2)

NC =N f lows

∑i=1

T Hi (4.3)

Delay

The end-to-end Delay or One-Way-Delay (OWD) in a network is the sum of the transmis-

sion, propagation, processing, and queuing delay. When the network is saturated the OWD

increases since packets can stay more time in queues, waiting to be transmitted. For each

packet j in a flow, OWD is calculated using Eq. (4.4). For each flow the average and median

OWD can be measured using Eq. (4.5) and Eq. (4.6), respectively.

OWD[ j] = Trx j −Ttx j (4.4)

4.2 Numerical Analysis 73

The average OWD is calculated by summing the OWD per successfully received packet

and dividing it by the number of received packets.

OWDavg =1

Npackets

Npackets

∑j=1

OWD[ j] (4.5)

To calculate the median of OWD, the vector OWD[ j] is sorted and then the middle value is

selected, as defined by Eq. (4.6).

OWDmedian = OWDsorted

[Npackets +1

2

](4.6)

Packet Loss Ratio

Packet loss in a wireless network can be caused by errors in the data transmission or by

network congestion. For each flow, the Packet Loss Ratio (PLR) is given by the ratio between

the number of lost packets and the number of transmitted packets. To measure the PLR for the

network, the total number of lost packets Plosti is divided by the total number of transmitted

packets Ptxi , as defined by Eq. (4.7).

PLR =∑

N f lowsi=1 Plosti

∑N f lowsi=1 Ptxi

(4.7)

Throughput Fairness

Since each WVS sends a flow to the gateway, we need to assess if all WVSs have equal

opportunity for transmitting. To measure throughput fairness, we use the well-known Jain’s

index [103]. Throughput fairness is calculated using Eq. (4.8), considering the throughput of

each flow, T Hi, given by Eq. (4.2). When the throughput fairness is close to 1 it means fairer

resource allocation for each WVS.

T F =

(∑

N f lowsi=1 T Hi

)2

N f lows ∑N f lowsi=1 T H2

i

(4.8)

4.2 Numerical Analysis

We assume a scenario in which all WVSs transmit data to the gateway. The total energy E

spent by the network considering the transmission of a successful frame from each WVS to the


gateway can be defined as follows [104]:

E = Etx +Erx +Eidle +Eoverhear (4.9)

where Etx and Erx are respectively the total energy required to transmit the frame and the

total energy required to receive one frame from all WVSs, Eidle is the total energy that all

WVSs spend in idle mode, and Eoverhear is the total energy spent by all WVSs when receiving

frames that are destined to other nodes. GREENNESS aims at minimising Eidle and Eoverhear

by switching OFF the Wi-Fi radio.

As can be observed from Eq. (4.9), the total energy consumption depends on the energy

consumption of the transceiver when it is either transmitting, receiving, or in idle mode. Wi-Fi

idle listening and receiving are the dominant modes of energy consumption under light and

moderate traffic conditions [54][104]. Ptx and Prx are calculated by adding to Pidle a value that

is a hundred times smaller, which means that Ptx, Prx, and Pidle are approximately equal [105].

This implies that the energy consumption of the network is mainly dependent on the time that

each node spends in one of the three Wi-Fi modes mentioned. Simplifying Eq. (4.9) we get:

E = Pidle · ttotal (4.10)

From Eq. (4.10) we can verify that the energy consumed by the network depends on the

total time ttotal spent by all WVSs in transmitting or receiving information, in idle mode waiting

for information, or in overhearing packets. Thus, our analysis is focused on the amount of time

WVSs spend in each mode. Given the power consumption Pidle provided by the manufacturer

of the Wi-Fi radio and the LPR, we simply have to estimate the total time ttotal for which the

WVSs have their Wi-Fi radio ON. We consider that there are no collisions since there is a

single packet flowing in the network at each instant of time. The analysis was first conducted

for three regular topologies (chain, binary tree, and grid) [10] and then generalised for random

topologies [102]. For each topology, we studied two solutions: PACE and GREENNESS. For

PACE the time is calculated assuming that all WVSs are with their Wi-Fi radios switched ON

during the transmission of all packets, as depicted in Fig. 4.1. For GREENNESS we performed

a similar analysis based on the NSM mechanism presented in Chapter 3.

4.2.1 Chain Topology

Fig. 4.1 depicts a WVSN with a chain topology. The gateway is represented in white (GW )

and the other nodes are WVSs that send data to the gateway. To simplify the analysis, we


assume that each WVS sends a packet to the gateway. Also, as described before, we focus on

the calculation of the total amount of time that all nodes spend in transmitting a single packet

to the gateway, assuming that T is the time required for a node to transmit a packet to its

neighbour and receive the IEEE 802.11 ACK. If IEEE 802.11e Enhanced Distributed Channel

Access (EDCA) is used, the frames with QosNoAck are not acknowledged. In that case, T

would simply be the time for a WVS to transmit a frame, thus saving the time required for the

acknowledgements and possible retransmissions. Herein, we assumed the worst-case scenario

where all frames are acknowledged. Fig. 4.1 presents a sequence diagram where the black

squares are time slots used by the nodes to transmit or receive frames, and the white squares

are time slots where the Wi-Fi radio interface of the node is switched OFF. This means that the

total amount of time that the Wi-Fi radio interfaces are ON for the network is given by the total

count of black squares in the sequence diagram for each solution.

For PACE, the time required to have all nodes sending a packet to the gateway is given by

the sum of the times each node needs to transmit or relay a packet from a neighbour. Fig. 4.1

shows that WV S#4 needs 4 time slots for the packet to be successfully delivered to the GW .

This means that WV S#4 needs 4 ·T , WV S#3 needs 3 ·T , WV S#3 needs 2 ·T , and WV S#1 needs

T . In the end, we get 4 ·T +3 ·T +2 ·T +T . This can be generalised to ∑hi=1 i ·T . So the time

required for all nodes to transmit a packet to the gateway is given by Eq. (4.11):

tPACE = N ·D

∑i=1

i ·T (4.11)

where D is the number of hops in the chain and N is the total number of nodes in the WVSN,

including the gateway. For instance, WV S#4 needs 4 time slots to transmit a packet to the

gateway, but at the same time all five WVSs are with the Wi-Fi radio ON, so the total time

spent is given by 4 · 5 · T . This applies for the other WVSs. The sum ∑Di=1 i · T is equal to

D·(D+1)2 ·T . As such, Eq. (4.11) can be translated into Eq. (4.12):

tPACE =D · (D+1)

2T ·N (4.12)

For GREENNESS, Fig. 4.1 shows that WV S#4 needs 4 time slots for the packet to be

successfully delivered to the GW . After that transmission, the Wi-Fi radio of WV S#4 is

switched OFF. This means that WV S#4 needs 4 ·T to transmit a packet to the GW . The other

nodes (WV S#3,WV S#2,WV S#1) relay the information to the gateway which means that the


12345678910

PACE

Tn5T

n5Tn5T

n5

Tn3T

n3

Tn4T

n4Tn4

Tn2

12345678910

GREENNESS

Tn5T

n5Tn5

Tn4T

n4

Tn4

Tn3T

n3Tn2

WVS#1

WVS#2

WVS#3

WVS#4

Tn5

GW

hop 1 hop 2 hop 3 hop 4

Figure 4.1: Time Sequence Diagram for Chain Topology

total time required to transmit the packet is 4 ·4 ·T . For WV S#3 the total time is 3 ·3 ·T . This

sequence can be represented by the sum ∑Di=1 i2 ·T and the total time is given by:

tGREENNESS =D

∑i=1

i2 ·T +D

∑i=1

i ·T, (4.13)

The second sum ∑Di=1 i · T refers to the number of time slots used by the gateway, GW .

Eq. (4.13) can be translated into Eq. (4.14):

tGREENNESS =D · (D+1) · (D+2)

3T (4.14)

The main difference between tPACE and tGREENNESS is that, instead of having all nodes

active during a full network transmission cycle, each WVS’s Wi-Fi radio is switched OFF

when they are not needed to transmit, receive, or relay data.


4.2.2 Binary Tree Topology

We consider a perfect binary tree where every node other than the leaves has two children,

as depicted in Fig. 4.2a. The number of nodes in a perfect binary tree is given by Eq. (4.15):

N = 2D+1−1, (4.15)

where D is the depth of the binary tree, i.e., the length from the gateway (white node) to the

deepest node in the tree (any leaf, e.g., WV S#6). For the binary tree presented in Fig. 4.2a,

D = 2 and N = 7. For perfect binary trees, the number of hops between the gateway and any

leaf is always the same and given by:

D = log2(N +1)−1. (4.16)

The number of leaf nodes is given by

Nleaves = 2D, (4.17)

For GREENNESS, each leaf creates a path to the gateway. So, in total, Nleaves paths are

created. As before, if we consider that there will be no collisions in the network, the analysis

performed for the chain topology can be reused. The main difference is that a binary tree will

have Nleaves different paths. For a binary tree topology, tPACE and tGREENNESS can be calculated

by multiplying the second term of the previous equations (Eq. (4.12) and Eq. (4.14)) by the total

number of paths. For tPACE , at each time interval T , all nodes are active; so, the contribution of

N nodes needs to be considered in Eq. (4.18).

tPACE = 2D · D · (D+1)2

N ·T −Fcor ·N ·T, (4.18)

If we consider only the first term in Eq. (4.18), the nodes acting as relays for different

chains will be incorrectly considered to transmit multiple local packets, since we are simply

multiplying the second term of the chain equations previously derived by Nleaves. Eq. (4.19)

defines Fcor, the second term in Eq. (4.18) that is used to cancel the additional transmissions of

each relay node considered in the first term.

Fcor =D

∑j=2

j−1

∑i=1

12·2 j · i (4.19)


For GREENNESS, the total network time is obtained by multiplying the second term of

Eq. (4.14) by the number of paths Nleaves and by applying Fcor, the correction factor.

tGREENNESS = 2D · D · (D+1) · (D+2)3

·T −FcorT (4.20)

Eq. (4.19) represents the second term in Eq. (4.20) to cancel the transmissions of each relay

node considered in the first term.

GW

WVS#1

WVS#4

WVS#3

WVS#2

WVS#6

WVS#5

(a) Binary tree topology

GW WVS#2

WVS#1

WVS#3

WVS#5

WVS#4

WVS#8

WVS#6

WVS#7

(b) Grid topology

Figure 4.2: Regular network topologies

4.2.3 Grid Topology

For a square grid network topology, the total number of nodes is given by C×C, where C

is the number of WVSs in one side of the grid. A grid network topology example is presented

in Fig. 4.2b with C = 3, and the total number of WVSs in the network given by Eq. (4.21) (with

N = 9 for the example).

N =C ·C, (4.21)

Since the WiFIX routing protocol uses hop count as the routing metric and creates a tree

rooted at the gateway, the number of hops between the leaves and the gateway is equal for all

paths for grid network topologies. The number of leaf nodes, Nleaves can be easily found to be

given by Eq. (4.22).

Nleaves = 2 ·√

N−1 = 2 · (D+1)−1 = 2D+1, (4.22)


where N is the total number of nodes in the grid network topology and D = C− 1 is the

tree depth, i.e., the average number of hops for all possible paths between leaf nodes and

the gateway. For the example depicted in Fig. 4.2b, the number of leaf nodes is calculated

using Eq. (4.22); for this example the number of leaves is 5. As before, a grid topology has

Nleaves different paths. tPACE and tGREENNESS can be calculated multiplying the second term

of Eq. (4.12) and Eq. (4.14), respectively by the last term of Eq. (4.22). In Eq. (4.23) tPACE

is defined for a grid topology, the term Fcor cancels the effect referred to the binary network

topology.

tPACE = (2D+1)D · (D+1)

2N ·T −N ·T Fcor, (4.23)

The Fcor for a grid topology can be demonstrated to be defined by Eq. (4.24).

