Technical Report UTD/EE/13/2005 November 2005

Achieving Reliable Networking for the Generic Autonomous Platform forSensor Systems (GAP4S)

P. Monti

Technical Report UTD/EE/13/2005November 2005

ACHIEVING RELIABLE NETWORKING FOR THE GENERIC AUTONOMOUS

PLATFORM FOR SENSOR SYSTEMS (GAP4S)

APPROVED BY SUPERVISORY COMMITTEE:

Dr. Andrea Fumagalli, Chair

Dr. Marco Tacca, Co-Chair

Dr. Franco Maloberti

Dr. Murat Torlak

Dr. Jason Jue

c© Copyright 2005

Paolo Monti

All Rights Reserved

To my parents, my brother Fabio, my grandmother, and my aunt Anna.



by

PAOLO MONTI, Dott. Eng.

DISSERTATION

Presented to the Faculty of

The University Of Texas At Dallas

in Partial Fulfillment

of the Requirements

for the Degree of

DOCTOR OF PHILOSOPHY IN ELECTRICAL ENGINEERING

THE UNIVERSITY OF TEXAS AT DALLAS

December 2005

PREFACE

This dissertation was produced in accordance with guidelines which permit the inclusion as

part of the dissertation the text of an original paper, or papers, submitted for publication.

The dissertation must still conform to all other requirements explained in the “Guide for

Preparation of Master’s Theses, Doctoral Dissertations, and Doctor of Chemistry Practica

Reports at The University of Texas at Dallas. It must include a comprehensive abstract,

a full introduction and literature review, and a final overall conclusion. Additional material

(procedural and design data as well as descriptions of equipment) must be provided in sufficient

detail to allow a clear and precise judgment to be made of the importance and originality of

the research reported.

It is acceptable for this dissertation to include as chapters authentic copies of papers already

published, provided these meet type size, margin, and legibility requirements. In such cases,

connecting texts which provide logical bridges between different manuscripts are mandatory.

Where the student is not the sole author of a manuscript, the student is required to make an

explicit statement in the introductory material to that manuscript describing the student’s

contribution to the work and acknowledging the contribution of the other author(s). The

signatures of the Supervising Committee which precede all other material in the dissertation

attest to the accuracy of this statement.

v

ACKNOWLEDGEMENTS

First and foremost I would like to thank my advisor, Dr. Andrea Fumagalli for his constant

patience, guidance and support. His positive attitude and endless encouragement made this

thesis possible. His invaluable advice will benefit my professional career for years to come.

Particular thanks go to my co-advisor Dr. Marco Tacca. He worked closely with me

throughout my studies at the University of Texas at Dallas and contributed to my work with

insightful suggestions and precious help.

I would also like to thank the other members of my committee, Dr. Franco Maloberti, Dr.

Murat Torlak, and Dr. Jason Jue for their constructive comments that helped me finalize this

dissertation.

I would like to express my thanks to all my former and current colleagues at the Optical

Network Advance Research (OpNeAR) laboratory. Working with them has been a pleasure.

A special thanks go to Isabella Cerutti, Marco Ghizzi, Stefano Gregori, Zsolt Pandi and Luca

Valcarenghi for their help and friendship in both my professional and personal life.

Finally, I own sincere gratitude to my family. They always encouraged and supported me.

Without them none of my academic goals would have been possible.

This work has been made possible by the financial support of NSF under grants No. CNS-

0082085, CNS-0435429 and ECS-0225528, the CPQd research foundation and the Italian Min-

istry of University (MIUR) under contract n. RBNE01KNFP.

This dissertation was submitted to the supervisory committee on October 2005.

vi



Publication No.

Paolo Monti, Ph.D.The University Of Texas At Dallas, 2005

Supervising Professor: Dr. Andrea Fumagalli

Networks of wireless integrated sensors are often used to monitor parameters distributed in the

environment. These parameters are related to a variety of applications such as security, patient

monitoring, chemical and biological hazard detection. Some solutions rely on replaceable

batteries with a limited life-time to provide long-term sensor operation. Others envision short

transmission range sensors (few meters) that harvest their energy from various environmental

sources (e.g., solar, vibrations, acoustic noise). The Generic Autonomous Platform for Sensors

(GAP4S) project explores an approach for wireless sensors that is complementary to these

and other pre-existing solutions.

In GAP4S, the wireless sensor micro-battery is remotely recharged via a microwave signal.

Medium transmission ranges in the tens to hundreds of meters are possible. Within these

wireless transmission ranges, a base-station collects data transmitted by the sensors and acts

as the access point to a wider (typically wired) communication network, e.g., the Internet.

The authorized user can, therefore, remotely connect to, monitor, and manage both the sensor

network and the individual sensors. An essential component of GAP4S is its end-to-end

vii

network reliability solution, which ensures the delivery of data generated at the sensor to the

interested user across both the wireless and wired segments.

This dissertation investigates ways to achieve reliable networking for GAP4S over both the

wireless and the wired segments. A specially designed solution is provided in each segment.

In the wireless segment, error-free transmissions from the sensor node to the base-station is

achieved using automatic repeat request (ARQ) protocols at layer 2. Two classes of ARQ pro-

tocols are designed and compared. The first is the conventional ARQ, whereby the data frame

is retransmitted by the originating sensor until successfully received by the base-station. The

second class takes advantage of cooperative radio communications, whereby multiple neigh-

boring sensor nodes may combine their efforts during the retransmission process. The ARQ

protocols are compared in terms of their saturation throughput, i.e., the maximum data flow

that the sensor node can sustain constrained to the available energy amount. In a variety of

scenarios — current and future expected circuit energy consumptions — the cooperative ARQ

protocols may more than double the saturation throughput when compared to conventional

ARQ protocols. Equivalently, it can be said that the energy required to operate the system

may be reduced by half.

In the wired segment, fault tolerant networking is achieved by means of protection switch-

ing at layer 3. Given the increasingly widespread use of Wavelength Division Multiplexed

(WDM) backbone networks, the protection switching scheme is designed to operate in con-

junction with WDM. Optical circuits are made reliable by means of a Shared Path Protection

(SPP) switching scheme. The SPP scheme is generalized to guarantee Differentiated levels of

Reliability (DiR) to the user. In the SPP-DiR combined scheme the desired level of reliability

may be guaranteed while minimizing the required network resources, i.e., wavelengths. This

viii

feature makes it possible to support more optical connections and users when compared to

other existing protection switching schemes.

ix

TABLE OF CONTENTS

PREFACE v

ACKNOWLEDGEMENTS vi

ABSTRACT vii

LIST OF TABLES xiii

LIST OF FIGURES xiv

CHAPTER 1 INTRODUCTION 1

1.1 Wireless Sensor Networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1.1 Design Considerations and Challenges . . . . . . . . . . . . . . . . . . 2

1.1.2 Wireless Sensor Networks Projects . . . . . . . . . . . . . . . . . . . . 5

1.2 The GAP4S Project . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.2.1 The GAP4S Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

1.3 Proposed Approach for Network Reliability . . . . . . . . . . . . . . . . . . . . 16

1.3.1 Cooperative ARQ Protocols . . . . . . . . . . . . . . . . . . . . . . . . 16

1.3.2 The SPP-DiR Protection Switching Scheme . . . . . . . . . . . . . . . 18

CHAPTER 2 BACKGROUND AND PREVIOUS WORK 20

2.1 Background on Cooperative ARQ Protocols . . . . . . . . . . . . . . . . . . . 20

2.1.1 Fading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

2.1.2 Channel Coding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

2.1.3 Channel Diversity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

2.1.4 Spatial Diversity Through Cooperative Communication . . . . . . . . . 26

2.1.5 The ARQ Protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

2.2 Previous Work on Cooperative ARQ Protocols . . . . . . . . . . . . . . . . . . 35

2.3 Background on Protection Switching Schemes . . . . . . . . . . . . . . . . . . 39

x

2.3.1 Multiplexing Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . 39

2.3.2 Second Generation Optical Networks . . . . . . . . . . . . . . . . . . . 41

2.3.3 Network Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

2.3.4 Network Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

2.3.5 Wavelength Routing . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

2.3.6 Survivability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

2.4 Previous Work on Protection Switching Schemes . . . . . . . . . . . . . . . . . 57

CHAPTER 3 ARQ-C: A CLASS OF COOPERATIVE ARQ PROTOCOLS 63

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

3.2 Two Classes of ARQ Protocols . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

3.2.1 The ARQ-NC Protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

3.2.2 The ARQ-C Protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

3.3 Assessing Saturation Throughput . . . . . . . . . . . . . . . . . . . . . . . . . 69

3.3.1 ARQ-NC Protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

3.3.2 ARQ-C Protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

3.4 One-Relay ARQ Protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

3.4.1 Sorting Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

3.4.2 The Greedy Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

3.5 Performance Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82

3.5.1 Wireless Network Assumptions . . . . . . . . . . . . . . . . . . . . . . 82

3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90

CHAPTER 4 ARQ-CN : A CLASS OF CASCADE COOPERATIVE ARQ PROTO-

COLS 91

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

4.2 ARQ-CN : a Cascade Cooperative Protocol . . . . . . . . . . . . . . . . . . . . 92

4.3 Assessing Saturation Throughput of the ARQ-CN Protocol . . . . . . . . . . . 95

4.4 Average Number of Transmissions for the Three Classes of ARQ Protocols . . 99

4.4.1 ARQ-NC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

4.4.2 ARQ-C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99

4.4.3 ARQ-CN . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100

xi


4.5.1 The Effect of Sensor Density on S . . . . . . . . . . . . . . . . . . . . . 103

4.5.2 The Effect of E(Rx)b on S . . . . . . . . . . . . . . . . . . . . . . . . . . 110

4.5.3 The Effect of Uplink Frequency and Path Loss Exponent on S . . . . . 112

4.5.4 Non Uniform Traffic Scenario . . . . . . . . . . . . . . . . . . . . . . . 114

4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

CHAPTER 5 SPP-DiR PROTECTION SWITCHING SCHEME 123

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

5.2 The SPP-DiR Model for WDM Networks with Dynamic Traffic . . . . . . . . 127

5.2.1 The Online RWA Problem for SPP-DiR . . . . . . . . . . . . . . . . . 130

5.3 Solving the Online RWA Problem for both the Working and Protection Paths 133

5.3.1 Step A: Construction of the DPM . . . . . . . . . . . . . . . . . . . . . 133

5.3.2 Step B: the RWA Algorithm . . . . . . . . . . . . . . . . . . . . . . . . 135


5.4.1 Comparison Between Pruning Techniques . . . . . . . . . . . . . . . . . 141

5.4.2 Comparison of SPP and SPP-DiR Schemes . . . . . . . . . . . . . . . . 143

5.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146

CHAPTER 6 CONCLUSIONS 152

REFERENCES 155

VITA

xii

LIST OF TABLES

3.1 Pseudo code of the greedy algorithm . . . . . . . . . . . . . . . . . . . . . . . 81

3.2 ARQ-C protocol: saturation throughput (S) as a function of the uplink fre-

quency (fup), footprint radius (R) and path loss exponent (n) . . . . . . . . . 89

4.1 ARQ-CN protocol: saturation throughput (S) as a function of the footprint

radius (R) and the number of sensor nodes (N) . . . . . . . . . . . . . . . . . 112

4.2 ARQ-CN protocol: saturation throughput (S) as a function of the uplink fre-

quency (fup), footprint radius (R) and path loss exponenet (n) . . . . . . . . . 113

5.1 Pseudo code of the algorithm to construct DPM . . . . . . . . . . . . . . . . . 134

5.2 Statistics on candidate paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141

5.3 LB solutions found by the SA algorithm . . . . . . . . . . . . . . . . . . . . . 142

5.4 DPM solutions found by the SA algorithm . . . . . . . . . . . . . . . . . . . . 143

xiii

LIST OF FIGURES

1.1 Possible GAP4S application areas . . . . . . . . . . . . . . . . . . . . . . . . . 8

1.2 GAP4S: sensor system and communication structure . . . . . . . . . . . . . . 9

1.3 GAP4S sensor node: block diagram . . . . . . . . . . . . . . . . . . . . . . . . 11

2.1 Received power vs. distance: effect of large-scale fading and small-scale fading [1] 20

2.2 Types of fading experienced by a signal as a function of: (a) symbol period,

(b) signal bandwidth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

2.3 Spatial diversity through radio cooperation [1] . . . . . . . . . . . . . . . . . . 27

2.4 Classification of ARQ protocols [1] . . . . . . . . . . . . . . . . . . . . . . . . 31

2.5 Non cooperative ARQ protocols [1] . . . . . . . . . . . . . . . . . . . . . . . . 33

2.6 Basic multiplexing techniques (T represent the bit duration) . . . . . . . . . . 40

2.7 WDM photonic network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

2.8 Optical component scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

2.9 Space switch . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

2.10 Optical crossconnect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

2.11 Wavelength routing: without wavelength conversion and with wavelength con-

version . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

2.12 Virtual topology drawn from the physical topology in Figure 2.11 . . . . . . . 49

2.13 Network layered structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

2.14 Path switching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

2.15 Line switching: span protection . . . . . . . . . . . . . . . . . . . . . . . . . . 56

2.16 Line switching: line protection . . . . . . . . . . . . . . . . . . . . . . . . . . . 56

2.17 Ring switching . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

2.18 Line protection in ring switching . . . . . . . . . . . . . . . . . . . . . . . . . . 58

3.1 ARQ-C protocol: cooperation between two sensor nodes . . . . . . . . . . . . 67

3.2 ARQ-NC protocol: flow model . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

3.3 ARQ-C protocol: flow model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73

xiv

3.4 Two sensor nodes network scenario: one source, one relay and the base-station 75

3.5 Saturation throughput (S) as a function of the position of the relay. The source

is placed in (50, 0) m and the base-station is placed in (0, 0) m . . . . . . . . . 78

3.6 Saturation throughput (S) as a function of the position of the relay. The source

is placed in (50, 0) m, the base-station is placed in (0, 0) m, and the y coordinate

of the relay is fixed and equal to 0 m . . . . . . . . . . . . . . . . . . . . . . . 79

3.7 Error probability (P (e)) as a function of SNRrec . . . . . . . . . . . . . . . . . 85

3.8 Saturation throughput (S) versus energy per bit radiated by the sensor nodes

(Eb). R = 50 m, G = 1, ξ = 300, F = 5 dB, E(Rx)b = 30 nJ, n = 3 . . . . . . . 86

3.9 Saturation throughput (S) versus energy per bit radiated by the sensor nodes

(Eb). R = 50 m, G = 1, ξ = 300, F = 5 dB, n = 3 . . . . . . . . . . . . . . . . 87