Fcor =D−1

∑j=1

j

∑i=1

2 · i (4.24)

For GREENNESS we get Eq. (4.25).

tGREENNESS = (2D+1) · D · (D+1) · (D+2)3

·T −Fcor ·T (4.25)

4.2.4 Random Network Topologies

An example of a random network topology is shown in Fig. 4.3. Although it is similar to

a binary tree topology, we no longer get a constant tree depth, D. In order to generalise the

previous formulas for regular network topologies we must consider a variable tree depth that

can change from branch to branch. Using this rationale the previous chain network topology

equations (Eq. (4.12) and Eq. (4.14)) can be used in a sum considering the different tree depths

per branch.

The total time required for a given WVS in a WVSN with N nodes to transmit one packet

to the gateway when using PACE is defined in Eq. (4.26).

tPACE =Nleaves

∑j=1

h j · (h j +1)2

N ·T −Fcor ·N ·T (4.26)

Fcor =N−1

∑i=1

hi ·(Nleaves

∑j=1

xi j−1)

(4.27)


GW

WVS#1

WVS#4

WVS#3

WVS#2

WVS#6

WVS#5

Figure 4.3: Random Network Topology

xi j =

1 =⇒ i ∈ B j

0 =⇒ i /∈ B j

(4.28)

In Eq. (4.26), h j is the number of hops for branch j, Nleaves is the number of branches of the

tree defining the WVSN active topology, N is the number of WVSs including the gateway, T is

the time for a WVS to transmit a frame and receive the corresponding acknowledgement from

its parent, and Fcor is the correction factor to cancel the transmission time of each relay WVS.

Fcor varies with the WVSN topology and is given by Eq. (4.27) and Eq. (4.28), which counts the

number of times a WVS is included in vectors B j, where j ∈ 1, ...,Nleaves. A simple example

can be given using Fig. 4.3. WV S#1 is included in the branches of WV S#3 and WV S#4. For

this example, Fcor equals 1, i.e., the number of repeated WVSs for the aforementioned branch.

Equation Eq. (4.29) defines the total time required for all WVSs to transmit one packet to

the gateway when GREENNESS is considered. Fcor can be calculated using Eq. (4.27) and

Eq. (4.28).

tGREENNESS =Nleaves

∑j=1

h j · (h j +1) · (h j +2)3

·T −Fcor ·T (4.29)

The total time required for all WVSs to transmit one packet to the gateway can now be used

to estimate the energy consumption of the WVSN for PACE and GREENNESS.


4.2.5 Numerical Analysis of Energy Consumption for PACE and GREENNESS

The total energy consumed by all the WVSs in the network for PACE and GREENNESS is

derived from Eq. (4.10) and can be calculated by using Eq. (4.30) and Eq. (4.31), respectively:

EPACE = Pidle · tPACE (4.30)

EGREENNESS =Pidle · tGREENNESS +ELPR +PWiFisleep · tWiFisleep(4.31)

In Eq. (4.31) ELPR is the energy consumption of the LPR, tWiFisleep is the time Wi-Fi radios

in the WVSs are in sleep mode and PWiFisleep is the power consumption of a Wi-Fi radio in sleep

mode. With the inclusion of the PWiFisleep and tWiFisleep we are considering the case where it is

not possible to switch OFF the Wi-Fi radio, and we switch it to sleep mode. In the example of

Fig. 4.1 the total number of white slots would represent tWiFisleep , which is obtained by tPACE −tGREENNESS. The total energy spent by the LPRs ELPR can be calculated using Eq. (4.32).

ELPR = PLPRrx · tLPRrx +PLPRtx · tLPRtx, (4.32)

where PLPRrx is the LPR power consumption for each receiver installed in the WVSs, PLPRtx is

the power of the transmitter at the gateway; tLPRtx together with tLPRrx represent the amount of

time that all the nodes spend receiving and transmitting the LPR signal, respectively. Assuming

that the LPR transmitter power equals the LPR receiver power, we can simplify the equation

to:

ELPR = PLPR · (tLPRrx + tLPRtx) (4.33)

The time that the network spends transmitting and receiving the LPR signal can be

estimated by observing Fig. 4.1. During the 10 time-slots, four nodes will be receiving the

LPR signal while the gateway transmits the LPR signal. This corresponds to the total time that

PACE needs for all nodes to transmit a frame to the gateway. Taking this into account, we can

derive Eq. (4.34).

EGREENNESS = Pidle · tGREENNESS +PLPR · tPACE +PWiFisleep · (tPACE − tGREENNESS) (4.34)


The energy saving enabled by GREENNESS can then be calculated using Eq. (4.35):

Esaving =(

1− EGREENNESS

EPACE

)×100 (%) (4.35)

By substituting in Eq. (4.35) EGREENNESS and EPACE respectively by Eq. (4.30) and Eq. (4.34)

we get:

Esaving =(

1−Pidle · tGREENNESS +PLPR · tPACE +PWiFisleep · tWiFisleep

Pidle · tPACE

)×100 (%) (4.36)

If we simplify the equation we get:

Esaving =

(1− tGREENNESS

tPACE− PLPR

Pidle−(

1− tGREENNESS

tPACE

)·

PWiFisleep

Pidle

)×100 (%) (4.37)

Esaving =

((1− tGREENNESS

tPACE

)·(

1−PWiFisleep

Pidle

)− PLPR

Pidle

)×100 (%) (4.38)

The ratio tGREENNESStPACE

can be expanded to Eq. (4.39) and simplified into Eq. (4.40).

tGREENNESS

tPACE=

∑Nleavesj=1

h j·(h j+1)·(h j+2)3 ·T −Fcor ·T

∑Nleavesi=1

h j·(h j+1)2 N ·T −Fcor ·N ·T

(4.39)

tGREENNESS

tPACE=

23 ·N

·∑

Nleavesj=1 h j · (h j +1) · (h j +2)−3 ·Fcor

∑Nleavesi=1 h j · (h j +1)−2 ·Fcor

(4.40)

From Eq. (4.40) we conclude that the energy saving does not depend on the variable T ,

which is associated with the Wi-Fi channel data rate. The ratiosPWiFisleep

Pidleand PLPR

Pidleare constants

characteristic of the Wi-Fi interface and the LPR selected. Furthermore, if we consider that the

Wi-Fi radios can be shutdown, Eq. (4.38) can be further simplified, only considering the ratios,tGREENNESS

tPACEand PLPR

Pidle. In order to save energy PLPR must be smaller than Pidle, as already referred

to in the LPR requirements. This ratio is further studied in Section 4.6 to understand exactly

how much smaller PLPR should be.

4.3 Simulation Setup 83

4.3 Simulation Setup

The implementation of GREENNESS in ns-3 (version 3.24.1) required three steps: 1)

implementation of the WATCM mechanism, which is used to calculate the polling vector; 2)

implementation of the NSM mechanism; 3) generation of 14,000 random wireless multi-hop

network topologies with 10, 20, and 30 WVSs. Alg. 1 defined in Chapter 3 was used to build

the node polling vector. From the 14,000 random wireless multi-hop networks, 8,874 where in

fact used, as only for these topologies all the WVSs could reach the gateway, either directly or

through a relay WVS. Furthermore, we needed to simulate this amount of WVSN topologies

in order to guarantee at least 60 active tree topologies rooted at the gateway for each of the 12

different tree depths simulated. An example of a network with 30 nodes randomly positioned

in a 500 m per 500 m space is illustrated in Fig. 4.4. MAC and IP addresses were added to

the ARP cache table of each node in order to avoid ARP requests during the registration phase.

The simulation parameters used are summarized in Table 4.2.

Table 4.2: Simulation parameters.

Simulation Variable ValueNo. of runs per test 30

Area where nodes are placed 500 m x 500 mRxGain 5 dB

TxPowerStart 16 dBmTxPowerEnd 16 dBm

Fragmentation DisabledPacket payload Size 1,200 bytes

RTS/CTS DisabledMobility model None

Propagation Loss model Friis Propagation ModelPropagation Delay model Constant Speed ModelCommunications standard IEEE 802.11b

Data Rate 11 Mbps

The power values for each technology are shown in Table 4.3. For the low power radio, we

used as reference the CC1200 transceiver that implements 802.15.4g with a maximum power

consumption of 57 mW. The Intel Wi-Fi, Link 5300 a/b/g/n wireless network adapter was

considered, which has a power consumption of 1.45 W in idle mode and 100 mW in sleep

mode [106]. The energy results for GREENNESS and PACE were obtained by using the ns-3

energy models. For the simulation, the traffic load was emulated as a raw H.264 video stream,


transmitted at a constant bit rate. The length of each packet was 1280 bytes, including 1200

bytes payload, 28 bytes IP and UDP header and 52 bytes MAC header. The transmission bit

rate for the WVSN was increased in steps of 450 kbps to increase the offered load and maintain

the same value for different network sizes.

Table 4.3: Values of the parameters PLPR, Pidle, and PWiFisleep considered in the numerical andsimulations analysis.

Constant ValuePLPR 57 mWPidle 1.45 W

PWiFisleep 100 mW

For the numerical analysis, the values of T , tLPR, Nleaves, Fcor and h j in Eq. (4.26) and

Eq. (4.29) were obtained from simulation for each random multi-hop topology. Besides, the

average number of hops was calculated for each multi-hop topology by averaging the depth of

each branch of the active tree topology rooted at the gateway.

Figure 4.4: Network simulation with 30 WVSs randomly positioned in a 500 m x 500 m areawith the gateway identified with GW.

4.4 Experimental Setup 85

4.4 Experimental Setup

The wireless video sensing scenario was implemented using Raspberry Pi (RPi) model B,

equipped with cameras and sending the video streams to the gateway using the Wi-Fi dongle

TP-LINK TL-WN823N. The RPi is an affordable platform adequate to be used as a WVS.

Besides being possible to add a camera, the two USB 2.0 ports together with the 28 General

Purpose Input Outputs (GPIOs) pins make it possible to add a Wi-Fi interface and an LPR,

respectively. As LPR, we selected Radio Data System (RDS) since it is widely available and

affordable, with the possibility to control sensors in a 100 km range. In the bootstrap process,

the gateway and WVSs were configured in the same ad-hoc network. After the nodes were

connected to the ad-hoc network, the WiFIX routing protocol was used to establish an active

tree topology rooted at the gateway, as explained in Section 3.1.

The assembled testbed was composed of a gateway and six WVSs (see Fig. 4.5), operating

in Wi-Fi ad-hoc mode. The goal of the experiments was to evaluate the performance and energy

consumption of the GREENNESS solution for three regular WVSN topologies. Afterwards,

its energy saving and performance were compared to the ones achieved by PACE in the same

scenario. The Iperf tool was used to generate traffic in each WVS at a constant bit rate and

provide statistics about the network performance. The experiments consisted in generating

traffic to the gateway simultaneously from all video sensors. Each sensor sent its data during

60 s. At the end of each experiment, the Iperf server calculated the statistics for each sensor

and saved them in a log file. Each experiment was repeated 8 times for the three topologies

shown in Fig. 4.6. Using the Student’s t-distribution, due to the small sizes of the data sets,

95 % confidence intervals were calculated. The following sections describe how the gateway

and WVSs were designed and developed, so that it is possible to control the data transmission

and turn ON/OFF the Wi-Fi interfaces using an LPR in a wireless video sensing scenario.

4.4.1 Gateway

The main gateway functions are: register each WVS, discover the network topology, and

poll each WVSs to transmit/receive a packet using the RDS control channel. The registration

starts when WVSs sends a Registration message (as shown in Fig. 3.3a) to the gateway

triggering the start of WATCM. As mentioned before, a nodeId is assigned to each WVS,

and the gateway replies to each WVSs with a Registration Acknowledgement message that

includes the nodeId. Using the registration messages, WATCM (as presented in Alg. 1) can


Figure 4.5: Testbed with a gateway and six WVSs using a Raspberry Pi model B boards asbasis for the GREENNESS proof-of-concept prototype.

now discover the network topology and create the polling vector of the WVSs. Next, the NSM

can start sending Poll messages using LPR. The RDS control channel was developed using the

RPi hardware and a program named Pi-FM-RDS, available in [107]. NSM communicates with

the Pi-FM-RDS software using Inter-Process Communication (IPC), passing as a parameter the

nodeId and a flag indicating whether the gateway has data to be transmitted to the WVSs. This

third party program turns the RPi into an RDS transmitter using a Pulse-Width Modulation

(PWM) generator to produce a Very High Frequency (VHF) signal, which is emitted on GPIO

4. To increase the transmitter’s range, a simple antenna was built with a 112 cm electrical

wire to couple with the Raspberry Pi’s GPIO 4. The Pi-FM-RDS software uses the RPi Direct

Memory Access (DMA) to assure that the samples used by the PWM generator are sent at

a precise bit rate of 1187.5 bit/s±0.125 bit/s, as required by the RDS standard [88]. The

software periodically wakes the DMA engine to replenish the DMA buffer. We have modified

the Pi-FM-RDS software to increase the polling speed since originally the samples were passed

to the DMA engine each 5 ms. We modified the Pi-FM-RDS software to decrease from 5 ms

period to 149 µs allowing the generation of a different RDS Poll message every 154 ms. As

specified by the standard [88], RDS consists of a constant and preferably, uninterrupted data

stream, generating messages with a period of ≈ 87.58ms. So, Pi-FM-RDS does not inform

the algorithm when an RDS message starts and stops, the software only assures a fixed guard

interval of 154 ms for the generation of a full RDS message.


GW

WVS#1

WVS#4

WVS#3

WVS#2

WVS#6

WVS#5

(a) Binary Tree Topology.

GW

WVS#1

WVS#5

WVS#3

WVS#2

WVS#6

WVS#4

(b) A balanced tree topology with two branchesand three hops.

GW

WVS#1

WVS#4

WVS#3

WVS#2

WVS#6

WVS#5

(c) Unbalanced tree topology.

Figure 4.6: The three regular WVSN topologies used to evaluate GREENNESS.

The RDS message format generated by Pi-FM-RDS is shown in Fig. 4.7. It is composed

of 4 blocks, with a total of 104 bits, each one with a 2 Bytes information word. Each block

includes a 10 bits offset word and a check-word used for block synchronisation and for error-

correction and detection, respectively. In GREENNESS we used the information word to carry

the 8 bits nodeId and 1 bit for the hasData flag. These fields are repeated across the 4 RDS

blocks to ensure information redundancy so that the RDS receivers can recover the information

from at least one of the blocks.

As soon as the NSM receives a packet from the nodeId, it will fetch the next nodeId

from vector V and send it to the Pi-FM-RDS software together with the indication whether it

hasData from the gateway to that nodeId. Additionally, after the NSM warm-up phase ends,

the gateway will replay the same traffic pattern by sending a nodeId through the RDS message


Block 1 Block 2 Block 3 Block 4

Information word Checkword + offset word

m15

m14

m13

m12

m11

m10

m6

m7

m5

m8

m9

m4

m0

m1

m2

m3

m6

m7

m5

m8

m9

m4

m0

m1

m2

m3

c’6

c’7

c’5

c’8

c’9

c’4

c’0

c’1

c’2

c’3

Block = 26 bits

Group = 4 blocks = 104 bits

Information word = 16 bits Checkword = 10 bits

nodeIdUnused flag

nodeId = 8 bitsUnused = 7 bits

hasData flag = 1 bit

Figure 4.7: GREENNESS modification to the RDS message.

accordingly.

4.4.2 WVS

The algorithms running in the WVS are responsible for registering the WVS at the gateway

and receive a nodeId, listen to the RDS control messages and switch ON/OFF the Wi-

Fi interface accordingly. First, the WVS runs WATCM and registers its MAC address in

the gateway. The gateway sends in the Registration Acknowledgement message the nodeId

assigned to the WVS. All nodes in the path between the current WVS and the gateway snoop

the Registration Acknowledgement message and store the nodeId in vector R. When all WVSs

are registered, NSM can start running by switching OFF the WVSs interfaces and listening to

RDS messages. Since the RPi does not possess an RDS receiver we have selected an affordable

evaluation board, Si4703 (shown in Fig. 4.8a) that is connected using SDIO and SCLK for

I2C communication to the RPi. To improve the RDS reception, we added a set of telescopic

antennas connected to Si4703 through a 3:5 mm audio jack connector and with a length of

245 mm. The WVS RDS receiver is composed of the Si4703 evaluation board and the RdSpi

software, which was modified to parse uninterruptedly the RDS data stream. Furthermore, the

RdSpi software was modified to retrieve the nodeId and the flag hasData from one of the 4

RDS blocks sent by the gateway. Only one valid pair of informations is enough, so when the

RdSpi software detects a different nodeId, it sends this information to NSM using IPC. When

the Poll message carries the nodeId assigned to the current WVS or it belongs to vector R,

the Wi-Fi interface is turned ON. If the hasData flag is set the WVS waits for data from the


gateway. When a packet from the gateway is received or the timeout TWV STout occurs, the WVS

will send its packet with the current queue size. The WVS queue size is used by the gateway

as a safety mechanism in case the rate changes. If the Poll message nodeId neither belongs to

the vector R nor equals the nodeId assigned to the WVS, the Wi-Fi radio of the current WVS is

turned OFF. Moreover, we have included the possibility to switch ON all Wi-Fi radios from the

WVSN, by sending a Poll message with nodeId 0. This Poll message is used by the gateway

to return the WVSs Wi-Fi radios to normal operation, allowing to refresh the topology and

register new WVSs.

(a) Si4703 FM Tuner Evaluation board. (b) Schematic showing the connection betweenRPi and the Si4703.

Figure 4.8: FM Tuner and connection with RPi.

GREENNESS requires that the change between states occurs within milliseconds, in order

to assure at least the same performance of in-band control, using Wi-Fi [13]. However, the RPi

takes about 4 s to switch ON and to start sending packets. This problem is related to the time

that is required to reconnect to the ad-hoc network. In the Linux Kernel, the mac80211 module

contains the MAC Sublayer Management Entity (MLME) where the MAC state machines are

included. In this implementation, the process of associating a device to an ad-hoc network

first checks if the provided Extended Service Set Identification (ESSID) already exists by

scanning and listening for beacons during a period. In order to decrease the association time

we removed this scanning period, but still, it took at least 1 s for the WVS to be able to transmit

packets. Since we were unable to switch OFF the Wi-Fi interface effectively, we studied the

possibility to switch it to sleep mode. In [55] a study of the state transition times for the


Wi-Fi card Atheros AR9280 is presented. The conducted experiments demonstrate that this

Wi-Fi card requires 250 µs to get ready to transmit/receive after returning from sleep mode. A

faster switch between sleep and transmit/receive states is possible since the association with

the ad-hoc network is maintained when changing from idle to sleep state. Based on this

assumption we have studied the possibility to adapt the IEEE 802.11 Power Saving Mode

(PSM), substituting the Ad-hoc Traffic Indication Map (ATIM) frames mechanism by our out-

of-band control channel to schedule the sleep periods. To succeed in this development, we

researched the possibility to force the Wi-Fi interface to enter sleep mode by issuing commands

implemented by the driver or by adapting the Mesh power save mechanism [108], performed

in the mac80211 module. Unfortunately, the drivers of the three chipsets available (RT5370,

RTL8192CU and AR9271) did not support the PSM in Independent Basic Service Set Identifier

(IBSS) mode. Another alternative was to use open80211s instead of WiFIX. open80211s is an

open-source implementation of the IEEE 802.11s wireless mesh standard supporting the Mesh

power save mechanism [108]. To bypass the implemented power saving mechanism would

require a significant implementation effort, so we decided not to go forward. To overcome

these implementation shortcomings, we have decided to profile the Wi-Fi card consumption

in different operation modes, estimate the time the Wi-Fi card was in each state during our

experimentation, and then calculate the total energy consumption. A multimeter was used to

measure the power consumption of the RPi when connected to the Wi-Fi card and the Si4703

board. Using this setup we were able to profile the power consumption of the Wi-Fi card and

the Si4703 board in their different modes: transmitting, receiving, and idle. The measured

values are presented in Table 4.4.

Table 4.4: Values of the measured parameters PLPR and Pidle.

Constant ValuePLPR 106 mWPidle 750 mW

4.5 Evaluation

In this section we present the numerical, simulation, and experimental results. We first

validate the numerical analysis using simulations and characterise the impact of different LPR

power consumptions. Furthermore, simulations are used to assess GREENNESS performance

and compare with experimental results.

4.5 Evaluation 91

4.5.1 Numerical and Simulation Results

The evaluation of GREENNESS and PACE is presented in this section. Fig. 4.9 shows the

energy saving for WVSNs with 10, 20, and 30 nodes calculated using Eq. (4.35). For each

topology, the average hop count and the energy saving is computed. The plots in Fig. 4.9 show

that the numerical and simulation curves are almost coincident for a confidence interval of

95 %, thus validating our numerical analysis. For N = 10, the energy saving ranges between

45 % and 65 %, while for N = 30 it ranges between 79 % and 85 %. So, the energy saving

increases as the network size increases. When the average number of hops rises, the energy

saving decreases since WVSs need more time to transmit a packet to the gateway as it is relayed

by more WVSs. Nevertheless, for higher network sizes, the gradient of the energy saving curve

is almost zero. This happens because the number of WVSs that are switched OFF tends to be

higher than the number of WVSs switched ON in a given moment.

0

0.2

0.4

0.6

0.8

1

2 2.5 3 3.5 4

Energ

y S

avin

g

Average number of hops

ESaving numericalESaving simulation

(a) N = 10

0

0.2

0.4

0.6

0.8

1

2 2.5 3 3.5 4

Energ

y S

avin

g



(b) N = 20

0

0.2

0.4

0.6

0.8

1

2 2.5 3 3.5 4

Energ

y S

avin

g



(c) N = 30

Figure 4.9: Energy saving achieved by GREENNESS with respect to PACE for WVSNs withdifferent sizes and average number of hops.


0

0.2

0.4

0.6

0.8

1

0.001 0.01 0.1 1

Energ

y S

avin

g

PLPR / Pidle

N=10N=20N=30

(a) Energy saving achieved by GREENNESS withrespect to PACE when PLPR is varied.