4.1 ARQ-CN protocol: cooperation among three sensor nodes . . . . . . . . . . . . 93

4.2 ARQ-CN protocol: flow model . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

4.3 R = 10 m, N = 100: (a) saturation throughput (S) versus energy per bit radi-

ated by the sensor nodes (Eb), (b) Eb = 6e−14: average number of transmissions

(E[Tx]) versus the distance of a sensor node from the base-station . . . . . . . 104
















4.9 R = 100 m, N = 100: (a) saturation throughput (S) versus energy per bit

radiated by the sensor nodes (Eb), (b) Eb = 1.5e−11: average number of trans-

missions (E[Tx]) versus the distance of a sensor node from the base-station . . 108

xv







4.12 E(Rx)b = 0.3 nJ, N = 300: saturation throughput (S) versus energy per bit

radiated by the sensor nodes (Eb) . . . . . . . . . . . . . . . . . . . . . . . . . 110

4.13 E(Rx)b = 0.3 nJ, N = 300: average number of transmissions (E[Tx]) versus the

distance of a sensor node from the base-station . . . . . . . . . . . . . . . . . . 111

4.14 Hot-spot in the center of the footprint: R = 50 m, N = 300, Rhot−spot = R/8 . 115



4.17 Hot-spot at the edge of the footprint: R = 50 m, N = 300, Rhot−spot = R/8 . . 117



4.20 Hot-spot in the middle of the footprint: R = 50 m, N = 300, Rhot−spot = R/8 118



4.23 Two rates traffic increase: R = 50 m, N = 300, 25% of all sensor nodes have

an increased value of generated traffic rate . . . . . . . . . . . . . . . . . . . . 120





5.1 SPP-DiR example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

5.2 The European optical network topology and the virtual single-slot input buffer 140

5.3 Blocking probability (Pb) versus arrival rate (λ): SPP-DiR . . . . . . . . . . . 144

5.4 Blocking probability (Pb) versus arrival rate (λ): SPP . . . . . . . . . . . . . . 145

5.5 Average hop length of the working path (|Hw|) versus arrival rate (λ): SPP-DiR146

5.6 Average hop length of the working path (|Hw|) versus arrival rate (λ): SPP . . 147

5.7 Average hop length of the protection path (|Hp|) versus arrival rate (λ): SPP-DiR148

5.8 Average hop length of the protection path (|Hp|) versus arrival rate (λ): SPP . 149

xvi

5.9 Average number of shared protection links (|Hs|) versus arrival rate (λ): SPP-DiR149

5.10 Average number of shared protection links (|Hs|) versus arrival rate (λ): SPP 150

5.11 Normalized average excess of reliability versus arrival rate (λ) . . . . . . . . . 150

5.12 Blocking probability (Pb) versus MCFP (d) . . . . . . . . . . . . . . . . . . . . 151

xvii

CHAPTER 1

INTRODUCTION

1.1 Wireless Sensor Networks

The deployment of sensor networks permits the distributed detection and estimation of pa-

rameters in our surrounding environment. Conventional sensor networks require wired links

between the sensing element (sensor) and a central processing unit, where the signal process-

ing takes place. In this case these wired networks are more commonly known as control

networks [2]. It must be noted that the wiring process is often expensive, time-consuming,

fault-prone, and potentially dangerous [3].

The development of relatively inexpensive and low-power wireless micro-sensors and the

advances in wireless networking has paved the way for a new class of ad hoc networks called

wireless sensor networks. These networks consist of a large number of relatively small sensors

deployed very close to or inside the monitored phenomenon. Networked sensor nodes can

organize themselves to jointly accomplish large sensing tasks, thus greatly improving the

accuracy of the information provided to the user. An overview of wireless sensor network

application might include but is not limited to [4, 5]:

• military applications: wireless sensor networks can play an important role on the battle-

field. They can be deployed as part of military command, control, communication, sur-

veillance and reconnaissance. Applications might span from monitoring friendly forces

and equipment to battlefield surveillance and battle damage assessment. In case of a

1

2

nuclear, biological or chemical attack they may be used for detection and reconnaissance,

• environmental sensing: many valuable applications can be found in agriculture where

wireless sensor networks could be equipped with a variety of chemical and biological

sensors. Monitoring the habitat and the environmental conditions that affects crops

and livestock is just one example of environmental wireless sensor networks deployment,

• health monitoring: wireless sensor networks may be used for monitoring blood, sugar

and pressure levels, heart beat and respiratory rate,

• commercial applications: wireless sensor networks may also be applied in many commer-

cial applications. Sensor nodes attached to goods in the form of electronic tags, may

potentially revolutionize the current asset tracking and supply management system.

1.1.1 Design Considerations and Challenges

Wireless sensor networks differ from conventional network systems in many aspects. They in-

volve a large number of spatially distributed, self-configuring, energy-limited sensor nodes.

They present challenges and design considerations influenced by many factors, which in-

clude [4, 5]:

• topology: a huge number of sensor nodes are densely deployed around the sensor field.

The sensor node deployment is usually performed randomly by scattering sensor nodes in

the monitored area. Sink nodes are responsible for collecting data from the sensor nodes,

and delivering the collected data to users. They also might be responsible for assigning

and starting new sensor node tasks. Multiple wireless sensor networks can be integrated

into a larger network through the use of some existing network infrastructure, e.g., the

3

Internet, making it possible for an authorized user to query and manage a wireless sensor

network located anywhere,

• fault tolerance: wireless sensor networks should be able to maintain their functionalities

in the presence of sensor node failures. Protocols and algorithms should be designed to

address the level of fault tolerance required by the application. In general, every aspect

influencing the design of wireless sensor networks should be considered as application

specific. It is not possible to derive a general and common design strategy that will fit

every specific application,

• scalability: the number of sensor nodes deployed may reach millions of units in some

applications. All schemes developed for wireless sensor networks have to be scalable

enough to fit with the sensor node densities and numbers that are higher than all the

other types of networks in the orders of magnitude,

• sensor node hardware: the four basic components of a wireless sensor node are: (i)

the sensing unit, usually composed of the sensor and of the analog-to-digital converter

(A/D). The analog signals produced by the sensor are converted to digital signals by the

A/D converters and then made available to the processing unit. A sensor node might use

more than a single sensor; (ii) the processing unit: this unit manages the procedures that

enable the sensor node to collaborate with other sensor nodes to carry out the assigned

sensing tasks. It is generally associated with a small storage unit; (iii) the transceiver

unit: it is responsible for keeping the sensor node connected to the rest of the network

through the wireless channel; (iv) the power unit: this unit may be supported with a

power-scavenging tool to harvest the necessary energy from the environment,

4

• transmission media: sensor nodes are linked by a wireless medium. These links can rely

on radio or infrared communication. Much of the current hardware for sensor nodes

is based upon radio frequency circuit design. Another alternate mode of operation for

wireless sensor networks is infrared. Infrared communication is license-free and robust

against interference from electrical devices. Infrared-based transceivers are cheaper and

easier to build. The main drawback of infrared is the requirement of a line of sight be-

tween sender and receiver. This makes infrared a less desirable choice as the transmission

medium in the wireless sensor network scenario,

• power consumption: sensor nodes are microelectronic devices and can only be equipped

with a limited energy source. In some application scenarios, replacing power resources

might be impossible. Sensor node lifetime is therefore strongly dependent on battery

lifetime. Power conservation and power management are of paramount importance.

Power consumption in wireless sensor networks can be divided into three domains: (i)

sensing, (ii) communication, and (iii) data processing. Sensor nodes spend most of their

energy in data communication. This involves both data transmission and reception. It

can be shown that for short-range communication with low radiation power, transmission

and reception energy costs are nearly the same [4]. In this computation it is important

to consider not only the active power but also the start-up power consumption in the

transceiver circuitry.

In recent years, new solutions have been introduced to satisfy the additional constraints

and design challenges introduced by wireless sensor networks design. The following section

describes some ongoing projects in this area.

5

1.1.2 Wireless Sensor Networks Projects

Much research has been done in the field of wireless sensor networks in an attempt to address

all or part of the challenges and design considerations mentioned in the previous section. Some

of the existing wireless sensor network projects are described below [6].

• The PicoRadio Project [7]. PicoRadio is a project at the Berkeley Wireless Research

Center. This project is a low-energy platform that configures itself to provide an ad

hoc network for a variety of applications, e.g., environmental control in office buildings,

warehouse inventory, patient monitoring and identification. Sensors harvest their energy

by converting vibrations from the surrounding environment into an energy source [8].

This project started in 1999 and was one of the first attempts to design a wireless sensor

network from scratch by carefully optimizing all protocol layers down to the hardware.

• The µAMPS Project [9]. µAMPS stands for Adaptive Multi-Domain Power-Aware Sen-

sors and is focused on the development of a complete system for wireless sensor networks,

with emphasis on the need for low-power operations. One of the main contributions of

this project is the development of a communication protocol called LEACH [10] which

stands for Low Energy Adaptive Clustering Hierarchy. LEACH is a node-clustering al-

gorithm that randomizes the assignment of the power-consuming cluster head function

among multiple sensor nodes in the network.

• WiseNET [11]. WiseNET stands for Wireless Sensor Networks and is a long-term re-

search project at the Swiss center for electronics and micro-technology (CSEM). WiseNET

is a wireless network of distributed sensors that combines sensing, signal processing, con-

trol, and short-range wireless communication capabilities. The project main objective is

6

to develop a low-power wireless ad hoc network where energetic autonomous miniature

sensor nodes are able to communicate among themselves and with the external world.

Sensors used in this project are usually battery operated.

• WINS [12]. WINS stands for Wireless Integrated Networks Sensors program. The WINS

project covers almost every aspect of wireless sensor network design. The project focuses

on the design of micro-electromechanical systems (MEMS), digital/analog integrated

circuits and protocols. Moreover WINS studies the fundamental principles of sensing

and detection theory.

• WSSN [6, 13]. WSSN stands for Wireless Self-sustaining Sensor Network. WSSN is a

project at ICT Vienna. WSSN is a wireless sensor network for home/building monitor-

ing that runs without batteries. Sensors can be powered by tiny solar cells and ultra

capacitors as energy storage elements. The prototype sensor nodes are small in size and

measure temperature, humidity, and light of the surrounding environment.

The appeal of wireless sensor networks has been also proven by the interest of the industry

world [6]. For example Enocean [14], a German spin-off company from Siemens, introduced

the first products that allowed wireless communication without batteries. Energy scavenging

techniques based on solar cells or piezo-elements were used to generate just enough energy to

power a very low power wireless sensor node. The first product introduced into the marketplace

was a wireless light switch able to scavenge its energy from the mechanical pressure derived

from pushing the button.

From the list of wireless sensor network projects and products it is clear how some of

the wireless sensor network solutions rely on replaceable batteries with limited life-time to

provide long-term sensor operation, while other solutions envision short transmission range

7

sensors (few meters) that harvest their energy from various environmental sources, e.g., solar

and vibrations.

The Generic Autonomous Platform for Sensors (GAP4S) [15] project explores an approach

for wireless sensors that is complementary to these and other pre-existing solutions. The

project envisions multiple integrated sensing functions, medium transmission range sensors

(tens of meters), remote re-charge of the sensor micro-battery, and end-to-end reliable access.

The following section will provide a detailed description of the GAP4S project.

1.2 The GAP4S Project

According to the GAP4S project vision, next generation sensor systems will evolve toward

a fully nomadic yet fully interconnected approach. Each electro-mechanical apparatus and

physical system will be supported by a number of electronic devices, all of them behaving as a

micro-system connected to a hierarchy of organized networks, aimed at improving efficiency,

increasing security, and reducing risk of malfunction. The basic components of networked

sensor-systems will integrate on the same micro-substrate sensing (and, when necessary, actu-

ating) devices with local processing capability and modular interconnection techniques, pro-

viding as much distributed intelligence as possible. The sensing modules will ensure mobility

without demanding huge communication and processing bandwidth. Moreover, the wireless

interconnection shall be responsible for getting and supplying data and power in a reliable, se-

cure, programmable manner. Power will be wireless received (microwave) by the sensor node,

transformed, stored in an integrated battery, and carefully managed to ensure the best quality

of service. Power consumption of all the electronics will be reduced significantly with respect

to present levels, and the functionality of sensor modules will be managed on the basis of the

8

Figure 1.1. Possible GAP4S application areas

amount of energy available and received. Therefore, sensing and processing techniques will

attain a power-conscious management for ensuring the optimum level and quality of service.

Any smart sensing device will be based on the same basic platform architecture. The digital

processing, the radio transceiver, the on-board battery, and the power management circuit

will have standardized features. A number of sensor interfaces will adapt the sensors to a

software programmable analog-to-digital converter and a software programmable processor,

thus enabling a quick “plug-in” of different kind of sensors and leading to a “software sensor”

methodology. The same module will include multi-sensors for data-fusion and multiple con-

trol and measurement. Different kinds of sensors will be integrated on the same substrate,

9

possibly together with the electronic section (silicon multi-sensor systems). The sensor nodes

will be in the footprint (with medium radii, e.g., 100 m) of a (mini) base-station, possibly

mobile, that will be the entry point to a wider communication network. The base-station will

use smart antennas to ensure power provisioning and full-duplex connectivity to the sensor

nodes.

Figure 1.1 depicts a number of possible of applications within various fields. They span

from building, airport, and monument control to industrial and agricultural activities, personal

safety, monitoring and alerting systems. The considered category of examples assumes the

sensors are grouped-up in the electro-mechanical apparatus or physical system at a medium

distance from the base-station.

Figure 1.2. GAP4S: sensor system and communication structure

Figure 1.2 shows the GAP4S network architecture. Autonomous sensor modules are placed

on given fixed or mobile positions. They send data via a wireless channel to the base-station

and receive acknowledgment signals and power via a microwave signal transmitted by the

10

base-station. A simple modulation of the microwave link enables data transfer from the base

station to the sensor module, thus permitting, in addition to acknowledgment communica-

tion, the system administrator to update the software or to set the sensing conditions. The

data communication and protocol used on the uplink channel must ensure reliability on both

segments of the network, i.e., wireless and wired. The base-station is connected to a computer

that receives the data collected and produced by the sensors. The computer acts as a gateway

to the Internet, allowing a remotely connected end-user to both control the sensor network

and receive such data.

Figure 1.3 shows the block diagram of the generic sensor module. A variable number

of sensors are connected to the multi-function sensor interface. The interface is capable of

suitably handling signals coming from different sensors’ categories. Namely, the interface can

process voltages, currents or capacitance modulations, and resistance modulations. Signals are

then amplified by a digitally controlled gain factor. The analog interface includes filters “on

demand” if some filtering action is required. The A/D converter is software programmable

and permits to change the resolution from 8 to 12-bits and to adapt the conversion rate

from low frequency up to 40 MHz. The power management section is integrated with the

sensors, analog interface, and data converter. The power management section controls the

battery charge and discharge, determines the frequency of main clock for an optimum use

of power, incorporates the AC-DC converter for re-charging the battery, and manages the

radio section. The power management module operates its functions on the basis of expected

charge-discharge operative conditions. The processing section and the memory take benefit

from existing low-power commercial components. The VCO radio sections are integrated

using a sub-micron Si-Ge BiCMOS technology. Moreover, the return link block recovers the

slot synchronism and extracts the data transmitted over the MW signal transmitted by the

11

base-station.