0

0.2

0.4

0.6

0.8

1

10 20 30

Energ

y S

avin

g

WVSN Size

no sleepsleep

(b) Considering that Wi-Fi radios are put in sleepmode instead of switched OFF.

Figure 4.10: Energy saving exhibited by GREENNESS when varying PLPR and consideringWi-Fi radios in sleep mode.

In order to analyse the way the energy saving achieved by GREENNESS changes with

PLPR, we have performed a sensitivity analysis using Eq. (4.30), Eq. (4.31), and Eq. (4.35).

From the curves presented in Fig. 4.10a, we can conclude that the energy saving is almost

constant for LPRs consuming less than 10 % of Pidle. Since the power consumption of candidate

LPRs is below 4 % of Pidle, the energy saving achieved for all the candidate LPRs referred to in

Chapter 3 will be similar. When designing the network, the user needs to perform a trade-off

between network coverage, which means selecting a radio with higher energy consumption to

achieve greater distances and energy saving.

For the case when the Wi-Fi radio of WVSs is switched to sleep mode instead of switched

OFF, from Fig. 4.10b we can conclude that for N = 30, the energy saving slightly reduces from

85 % to 79 %. The other WVSNs sizes also suffer a similar reduction in the energy saving.

Even when it is not possible to switch OFF the Wi-Fi radio of the WVSs, GREENNESS can

achieve significant energy savings. This was expected since PWiFisleep represents 7 % of Pidle

and thus the impact in the energy saving is negligible.

The simulation results in Fig. 4.11 show that GREENNESS can achieve energy saving up to

92 % for random topologies with 10, 20, and 30 nodes. For low offered loads, since the WVSs

do not need to transmit data so often, Wi-Fi radios can be switched OFF more time, thus saving

more energy. For offered loads higher than 3 Mbit/s, the energy saving is constant, being equal

to 85 % for N = 30. For bigger WVSNs sizes, the gradient of energy saving curve is small

because, as in Fig. 4.9, the number of WVSs that are switched OFF tends to be higher than the

number of WVSs switched ON in a given moment. For smaller WVSNs (e.g., N = 10), the

traffic-aware mechanism causes an increase of 20 % in the energy saving for low offered loads.

4.5 Evaluation 93

0

0.2

0.4

0.6

0.8

1

0 1 2 3 4 5 6

Energ

y S

avin

g

Offered Network Load (Mbps)

N=10N=20N=30

Figure 4.11: Impact of different offered network loads and WVSNs sizes in the energy savingof GREENNESS.

As explained before, NSM does not send a Poll message when the WVS does not have data to

be sent. Otherwise, NSM running in the gateway waits for a timeout TGWTout to send a new Poll

message. By not having to wait TGWTout for small WVSNs sizes, GREENNESS increases the

energy saving.

0

0.5

1

1.5

2

2.5

3

3.5

4

0 1 2 3 4 5 6 0

0.2

0.4

0.6

0.8

1

Netw

ork

Capaci

ty (

Mbps)

Jain

's Index


CSMA Network CapacityGREENNESS Network Capacity

CSMA Jain's IndexGREENNESS Jain's Index

(a) N = 10

0

0.5

1

1.5

2

2.5

3

3.5

4

0 1 2 3 4 5 6 0

0.2

0.4

0.6

0.8

1

Netw

ork

Capaci

ty (

Mbps)

Jain

's Index




(b) N = 20

0

0.5

1

1.5

2

2.5

3

3.5

4

0 1 2 3 4 5 6 0

0.2

0.4

0.6

0.8

1

Netw

ork

Capaci

ty (

Mbps)

Jain

's Index




(c) N = 30

Figure 4.12: Network Capacity of GREENNESS and CSMA/CA for different offered networkloads with average number of hops equal to 2.


0

0.5

1

1.5

2

2.5

3

3.5

4

0 1 2 3 4 5 6

Ave

rage O

ne-w

ay-

dela

y (s

)


CSMAGREENNESS

(a) N = 10

0

0.5

1

1.5

2

2.5

3

3.5

4

0 1 2 3 4 5 6

Ave

rage O

ne-w

ay-

dela

y (s

)


CSMAGREENNESS

(b) N = 20

0

0.5

1

1.5

2

2.5

3

3.5

4

0 1 2 3 4 5 6

Ave

rage O

ne-w

ay-

dela

y (s

)


CSMAGREENNESS

(c) N = 30

Figure 4.13: One-way-delay of GREENNESS and CSMA/CA for different offered networkloads with an average number of hops equal to 2.

Fig. 4.12 presents the Network Capacity achieved by GREENNESS and a CSMA/CA-

based solution for 30 random network topologies with and average hop count of 2 hops. As

expected, GREENNESS network throughput capacity is constant when saturation is reached;

also, the saturation point is reached for offered loads greater than those obtained when CS-

MA/CA is used. Since in GREENNESS each node has a time slot to transmit information, the

Jain’s index, which measures the level of fairness, is constant. CSMA/CA fairness decreases

when network maximum throughput reaches the limit, meaning that nodes closer to the gateway

will have more access to the medium than the others.

For the OWD metric presented in Fig. 4.13, GREENNESS increases the OWD up to 3 s

for higher offered network loads, but CSMA/CA is less than 0.5 s. CSMA/CA outperforms

GREENNESS but with an high PLR, as it can be observed in Fig. 4.14. The GREENNESS

high OWD still can offer good video quality with a constant delay.

The PLR in Fig. 4.14 shows that GREENNESS assures the delivery of every packet, but

CSMA/CA starts to lose packets at an offered load of 1 Mbit/s. Since for GREENNESS the

4.5 Evaluation 95

0

0.2

0.4

0.6

0.8

1

0 1 2 3 4 5 6

Ave

rage P

ack

et

Loss

Rati

o


CSMAGREENNESS

(a) N = 10

0

0.2

0.4

0.6

0.8

1

0 1 2 3 4 5 6

Ave

rage P

ack

et

Loss

Rati

o


CSMAGREENNESS

(b) N = 20

0

0.2

0.4

0.6

0.8

1

0 1 2 3 4 5 6

Ave

rage P

ack

et

Loss

Rati

o


CSMAGREENNESS

(c) N = 30

Figure 4.14: Packet Loss Ratio for GREENNESS and CSMA/CA for different offered networkloads and average number of hops equal to 2.

PLR is 0 for the different offered network loads, the curve is displayed on top of the x-axis.

For higher WVSN sizes the PLR also increases for CSMA/CA based solutions.

GREENNESS not only can achieve energy saving up to 92 % for low offered network loads

but it also maintains the same level of performance and fairness for higher offered network

loads.

4.5.2 Experimental Results

GREENNESS was also evaluated based on the results obtained from experimental tests.

For the three regular topologies shown in Fig. 4.6, GREENNESS was compared with PACE.

Fig. 4.15 compares the simulation and experimental results regarding the total energy spent

when each node transmits a video stream of 350 kbit/s to the gateway. In fact, for Wi-Fi

interfaces in ad-hoc mode at a rate of 11 Mbit/s, 350 kbit/s is enough to saturate the network in

CSMA/CA. The power consumption of the Wi-Fi dongle in idle state (750mW) and of the FM

tuner (106mW) considered in the simulations were measured experimentally. GREENNESS


can achieve energy savings up to 53 % for a binary tree topology. The numerical and simulation

results are coincident for a confidence interval of 95 %.

0

20

40

60

80

100

GREENNESS exp.

GREENNESS sim.

GREENNESS num.

PACEGREENNESS exp.

GREENNESS sim.

GREENNESS num.

PACEGREENNESS exp.

GREENNESS sim.

GREENNESS num.

PACE

Energ

y (

mW

h)

Wi-FiFM-RDS

Unbalanced2 Branches 3 hopsBinary

Figure 4.15: Energy consumption of GREENNESS and PACE considering testbed, simulation,and numerical evaluations for three scenarios

0

0.5

1

1.5

2

2.5

3

3.5

4

Binary Tree 2 Branches with 3 Hops

Unbalanced Tree

Netw

ork

Capaci

ty (

Mbps)

GREENNESS testGREENNESS Simulation

PACE test

Figure 4.16: Network capacity achieved by GREENNESS and PACE during testbedexperiments and simulations for the three scenarios.

Fig. 4.16 shows the network capacity results obtained for GREENNESS and PACE. GREEN-

NESS always achieves a slightly better network capacity than PACE. Since GREENNESS

proof-of-concept was modified to allow the transmission of more than one packet on each

polling interval when using the RDS LPR, NSM assigns equal time slots to all nodes. As a

consequence, nodes closer to the gateway can transmit more information than leaf nodes. In

PACE, every node receives exactly one polling opportunity per round, and all nodes are given

4.6 Discussion 97

the opportunity to send one packet. GREENNESS guarantees equal access to the medium

to every node by assigning the same time slot size to every node; the nodes closer to the

gateway can transmit more packets in the same slot than the leaf nodes. By using this strategy

GREENNESS enables higher aggregated network throughput.

We can conclude that the simulation results are consistent with the experimental results.

4.6 Discussion

GREENNESS energy saving increases as the network size increases. Yet, for WVSNs with

more than 20 nodes, energy saving is kept almost constant, independently of the variation of

the average number of hops, number of nodes, and offered load. Moreover, the impact of the

LPR power consumption in GREENNESS energy saving is approximately constant for LPRs

consuming less than 10 % of Pidle. This means that GREENNESS attains the same energy

saving for all candidate LPRs identified in this thesis. By switching the Wi-Fi radio to sleep

mode instead of switching it OFF does not affect GREENNESS energy saving significantly

since PWiFisleep only represents 7 % of Pidle. NSM was designed to be traffic-aware, but from

our evaluation for WVSNs with sizes above 20 nodes the additional energy saving gain is

negligible, thus the algorithm complexity can be reduced for these scenarios. For lower

WVSN sizes and low offered loads, the traffic-aware feature can save up to 20 %. Moreover,

GREENNESS improves network capacity and throughput fairness when compared to state-of-

the-art CSMA/CA-based WVSN solutions. GREENNESS focuses on the link and network

layers. If a cross-layer approach is employed, for instance, considering video compression or

source coding, the amount of information transmitted would diminish, especially in a multi-

hop network where a video stream is transmitted and received by several nodes before reaching

the sink, thus the energy gains can be even higher. GREENNESS can also be employed in a

different scenario than wireless video sensing and generalised to other multi-hop scenarios

since the traffic-aware mechanism allows energy savings for low data rates. Since 2016, a new

task group was created in the 802.11 Working Group for the development of an amendment to

the standard, named "Wake-up Radio Operation". The working group is currently addressing

Wi-Fi networks in infrastructure mode, but the specified wake-up radio can be adopted and

integrated with GREENNESS mechanisms to support multi-hop networks.

GREENNESS evaluation could be improved in four main directions: 1) study the solution

for different LPR inter-frame intervals; 2) evaluate the performance of WVSNs with average

hop count higher than two; 3) study the WVSN for sizes higher than 30 nodes; and 4) evaluate


the solution when using the IEEE 802.11ac standard. We expect that longer LPR inter-frame

intervals will increase the video delay and jitter. Preliminary results we have obtained for

network topologies having high average hop count showed that the GREENNESS gains would

still exist. The WVSN size was limited to 30 nodes in the evaluation because we consider it

to be a reasonable size for a WVSN; moreover, GREENNESS divides the channel capacity

by the number of nodes, meaning that above 30 nodes the bit rate for video will be low. In

our simulation scenario for N = 30 each WVS only obtains 67 kbit/s. In the simulations we

adopted the communications standard IEEE 802.11b and Friis propagation model since IEEE

802.11ac for multi-hop CSMA/CA networks and high bit rates was not functioning properly in

ns-3 by the time we made the study. Nevertheless, in the numerical analysis, we can observe

that the communications standard only affects T , the time for a WVS to transmit a frame and

receive the corresponding acknowledgement from its parent. As such, from Eq. (4.35) we can

conclude that the GREENNESS energy saving is independent of T .

GREENNESS has a set of important features, including the following: 1) it was designed

for multiple traffic scenarios, namely video streaming; 2) incorporates failure detection mech-

anisms to overcome the loss of LPR messages; and 3) supports high inter-frame interval,

which enables the usage of FM-RDS, IEEE 802.15.4, and BLE. Although GREENNESS was

evaluated for a scenario where each WVS sends a video stream to a server located in the cloud,

it is designed to support traffic from the gateway to the WVSs, for instance, to enable the Real-

Time Transport Protocol Control Protocol (RTCP) typically used to control RTP sessions; in

each LPR control message we have included a flag that indicates whether the gateway has

data to be transmitted. When the flag is true, the WVS will first wait to receive data from

the gateway. The NSM has built-in failure detection methods to overcome losses of control

messages sent over LPR and failures in the transmission of data packets over Wi-Fi. By

configuring timeouts in both the gateway and the WVSs, the algorithm will recover from these

failures. NSM was designed to allow higher inter-frame intervals, which can potentiate the

adoption of other LPRs such as IEEE 802.15.4 and BLE that support multi-hop topologies to

extend their coverage and achieve the same range as Wi-Fi. LPR candidates should have inter-

frame intervals at least equal to the one achieved by the in-band control using Wi-Fi [13]. In

fact, this was a simplification since in [13] the control channel used Wi-Fi, and good results

regarding network capacity and fairness were attained. We believe that IEEE 802.15.4 and

BLE LPRs can also achieve the same results as in [11] since for a WVSN with an average hop

count of four this will mean an inter-frame interval of 28 ms, which is much less than the one

for FM-RDS.

4.7 Summary 99

The GREENNESS solution has three limitations: 1) it needs the installation of an additional

radio in each WVS, 2) the traffic-aware NSM only supports CBR traffic, and 3) OWD is higher

than in CSMA/CA-based solutions. Although GREENNESS requires the installation of an

additional radio, its cost is low and negligible when compared to the rest of the WVS cost

[11]. The traffic-aware NSM was designed for CBR traffic and has to be enhanced using

WVS queue’s size information to support Variable Bit Rate (VBR) traffic. The OWD metric

is high when compared with CSMA/CA-based solutions, due to GREENNESS reservation of

a time slot for each WVS. Nevertheless, this can be mitigated by adopting frame aggregation

or spatial reuse, to decrease the number of time slots exclusively reserved, but still assuring the

same performance.

4.7 Summary

The GREENNESS solution proposed in this thesis was evaluated in this chapter using

numerical analysis, simulations, and experiments employing a proof-of-concept prototype.

An evaluation methodology was designed to study the energy saving and performance of

GREENNESS when compared to state-of-the-art CSMA/CA-based WVSN solutions. The en-

ergy saving and performance of GREENNESS in multi-hop scenarios was evaluated applying

the following figures of merit: energy saving, network capacity, delay, packet loss ratio, and

throughput fairness.

The evaluation considered a numerical analysis of GREENNESS and PACE energy con-

sumption for three regular network topologies: chain, binary, and grid. The numerical analysis

was then generalised for random topologies and compared with ns-3 simulations, which

validated the numerical analysis. The results showed the energy saving increases when network

size increases and decreases when the average number of hops increases. For high network

sizes, the energy savings do not change when the average number of hops increases since the

number of WVSs, which are switched OFF is much higher than those switched ON.

The impact of LPR power consumption and switching of Wi-Fi radios to sleep mode on

GREENNESS energy saving was calculated using the equations validated previously. For

LPRs consuming less than 10 % of Pidle, the energy saving is nearly constant. The energy

saving achieved for all the candidate LPRs, identified in Section 3.6.2, is similar since their

power consumption is below 4 % of Pidle. GREENNESS still achieves significant energy

savings when WVSs Wi-Fi radio is switched to sleep mode instead of switched OFF because

PWiFisleep represents 7 % of Pidle.


The energy saving was measured for different offered loads in order to evaluate the NSM

traffic-aware feature. The simulation results demonstrate that GREENNESS achieves energy

saving up to 92 % for random topologies with up to 30 nodes. GREENNESS is energy efficient

for low offered loads because Wi-Fi radios are switched OFF when the WVSs do not transmit

data.

GREENNESS performance was evaluated through simulations for different offered traffic

loads and compared with CSMA/CA-based WVSN solutions. The simulation results for

network capacity, packet loss ratio, and throughput fairness showed that GREENNESS out-

performs CSMA/CA-based WVSN solutions. CSMA/CA has lower OWD than GREENNESS

but at the cost of higher PLR.

Finally, the results obtained using a testbed with 6 WVSs were reported. The results were

consistent with those obtained in simulation for the three regular topologies considered: binary

tree, balanced tree topology with two branches and three hops, and unbalanced tree.

In conclusion, GREENNESS can offer high energy savings for multi-hop WVSNs and im-

prove network capacity and throughput fairness when compared to state-of-the-art CSMA/CA-

based WVSN solutions. Moreover, it incorporates a traffic-aware algorithm that enables

substantial energy savings, namely for small size WVSNs with low offered loads. The impact

on the energy saving for not switching OFF the Wi-Fi radios but only changing them to sleep

mode can be neglected and may simplify the implementation of GREENNESS.

Chapter 5

Conclusion

In this chapter, we provide an overview of the developed work, summarise our original

contributions, and discuss topics for future work.

5.1 Work Review

WVSNs are being applied in scenarios ranging from healthcare to surveillance. This is

motivated by the high availability of low cost networked wireless devices and video cameras.

Multi-hop IEEE 802.11-based WVSNs suffer from three problems: low network capacity,

throughput unfairness, and energy inefficiency.

In Chapter 2, we reviewed a set of green WVSN solutions that were proposed in the

literature to address these problems. Our review aimed at studying energy efficient solutions for

multi-hop topologies that could offer QoS guarantees – delay, packet loss ratio, and throughput

fairness – for wireless video sensing scenarios. The solutions were classified into three

categories: 1) out-of-band control oriented, 2) MAC oriented, and 3) routing oriented. From

the green WVSN solutions studied, none was found capable of turning OFF, or switching to

sleep mode, the Wi-Fi radio of the node when it is not transmitting data. Furthermore, the

presented solutions could not guarantee throughput performance and fairness per node in a

wireless video sensing scenario with all WVSs transmitting to a sink located in the cloud.

GREENNESS and its companion algorithms and mechanisms were described in Chapter 3.

First, the GREENNESS concept and architecture were presented. GREENNESS consists of

the following modules: 1) Routing module, 2) WATCM, 3) NSM, 4) FRM, and 5) LPR.

The Routing module runs during the network bootstrap to create the routes between each

WVS and the gateway. The WATCM mechanism runs in the gateway and WVSs, being

101

102 Conclusion

responsible for discovering the network topology using the IEEE 802.11 interface. The NSM

mechanism manages the poll messages through the LPR and turns the WVS Wi-Fi radios

ON/OFF accordingly. The FRM mechanism runs in the gateway and WVSs when a failure

is detected. The LPR module is responsible for receiving the poll messages and waking-

up the Wi-Fi radio interfaces. For the control channel, it is essential to select a low power

technology that fulfils a set of requirements. A few candidate LPRs were identified; when

designing the network, LPR can be chosen based on a compromise between radio coverage

and energy consumption.

In Chapter 4, the methodology used to evaluate GREENNESS was presented. First, we

conducted an analysis of GREENNESS and PACE energy consumption for random network

topologies to evaluate the energy saving attained. Next, we measured the impact of different

LPR power consumption and the switching of Wi-Fi radios to sleep mode on GREENNESS

energy savings. Finally, we conducted simulations to study the energy savings for different

offered traffic loads and compared the performance of GREENNESS with CSMA/CA-based

WVSN solutions. The simulation results were confirmed for regular topologies using a testbed

with 7 nodes. GREENNESS has showed energy savings up to 92 % when compared to state-of-

the-art CSMA/CA-based WVSNs, while improving network capacity and throughput fairness.

5.2 Contributions Summary

The aim of this thesis was to develop a green multi-hop WVSN solution for a wireless video

sensing scenario. The goal was to minimise energy consumption and guarantee performance,

considering that every WVSs transmits a video stream to a cloud server. The main original

contribution of this thesis is GREENNESS, which features a low power out-of-band con-

trol channel and a traffic-aware Node Scheduling Mechanism (NSM). GREENNESS enables

significant energy savings while improving network capacity and throughput fairness when

compared to CSMA/CA-based WVSNs. The GREENNESS solution includes the following

specific contributions:

• GREENNESS concept and architecture: GREENNESS is a holistic solution com-

posed of the WATCM, a low power out-of-band control channel, and a traffic-aware node

scheduling mechanism. GREENNESS enables significant energy savings while improv-

ing network performance when compared to CSMA/CA-based WVSNs. GREENNESS

differs from related work by supporting multi-hop networks for video streaming sce-

nario, without changing the current IP stack, and it is capable of turning OFF entirely or

5.3 Future Work 103

switching to sleep mode the Wi-Fi radio of the WVSs node when it is not transmitting

data.

• WATCM: is responsible for discovering the network topology, using the IEEE 802.11

interface. Since poll messages are limited to one per WVS and are not required to poll

relay nodes, the signalling to control the WVS Wi-Fi radios is minimised. Moreover, the

size of the Poll messages is also reduced since a shorter identifier than the MAC address

is adopted.

• NSM: manages the transmission of poll messages through LPR, turning the WVS

Wi-Fi radios ON/OFF accordingly. Energy savings are almost not affected when Wi-Fi

radios are changed to sleep mode instead of turned OFF. The implemented traffic-aware

mechanism adapts the WVSs polling order according to the traffic pattern, and the results

demonstrate that it improves energy savings for small size WVSNs with low offered

loads. The energy savings reached up to 92 % while improving network capacity and

throughput fairness when compared to CSMA/CA-based WVSNs.

• FRM: the WVSN topology can change, but GREENNESS detects a failure using

timeouts and tries to recover by signalling all WVSs to turn ON their Wi-Fi interfaces

and use WiFIX TR messages and WATCM to discover the new network topology and

compute the polling order for NSM. The developed prototype demonstrated that FRM

detects topology changes and can recover the network topology.

5.3 Future Work

As future work, we will consider the testing of bidirectional LPRs which would enable

faster recovery mechanisms and improved node scheduling mechanisms. The usage of frame

aggregation mechanisms in relay nodes and the spatial reuse would improve WVSN perfor-

mance and energy saving. We also expect to enhance the traffic-aware mechanism to support

Variable Bit Rate (VBR) traffic and to take advantage of the WVS queue’s size information.

Next, we detail each topic considered for future work:

• Bidirectional LPRs: By using bidirectional LPRs, the energy saving decreases as more

power is necessary to transmit data. The poll messages could now include a confirmation

message to avoid the usage of timeouts, included in the WVSs and the gateway. The

nodes can also signal when they need to send a frame to the gateway. This can be

104 Conclusion

important for scenarios when WVSs need to report asynchronous events. For the traffic-

aware mechanism, bidirectional communication can be used to signal the WVS queue

size to offer OWD guarantees.