Figure 1.3. GAP4S sensor node: block diagram

1.2.1 The GAP4S Modules

This section is intended to provide a brief description of each one of the GAP4S modules.

1.2.1.1 Monolithic Micro-Sensors

The sensors that this project addresses are: a temperature sensor, an array of optical sen-

sors, and a humidity sensor. They are integrated on a sub-micron CMOS technology with

the entire processing function. The temperature sensor enables the digital compensation for

the temperature dependence of many physical or chemical sensors. Moreover, the knowledge

of the environment temperature allows one to optimize the power management of on-board

batteries. Since the above-mentioned applications do not require very high accuracy, thermal

sensors based on p-n junction are used for the present project. A conventional CMOS tech-

12

nology enables the design of optical photo-diodes with good sensitivity in the visible and near

infrared spectrum. The photo-diodes exploit the p-n junction made by the substrate and the

drain/source diffusion of MOS transistors. For vision or pattern recognition the photodiode

array is arranged in an x-y matrix. The photo-charges are periodically read and transformed

into voltage. One or more A/D converters, integrated on the same chip, provide the output in

digital format. State-of-the-art solutions use active pixels for very high resolutions. However,

the class of applications addressed by this project (less than 256 x 256 pixel) suggests using

passive pixels [16]. Various physical or chemical mechanisms sense humidity. The one being

used in this project (that allows on-chip integration and good reliability) exploits the varia-

tion of the dielectric constant of a thin polyimide film deposited after a conventional CMOS

process. The sensor is a fringing capacitor made by inter-digitized electrodes (poly-1, poly-2

or metal) covered by the polyimide layer. The capacitor value, using minimum size elements

and a relatively small area, is a fraction of a pF. The humidity produces capacitance variations

of about hundred parts per million; therefore, a 10-15F - 10-16F sensitivity is required. The

analog interface can use the sensor as input capacitor of a sigma-delta modulator. A simple

digital filter generates the output. A look-up table placed inside the digital filter corrects

possible non-linearity. Therefore, the measurement of the capacitance and the conversion of

the result into the digital domain may be obtained at the same time [17].

1.2.1.2 Low Power Signal Processing

The power consumption of a digital signal processing circuit is given by the sum of switch-

ing, glitching, shorting, and leaking power contributions. Switching is proportional to the

percentage activity, the square of the portion of supply voltage used, the capacitance being

13

switched, and the frequency. Glitching is the loss due to false switching. Shorting is due to

shorting paths between supply lines. Leaking is due to the current flowing when circuit is

at rest (proportional to the complement to the percentage activity and the supply voltage).

Leaking worsens by reducing voltage because of consequent reduction in threshold and in-

complete blocking of gates when steady. This project addresses each of the areas influencing

power consumption. This requires studies of architectures tuned to the requested algorithms

(for power administration, sensor monitoring, rect-antenna control) with optimization of the

circuitry to avoid unnecessary switching: the algorithm flows with minimum transitions, and

those requested transitions take place on the minimum amount of gates, thus optimizing static

and dynamic dissipation. “Sleep mode” power down should be avoidable if the architecture

could be properly optimized after “optimal coding” of software on the chosen instruction

set. The operating frequency is kept to the lowest possible, exploiting maximum parallelism

(VLIW structure). The major effort, however, is focused on architectural studies, with proper

choice of the HW versus SW and reduction of memory/processor traffic; also, the tasks in the

system are partitioned (as it is traditional) not for reducing the area but for trading area with

low power.

1.2.1.3 Rectifying Micro-Antenna

The rect-antenna (rectifier + antenna) that receives and converts microwave power into dc

power is a key element in this autonomous sensor-system. The remote charging of a micro-

battery must be performed with the paramount objective of maximizing the energy transfer

from the microwave beam to the actual storage device. In addition, this element must fulfill

a number of environmental and operational constraints in terms of size, compactness and

14

manufacturing capability. The target of this project is to scale the power level to a few

hundreds of W. As a result, the structure of the rect-antenna will evolve from discrete elements

to a planar thin-film technology greatly reducing the weight to power output radio module.

It is expected that the use of a few tens of GHz rect-antenna implemented with a microstrip

patch antenna can be designed with an area of a few mm2. Furthermore, it is possible to pursue

a monolithic integration of active devices, passive network, and planar antenna structures on

high resistivity substrates, i.e., on a single chip.

1.2.1.4 The Micro-Battery

Up to a few years ago, the integrated batteries for microelectronics were not common yet.

The telephone handset and the portable computer market have driven the technology: the

reduction in battery size not so critical as their energy capacity. Batteries for microelectronics

were non rechargeable and many times larger than the chip itself. In recent times, however,

a significant effort has been made to scale down the size of batteries to match the size of

microelectronic components [18, 19]. The aim of this project is to do a feasibility study for

fabricating a thin-film micro-battery, which fulfills the technological needs. Both conventional

lithium metal anode and polymers are investigated. The work splits between three focused

activities: electrolyte, electrode, and the whole battery integration.

1.2.1.5 Radio Transceiver

Low power consumption is the major research effort for a nomadic sensor system. Radio

communications require energy for analog signal processing (amplifiers, mixers, oscillators)

and the output power amplifier. In the GHz band, the frequency synthesizer and the power

15

amplifier dominate energy consumption. In order to save power, the data are sent at the

maximum rate so that the duty-cycle in the power hungry sections can be minimized. For

short-range communication with low-radiation power (0 dBm) transmission and reception,

energy costs are nearly the same [4, 20]. The data traffic is mostly from the sensor node.

However, communications to sensor nodes may be necessary for secure and reliable protocols.

The radio transceiver module consists of three main blocks: the VCO, the radio transmitter

section, and the MW return link data extraction. The VCO will be digitally tunable so that

the return link communication will enable frequency tuning and, when necessary, frequency

hopping. The radio transmitter section has a conventional architecture. However, it requires

the minimum power consumption, especially in the power amplifier. Moreover, a digital control

from the power management section will determine the best trade-off between power budget

and transmission efficiency. A unique feature of GAP4S sensor is the use of a microwave

link, in addition to its being energy source, as the forward link. The microwave link (MW)

from the base-station can be modulated using the on-off keying (OOK) similar to optical

communications. Because of the short range (100m) and a medium data speed on the MW

link (1 MHz), the time dispersion on the wireless channel is negligible. Establishing a reliable

transmission link from sensors to the base-station can be provided either by increasing the

output transmit power or by employing error correction coding with the data. Since the

energy required to encode the data (for convolutional codes) is negligible, many types of

error-correcting codes can be used to improve the probability of bit error.

In the GAP4S original vision, the radio transceiver module does not include a radio receiver

section because the sensed information is delivered from each sensor node directly to the base-

station. For this reason, interaction among sensor nodes in not necessary. However, if the

16

GAP4S is used with protocols based on radio cooperation, e.g., the kind of protocols presented

in Chapter 3 and Chapter 4, a radio receiver section needs to be added to the radio transceiver

module as shown in Figure 1.3. This new additional section is necessary to enable each sensor

node to receive data frames from the neighboring sensor nodes.

1.2.1.6 Network Reliability

As already mentioned in Section 1.1.1, protocols and algorithms in a wireless sensor network

should be designed to address the level of reliability required by the application. In GAP4S

the base-station collects data transmitted by the sensor nodes and acts as the access point

to a wider (typically wired) communication network, e.g., the Internet. The authorized user

can therefore remotely connect to monitor and manage, both, the wireless sensor network

and the individual sensor nodes. An essential component of GAP4S is its end-to-end network

reliability solution, which ensures the delivery of data generated at the sensor node to the

interested user across both the wireless and wired segments.

1.3 Proposed Approach for Network Reliability

This dissertation investigates ways to achieve reliable networking for GAP4S over both the

wireless and the wired segments. A specially designed solution is provided for each segment,

as explained next.

1.3.1 Cooperative ARQ Protocols

In the wireless segment, error-free transmissions from the sensor node to the base-station is

achieved using automatic repeat request (ARQ) protocols at layer 2.

17

One of the factors that affects the performance of the wireless channel from the sensor

node to the base-station is fading. There are several techniques to overcome the effect of mul-

tipath fading. One technique, known as diversity, relies on providing statistically independent

channels from the source to the destination.

Cooperative radio communication is an innovative concept that can provide diversity using

one or more cooperating partner sensor nodes. This concept can be incorporated into the data

link layer. Reliable data delivery over a single hop radio channel requires the retransmission

of the data frame until its reception is successful at the destination. Traditional Automatic

Repeat Request (ARQ) protocols specify the frame exchange that takes place between the

source and the destination, represented by the base-station in the GAP4S.

The usefulness of ARQ protocols in wireless sensor network applications is usually limited

by the additional retransmission cost and overhead [4]. Nonetheless, ARQ protocols are

particularly appealing for the GAP4S due to powering MW flow used as a feedback channel

from the sensor node to the base-station.

In this dissertation, three classes of ARQ protocols are designed and compared. The first

is the conventional ARQ, whereby the data frame is retransmitted by the originating sensor

until successfully received by the base-station. The other two classes takes advantage of

cooperative radio communications, whereby multiple neighboring sensor nodes may combine

their efforts during the retransmission process. The ARQ protocols are compared in terms

of their saturation throughput, i.e., the maximum data flow that the sensor node can sustain

constrained to the available energy amount. The objective is accomplished by taking into

account the effect of path loss on the MW recharging channel and, both, path loss and

fading on the uplink wireless channel. Single and multi-hop transmission, current and future

18

expected circuit energy consumptions, are considered under the assumption of limited sensor

node energy. It was found that the cooperative ARQ protocols may more than double the

saturation throughput when compared to conventional ARQ protocols. Equivalently, it can

be said that the required energy to operate the system may be reduced by half.

1.3.2 The SPP-DiR Protection Switching Scheme

In the wired segment, fault tolerant networking is achieved by means of protection switching

at layer 3. Given the increasingly widespread use of Wavelength Division Multiplexed (WDM)

backbone networks, the protection switching scheme is designed to operate in conjunction with

WDM.

WDM networks are able to offer a larger bandwidth, low attenuation value, and bit error

rates that can be as low as 10−15. This extremely low value of the bit error rate ensures a

smooth transmission from the source to the destination. Unfortunately, WDM networks may

be affected by failures — cable cuts, power outage, etc.— that may interrupt a large number

of communication sessions in progress, consequently losing a large amount of information.

Therefore, a survivable WDM network design is mandatory.

Survivability may be guaranteed by assigning spare resources available at the optical links,

for protection purposes. For an optical circuit, a protection scheme consists of assigning a

working and a protection path between the source and the destination. The working path

carries the offered traffic during normal network operations. When the working path is dis-

rupted by a fault, the interrupted traffic is re-routed over the protection path until the fault

is repaired. Different protection schemes have been proposed in the literature, trading off fast

switching time and fairly simple signaling protocols (1 + 1 and 1 : 1 protection schemes) with

19

a more efficient reuse of the spare capacity (1 : N protection schemes). These conventional

protection schemes are able to provide full protection in the case of a single fault, however,

when over-reservation of network resources is not acceptable, some of these solutions may not

be adequate.

Differentiated Reliability (DiR) is an innovative concept able to provide multiple reliability

degrees (or classes) at the same network layer, using a common protection mechanism, e.g.,

path switching. By means of the DiR concept, a network can be designed to provide multiple

degrees of reliability and efficiently satisfy the user-specific requirements, yet minimizing the

network total cost.

In this dissertation, WDM optical circuits are made reliable by means of a Shared Path

Protection (SPP) switching scheme. First the SPP switching scheme is generalized to guaran-

tee the required (differentiated) level of reliability to users, while ensuring that all connections

are provided with the required amount of reserved network resources, thus avoiding unneces-

sary provisioning of spare resources. This generalization is referred to as SPP-DiR. Second, an

approach for choosing the working and protection path-pair routing for the arriving demand is

proposed. The approach is based on a matrix of preselected path-pairs: the Disjoint Path-Pair

Matrix (DPM). Results show that when the SPP-DiR scheme is applied, by more accurately

controlling each connection reliability, a significant reduction in the amount of required net-

work resources can be achieved when compared to the conventional SPP switching scheme.

In turn, the demand blocking probability may be reduced more than one order of magnitude.

It is also shown that the DPM approach is suitable for obtaining satisfactory Routing and

Wavelength Assignment (RWA) solutions in both DiR-SPP and conventional SPP networks.

CHAPTER 2

BACKGROUND AND PREVIOUS WORK

2.1 Background on Cooperative ARQ Protocols

In a wireless channel, the transmission path from the source to the destination may contain

obstacles which affect the transmitted signal. Multiple components of the transmitted signal

may follow different path lengths, thus arriving at the destination out of step. This phenom-

enon creates interference and introduces a variety of impairments in the wireless channel such

as fading, delay spread, and attenuation [21].

2.1.1 Fading

One of the main source of impairments in a wireless channel is the occurrence of fading [21].

Fading in the wireless channel can be classified as shown in Figure 2.1.

Figure 2.1. Received power vs. distance: effect of large-scale fading and small-scale fading [1]

20

21

2.1.1.1 Large-Scale Fading

Large-scale fading is also referred as to path loss. Path loss is defined as the difference between

the transmitted power and the received power. Path loss is due mainly to shadowing and

the distance between the source and the destination. There are several models available in

the literature to estimate the value of the path loss. Examples are: free space model, two

rays model, long distance propagation model, and the log-normal shadowing propagation

model [21]. The most widely used model for signal prediction in urban areas is the Okumura

model. Power control techniques may be used to control and attenuate the effects of large-scale

fading.

2.1.1.2 Small-Scale Fading

Small-scale fading describes the rapid fluctuations of the amplitude and phase of the trans-

mitted signal over a short period of time or travel distance. Small-scale fading is due to the

multipath propagation of the signal. Two or more versions of the signal may arrive at the

destination at different times. When combined, they may result in a received signal widely

varying in amplitude and phase, depending on the distribution of the intensity and the prop-

agation time of each multipath component. In small-scale fading, the received signal power

may vary by as much as three or four orders of magnitude when the destination is moved only

a fraction of a wavelength.

The type of fading experienced by a signal propagating through a mobile radio channel

depends on the nature of the transmitted signal when compared to the characteristics of

the channel. Parameters normally used to characterize the transmitted signal are the signal

bandwidth (BS) and the symbol period (TS). Conversely, the channel may be characterized

22

using the rms delay spread (στ ) and the Doppler spread (BD) [21].