• Frame aggregation: Frame aggregation can be applied to reduce the number of poll

messages sent by the gateway, thus reducing the communication OWD. Since GREEN-

NESS needs to poll every WVS, the OWD metric increases for high offered loads. An

alternative is to aggregate the video frames from relay WVS with the frame from the leaf

WVS, i.e., a single branch could be controlled by a single poll message. By reducing the

number of poll messages, we diminish the communication OWD. Moreover, aggregation

will not increase the communication OWD since all WVS transmit video frames to the

gateway with a CBR, and a packet does not wait more time within a node for other

packets.

• Spatial reuse: GREENNESS can reuse channels by adopting a Distributed RAND

[41], similar to the one implemented by Zebra MAC (Z-MAC). DRAND is an efficient,

scalable channel-scheduling algorithm, which allocates time slots for all the nodes in the

network. DRAND allows higher channel utilisation efficiency since two-hop neighbours

of a WVS can own the same slot at the same time. GREENNESS could run in the

gateway a DRAND algorithm, enabling to switch ON/OFF more than one WVS per

time slot. Since the number of poll messages decreases, the communication OWD also

diminishes.

• Traffic-aware mechanism: NSM was designed to be traffic-aware to further improve

energy saving. In the warm-up phase, NSM learns the traffic pattern so that it can be

reproduced later. Since NSM knows there will be no traffic from a given WVS, the WVS

is not polled, and the Wi-Fi radio is kept OFF. For CBR traffic, this mechanism is easily

implemented but will not work for VBR. Since GREENNESS already provides WVS

queue’s size information periodically to the gateway, NSM can be enhanced to include

this parameter and change the scheduling of WVSs according to their queue’s size.

References

[1] Ericsson, “Ericsson Mobility Report:On the pulse of the networked society,” Ericsson,Tech. Rep., 2016. [Online]. Available: https://www.ericsson.com/assets/local/mobility-report/documents/2016/ericsson-mobility-report-november-2016.pdf

[2] Technavio, “Global Wireless Video Surveillance Market 2017-2021,” Technavio, Tech.Rep., 2017. [Online]. Available: https://www.technavio.com/report/global-it-security-global-wireless-video-surveillance-market-2017-2021?utm_source=T3&utm_campaign=Media&utm_medium=BW

[3] B. Walsh, “New Greenpeace Report Shows the Environmental Impact of the Internet |Time,” 2014. [Online]. Available:http://time.com/46777/your-data-is-dirty-the-carbon-price-of-cloud-computing/(Accessed 2018-06-02).

[4] A. Seema and M. Reisslein, “Towards Efficient Wireless Video Sensor Networks: ASurvey of Existing Node Architectures and Proposal for A Flexi-WVSNP Design,”IEEE Communications Surveys & Tutorials, vol. 13, no. 3, pp. 462–486, 2011.

[5] F. Al-Turjman, “Cognitive routing protocol for disaster-inspired Internet of Things,”Future Generation Computer Systems, mar 2017.

[6] L. Mainetti, L. Patrono, and A. Vilei, “Evolution of wireless sensor networks towardsthe Internet of Things: A survey,” SoftCOM 2011, 19th International Conference onSoftware, Telecommunications and Computer Networks, pp. 1–6, 2011.

[7] J.-P. Vasseur and A. Dunkels, Interconnecting smart objects with IP: the next Internet.Morgan Kaufmann Publishers/Elsevier, 2010. [Online]. Available:http://linkinghub.elsevier.com/retrieve/pii/B9780123751652000272

[8] I. O. for Standardization., I. E. Commission., I. o. E. Engineers., Electronics, andI.-S. S. Board., “ISO/IEC/IEEE International Standard - Information technology –Telecommunications and information exchange between systems – Local andmetropolitan area networks – Specific requirements – Part 11: Wireless LAN mediumaccess control (MAC) and physical,” pp. 1–430, 2015. [Online]. Available:https://ieeexplore.ieee.org/document/7226764

[9] International Organization for Standardization., International ElectrotechnicalCommission., Institute of Electrical and Electronics Engineers., and IEEE-SA

105

https://www.ericsson.com/assets/local/mobility-report/documents/2016/ericsson-mobility-report-november-2016.pdf

https://www.ericsson.com/assets/local/mobility-report/documents/2016/ericsson-mobility-report-november-2016.pdf

https://www.technavio.com/report/global-it-security-global-wireless-video-surveillance-market-2017-2021?utm_source=T3&utm_campaign=Media&utm_medium=BW



http://time.com/46777/your-data-is-dirty-the-carbon-price-of-cloud-computing/

http://linkinghub.elsevier.com/retrieve/pii/B9780123751652000272

https://ieeexplore.ieee.org/document/7226764

106 REFERENCES

Standards Board., “ISO/IEC/IEEE 8802-11: Wireless LAN Medium Access Control(MAC) and Physical Layer (PHY) specifications. Amendment 3: Enhancements forvery high throughput in the 60 GHz band,” pp. 1–634, 2014. [Online]. Available: https://ieeexplore.ieee.org/document/6774849http://ieeexplore.ieee.org/document/6774849/

[10] F. Sousa, R. Campos, and M. Ricardo, “Energy-efficient wireless multimedia sensornetworks using FM as a control channel,” 2014 IEEE Symposium on Computers andCommunications (ISCC), pp. 1–7, jun 2014.

[11] J. Dias, F. Sousa, F. Ribeiro, R. Campos, and M. Ricardo, “Green wireless video sensornetworks using FM radio system as control channel,” in 2016 12th Annual Conferenceon Wireless On-demand Network Systems and Services (WONS). IEEE, 2016, pp. 1–8.

[12] R. Campos, R. Duarte, F. Sousa, M. Ricardo, and J. Ruela, “Network infrastructureextension using 802.1D-based wireless mesh networks,” Wireless Communications andMobile Computing, vol. 11, no. 1, pp. 67–89, jan 2011.

[13] F. Ribeiro, R. Campos, D. Rua, C. Pinho, and J. Ruela, “PACE: Simple multi-hopscheduling for single-radio 802.11-based Stub Wireless Mesh Networks,” inInternational Conference on Wireless and Mobile Computing, Networking andCommunications. IEEE, oct 2013, pp. 103–110.

[14] V. Jelicic, M. Magno, D. Brunelli, V. Bilas, and L. Benini, “Benefits of Wake-Up Radioin Energy-Efficient Multimodal Surveillance Wireless Sensor Network,” IEEE SensorsJournal, vol. 14, no. 9, pp. 3210–3220, sep 2014.

[15] H. Karvonen, J. Petajajarvi, J. Iinatti, M. Hamalainen, and C. Pomalaza-Raez, “Ageneric wake-up radio based MAC protocol for energy efficient short rangecommunication,” in 2014 IEEE 25th Annual International Symposium on Personal,Indoor, and Mobile Radio Communication (PIMRC), vol. 2014-June. IEEE, sep 2014,pp. 2173–2177.

[16] J. H. Antti Lappeteläinen, Tapani Ryhänen, Mauri Honkanen, “Method and device fortransponder aided wake-up of a low power radio device by a wake-up event,” oct 2002.[Online]. Available: https://patents.google.com/patent/US7072697B2/en

[17] S. Tang, C. Zhang, H. Yomo, and S. Obana, “Energy and spectrum efficient wirelessLAN by tightly integrating low-power wake-up radio,” IEEE International Symposiumon Personal, Indoor and Mobile Radio Communications, PIMRC, no. i, pp. 1–6, 2016.

[18] T. Mekonnen, P. Porambage, E. Harjula, and M. Ylianttila, “Energy ConsumptionAnalysis of High Quality Multi-Tier Wireless Multimedia Sensor Network,” IEEEAccess, vol. 5, pp. 15 848–15 858, 2017.

[19] A. Rowe, R. Mangharam, and R. Rajkumar, “FireFly: A time synchronized real-timesensor networking platform,” in Wireless Ad Hoc Networking: Personal-Area,Local-Area, and the Sensory-Area Network, S.-L. Wu and Y.-C. Tseng, Eds.Auerbach Publications, 2006, pp. 65–104.

https://ieeexplore.ieee.org/document/6774849 http://ieeexplore.ieee.org/document/6774849/

https://ieeexplore.ieee.org/document/6774849 http://ieeexplore.ieee.org/document/6774849/

https://patents.google.com/patent/US7072697B2/en

REFERENCES 107

[20] L. Gu and J. A. Stankovic, “Radio-triggered wake-up for wireless sensor networks,”Real-Time Systems, vol. 29, no. 2-3, pp. 157–182, 2005.

[21] B. V. der Doorn, W. Kavelaars, and K. Langendoen, “A prototype low-cost wakeupradio for the 868 MHz band,” International Journal of Sensor Networks, vol. 5, no. 1,p. 22, 2009.

[22] I. Haratcherev, M. Fiorito, and C. Balageas, “Low-Power Sleep Mode andOut-Of-Band Wake-Up for Indoor Access Points,” in 2009 IEEE Globecom Workshops.IEEE, nov 2009, pp. 1–6.

[23] E. Shih, P. Bahl, and M. J. M. Sinclair, “Wake on wireless: an event driven energysaving strategy for battery operated devices,” Proceedings of the 8th annualinternational ACM MobiCom, vol. pp, pp. 160–171, 2002.

[24] H. Dubois-Ferriere, R. Meier, L. Fabre, and P. Metrailler, “TinyNode: a comprehensiveplatform for wireless sensor network applications,” 5th International Conference onInformation Processing in Sensor Networks, pp. 358–365, 2006.

[25] i. D. P. Paramvir Bahl, Atul Adya, Alastair Wolman, “Energy-aware communicationsfor a multi-radio system,” jun 2003. [Online]. Available:https://patents.google.com/patent/US7099689

[26] T. Mekonnen, E. Harjula, A. Heikkinen, T. Koskela, and M. Ylianttila, “EnergyEfficient Event Driven Video Streaming Surveillance Using SleepyCAM,” in 2017IEEE International Conference on Computer and Information Technology (CIT).IEEE, aug 2017, pp. 107–113.

[27] T. Mekonnen, E. Harjula, T. Koskela, and M. Ylianttila, “SleepyCAM: Powermanagement mechanism for wireless video-surveillance cameras,” in 2017 IEEEInternational Conference on Communications Workshops, ICC Workshops 2017, 2017,pp. 91–96.

[28] R. Bhatt and R. Datta, “A two-tier strategy for priority based critical event surveillancewith wireless multimedia sensors,” Wireless Networks, vol. 22, no. 1, pp. 267–284, jan2016.

[29] W. Ye, J. Heidemann, and D. Estrin, “An energy-efficient MAC protocol for wirelesssensor networks,” Proceedings of the Twenty-First Annual Joint Conference of theIEEE Computer and Communications Societies, vol. 3, pp. 1567–1576, 2002.

[30] W. Ye, J. Heidemann, and D. Estrin, “Medium Access Control With CoordinatedAdaptive Sleeping for Wireless Sensor Networks,” IEEE/ACM Transactions onNetworking, vol. 12, no. 3, pp. 493–506, jun 2004.

[31] K. Joice Olempia, C. Pandeeswaran, and P. Natarajan, “A Survey on Energy EfficientContention based and Hybrid MAC Protocols for Wireless Sensor Networks,” IndianJournal of Science and Technology, vol. 9, no. 12, pp. 1–10, 2016.

https://patents.google.com/patent/US7099689

108 REFERENCES

[32] T. van Dam, T. V. Dam, K. Langendoen, and K. Langendoen, “An AdaptiveEnergy-Efficient MAC Protocol for Wireless Sensor Networks,” Proceedings of the 1stInternational Conference on Embedded Networked Sensor Systems (SenSys), pp.171–180, 2003.

[33] J. Polastre, J. Hill, and D. Culler, “Versatile Low Power Media Access for WirelessSensor Networks,” Proceedings of the 2nd International Conference on EmbeddedNetworked Sensor Systems (SenSys), pp. 95–107, 2004.

[34] A. El-Hoiydi, “Aloha with preamble sampling for sporadic traffic in ad hoc wirelesssensor networks,” in 2002 IEEE International Conference on Communications.Conference Proceedings. ICC 2002 (Cat. No.02CH37333), vol. 5. IEEE, 2002, pp.3418–3423.

[35] Gang Lu, B. Krishnamachari, and C. Raghavendra, “An adaptive energy-efficient andlow-latency MAC for data gathering in wireless sensor networks,” in 18th InternationalParallel and Distributed Processing Symposium, 2004. Proceedings. IEEE, 2004, pp.224–231.

[36] R. Yadav, S. Varma, N. Malaviya, I. Varma, S (Indian Institute of InformationTechnology, Allahabad, I. Malaviya, N. (Institute of Engineering & Technology,Lucknow, S. Varma, and N. Malaviya, “A survey of MAC protocols for wireless sensornetworks,” UbiCC journal, vol. 4, no. 3, pp. 827–833, 2009.

[37] S. S. Ray, I. Demirkol, and W. Heinzelman, “ADV-MAC: Advertisement-Based MACProtocol for Wireless Sensor Networks,” in 2009 Fifth International Conference onMobile Ad-hoc and Sensor Networks. IEEE, 2009, pp. 265–272.

[38] M. Buettner, G. V. Yee, E. Anderson, and R. Han, “X-MAC: a short preamble MACprotocol for duty-cycled wireless sensor networks,” in Proceedings of the 4thinternational conference on Embedded networked sensor systems - SenSys ’06. NewYork, New York, USA: ACM Press, 2006, p. 307.

[39] A. Ullah, J.-S. Ahn, and Gayoung Kim, “X-MAC protocol with Collision Avoidancealgorithm,” in 2013 Fifth International Conference on Ubiquitous and Future Networks(ICUFN). IEEE, jul 2013, pp. 228–233.

[40] Y. Kim, H. Shin, and H. Cha, “Y-MAC: An Energy-Efficient Multi-channel MACProtocol for Dense Wireless Sensor Networks,” in 2008 International Conference onInformation Processing in Sensor Networks (ipsn 2008). IEEE, apr 2008, pp. 53–63.

[41] Injong Rhee, A. Warrier, M. Aia, Jeongki Min, and M. M. Sichitiu, “Z-MAC: A HybridMAC for Wireless Sensor Networks,” IEEE/ACM Transactions on Networking, vol. 16,no. 3, pp. 511–524, jun 2008.

[42] L. Sitanayah, C. J. Sreenan, and K. N. Brown, “ER-MAC: A Hybrid MAC Protocol forEmergency Response Wireless Sensor Networks,” in 2010 Fourth InternationalConference on Sensor Technologies and Applications. IEEE, jul 2010, pp. 244–249.

REFERENCES 109

[43] K. T. Kim, W. J. Choi, M. J. Whang, and H. Y. Youn, “CA-MAC: Context AdaptiveMAC Protocol for Wireless Sensor Networks,” in 2009 International Conference onComputational Science and Engineering. IEEE, 2009, pp. 344–349.

[44] C. Pandeeswaran, N. Papa, and S. G. Jayesh, “EE-Hybrid MAC Protocol for WirelessSensor Networks,” Applied Mechanics and Materials, vol. 573, no. 6, pp. 407–411, jun2014.

[45] Q. Zhou, Y. Xu, and X. Li, “HTSMAC: High throughput sensor MAC for wirelessvideo networks,” 2007 International Conference on Wireless Communications,Networking and Mobile Computing, WiCOM 2007, no. 2006, pp. 2428–2431, 2007.

[46] A. Shahzad, G. a. Shah, and A. U. Khattak, “QoS-supported energy-efficient MAC(QEMAC) protocol based on IEEE 802.11e for Wireless Multimedia SensorNetworks,” The 5th International Conference on New Trends in Information Scienceand Service Science, vol. 1, pp. 200–204, 2011.

[47] IEEE Computer Society. LAN/MAN Standards Committee. and Institute of Electricaland Electronics Engineers., IEEE standard for information technology part 11, wirelessLAN medium access control (MAC) and physical layer (PHY) specifications. Instituteof Electrical and Electronics Engineers, 2007.

[48] Eun-Sun Jung and N. Vaidya, “An energy efficient MAC protocol for wireless LANs,”in Proceedings.Twenty-First Annual Joint Conference of the IEEE Computer andCommunications Societies, vol. 3. IEEE, 2002, pp. 1756–1764.

[49] Y. Shi and T. A. Gulliver, “An Energy-Efficient MAC Protocol for Ad Hoc Networks,”Wireless Sensor Network, vol. 01, no. 05, pp. 407–416, dec 2009.

[50] V. Vukadinovic, I. Glaropoulos, and S. Mangold, “Ad Hoc Networks Enhanced powersaving mode for low-latency communication in multi-hop 802 . 11 networks,” Ad HocNetworks, vol. 23, pp. 18–33, dec 2014.

[51] N. Rajagopalan and C. Mala, “Modified Power Save Model for Better EnergyEfficiency and Reduced Packet Latency,” American Journal of Engineering andApplied Sciences, vol. 5, no. 3, pp. 237–242, 2012.

[52] S. P. S. Prakash, T. N. Nagabhushan, and K. Krinkin, “Energy aware power save modemanagement in wireless mesh networks,” in 14th Conference of Open InnovationAssociation FRUCT. IEEE, nov 2013, pp. 122–131.

[53] V. Bernardo, M. Curado, and T. Braun, “An IEEE 802.11 energy efficient mechanismfor continuous media applications,” Sustainable Computing: Informatics and Systems,vol. 4, no. 2, pp. 106–117, jun 2014.

[54] X. Zhang and K. G. Shin, “E-MiLi: energy-minimizing idle listening in wirelessnetworks,” in Proceedings of the 17th annual international conference on Mobilecomputing and networking - MobiCom ’11. New York, New York, USA: ACM Press,sep 2011, p. 205.

110 REFERENCES

[55] A. Azcorra, I. Ucar, F. Gringoli, A. Banchs, and P. Serrano, “µNap: Practicalmicro-sleeps for 802.11 WLANs,” Computer Communications, vol. 110, pp. 175–186,sep 2017.

[56] V. Bhandary, A. Malik, and S. Kumar, “Routing in Wireless Multimedia SensorNetworks: A Survey of Existing Protocols and Open Research Issues,” Journal ofEngineering, vol. 2016, pp. 1–27, 2016.

[57] Z. Ling and W. Jian-xin, “Delay-constrained maximized lifetime routing algorithm inwireless multimedia sensor networks,” in 2010 2nd International Conference on FutureComputer and Communication, vol. 1. IEEE, 2010, pp. 215–219.

[58] K. Akkaya and M. Younis, “An energy-aware QoS routing protocol for wireless sensornetworks,” Distributed Computing Systems Workshops, 2003. Proceedings. 23rdInternational Conference on, pp. 710–715, 2003.

[59] Tian He, J. Stankovic, Chenyang Lu, and T. Abdelzaher, “SPEED: a stateless protocolfor real-time communication in sensor networks,” in 23rd International Conference onDistributed Computing Systems, 2003. Proceedings. IEEE, 2003, pp. 46–55.

[60] O. Chipara, Z. He, G. Xing, Q. Chen, X. Wang, C. Lu, J. Stankovic, and T. Abdelzaher,“Real-time Power-Aware Routing in Sensor Networks,” in 200614th IEEEInternational Workshop on Quality of Service. IEEE, jun 2006, pp. 83–92.

[61] S. Ratnaraj, S. Jagannathan, and V. Rao, “OEDSR: Optimized energy-delaysub-network routing in wireless sensor network,” in Networking, Sensing and Control,2006. ICNSC ’06. Proceedings of the 2006 IEEE International Conference on. IEEE,2006, pp. 330–335.

[62] S. Medjiah, T. Ahmed, and F. Krief, “AGEM: Adaptive Greedy-CompassEnergy-Aware Multipath Routing Protocol for WMSNs,” in 2010 7th IEEE ConsumerCommunications and Networking Conference. IEEE, jan 2010, pp. 1–6.

[63] J. Ben-Othman and B. Yahya, “Energy efficient and QoS based routing protocol forwireless sensor networks,” Journal of Parallel and Distributed Computing, vol. 70,no. 8, pp. 849–857, aug 2010.

[64] B. Li and K.-I. Kim, “A novel real-time scheme for (m,k)-firm streams in wirelesssensor networks,” Wireless Networks, vol. 20, no. 4, pp. 719–731, may 2014.

[65] M. A. Razzaque, M. M. Alam, M. Mamun-Or-rashid, and C. S. Hong,“Multi-constrained QoS geographic routing for heterogeneous traffic in sensornetworks,” IEICE Transactions on Communications, vol. E91-B, no. 8, pp. 2589–2601,2008.

[66] A. A. Ahmed and N. Fisal, “A real-time routing protocol with load distribution inwireless sensor networks,” Computer Communications, vol. 31, no. 14, pp. 3190–3203,2008.

REFERENCES 111

[67] H. Shen, G. Bai, Z. Tang, and L. Zhao, “QMOR: QoS-Aware Multi-sink OpportunisticRouting for Wireless Multimedia Sensor Networks,” Wireless PersonalCommunications, vol. 75, no. 2, pp. 1307–1330, mar 2014.

[68] X. Yan, L. Li, and F. An, “Multi-constrained routing in wireless multimedia sensornetworks,” in 2009 International Conference on Wireless Communications & SignalProcessing. IEEE, nov 2009, pp. 1–5.

[69] M. A. Rahman, R. GhasemAghaei, A. El Saddik, and W. Gueaieb, “M-IAR:Biologically Inspired Routing Protocol for Wireless Multimedia Sensor Networks,” in2008 IEEE Instrumentation and Measurement Technology Conference. IEEE, may2008, pp. 1823–1827.

[70] Y. Sun, H. Ma, L. Liu, and Y. Zheng, “ASAR: An ant-based service-aware routingalgorithm for multimedia sensor networks,” Frontiers of Electrical and ElectronicEngineering in China, vol. 3, no. 1, pp. 25–33, jan 2008.

[71] M. Abazeed, N. Faisal, S. Zubair, and A. Ali, “Routing protocols for wirelessmultimedia sensor network: a survey,” Journal of Sensors, vol. 2013, p. 11, 2013.

[72] X. Yu, J. Luo, and J. Huang, “An Ant Colony Optimization-based QoS routingalgorithm for wireless multimedia sensor networks,” 2011 IEEE 3rd InternationalConference on Communication Software and Networks, pp. 37–41, 2011.