The rms delay spread is a natural phenomenon caused by the scattered and reflected

propagation paths in the radio channel. This parameter describes the time dispersive nature

of the channel in a local area. The measure of the delay spread may be useful to define a

parameter called coherence bandwidth (Bc). The coherence bandwidth is used to characterize

the channel in the frequency domain. It represents the range of frequencies over which two

frequency components have a strong potential of amplitude correlation. In other words, it

describes the range of frequencies over which the channel passes all spectral components with

approximately equal gain and linear phase.

The Doppler spread measures the spectral broadening caused by the time rate change of

the mobile radio channel. When a sinusoidal signal is transmitted at a frequency fc, the

received signal spectrum, called the Doppler spectrum, will have components in the range

fc − fd to fc + fd, where fd — called the Doppler shift — is a function of the relative speed

of the source and the angle between the direction of motion of the source and the direction of

arrival of the signal to the destination. The Doppler spread, that is a function of fd, is defined

as the range of frequencies over which the received Doppler spectrum is not equal to zero.

The measure of the Doppler spread may be useful to define a parameter called coherence time

(TC). The coherence time is the time duration over which two received signals have a strong

potential for amplitude correlation.

2.1.1.3 Types of Small-Scale Fading

The time dispersion and the frequency dispersion mechanism in a mobile radio channel lead

to four distinct small-scale fading effects [21].

23

The multipath delay spread leads to time dispersion and frequency selecting fading. If

BS << BC and TS >> στ a signal undergoes flat fading. With this type of fading the spectral

characteristics of the transmitted signal are preserved at the receiver. However, the strength

of the received signal changes in time due to the variation of the gain caused by the multipath

propagation. If BS > BC and TS < στ a signal undergoes frequency selective fading. With this

type of fading, the received signal includes multiple attenuated versions of the transmitted

waveform which are delayed in time. The received signal is thus distorted. Viewed in the

frequency domain, certain frequency components of the received signal experience greater

gains than others.

The Doppler spread leads to frequency dispersion and time selecting fading. If TS > TC

and BS < BD a signal undergoes fast fading. With this type of fading, the channel impulse

response changes rapidly within the symbol duration. This lead to frequency dispersion due to

the Doppler spreading that in turn leads to signal distortion. If TS << TC and BS >> BD a

signal undergoes slow fading. With this type of fading, the channel impulse response changes

at a rate much slower than the transmitted signal. The channel may be assumed static over

one or more symbol intervals.

These four types of fading are independent of one another. Figure 2.2 summarizes the

types of small-scale fading a signal may experience.

Power control techniques are not effective in controlling small-scale fading. Techniques

that have been proven to be effective in controlling small-scale fading are channel coding and

diversity.

24

TS

στ

TS

TC

Frequency SelectiveSlow fading

Frequency SelectiveFast fading

Flat SlowFading

Flat FastFading

(a) Symbol period

BD

BC

BS

BS

Frequency Selective

Flat Fast

Frequency SelectiveFast fading

Fading

Slow fading

Flat SlowFading

(b) Signal bandwidth

Figure 2.2. Types of fading experienced by a signal as a function of: (a) symbol period, (b)signal bandwidth

2.1.2 Channel Coding

Channel coding is one of the techniques used to overcome transmission errors over a wireless

channel affected by fading. With channel coding, redundancy bits are added in the transmitted

signal so that if an instantaneous fade occurs in the channel, data may still be recovered.

Coding is considered a post detection technique [21] because detection is performed after the

demodulation portion of the receiver. This technique is used to detect or correct some (or

all) the errors introduced by the wireless channel in a given sequence of bits. In general

there are three types of channel codes: block codes, convolutional codes, and turbo codes.

Coding is effective when trying to correct independent random symbols, i.e., when errors

occurs randomly or in a bursty manner. However, when fading in the channel is correlated,

the effectiveness of channel coding diminishes and interleaving techniques must be employed.

With interleaving methods, the source bits are spread over time before being coded. This, in

turn, guarantees that in the presence of a deep fade or noise burst the important bits in a block

of source data are not corrupted at the same time. When interleaving techniques are employed,

extra costs in terms of delay and hardware complexity are expected. In addition, both channel

25

coding and interleaving techniques expand the bandwidth occupied by a particular message.

A mitigated effect of fading in the wireless channel is, therefore, achieved with a reduced raw

data transmission rate through the channel.

2.1.3 Channel Diversity

Diversity is another possible technique to mitigate the effects of a wireless fading channel.

Diversity is able to improve the overall reception by making use of more than one independent

faded version of a transmitted signal without increasing the transmitted power or bandwidth.

Diversity is based on the concept that if several copies of the original signal are sent through

different paths, they encounter different channel characteristics and, therefore, the probability

that all paths experience deep fading at the same time is greatly reduced. Diversity can be

achieved using the following techniques [21]:

• spectral diversity: the signal is transmitted simultaneously over several frequency slots.

The rationale behind this technique is that frequencies separated by more than the co-

herence bandwidth of the channel will be uncorrelated and therefore will not experience

the same fades. Examples of systems employing spectral diversity are frequency hopped

spread spectrum communication systems,

• temporal diversity: the signal is transmitted over several time slots. This technique is

effective in the presence of time selective fading. Time slots must exceed the coher-

ence time of the channel so that the channel fading experienced by one transmission

is independent of the channel fading experienced by other transmissions. This form of

diversity, therefore, introduces a significant delay in the system. Temporal diversity can

be employed using techniques such as interleaving, forward error correction code, or au-

26

tomatic repeat request (ARQ) protocols. Another example of technique used to provide

space diversity is the RAKE receiver for spread spectrum CDMA systems. With this

technique, the multipath channel provides the redundancy in the transmitted message.

By demodulating each replica of the CDMA signal, where each one experienced a dif-

ferent multipath delay, the receiver is able to align the replicas in time so that a better

estimate of the original signal may be performed at the receiver,

• spatial diversity: the signal is transmitted and received using multiple transmitting/receiving

antennas. The separation among adjacent antennas should be large enough to allow

signals from different antennas to undergo independent fading. In environments with

Rayleigh fading and shadowing, adjacent antennas need to be at least 10 carrier wave-

length apart. This diversity technique is not, therefore, suitable for application involving

small devices such as handsets and sensor nodes.

2.1.4 Spatial Diversity Through Cooperative Communication

Cooperative communication is an innovative concept able to provide spatial diversity in ap-

plication where the use of multiple antennas is not suitable. Figure 2.3 gives an example of a

cooperative radio communication system.

One of the most peculiar characteristics of the radio medium is its inherent broadcast

nature. When a source transmits, aside from the intended destination, other nodes within

earshot may receive the transmitted signal. Depending on various factors, the received signal

may be affected by different amounts of fading, power attenuation, and noise. In single-hop

networks, this characteristic is traditionally treated as interference at the physical layer and

— unless scheduling in time has been accommodated — as collision at the Medium Access

27

Figure 2.3. Spatial diversity through radio cooperation [1]

Control (MAC) layer. Both may prevent correct reception at the destination. In essence, the

broadcast nature of the wireless link and the fading channel, typical of the radio transmission,

has been historically treated as a nuisance.

Cooperative communication can turn this liability into an advantage. The idea works

as follows: the source tries to transmit to the destination. A second node, or relay, having

overheard the source transmission, will send reinforcement to the destination, in different forms

depending on the cooperative strategy. The destination takes advantage of the additional

information — in forms that may vary with the adopted cooperative strategy — thus getting

a better overall reception quality. The essence of the idea lies in that the destination benefits

from messages arriving via two statistically independent paths, the concept of spatial diversity

introduced in Section 2.1.3. This additional information, distributed throughout the network,

may be exploited to yield a more efficient communication system.

28

Cooperative communication schemes can be categorized according to a variety of parame-

ters. Based on their forwarding strategies, cooperative communication schemes can be divided

into [22,23]:

• amplify-and-forward: in amplify-and-forward schemes, the relay simply amplifies the

packet received from the source before transmitting it to the destination. At the desti-

nation, the packet received from the source and the packet received from the relay are

combined, and a decision is made on whether the resulting packet is correct or not. The

advantage of these schemes is that the destination receives packets from two indepen-

dent fading paths. The drawback is that the relay receives and amplifies a noisy version

of the source packet,

• decode-and-forward: in decode-and-forward schemes the relay is required to demodulate,

decode and re-encode the packet received from the source before transmitting it to the

destination. While these schemes do not suffer from noise amplification, they pose the

danger of error propagation that may occur if the relay sends an incorrectly decoded

version of the packet to the destination. Both amplify-and-forward and detect-and-

forward methods, retransmit an exact replica of the transmitted packet.

• coded cooperation: coded cooperation is a method that integrates cooperation into chan-

nel coding. Coded cooperation works by sending different portions of each users code-

word via two independent fading paths. The basic idea is that each user tries to trans-

mit incremental redundancy to its partner. Whenever that is not possible, the users

automatically revert to a non cooperative mode. The key to the efficiency of coded

cooperation is that everything is managed automatically through code design, with no

feedback between users.

29

Almost independent of the above classification, cooperative schemes can be static or adap-

tive with respect to their transmission mechanism [23].

In static protocols, transmissions from the source and from the relay follow a two-phase

process. The source broadcasts in the first phase. The relay and the destination receive

the broadcasted signal. In the second phase, the relay retransmits the signal, using either

amplify-and-forward or detect-and-forward. The drawbacks of static protocols is their repet-

itive nature. For some channel conditions, the destination may have successfully received

the source message in the first phase, thus making the second phase a waste of resources.

Similarly, when the relay did not receive the source message with a quality that allows for

retransmission, then the second phase is useless as well. On the other hand, static protocols

are simple, and therefore commonly used.

In adaptive protocols, the relay can prevent error propagation by deciding at each trans-

mission cycle whether or not to retransmit the source message. A variety of protocols can be

designed. One of the possible examples are protocols based on feedback.

Protocols based on feedback can overcome the drawbacks of static protocols by attempting

to exploit variable channel conditions. Retransmissions occur only if the destination requires

additional information. Such adaptive protocols may require some form of feedback, e.g.,

ARQ protocols.

2.1.5 The ARQ Protocol

Two existing error control schemes are Automatic Repeat Request (ARQ) and Forward Error

Correction (FEC) [4].

FEC is a method of obtaining error control in data transmission in which the source sends

30

redundant data and the destination recognizes only the portion of the data that contains no

apparent errors. FEC does not require handshaking between the source and the destination,

therefore it can be used for broadcasting of data to many destinations simultaneously from

a single source. In the simplest form of FEC, each character is sent twice. The destination

checks both instances of each character for adherence to the protocol being used. If conformity

occurs in both instances, the character is accepted. If conformity occurs in one instance and

not in the other, the character that conforms to the protocol in use is accepted. If conformity

does not occur in either instance, the character is rejected.

ARQ protocols are used to provide a feedback channel from the source to the destination.

The main idea behind ARQ is to detect packets with errors at the receiving Data Link Control

(DLC) module and, then, to request the retransmission of the packet by the transmitting DLC

module. Parity checks codes and Cyclic Redundancy Check (CRC) codes are possible examples

of techniques used by the receiving DLC module to detect an erroneous packet [24]. There

are a number of variations of the the basic ARQ protocol as shown in Figure 2.4. Protocols

which involve only the source and the destination are referred to here as non-cooperative ARQ

protocols, while those that involve neighboring or partner nodes in the retransmission process

are referred to as cooperative ARQ protocols.

2.1.5.1 Non Cooperative ARQ Protocols

As explained in the previous section, in non cooperative ARQ protocols the destination relies

only on retransmission from the source to retrieve the correct version of the packet. If the

retransmission from the source does not involve any use of channel coding, these ARQ proto-

cols are known as non-hybrid. As shown in Figure 2.5 available implementation of non-hybrid

31

Figure 2.4. Classification of ARQ protocols [1]

ARQ protocols are:

• stop and wait: this is the simplest type of retransmission protocol. The basic idea is to

ensure that each packet has been correctly received before transmitting the next packet.

In transmitting packets from the source to the destination, the first packet is transmitted

in the first frame and, then, the source DLC waits. If the packet is correctly received

at the destination, a positive acknowledgment (ACK) is sent back to the source. If the

packet is not error free the destination sends a negative acknowledgment (NAK) back to

the source. ACK and NAK are protected with CRC in order to avoid errors during the

transmission. If the source receives a positive acknowledgment, it starts the transmission

of a new packet in a new frame. If the source receives a negative acknowledgment, it

retransmits the old packet in a new frame. Finally, if the acknowledgment packet is lost

the source eventually times-out and re-sends the packet to the destination,

32

• go back n: with this protocol several packets can be transmitted from the source to

the destination without waiting for the next packet to be requested. Each packet at

the source DLC is numbered sequentially, using a sequence number SN. This sequence

number is sent in the header of the frame transmitting the packet. The DLC at the

destination essentially works in the same way as stop and wait ARQ. It accepts error

free packets in the correct order and sends back request numbers RN to the source to

acknowledge all packets prior to RN. n is a parameter that determines the number of

successive packets that can be sent in the absence of a request for a new packet. The

source is not allowed to transmit packet i + n before packet i has been acknowledged.

There is a limit — or window — on the number of packets that can be transmitted

before receiving a request for the next packet. For this reason, go back n ARQ protocols

are also referred to as sliding window ARQ protocols,

• selective repeat: is another variation of ARQ protocol. The basic idea behind selective

repeat ARQ is to accept out-of-order packets at the destination DLC and to request

retransmission from the source, only for those packets that are not successfully received.

There is still a window of size n to specify how many packets the source can transmit

after RN, the lowest numbered error free packet received at the destination.

If the retransmission from the source involves the use of channel coding, then these ARQ

protocols are known as hybrid. Hybrid ARQ (H − ARQ) protocols combine FEC and ARQ

techniques to provide increased throughput over the conventional ARQ protocols. There are

three types of H-ARQ protocols:

• type I H-ARQ: type I H − ARQ protocol makes use of both error detection and error

correction capabilities. When only error detection capability is used, this is the conven-

33

Figure 2.5. Non cooperative ARQ protocols [1]

tional stop and wait ARQ protocol. Upon detecting an error, this ARQ scheme discards

the erroneous packet and sends a retransmission request to the source,

• type II H-ARQ: this is the most commonly used hybrid ARQ technique. With this

protocol, a message of m symbols and c CRC bits produce a codeword of m + c + r

bits. The codeword is punctured before transmission removing some or all of the error

correction redundancy bits and only transmitting redundancy bits for error detection.

If the destination does not correctly receive the packet, it sends a NAK to the source.

The source, instead of retransmitting the entire packet, transmits an additional block of

redundancy bits which the destination combines with the original packet and reattempts

to decode the original word. This process continues until the original packet is correctly

decoded. In this way redundancy bits are transmitted only when required. Type II

H − ARQ is more advantageous than type I H − ARQ as it does not discard the

34

previous erroneous information,

• type III H-ARQ: in type III H − ARQ schemes, individually transmitted packets are

self-decodable and each packet differs in coded bits from the previous transmission. In

type III H − ARQ, packets are only combined after decoding has been attempted on

the individual packet.

2.1.5.2 Cooperative ARQ Protocols

Cooperative ARQ protocols are based on the principle of providing diversity through coopera-

tive communication. With this aim in mind, nodes, other than the source and the destination,

actively help deliver packets to the destination correctly. The rationale is that nodes which are

within earshot from the source and the destination may cooperate by making use of received

interference to improve the overall capacity of the source-destination wireless channel. Upon

the unsuccessful reception of a packet transmitted by the source, the destination requests that

the packet be retransmitted by another “cooperating” node (i.e., the relay). The assumption

here is that the relay may have successfully overheard the packet transmission performed

by the source earlier and stored a copy of that packet temporarily. A wisely chosen relay

may increase the probability of successfully delivering the packet to the destination, without

requiring additional retransmission attempts from the source. The relay offers both spatial

diversity and its own energy to help with the source’s packet transmission.

Available options for cooperative ARQ protocol are: stop and wait, go back n and se-

lective and repeat. If the retransmission from the relay to the destination involves the use

of channel coding, cooperative ARQ protocols are referred to as Coded Cooperative ARQ

(CC − ARQ) [25].

35

2.2 Previous Work on Cooperative ARQ Protocols

Focusing on cooperative communications, it must be noted that a number of results have been

published on this promising topic. A recent survey on cooperative radio communications can

be found in [22,23].

Initial work on cooperative communications on the Gaussian relay channel is reported

in [26]. The relay role is to assist the source, i.e., single-source cooperation. While this

example focuses on the three-terminal case, a more general approach is taken in [27, 28],

which establishes performance bounds by examining the situation in which a single source-

destination pair is assisted by a network of relay terminals.

More recent works [22, 23, 29, 30, 30–35] have extended the concept of cooperative com-

munications by taking into account fading and allowing two sources to cooperate with one

another at the same time. This case is referred to as double-source cooperation, whereby each

source interleaves the transmission of its own packets with the retransmission of the other

source’s packets. As already explained in Section 2.1.4 these works can be divided into three

categories: amplify-and-forward, decode-and-forward and coded cooperation.

Amplify-and-forward methods are proposed and analyzed in [29]. The authors develop two

variants of an energy-efficient cooperative diversity protocol that combats fading induced by

multipath propagation. The underlying techniques build upon the classical relay channel and

related work and exploit space diversity available at distributed antennas through coordinated

transmission and processing by cooperating nodes.

An example of detect-and-forward signaling can be found in [31, 32]. This work presents

an analysis and a simple code-division multiple access (CDMA) implementation of decode-

36

and-forward cooperative signaling. [29] proposes an hybrid decode-and-forward method where,

when the fading channel has high instantaneous signal to noise ratio (SNR), users detect and

forward their partners’ data but, when the channel has low instantaneous SNR, users revert

to a non cooperative signaling mode.

More recently, [36] proposes various cooperative protocols for the double-source cooperation

case. The authors present static and adaptive protocols for both the amplify-and-forward and

the decode-and-forward case. They also present a cooperative protocol based on feedback.

They studied protocols performances in terms of the outage probabilities under the assumption

of limited bandwidth and constrained end-to-end delay. Results show that the relay may suffer

from the necessity of providing orthogonal resources for reception and transmission.

In coded cooperation methods, cooperation is achieved in the framework of channel cod-

ing. [37, 38] suggest the use of distributed Alamouti coding. A slightly different approach for

a decode-and-forward case is proposed in [39].

The approaches in [30,34] discuss cooperation through codeword partitioning. Each node

divides its source data into blocks which are augmented with a CRC code. Codewords of N bits

are then partitioned into N1 and N2 = N - N1 bits, where the N2 bits can be determined from

N1 (parity). The data transmission period for each node is divided into two time segments

of N1 and N2 bit intervals, respectively. These time intervals are called frames. For the first

frame, each node transmits a codeword consisting of the N1 bit code partition. Each node also

attempts to decode the transmission of its partner. If this attempt is successful (determined

by checking the CRC code), in the second frame, the node calculates and transmits the second

code partition of its partner, containing N2 code bits. Otherwise, the node transmits its own

second partition, again containing N2 bits. Thus, each node always transmits a total of N

37

= N1 + N2 bits per source block over the two frames. The nodes act independently in the

second frame, with no knowledge of whether their own first frame was correctly decoded. As

a result, there are four possible cooperative cases for the transmission of the second frame:

both nodes cooperate, neither node cooperates, Node 1 cooperates and Node 2 does not, and

vice versa. The authors suggest the use of rate-compatible punctured convolutional codes

(RCPC) [40] and analyze the pairwise error probability and block error rates.

[41–43] were the first to discuss distributed turbo coding . The source encodes its data

using a recursive systematic convolutional code with rate 1/2. The relay decodes data received

from the source and codes them again after interleaving. The destination decodes using a

standard turbo decoder. This scheme achieves coding gains at the cost of additional complexity

at the relay.

In [23,35], adaptive decode-and-forward and more complex decode-and-re-encoding schemes

that realize distributed coding strategies are presented. Both two-hop and multi-hop scenar-

ios are considered. The analysis is conducted over fading channels under limited bandwidth,

energy, and end-to-end delay.

Most of these results focus on the physical layer aspects of cooperation. Only a few works

have considered related ARQ protocol aspects [1, 25, 33, 35, 44–46]. In [33], SNR gain and

average number of retransmissions of a single-source cooperative ARQ protocol is studied.

In [44], the performance of different cooperative protocols is derived in terms of outage prob-

ability and SNR gain and compared against non-cooperative protocol performance. In [46],

the saturation throughput and latency of three double-source cooperative ARQ protocols are

studied. In [45], a relaying protocol for multiple relays, operating over orthogonal time slots,

is proposed as a generalization of hybrid ARQ protocols. Throughput, energy consumption,

38

and outage probability of the proposed protocol are compared against multi-hop protocol

performance.

In [25], the first delay model of four single-source and single-relay Cooperative ARQ pro-

tocols (C−ARQ) is presented. Two single-source single-relay cooperative ARQ protocols are

considered. In the first protocol, stop and wait C − ARQ, the source can only transmits a

new packet once the previous one has been successfully acknowledged by the destination. In

the second protocol, selective repeat C −ARQ, the source can transmit a new packet only in

the absence of packets requiring retransmission because of timeout expiration. In addition,

two coding strategies for single-source single-relay cooperative ARQ protocols are considered.

In the first protocol, type I C − ARQ, the relay transmits an exact replica of the packet, as

it was transmitted by the source. In the second protocol, type II C − ARQ, the relay trans-

mits a packet, which contains incremental redundancy bits. These bits are computed by the

relay and used by the destination by means of a code combining strategy for decoding. The

latter case is based on the coded cooperation framework discussed in [22]. The performance

of the cooperative ARQ protocols is compared against two non-cooperative ARQ protocols.

Saturation throughput, expected packet latency, and expected buffer occupancy at the source

and at the relay are evaluated and compared. Various scenarios of offered load, radio channel

attenuation, and geographical distribution of the nodes are considered.

39

2.3 Background on Protection Switching Schemes

Optical fiber began to spread because they offer larger bandwidth than copper wires and they

are not affected by various kinds of electromagnetic interferences and other undesirable effects.

Fibers are less sensitive to noise, because photons in the light transmission do not interact

with other moving photons, the interference from external light can be avoided, and the

transmitter and receiver internal noise levels as well as the power level of the light signal, can

be controlled with sufficient accuracy. It is thus possible, with optical fibers, to reach BERs

as low as 10−15, while in copper cables BERs commonly range in the 10−5.

During the 70’s the high loss of optical fiber could be reduced to about 20 dBKm (for

wavelengths operating near 1 µm), making them suitable for transmission. As a result they

were used essentially as a transmission medium used as improved replacement of copper cables.

The transmission was performed in optics while all data switching and processing was done

electronically. These networks are referred to as First Generation optical networks and today

they are widely deployed, except for residential access networks. Examples of first generation

networks are SONET (Synchronous Optical NETwork) based networks in North America and

SDH (Synchronous Digital Hierarchy) based networks in Europe and Asia.

2.3.1 Multiplexing Techniques

Optic offer a high transmission bit rate but, actually, the access to the network is limited by the

electronic speed (few Gbps) of the end terminals and line terminals. Multiplexing strategies

allow to overcome the co called electronic bottleneck . Possible multiplexing strategies in optical

networks are based on wavelength division or frequency division (WDM), time division (TDM),

or code division (CDM). These alternative options are illustrated in Figure 2.6.

40

... N4321C

onnection

Power

tT

2

Connection 1

3

4

...

N

λ

Tt

T

Connection 1 N2 ...

Voltage

WDMA (FDMA) TDMA CDMA

t

or f

Figure 2.6. Basic multiplexing techniques (T represent the bit duration)

In TDM systems, bits associated with different channels are interleaved in the time domain

to form a composite bit stream. This technique allows multiplexing of a number of lower-speed

traffic streams into a higher speed channel. The multiplexing can be performed in a fixed or

statistical way. An example of fixed TDM is SONET. Today, commercially available systems

based on fixed multiplexing have bit rates that are around 40 Gbps, while higher rates have

been reached in research laboratories and test-beds [47].

In the CDM system, users transmit simultaneously and in the same frequency band. How-

ever, each user transmits using a signal waveform that is non-interfering with the waveforms

transmitted by the other users. To overtake the limits, researchers are pushing the develop-

ment of optical multiplexing and demultiplexing functionalities. Optical TDM and CDM net-

works are futuristic, because they require much higher rates than present electronic processing

speeds.

WDM technology works on the principle of frequency division multiplexing, which has

been intensively deployed in radio systems. It allows communication at the desired bit rate

on each WDM channel, e.g., peak electronic processing speed. This multiplexing technique is

designed to become the natural evolution of the actual optical network in the next years. The

41

ITU has, for example, already standardized the Optical Layer, mainly taking into account

WDM transmission [48].

Combinations of the previous techniques are possible and allow efficient increase of the

transmission capacity. For example, WDM systems using TDM on each channel are commer-

cially available and used in WDM SONET networks.

2.3.2 Second Generation Optical Networks

Next generation optical networks should be able to extend beyond the possibility of SDH/SONET

and ATM. In recent years, it has been demonstrated that optical fibers are capable of provid-

ing more functions than just point-to-point transmission. For this reason, the future upgrade,

able to increase the bandwidth and to simplify the transport and the processing of information,

will employ WDM-based optical system. Moreover, radical changes are necessary in routing

schemes in order to simplify the transport and the processing of information, while simultane-

ously limiting the risk of large scale failures arising from the complexity of the structure. This

improvement is made possible by implementing the switching and routing functions in the

optical (physical) layer instead of requiring the Optical/Electrical/Optical (OEO) conversions

at each node. Therefore, the electronics at a node only have to handle incoming or outgoing

data at such a node, while all passing through data is routed in the optical domain, allowing

transparent routing. In first-generation networks, all the data routed through the node had

to be dealt with electronics, thus reducing the transmission speed and increasing the node

hardware complexity.

The availability of promising optical technology makes the design of new cost-effective

transport networks possible and compatible with the existing ones (e.g., ATM, SDH/SONET)

42

with the following features [49]:

• high capacity : core networks should support the increase of traffic from the access net-

work. Fibers should be able to transport high bit rates (above 10 Gbps per fiber),

• high survivability and reliability : networks should be able to circumvent failures (pro-

tection) or recover with minimum delay (restoration),

• flexibility : in case of traffic demand variations or network failures, it should be possible

to easily accommodate changes with click and point operations,

• high quality of services : networks should be able to support high bit rate client services

and to allow fast restoration and reliability,

• universality : networks must be capable of supporting a wide range of current and future

services,

• interoperability(multi-vendor networks): transport systems from different vendors should

be compatible,

• scalability : optical networks must be designed to allow expansion in number of nodes,

links, users, etc., with small incremental costs,

• upgradability : it should be possible to evolve in terms of architecture, performance,

managements, technology, standards, protocols, etc.,

• signaling and monitoring capabilities,

• efficient network management : allow an easy interface to other network managers,

43

• protocol transparency : new different classes of non-standard high bit rate signals should

be accommodated in the networks.

At the beginning, this sort of WDM network will be employed mainly as a backbone network

for a large region, such as nationwide or global coverage. End-users will comprise not only

terminal equipment but, also, regional or local subnetworks. Then, the second generation

optical networks will also appear in local-exchange and access networks.

2.3.3 Network Elements

We consider WDM photonic networks consisting of nodes interconnected by optical links. As

shown in Figure 2.7, the physical network topology is composed of:

D

E

B

A

F

1

2

3

4

5

6

7

λ2

λ3

λ1

λ1

Access station: contains receivers and trasmittersOptical nodeLink λ1 λ2 λ3

C

WDM

photonic network

carring lighpaths at

Figure 2.7. WDM photonic network

44

• nodes : capable of routing any incoming wavelength channel on any of the incoming

fibers to any wavelength channel on any of the outgoing fibers. Each node in the net-

work consists of multi-wavelength transmitters and receivers, optical space switches, and

optical amplifiers, configured to form an optical crossconnect (see Section 2.3.4),

• network port : a node element which gives access to the photonic network. Ports con-

nect low-speed end-user communication to the high-speed optical network, by means

of optoelectronic devices. They can be in connection with SONET/SDH crossconnects,

ATM nodes or IP router, called higher layer switches/routers,

• optical line or simply line: a bidirectional physical connection between two adjacent

nodes. It can be composed of many cables, each of them with several unidirectional

fibers.

• channel : a logical link between nodes that uses a specific wavelength,

• lightpath: one channel or a concatenation of channels, each of them having a reserved

wavelength, so it can pass through one (single-hop) or more (multi-hop) nodes. It sets

up a direct connection between two nodes.

2.3.4 Network Components

Figure 2.8 shows the basic technological components in fiber optics networks. A detailed

description of each of these components may be found in [47].

In WDM-optical networks, these basic components are combined to build more complex

structures. The most important components for the signal transmission are the Optical

Add/Drop Multiplexer (OADM) and the Optical Cross-Connect (OXC).