[73] S. Selvakennedy, S. Sinnappan, and Y. Shang, “A biologically-inspired clusteringprotocol for wireless sensor networks,” Computer Communications, vol. 30, no. 14-15,pp. 2786–2801, oct 2007.

[74] L. Cobo, A. Quintero, and S. Pierre, “Ant-based routing for wireless multimedia sensornetworks using multiple QoS metrics,” Computer Networks, vol. 54, no. 17, pp.2991–3010, 2010.

[75] H. Huang, X. Cao, R. Wang, and Y. Wen, “A QoS-aware routing algorithm based onant-cluster in wireless multimedia sensor networks,” Science China InformationSciences, vol. 57, no. 10, pp. 1–16, oct 2014.

[76] D. Niculescu and B. Nath, “DV based positioning in ad-hoc networks,” SpringerJournal on Telecommunication Systems, vol. 22, no. 1, pp. 267–280, 2003.

[77] Yao Lan, Wen Wenjing, and Gao Fuxiang, “A real-time and energy aware QoS routingprotocol for Multimedia Wireless Sensor Networks,” in 2008 7th World Congress onIntelligent Control and Automation. IEEE, 2008, pp. 3321–3326.

[78] A. Nayyar, F. Bashir, and Ubaid-UR-Rehman, “Load based energy aware multimediarouting protocol-(LEAR),” in 2011 3rd International Conference on ComputerResearch and Development, vol. 2. IEEE, mar 2011, pp. 427–430.

112 REFERENCES

[79] D. Kandris, M. Tsagkaropoulos, I. Politis, A. Tzes, and S. Kotsopoulos, “Energyefficient and perceived QoS aware video routing over Wireless Multimedia SensorNetworks,” Ad Hoc Networks, vol. 9, no. 4, pp. 591–607, jun 2011.

[80] M. Chen, V. C. M. Leung, S. Mao, and Y. Yuan, “Directional geographical routing forreal-time video communications in wireless sensor networks,” ComputerCommunications, vol. 30, no. 17, pp. 3368–3383, nov 2007.

[81] L. Shu, Y. Zhang, L. T. Yang, Y. Wang, M. Hauswirth, and N. Xiong, “TPGF:Geographic routing in wireless multimedia sensor networks,” TelecommunicationSystems, vol. 44, no. 1-2, pp. 79–95, 2010.

[82] I. Bennis, O. Zytoune, D. Aboutajdine, and H. Fouchal, “Low energy geographicalrouting protocol for wireless multimedia sensor networks,” in 2013 9th InternationalWireless Communications and Mobile Computing Conference (IWCMC). IEEE, jul2013, pp. 585–589.

[83] J. Wang, Y. Zhang, J. Wang, Y. Ma, and M. Chen, “PWDGR: Pair-Wise DirectionalGeographical Routing Based on Wireless Sensor Network,” IEEE Internet of ThingsJournal, vol. 2, no. 1, pp. 14–22, feb 2015.

[84] R. Piyare, A. L. Murphy, C. Kiraly, P. Tosato, and D. Brunelli, “Ultra Low PowerWake-Up Radios: A Hardware and Networking Survey,” pp. 2117–2157, 2017.[Online]. Available: http://ieeexplore.ieee.org/document/7990121/

[85] V. Jelicic, M. Magno, D. Brunelli, V. Bilas, and L. Benini, “Analytic comparison ofwake-up receivers for WSNs and benefits over the wake-on radio scheme,” inProceedings of the 7th ACM workshop on Performance monitoring and measurement ofheterogeneous wireless and wired networks - PM2HW2N ’12. New York, New York,USA: ACM Press, 2012, p. 99.

[86] J. Blanckenstein, J. Klaue, and H. Karl, “A survey of low-power transceivers and theirapplications,” IEEE Circuits and Systems Magazine, vol. 15, no. 3, pp. 6–17, 2015.

[87] U. Raza, P. Kulkarni, and M. Sooriyabandara, “Low Power Wide Area Networks: AnOverview,” IEEE Communications Surveys and Tutorials, vol. 19, no. 2, pp. 855–873,2017.

[88] E. B. U. (EBU), Specification of the radio data system (RDS) for VHF/FM soundbroadcasting in the frequency range from 87,5 to 108,0 MHz Spécifications.CENELEC, 1998.

[89] B. Latré, P. D. Mil, and I. Moerman, “Maximum throughput and minimum delay inIEEE 802.15. 4,” Mobile Ad-hoc and Sensor Networks, pp. 866–876, dec 2005.

[90] K. Zheng, S. Zhao, Z. Yang, X. Xiong, and W. Xiang, “Design and Implementation ofLPWA-Based Air Quality Monitoring System,” IEEE Access, vol. 4, pp. 3238–3245,2016.

http://ieeexplore.ieee.org/document/7990121/

REFERENCES 113

[91] “Datasheet for FEDL7404-03,” Tech. Rep., 2018. [Online]. Available:http://www.lapis-semi.com/en/data/datasheet-file_db/telecom/FEDL7404-03.pdf

[92] B. G. Gebremedhin, J. Haapola, and J. Iinatti, “Performance Evaluation of IEEE 802 .15 . 4k Priority Channel Access with DSSS PHY,” in Proceedings of EuropeanWireless 2015; 21th European Wireless Conference, 2015, pp. 368–373.

[93] Silicon Labs, “AN1137: Bluetooth R©Mesh Network Performance,” Silicon Labs,Tech. Rep., 2018. [Online]. Available: http://www.silabs.com/mesh-performance.

[94] ISO, ISO/IEC 18000-7:2014 - Information technology – Radio frequency identificationfor item management – Part 7: Parameters for active air interface communications at433 MHz. ISO, 2014. [Online]. Available:https://www.iso.org/standard/57336.htmlhttp://www.iso.org/iso/catalogue_detail.htm?csnumber=53424http://www.iso.org/iso/home/store/catalogue_tc/catalogue_detail.htm?csnumber=59643

[95] Haystack Technologies, “How The Next Generation Will Connect Wirelessly,”Haystack Technologies, Tech. Rep., 2012. [Online]. Available:http://haystacktechnologies.com/wp-content/uploads/2012/08/Haystack-Presentation-Tutorial-082112.pdf

[96] Wizzilab, “Ultra low power DASH7 Arduino Shield Modem for 868 MHz.” [Online].Available: https://technodocbox.com/PC_Support/67508019-Sh1030-rev-introduction-ultra-low-power-dash7-arduino-shield-modem-applications-description-868-mhz-features.html(Accessed 2018-10-20).

[97] “Weightless - Setting the Standard for IoT.” [Online]. Available:http://www.weightless.org/ (Accessed 2018-10-13).

[98] “Ingenu - Dedicated Machine Connectivity for IoT.” [Online]. Available:https://www.ingenu.com/ (Accessed 2019-01-04).

[99] “Weightless vs Other LPWAN.” [Online]. Available:https://www.ubiik.com/lpwan-comparisons (Accessed 2018-10-13).

[100] Sami Tabbane, “Internet of Things: A technical overview of the ecosystem.” [Online].Available: https://www.itu.int/en/ITU-D/Regional-Presence/Africa/Documents/IoT_TechnicalOverviewoftheEcosystem_v2_27062017.pdf (Accessed2019-01-04).

[101] 3GPP TS 36.300, “Evolved Universal Terrestrial Radio Access ( E-UTRA ) Overalldescription,” Tech. Rep., 2015. [Online]. Available: http://www.etsi.org/deliver/etsi_ts/136300_136399/136300/11.08.00_60/ts_136300v110800p.pdf

[102] F. Sousa, J. Dias, F. Ribeiro, R. Campos, and M. Ricardo, “Green Wireless VideoSensor Networks Using Low Power Out-of-Band Signalling,” IEEE Access, vol. 6, pp.30 024–30 038, 2018.

http://www.lapis-semi.com/en/data/datasheet-file_db/telecom/FEDL7404-03.pdf

http://www.silabs.com/mesh-performance.

https://www.iso.org/standard/57336.html http://www.iso.org/iso/catalogue_detail.htm?csnumber=53424 http://www.iso.org/iso/home/store/catalogue_tc/catalogue_detail.htm?csnumber=59643



http://haystacktechnologies.com/wp-content/uploads/2012/08/Haystack-Presentation-Tutorial-082112.pdf

http://haystacktechnologies.com/wp-content/uploads/2012/08/Haystack-Presentation-Tutorial-082112.pdf

https://technodocbox.com/PC_Support/67508019-Sh1030-rev-introduction-ultra-low-power-dash7-arduino-shield-modem-applications-description-868-mhz-features.html



http://www.weightless.org/

https://www.ingenu.com/

https://www.ubiik.com/lpwan-comparisons

https://www.itu.int/en/ITU-D/Regional-Presence/Africa/Documents/IoT_Technical Overview of the Ecosystem_v2_27062017.pdf

https://www.itu.int/en/ITU-D/Regional-Presence/Africa/Documents/IoT_Technical Overview of the Ecosystem_v2_27062017.pdf

http://www.etsi.org/deliver/etsi_ts/136300_136399/136300/11.08.00_60/ts_136300v110800p.pdf

http://www.etsi.org/deliver/etsi_ts/136300_136399/136300/11.08.00_60/ts_136300v110800p.pdf

114 REFERENCES

[103] R. Jain and D. Chiu, “A Quantitative Measure of Fairness and Discrimination forResource Allocation in Shared Systems, Digital Equipment Corporation,” EasternResearch Lab, Tech. Rep., 1984. [Online]. Available:http://www.academia.edu/download/9569743/fairness.pdf

[104] C. Aydogdu and E. Karasan, “An Analysis of IEEE 802.11 DCF and Its Application toEnergy-Efficient Relaying in Multihop Wireless Networks,” IEEE Transactions onMobile Computing, vol. 10, no. 10, pp. 1361–1373, oct 2011.

[105] L. M. Feeney and M. Nilsson, “Investigating the energy consumption of a wirelessnetwork interface in an ad hoc networking environment,” INFOCOM 2001: TwentiethAnnual Joint Conference of the IEEE Computer and Communications Societies, vol. 3,pp. 1548–1557, 2001.

[106] D. Halperin, B. Greenstein, A. Sheth, and D. Wetherall, “Demystifying 802.11n PowerConsumption,” in Proceedings of the 2010 Workshop on Power Aware Computing andSystems (HotPower’10). Vancouver, BC, Canada: USENIX Association, 2010, pp.2–6.

[107] C. Jacquet, “FM-RDS transmitter using the Raspberry Pi,” 2015. [Online]. Available:https://github.com/ChristopheJacquet/PiFmRds (Accessed 2018-09-15).

[108] T. Pedersen, “Mesh Powersave Implementation Notes.” [Online]. Available:https://github.com/o11s/open80211s/wiki/Mesh-Powersave-Implementation-Notes#scheduling (Accessed 2018-10-06).

http://www.academia.edu/download/9569743/fairness.pdf

https://github.com/ChristopheJacquet/PiFmRds

https://github.com/o11s/open80211s/wiki/Mesh-Powersave-Implementation-Notes#scheduling

https://github.com/o11s/open80211s/wiki/Mesh-Powersave-Implementation-Notes#scheduling

Green Wireless Video Sensor Networks using a Low-Power ...Doutora Marília Pascoal Curado Professora...

Documents

Transcript of Green Wireless Video Sensor Networks using a Low-Power ...Doutora Marília Pascoal Curado Professora...