45

λ1,λ2,...λΝ F i l t e r λ1,λ2,...λΝMux / Dmux

λ1λ2λΝ

λ1 Converter λ2λ1

λ2,..,λΝ

Figure 2.8. Optical component scheme

An optical add/drop multiplexer (OADM) is essentially a simple form of a wavelength

router with one input and one output port, with additional local ports wherein wavelengths are

added to/dropped from the incoming light stream. The OADM can have built-in wavelength

conversion capabilities.

bar statecross state

Figure 2.9. Space switch

A basic crossconnect element is a switch like the one shown in Figure 2.9. A 2 × 2 Optical

CrossConnect (OXC) routes optical signals from two input ports to two output ports and

has two states: bar state and cross state. In the bar state, the signal from the upper input

port is routed to the upper output port while, in the cross state, the upper port is routed

to the lower output port. A more complex structure is shown in Figure 2.10, as it includes

optical amplifiers and demultiplexers, in addition to the crossconnect capabilities. This OXC

has M = 2 incoming fiber links (input ports) and M = 2 outgoing fiber links (output ports),

each carrying W = 3 wavelengths. The incoming signal on a link is demultiplexing by a

46

OA

WDM WDM

Optical switch matix

In

OA

OXC

λ− converter

Out

Mux

1

2

λ1,λ2,λ3

λ1

λ2

λ3

λ1

λ2

λ3

λ1,λ2,λ3

λ1,λ2,λ3λ1,λ2,λ32

1

(optional)

Local ADM input/output ports

Demux

Figure 2.10. Optical crossconnect

wavelength demultiplexer. For each wavelength, there is a dedicated M × M optical switch,

so the signals on wavelength λ1 are sent from each input port to an optical switch dedicated

only to that wavelength. From the switch to the output port, the wavelengths are recombined

together by a wavelength demultiplexer. This OXC requires M wavelength demultiplexers

and multiplexers, and W M × M optical switches. The OXC can have also local ports,

others than input and output ports, called trunk ports . The implementation of such OXC

requires four major optical function [50,51]:

• space selection: the signal can be routed on a different output port by switching state.

This is useful in case of fault (see Section 2.3.6.1): control signals can activate the

switching mechanism, allowing lightpaths to be routed through a different port. The

failed lightpath can be restored on a different channel, avoiding information losses.

• tunable wavelength-selection: is performed using tunable filters.

• wavelength-conversion: should be performed in the optical domain in order to assure

transparency. The design of a wavelength converter could also embed signal reshaping

47

functionalities,

• multiwavelength amplification: compensates losses accumulated during the routing process.

Experiments on OXC cascadability [49] demonstrate the degeneration of the signal after

each passage. Thus, the number of OXC which can be cascaded should be limited.

2.3.5 Wavelength Routing

Routing strategies aim to fulfill all the traffic demands, between each node pair generating

traffic, using the available network resources. The routing problem is addressed in the optical

layer, i.e., connections between nodes are set up using available wavelengths [52]. The role of

the optical layer is to provide lightpaths to the higher layers, satisfying the traffic demands.

In WDM-optical networks, the same wavelength can not be assigned to two distinct working

connections on any given link but can be reused in other links. This enables many simultaneous

lightpaths to be set up using the same wavelength in the network (wavelength reuse). The

number of wavelengths each link can support depends on the optical components. In WDM

networks, a lightpath can be assigned a single wavelength or several different wavelength as

in Figure 2.11. The left part of Figure 2.11 shows the case when the network nodes are not

λ1

λ1

λ2

λ2λ1

λ2λ1

λ1

λ3

λ3λ2

λ2

C

D AA

B

C

D

B

Figure 2.11. Wavelength routing: without wavelength conversion and with wavelength con-version

equipped with wavelength conversion facilities; a lightpath is assigned the same wavelength

48

for every link on which it flows. So the simple network shown needs at least three wavelengths

to avoid wavelength blocking. The right part of Figure 2.11 shows the same simple network

whose nodes are equipped with wavelength converters; a lightpath does not have to be assigned

the same wavelength on every link it flows on, so only two wavelengths can accommodate the

traffic flows shown. The network nodes should be capable of routing different wavelengths

from an input port to a different output port. In the first case, the intermediate nodes have to

route transparently the lightpaths, without changing their wavelengths. In the second case,

the intermediate nodes must offer wavelength conversion capabilities. The interesting features

of wavelength routing can be listed [47]:

• transparency : lightpaths can carry data using a variety of bit rates, protocols, etc. So the

optical layer can support many different protocol concurrently, because of its protocol

insensitive behavior,

• wavelength reuse: as shown in Figure 2.11 wavelengths can be spatially reused in the

network, therefore two separated lightpaths not sharing any link can be assigned the

same wavelength. Thus, even if the number of wavelengths available is limited, the

number of feasible lightpaths can be larger than the number of available wavelengths

per fiber,

• reliability : in the event of a component failure, the lightpaths can be rerouted over

alternative paths automatically. Many optical components are passive. This provides a

high degree of reliability in the network,

• virtual topology : consists of a graph with links between nodes connected by a lightpath.

Figure 2.12 is the virtual topology drawn out from Figure 2.11. The virtual topology is

49

the graph topology higher layers see,

B

C

A

Figure 2.12. Virtual topology drawn from the physical topology in Figure 2.11

• circuit switching : the lightpaths can be set up and shut down according to the traffic

demand. This behavior looks like circuit-switching networks but at higher bit rates.

Current WDM technology does not allow implementation of the packet switching func-

tionality in the optical layer, however, higher layers, such as ATM, MPLS, or IP, can

perform it.

The implementation of the wavelength routing technique requires two selections:

• which route the lightpaths follow, i.e., the link upon which they lay (routing problem),

• which wavelength/wavelengths should be assigned to the lightpaths (wavelength assign-

ment problem).

An efficient use of the available resources implies a correlation between the two problems. To

reduce the complexity of the whole problem the wavelength routing function is done indepen-

dently of the wavelength assignment. When wavelength conversion capability is available, the

separation is achieved by introducing wavelength conversion capabilities at node level. Wave-

lengths are, therefore, assigned on a link basis, and the same wavelength is reusable on all

links. Wavelength conversion within the node resolves blocking due to wavelength conflicts.

50

Routing strategies can be divided into static or dynamic. In the static case, routing paths

are fixed, and the hardware is configured, i.e., the wavelength converters are fixed, during the

set up process of the network. Dynamic routing, also called reconfigurable routing, requires

setting up a connection and using the available resources whenever a new traffic demand

appears. In reconfigurable networks, switches and/or dynamic wavelength converters are

necessary to change the set of lightpaths. Static networks are expected to be more economical

than a reconfigurable one because they can be built of passive components. Moreover, passive

components make the network more reliable than reconfigurable networks that need active

components inside. The problem is that, in most networks, the traffic demands are not known

a priori or change with time.

2.3.6 Survivability

WDM technique has been primarily used to increase point to point transmission capability.

Fiber optics supporting WDM technology can carry a large amount of information on each

channel. In such a system, network failures cause a loss of data much larger than the loss in a

traditional network. A failure interrupts a large number of communication sessions in progress

and, as a consequence, a large amount of information is lost. Unfortunately, these failures, such

as cable cuts, power outages, component failure, software problems, central offices fires, etc.,

are common in telecommunication systems. Therefore, a survivable WDM network design is

mandatory. It is required from both the service user perspective, since outages cause revenue

loss, as well as from the network provider prospective, since service disruption causes revenue

loss, asset loss, legal costs, loss of competitive advantage, and credibility [53]. A network is

referred to as survivable if it provides some ability to restore the ongoing connections in the

51

event of a catastrophic failure of a network component, such as a cable interruption or a node

failure. The protection techniques use redundant capacity available in WDM network (spare

capacity) to reroute the traffic in presence of a failure.

2.3.6.1 Failure Scenarios

Network failures can be identified in the following three types [54,55]:

• channel fault : a single channel on a link has been interrupted as a consequence of a

failure of the laser or the receiver designed for the channel or a wire disconnection. This

failure can be recovered by rerouting the traffic from the faulted channel to a spare

channel on the same link,

• link fault : all channels routed on the link are faulted, usually as a result of a fiber cut.

The traffic on the faulted link can be restored using separate protection fibers or using

protection wavelength, as explained in Section 2.3.6.2,

• node fault : hardware in the node has failed, resulting from power outages or catastrophic

events. It is the most severe fault of the three and the most complex to handle.

It is expected that the configuration of future transport networks will present fewer nodes but

longer transmission links. According to this tendency, cable failures are supposed to be more

probable than node failure [56, 57]. Thus, this dissertation concentrates on survivability in

case of link failure.

52

2.3.6.2 Protection Mechanisms

Fiber optic networks are multiple layered structures, each layer serving or being a client of

another, as shown in Figure 2.13.

Lightpaths

Other layers(SONET, ATM, IP ...)

Optical layer

Figure 2.13. Network layered structure

The multi-layer network can be divided into one or more sub-networks consisting of nodes

and links. In layered architecture, the property of any layer can be defined independently of

others, so it is possible to add or change a layer without affecting others. Therefore, layers

can support different protocols and services, such as SDH/SONET, ATM, IP, etc. A layered

structure offers the designer a variety of survivability options. The first decision concerns what

layers should provide the protection mechanism. The protection techniques involve providing

some redundant capacity within the network, used to reroute traffic in the presence of failures.

The choice is between lower or higher layers or multiple layers [58,59]. Survivability provided

for a subnetwork at a given layer protects higher layer subnetworks [53]. The benefits of

providing lower layer resilience are [59, 60]:

• the recovery mechanism is activated immediately, for failures occurring in lower layers,

while higher layers have to wait for the alarms to propagate up from lower layers (unless

higher layers have their own fault detection mechanism),

• fewer, larger blocks of traffic need to be rerouted, resulting in a easier and faster restora-

53

tion. At a higher layer, a single fault may cause multiple logical failures, making it

sometimes impossible to perform a complete restoration [61],

• higher layers may not be able to ensure sufficient resilience,

• protection mechanisms implemented in lower layers may be more cost-effective.

On the other hand, higher layer resilience:

• can protect service-specific equipment or particular services,

• should be fast if it provides its own fault detection scheme,

• should not need lower layer resilience, and

• requires each layer to have a separate resilience mechanism and, therefore, ability to

coordinate the recovery actions between different layers, is necessary [59].

Each layer may have its own built-in protection mechanism, independent of the protection

mechanisms of other layers. Usually, its implementation does not interact with the protection

mechanism of other layers. In general, protection mechanisms should provide restoration for

layers without a built-in protection-mechanism. Unfortunately, the restoration mechanism

can be activated simultaneously in different layers, resulting in a large number of unnecessary

alarms. Thus, it is desirable to have coordination among protection mechanisms in different

layers. One way is to give different priorities to protection mechanisms in different layers.

Another is to ensure a complete and fast restoration in one layer, before protection mechanisms

of other layers can detect the failure.

54

2.3.6.3 Protection Concepts: Spare Capacity Allocation

The unused capacity, available in optical links, can be assigned for protection purposes, making

the network survivable. Different protection mechanisms are used for simple point-to-point

link: 1 + 1, 1 : 1, and 1 : N protection [47, 62]. In 1 + 1 protection, the traffic is transmitted

simultaneously on two different fibers (usually over distinct routes) from the source to the

destination. The destination node selects from which fiber it receives the incoming traffic.

In case of a fiber cut, the destination node has to switch over to the other fiber to keep on

receiving data. This protection mechanism is very fast and does not require any signaling

protocol between the two network end nodes. In 1 : 1 protection, there are still two separate

fibers between the end nodes. In this case, the transmission takes place only on one fiber, the

working fiber, avoiding 3 dB loss for signal splitting. In case of a fiber cut, both nodes have

to switch to the other fiber, the protection fiber. In a unidirectional communication system,

fiber cuts can be detected only by the destination node, so a signaling protocol is necessary

to inform the source node of the fault. This protocol is called APS (Automatic Protection

Switching) and does not allow a restoration as fast as the 1 + 1 protection mechanism. In

bidirectional systems, no protocols are required because both source and destination nodes

can test the possible failures. In 1 : 1 protection, the protection fibers are unused, so they

can be used to carry low priority traffic. In case of a failure, this traffic will be preempted

to activate the protection. With the 1 : N protection scheme the same protection fiber can

support different protection, i.e., N working fibers share the same protection fibers. Only

single fiber failures can be protected while, in the event of multiple failure, the restoration

is not guaranteed. Spare capacity allocation can be dynamic or static. Static protection is

foreseen at the network configuration time. Dynamic restoration finds the restoration path

55

when the failure is detected. Despite the better use of spare resources, dynamic restoration is

relatively slow, since more information and messaging needs to be exchanged in order to find

the alternative routes.

2.3.6.4 Resilience Techniques

There are two ways of protecting traffic: path switching and line switching [47]. The path

switching mechanism recovers the disrupted traffic by rerouting the interrupted traffic along

a different path between the source and destination nodes. Therefore, each node pair requires

another link disjoint path to reroute the traffic in case of a link failure, as shown in Figure 2.14.

In a path restorable networks, it is advantageous to release the surviving portions of a failed

Sourcenode

Destinationnode

Path

Path switching

Figure 2.14. Path switching

working path and make those links available for restoration. In this case, the path switching

restoration technique is defined with stub release [63], otherwise, without stub release. The

second option requires link disjoint paths, and the restoration technique can start when the

failure is detected, even without knowing the exact location of the failed link but only knowing

which are the working lightpaths affected by such a failure.

On the other hand, line switching protection is initiated by the end nodes of a failed link,

56

rather than by the source/destination nodes. The line switching implementation might be

span protection or line protection. Span protection switches the traffic from the failed cable to

another cable on the same link, like in Figure 2.15. Line protection reroutes the traffic along

Sourcenode

Destinationnode

Path

Line switching:Line switching:span protection

Figure 2.15. Line switching: span protection

another path between the end nodes of the failed cable, as in Figure 2.16. Path restoration

Sourcenode

Destinationnode

Path

Line switching:Line switching:line protection

Figure 2.16. Line switching: line protection

is more complex than line restoration, because it requires finding a set of protection paths

between numerous source and destination pairs. However, this technique allows minimization

of the spare capacity requirements, especially when the stub release path restoration is per-

57

formed [63]. A modified version of line and path protection is the ring protection scheme [64].

In case of failure of any link belonging to a predefined ring1, the restoration path is routed

through the opposite direction of such a ring. Two different restoration techniques can be

activated: path restoration or line restoration, both inside the ring, as shown in Figure 2.17.

In case of path switching, the back and forth propagation is avoided. The nodes delimiting the

Destinationnode

Sourcenode

Path

Path restoration

Line restoration

Figure 2.17. Ring switching

failed link signal to all the nodes belonging to the ring that a failure occurred. Line switching

is easier and faster, because the switching mechanism is activated immediately, as soon as

the nodes detect the failure and without any further signaling, like in Figure 2.18. But, the

restoration paths will be longer because of the back and forth propagation. This results in a

wasteful bandwidth use.

2.4 Previous Work on Protection Switching Schemes

In this section, we review the existing work in the field of protection switching, with particular

emphasis on solutions proposed for optical networks.

1A ring or cycle is a undirected closed path with no repeated nodes other than the first and the last one.

58

Destinationnode

Sourcenode

Line restoration

Path

Figure 2.18. Line protection in ring switching

In recent years, several studies addressed the problem of optimal resource allocation in

optical survivable networks2.

The resources under optimization span from the total bandwidth provisioned to more

complex cost functions that take into consideration the number of expensive optical devices,

such as optical crossconnects (OXC’s) and optical add-drop multiplexers (OADM’s).

The protection techniques reviewed here include Dedicated-Path Protection (DPP), Shared-

Path Protection (SPP), Shared-Line Protection (SLP), and ring protection. Various approach-

es have been proposed to implement each type of protection technique [56,57,65–75].

Among the first studies, [65] compares three different approaches to achieve path pro-

tection, namely, Minimal Wavelength (MW), Disjoint Path (DP), and restoration on Single

Line Basis (SLB). The optimization of the total number of wavelengths is carried out using

a shortest path approach applied separately to the working lightpaths and to the protection

wavelengths.

The efficient use of spare resources in path protection techniques is demonstrated in [56,

2A network is referred to as survivable if it provides ability to restore the ongoing traffic in case of acatastrophic network element failure.

59

57]. The problem of optimizing cable, fiber, and wavelength costs is described using an

Integer Linear Programming (ILP) formulation [76], which is solved using both an ILP solver

and a Simulated Annealing (SA) algorithm [77]. The optimal network design problem is

approximated in two steps. First working lightpaths are set up, then spare protection resources

are provisioned.

An approach that jointly optimizes working and protection resources, while guaranteeing

protection for any single line failure, is presented in [66]. The problem is formulated as an ILP

task that aims to optimize the number of fibers and the number of OXC’s. Only a limited

number of candidate working/protection path pairs is considered in the combinatorial problem

to reduce the solution space of the ILP problem.

In [67], DPP, SPP, and SLP are compared with the objective of minimizing the total

number of wavelengths in the network, counting both working and backup (protection) paths.

An ILP formulation of the problem is proposed to achieve full traffic protection against any

single line failure. In the optimization process, only a pre-selected set of working and backup

paths is used. The paper shows that SPP yields significant savings in terms of capacity

utilizations over DPP and SLP, while DPP and SLP show similar behavior to each other.

In [68], SPP, and SLP are considered under different network architecture constraints.

OXC’s with and without wavelength conversion capability are considered (no optimization

is performed on the selection of the OXC type at each node). An exact ILP formulation

is presented. Heuristic algorithms are provided to sequentially solve the optimization sub-

problems, i.e., first working resources are allocated, then backup bandwidth is assigned to the

working connection.

Another proposed protection technique superposes a set of logical rings over a mesh net-

60

work [69]. In each logical ring a SONET-like protection mechanism [47, 62] is used to make

the mesh survivable. Resource optimization is done sequentially. First the working resources

are set up, then the backup resources are provisioned within each ring.

In [70–72], the problem of designing optimal survivable networks based on WDM bidirec-

tional line switched rings is carried out using both ILP and Simulated Annealing approaches

that jointly solve the resource assignment for, both, working and protection wavelengths.

A different approach is presented in [78]. The design process requires three steps: finding

all the possible rings in the given network, selecting an appropriate subset, and distributing

the traffic over those rings. Although interrelated, the three tasks are solved sequentially. The

solution is based on the simplifying assumption that every node pair exchanging traffic can

be connected using one ring only.

The first attempt to define varying degrees of QoS at the WDM layer, can be found

in [73–75,79]. In [79] the concept of QoS wavelength routing is introduced, in which the path

and the wavelength(s) chosen for an optical connection are dependent upon the requested

QoS for the connection and the physical characteristics of the optical devices. For example,

in the same fiber, distinct wavelengths may have different Optical Signal to Noise Ratios (OS-

NRs), depending on the wavelength spectral position and transmitted power. High priority

connections will be routed using only wavelengths whose OSNR is high. Low priority connec-

tion may be routed using any wavelength, but possibly a wavelength whose OSNR is low, in

order to provide availability of precious wavelengths with high OSNR to serve future traffic

demands. In [73] the QoS wavelength routing concept is extended to take into account connec-

tion restoration in the optical layer. If interrupted by a fault, an optical connection will have

to be rerouted using spare resources that can provide the necessary QoS level. For example,

61

only spare wavelengths whose OSNR is high can be used to reroute a high priority connec-

tion. The so called QoS-restoration can therefore be defined as the capability to guarantee

restoration to a particular service without violating its QoS routing requirements. The concept

of QoS-restorability is used in [74] to identify various service-specific restoration techniques

that may coexist in the same optical layer. In [73] different restoration schemes are applied

in distinct failure scenarios: fiber fault, single wavelength fault, multi-wavelength fault. In

this study the network performance is characterized using the per-service restorability3, the

per-service blocking probability4, and the resource utilization efficiency5.

[80] provides some analysis and simulation results to compare the tradeoff between cost

and availability that result from using DPP or SPP.

In [81] QoS is introduced as a survivability requirement. Three classes of traffic demands

are envisioned. The first class uses the DPP protection scheme, the second uses the SPP

scheme, and the last class is not protected. While this approach is a first step toward reliability

differentiation, class are differentiated using different protection schemes according to the

results in [80]. It must be observed that using service-specific (multiple) restoration techniques

as proposed in [73,81] may complicate and delay the protection management process.

In order to provide multiple reliability degrees (or classes) at the same network layer

using a common protection mechanism e.g., path switching, the authors in [82] introduced

the concept of Differentiated Reliability (DiR). By means of the DiR concept, a network can

be designed to provide multiple degrees of reliability and efficiently satisfy the user-specific

3Restorability is defined as the ratio between the successfully restored services and the number of attemptsto restore a service.

4Blocking probability is defined as the ratio of unsuccessfully routed paths to the total number of requeststo route a path.

5Resource utilization efficiency is defined as the ratio between the number of wavelengths allocated to theworking paths and the total number of allocated wavelengths in the network.

62

requirements, yet minimizing the network total cost. In [83], the DiR concept is applied

— using a DPP protection scheme — to designing the WDM layer of a ring network in

which wavelength conversion is not available. To solve the routing and wavelength assignment

problem, an efficient algorithm is proposed that resorts to reusable protection wavelengths

while guaranteeing the required reliability degree of each connection. [84] extends the DiR

concept to a SPP protection scheme applied to an arbitrary (mesh) topology. A two-step

algorithm based on Simulated Annealing is proposed. The algorithm searches for both primary

and backup paths under two assumptions: (i) only one link can fail at once and (ii) the set

of connection demands that must be routed across the network are known a priori. Each

demand is assigned the desired level of reliability while minimizing the required network

resources. In [85] the authors extended the DiR concept to restoration schemes in which

network resources for a disrupted connection along secondary paths are sought in the event of

a failure. In [86, 87], the DiR concept based on a SPP scheme has been proven to be feasible

on all-optical test-bed.

CHAPTER 3

ARQ-C: A CLASS OF COOPERATIVE ARQ PROTOCOLS

3.1 Introduction

Networks of wireless integrated sensors are often used to monitor distributed parameters in

the environment. These parameters are related to a variety of applications, such as securi-

ty, patient monitoring, chemical and biological hazard detection [4]. The benefits of using

wireless sensor networks include reduced installation costs, ability to rapidly reconfigure the

data acquisition procedure, and safe deployment in hostile environments [7, 9, 12, 13, 88, 89].

Networked sensor nodes can jointly perform large sensing tasks, thus greatly improving the

accuracy and scope of the information provided to the user.

Some solutions rely on sensor nodes that are powered up by limited life-time batteries,

which are periodically replaced to provide long-term sensor operation. Others envision short

transmission range sensors (2-10 m), which harvest their energy from various environmental

sources (e.g., solar, vibrations, acoustic noise). These maintenance-free solutions — examples

are the PicoRadio project at Berkeley, the µAMPS project (with base-station) at MIT, and

the WSSN project at ICT Vienna — aim at low cost sensor nodes densely deployed across

the area of interest. The foreseen power dissipation level at the sensor node is in the order

of 100 µW. At these power levels, it may be possible to energy-scavenge or harvest directly

from the environment [6,8]. To cope with the relatively short transmission range, ad-hoc and

multi-hop networking is envisioned.

63

64

The Generic Autonomous Platform for Sensors (GAP4S) project [15] explores an approach

to wireless sensors that is complementary to the above and other existing solutions. GAP4S is

suitable for those applications in which the energy harvesting from the environment is neither

possible, nor efficient, nor sufficient. The required power is provided by a base-station that

remotely recharges the sensor node on-board battery via a microwave (MW) signal. For the

purpose of both recharging from and transmitting directly to (single-hop) the base-station,

the sensor node must be located inside the base-station footprint. The base-station — which

may be mobile — constitutes the access point to a wider communication network, e.g., the

Internet.

A challenge common to GAP4S and other wireless sensor network solutions is to maximize

the saturation throughput given the energy budget constraint at the sensor node.

The objective in this chapter is to maximize the sensor network saturation throughput

while offering a reliable and fair single-hop solution in GAP4S. The term reliable is used to

indicate that an automatic repeat request (ARQ) protocol is used to cope with transmission

errors. The term fair is used to indicate that all the sensor nodes must be able to transmit an

amount of data that is proportional to a reference value. Throughput is maximized by carefully

choosing the transmitted energy at the sensor node once the ARQ protocol is selected.

Two classes of ARQ protocols are considered. The first is the conventional class of ARQ

protocols, whereby the (sensor node) source retransmits its own data frames until they are

successfully delivered to the base-station. The second class takes advantage of cooperative

radio communications [22, 23, 26–35]. In this second class of ARQ protocols, one or more

relays (i.e., selected sensor nodes) assist the source during the retransmission process. In

simple terms, the relay — instead of the source — is requested to retransmit the data frame

65

to the base-station when the earlier transmission attempt made by the source is not successful.

With this retransmission mechanism, it is as if sensor nodes could borrow energy from one

another and balance their energy consumption to match their own battery recharging rate. In

turn, a balanced energy consumption-to-recharge rate ratio has the potential to improve the

network throughput. A second advantage of using cooperative ARQ protocols derives from

their inherent two-hop retransmission mechanism. If the relay is located between the source

and the base-station, data frames are (re)transmitted over a shorter range, thus reducing the

required amount of transmitted energy — a known advantage from multi-hop networking.

For the cooperative ARQ protocol, relays must be chosen for each source to best balance

energy usage at the sensor nodes. This problem is formulated in its most general form —

i.e. allowing multiple relays to be assigned to the same source — using Linear Programming

(LP). A special subcase is obtained by constraining each source to make use of one relay only,

i.e., the one-relay ARQ protocol. For the one-relay subcase an approximate greedy solution

is presented that is based on a sorting function applied to choosing the relay for each source.

The saturation throughput of both cooperative and non-cooperative ARQ protocols is

discussed for a number of network scenarios, e.g., varying sensor node transmission power,

footprint size, path loss exponent, and radio frequency used on the uplink channel. Results

indicate that the radius of the base-station footprint may range up to hundred meters. It

is confirmed that introducing cooperation in the ARQ protocol enables the sensor node to

transmit data efficiently at lower energy levels. Under constrained energy budget at the sensor

node, higher saturation throughput can thus be achieved when compared to non-cooperative

ARQ protocols. The one-relay ARQ protocol is shown to be a good tradeoff between network

throughput and protocol complexity.

66

3.2 Two Classes of ARQ Protocols

Two classes of ARQ protocols are considered to overcome transmission errors on the uplink

channel of GAP4S. Transmission errors may be frequent on the uplink channel as the sensor

node power budget limits the SNR value. Transmission errors on the downlink MW channel

are not a concern, due to the relatively high power MW signal, and can be overcome with a

timeout-triggered retransmission at the base-station.

The considered ARQ protocols take into account the unique nature of the GAP4S, whereby

the base-station — characterized by non-stringent power constraints — remotely recharges the

sensor nodes and schedules uplink transmissions via the MW downlink control channel. Sensor

nodes simply obey the control frames received from the base-station. Available options for the

two classes of protocols are stop and wait, go back N, selective repeat, etc., (Section 2.1.5).

3.2.1 The ARQ-NC Protocol

The ARQ-NC is the class of (conventional) non-cooperative protocols. Upon request from

the base-station, the sensor node transmits its data directly to the base-station. The data

frames are encoded to provide error detection and, optionally, correction capabilities at the

base-station. Upon the successful reception of a data frame, the base-station replies with a

positive ACK frame. Upon reception of a data frame which contains errors that cannot be

corrected, the base-station sends a NAK frame to the sensor node. In turn, the sensor node

retransmits the data frame until reception at the base-station is successful.

67

3.2.2 The ARQ-C Protocol

The ARQ-C is the class of cooperative protocols that take advantage of the broadcast nature

of the uplink channel to reach the base-station via spatial diversity. Figure 3.1 depicts how

the ARQ-C protocol works. Upon the unsuccessful reception of a data frame transmitted by a

sensor node (i.e., the source), the base-station requests that the data frame be retransmitted by

another “cooperating” sensor node (i.e., the relay). To accomplish this task, the optional RX

section in the radio transceiver module shown in Figure 1.3 is required at each sensor node. The

assumption here is that the relay may have successfully overheard the data frame transmission

performed by the source earlier and stored a copy of that data frame temporarily. As already

mentioned, a wisely chosen relay may increase the probability of successfully delivering the

data frame to the base-station without requiring additional retransmission attempts.

Base−station

Relay

Source

Figure 3.1. ARQ-C protocol: cooperation between two sensor nodes

In the ARQ-C protocol simplest version, the base-station broadcasts the following infor-

mation: the identifier of the source that must transmit next, the data frame to be transmitted,

and the identifier of the sensor node that acts as the relay for this data transmission. If the

68

data frame transmission from the source to the base-station is successful, the base-station

begins a new transmission cycle, and the relay discards the temporary data frame copy. If

the transmission to the base-station is unsuccessful, the relay is invited to retransmit the data

frame. If the data frame transmission from the relay to the base-station is successful, the

base-station begins a new transmission cycle.

Of course, the relay can help only when it has correctly overheard the data frame trans-

mitted by the source. If the base-station does not hear from the relay, it safely assumes that

the transmission of the data frame from the source to the relay was not successful. It is

also possible that the relay retransmission attempt to the base-station is unsuccessful. Under

either circumstance, the base-station begins a new cycle by requesting the source to transmit

the data frame again.

The relay offers both spatial diversity and its own energy to help with the source’s data

frame transmission. For improved load and energy consumption balancing, multiple sensor

nodes may be chosen to act as relay for the same source. Likewise, the same sensor node may

act as relay for multiple sources. When multiple relays are assigned to the same source, it is

assumed that only one of them will be “active” during one transmission cycle. This choice

eliminates unnecessary reception energy consumption at the relays. In practice, the base-

station selects the active relay for the next data frame transmission proportionally to some

predefined distribution values. Note that the required intelligence is still entirely residing at

the base-station. Sensor nodes are simply ordered to overhear and transmit by the base-station

via the MW control channel.

It must be noted that the ARQ-C protocol solution just described has some interesting

similarity with multi-hop solutions. However, these two solutions should not be confused. In

69

fact, the latter is a layer 3 solution, which requires routing protocols at the sensor nodes. The

former is a layer 2 solution, whereby the base-station determines which sensor node makes

the next (re)transmission attempt.

3.3 Assessing Saturation Throughput

The two classes of ARQ protocols are compared using saturation throughput as the metric of

interest. The saturation throughput is defined as the maximum offered load — or data frame

generation rate — that can be sustained by the network under two constraints:

• the average energy consumption does not exceed the energy recharge rate at each sensor

node,

• data frames are generated at each sensor node proportionally to a given reference value

and cannot be dropped.

It is assumed that other system parameters, e.g., transmission rate, wireless channel and

electronics bandwidth, buffer capacity at the base-station and sensor nodes, network latency,

QoS, etc., do not limit the network throughput. The sensor nodes and the base-station are

stationary. The sensor node battery is ideal, i.e., linear recharge rate and unbounded storage

capacity.

The problem of maximizing the offered load under the two previously stated constraints

can be formulated using a flow model that relies on the following input parameters and vari-

ables.

70

Input:

• N : the set of sensor nodes,

• N = |N |: the number of sensor nodes,

• PBS: recharge power radiated on the MW channel at the base-station,

• P reci : amount of recharge power that reaches the battery of sensor node i,

• di,j: distance between sensor node i and j,

• di,BS: distance between sensor node i and the base-station,

• Ebi: energy per bit radiated by sensor node i,

• E(Rx)b : energy necessary to receive one bit at the relay,

• L: number of bits per data frame,

• ξ: the transmission energy overhead at the sensor node, defined as the ratio between

the energy consumption and the radiated energy Ebi,

• P(e)(i,j): probability of unsuccessful reception at sensor node j of a data frame sent by

sensor node i,

• P(e)(i,BS): probability of unsuccessful reception at the base-station of a data frame sent by

sensor node i,

• λi: normalized offered load at sensor node i.

71

Variables:

• S: network saturation throughput,

• S · λi: offered load1 at sensor node i,

• p(i,j): probability that the base-station chooses sensor node j to be the relay for sensor

node i,

• λ(i,j): normalized rate of data frames transmitted at source i while having relay j active,

• λ(i,j): normalized rate of data frames retransmitted by relay j on behalf of source i,

• λ∗i : normalized data frame transmission rate at sensor node i counting both transmission

and retransmission attempts.

3.3.1 ARQ-NC Protocol

S λ

λ

λ∗

P

ii

(e)

(i,BS)(i,i)

Figure 3.2. ARQ-NC protocol: flow model

By definition, in the ARQ-NC protocol p(i,i) = 1 ∀i ∈ N and p(i,j) = 0 ∀i 6= j ∈ N . Fig-

ure 3.2 shows the flow model of the ARQ-NC protocol. The figure represents the transmission

queue at sensor node i. With probability P(e)(i,BS) the data frame transmission is not successful,

1This definition of the offered load permits to model the traffic intensity at every sensor node using a singlevariable, i.e., S.

72

in which case the data frame remains in the transmission queue for the next transmission

attempt.

The problem of maximizing the saturation throughput (S) can be formulated as follows:

max : S (3.1)

subject to:

(λ∗i · L · Ebi· ξ) ≤ P rec

i ∀i ∈ N (3.2)

λ∗i =S · λi

1− P(e)(i,BS)

∀i ∈ N . (3.3)

The constraints in (3.2) ensure that the recharge power at each sensor node is sufficient to

sustain the total number of (re)transmission attempts.

3.3.2 ARQ-C Protocol

Figure 3.3 shows the flow model of the ARQ-C protocol. For simplicity, only two sensor nodes

are shown. N + 1 queues are used at each sensor node. Sensor node i has one transmission

queue for its own data frames and a dedicated queue to contain the data frame copies received

from source j ∈ N . Note that j = i represents the case in which source i does not have a

relay. Having separate queues in the figure simplifies the description of the flow model but is

not strictly necessary in the protocol implementation.

Upon the occurrence of an unsuccessful transmission at source i, flow λ∗i is split to reach

multiple relays. Then, p(i,j) = λ(i,j)

λ∗i P(e)(i,BS)

. Recall that sensor node j can act as relay only when the

transmission from sensor node i to sensor node j is successful. Then, λ(i,j) =(1− P

(e)(i,j)

)·λ(i,j).

The data frame transmission rate at sensor node i is then equal to λ∗i +∑

j∈N λ(j,i).

73

λ 1

∗

λ 2

λ 1

λ 2

P(1,BS)

(e)

P(1,BS)

(e)

P(1,BS)

(e)

P(2,BS)

(e)

P(2,BS)

(e)

P(2,BS)

(e)

1− P(1,2)

(e)

P(1,2)

(e)

λ (1,2)

λ (1,1) λ (1,1)

λ (2,1)

λ (2,2)

λ (1,2)

1− P(2,1)

(e)

λ (2,2)

λ (2,1)

λ (1,2)

∗

S

S

( )

( )

Figure 3.3. ARQ-C protocol: flow model

The amount of power dissipated at relay j to overhear data frames transmitted by source

i is proportional to p(i,j) · λ∗i =λ(i,j)

P(e)(i,BS)

. The total amount of power dissipated at relay j to

overhear data frames transmitted by all sources is proportional to∑

i∈N ,i 6=j

λ(i,j)

P(e)(i,BS)

. Note that

the amount of power dissipated while overhearing is irrespective of the successful reception at

the relay.

74

The problem of maximizing the saturation throughput (S) can be formulated as follows:

max : S (3.4)

subject to:( ∑

j∈N ,j 6=i

λ(j,i)

P(e)(j,BS)

)· L · E(Rx)

b +

(λ∗i +

∑j∈N

λ(j,i)

)· L · Ebi

· ξ ≤ P reci ∀i ∈ N (3.5)

λ∗i = S · λi +∑j∈N

(P

(e)(i,j) · λ(i,j)

)+

∑j∈N

(P

(e)(j,BS) · λ(i,j)

)∀i ∈ N (3.6)

λ(i,j) =(1− P

(e)(i,j)

)· λ(i,j) ∀i, j ∈ N (3.7)

∑j∈N

λ(i,j) = P(e)(i,BS) · λ∗i ∀i ∈ N . (3.8)

The expression in (3.5) balances the total energy used to transmit and overhear data frames

with the energy received from the base-station. The expression in (3.6) represents the nor-

malized transmission rate of its own data frames at sensor node i. This is the sum of three

terms: new data frames, data frames that the base-station designated to be retransmitted by

relay j but were not successfully received at relay j, and data frames that source i has to

retransmit because the retransmission attempt made by relay j was not successful. Expres-

sions in (3.7) and (3.8) are flow conservation constraints. The expression in (3.8) ensures that

λ(i,j) ≤ P(e)(i,BS) · λ∗i , i.e., p(i,j) ∈ [0, 1].

3.4 One-Relay ARQ Protocol

The formulation of the maximum saturation throughput for the cooperative ARQ protocol

(Section 3.3.2) makes the assumption that multiple sensor nodes may be relays for the same

75

source. It may be advantageous to simplify the cooperative ARQ protocol by limiting the

number of relays per source to one. The LP problem in Section 3.3.2 can be modified to be

an integer (ILP) problem that takes into account these new constraints, i.e.,

∑j∈N

p(i,j) = 1 ∀i ∈ N (3.9)

M · p(i,j) ≤ λ(i,j) ∀i, j ∈ N (3.10)

p(i,j) ∈ {0, 1} ∀i, j ∈ N (3.11)

where M is a constant that takes a large value. To circumvent the computationally intensive

ILP solution, an approximate greedy solution is proposed. First, a function is introduced to

sort the set of potential relay candidates. Then, the greedy algorithm is described.

3.4.1 Sorting Function

Source traffic

Relay traffic

Relay

Base−stationSource

Figure 3.4. Two sensor nodes network scenario: one source, one relay and the base-station

A criterium must be found to identify the relay that is most suitable to cooperate with

any given source. The simplified network scenario shown in Figure 3.4 is used to construct

76

a sorting function. The network consists of one base-station (BS), one source (s), and one

relay (r). The following assumptions are made:

• when r transmits its own data frames to BS, the transmission is always successful,

• when s transmits its own data frames, the transmission may be unsuccessful,

• upon the unsuccessful data frame transmission from s to BS, r is always the relay chosen

to retransmit the data frame, i.e., p(s,s) = 0, p(s,r) = 1,

• when r acts as a relay for s, its transmission to BS may be unsuccessful,

• s does not act as a relay for r, i.e., p(r,s) = 0,

• s and BS are in fixed locations,

• r may be anywhere in the footprint.

The objective is to define a function that takes, as input, both the coordinates of r and the

value of Ebiand returns the saturation throughput (S) subject to energy budget constraints.

Two cases are considered.

Unlimited energy budget at r. In this case, the limiting factor is the energy budget at s.

The saturation throughput (S(s)s,r ) can be derived from (3.5)–(3.8) as follows:

S(s)s,r =

λ∗s(1− P

(e)(s,r) · P (e)

(s,BS) − P(e)(r,BS) ·

(1− P

(e)(s,r)

)· P (e)

(s,BS)

)

λs

(3.12)

where:

λ∗s =P rec

s

L · Ebs · ξ. (3.13)

Equation (3.12) takes into account three energy consumption factors at s that are proportional

to the following transmission flows: (i) the flow of data frames transmitted to BS, i.e., λ∗s, (ii)

77

the flow of data frames not successfully received at BS nor at r, i.e., λ∗s · P (e)(s,r) · P (e)

(s,BS), and

(iii) the flow of data frames that s must retransmit because the retransmission performed by

r was not successful, i.e., λ∗s · P (e)(r,BS) ·

(1− P

(e)(s,r)

)· P (e)

(s,BS).

Unlimited energy budget at s. In this case, the limiting factor is the energy budget

at r. The saturation throughput (S(r)s,r ) can be derived from (3.5)–(3.12) as follows:

S(r)s,r =

P recr

λs · α−1 · L · E(Rx)b +

[λr + λs · α−1 · P (e)

(s,BS) ·(1− P

(e)(s,r)

) ]· L · Ebr · ξ

(3.14)

where:

α = 1− P (e)s,r · P (e)

(s,BS) − P(e)(r,BS) ·

(1− P

(e)(s,r)

)· P (e)

(s,BS). (3.15)

Equation (3.14) takes into account three energy consumption factors at r. They are pro-

portional to: (i) the energy spent to receive the flow of data frames from s while taking part

in the cooperating process, i.e., λs · α−1 · L · E(Rx)b , (ii) the energy spent to transmit its own

data frames to BS, i.e., λr ·L ·Ebr · ξ, and (iii) the energy spent to transmit the copies of the

data frames successfully received from s, i.e., λs · α−1 · P (e)(s,BS) ·

(1− P

(e)(s,r)

)· L · Ebr · ξ.

Since the only limiting factor in evaluating the saturation throughput is the energy budget

available at both s and r, the saturation throughput (S) is:

S = min{S(s)s,r , S

(r)s,r}. (3.16)

The result in (3.16) is shown in Figure 3.5, where S is plotted as a function of the coordinates of

r. Coordinates of s and BS are (50, 0) m and (0, 0) m, respectively. The following parameter

values are used: λs = λr = 1 packet/s, Ebs = Ebr = 2e−12 J, E(Rx)b = 0 J, ξ = 1, uplink

78

−200−100

0100

200300

−2000

200

0

2

4

6

8

10

12

yx

S

Figure 3.5. Saturation throughput (S) as a function of the position of the relay. The sourceis placed in (50, 0) m and the base-station is placed in (0, 0) m

frequency = 2.4 GHz, downlink frequency = 2.4 GHz, L = 256 bits, power radiated at BS,

PBS = 10 W, and path loss exponent n = 3. Figure 3.6 shows the y = 0 section of the three

dimensional plot in Figure 3.5. The value of the data frame error probability used to obtain

these results is shown in Figure 3.7. Both figures indicate that cooperation is most effective

when r is placed between s and BS.

For comparison, in Figure 3.6, the saturation throughput of s when making use of the

ARQ-NC protocol is reported using a dashed line. It is easy to evaluate SARQ−NC from (3.2)

and (3.3) as:

SARQ−NC =λ∗s

(1− P

(e)(s,BS)

)

λs

=P rec

s

L · Ebs · ξ·

(1− P

(e)(s,BS)

)

λs

(3.17)

79

−200 −100 0 100 200 3000

2

4

6

8

10

12

Relay position

S

SS

ARQ−NC

Base−station Source

Figure 3.6. Saturation throughput (S) as a function of the position of the relay. The sourceis placed in (50, 0) m, the base-station is placed in (0, 0) m, and the y coordinate of the relayis fixed and equal to 0 m

where λ∗s is the total rate of data frames transmitted by s to BS as indicated in (3.13). For

the set of parameter values chosen in this example, cooperation may increase the saturation

throughput up to three times.

3.4.2 The Greedy Algorithm

The function in (3.16) takes into account the retransmission attempts made for one sensor

node only and ignores the other sensor nodes’ retransmission attempts. For example, it does

not capture the practical case whereby one sensor node may act as the relay for two or more

sources.

The greedy algorithm in this section is designed to guarantee that, as relays are assigned to

sources, the energy budget available at the sensor nodes is met, while maximizing the minimum

80

source saturation throughput. In achieving its goal, the algorithm makes use of (3.16) to sort

the relay candidates for every given source.

The algorithm works as follows. Each sensor node s ∈ N is assigned a value d = P recs

λsEbsLE[txs].

E[txs] is the average number of transmission attempts per data frame if s makes use of an

ARQ-NC protocol. The value of E[txs] = λ∗sS·λs

= 1

1−P(e)s,BS

can be derived using (3.2) and (3.3).

Let D be the set of sensor nodes ordered by non decreasing values of d. Note that when all

sensor nodes use the same value of Ebs and λs, the first sensor node in set D is the sensor

node that is the farthest away from the base-station. The algorithm considers one sensor node

at a time following the order in D. For each i ∈ D the algorithm computes Si,j defined as

the saturation throughput of i when j ∈ N is the relay for i. The value of Si,j is derived

using (3.16)2. Sensor node ri such that Si,ri= maxj∈N (Si,j) is chosen to be the relay for i.

The algorithm must consider that a sensor node r may be chosen to be the relay for two

(or more) sources, i.e., r = ri = rj i 6= j. This affects the power budget available at r as

follows. Every time r is chosen to be a relay, its remaining power budget is derived from (3.14)

as follows:

P rec(new)

r = P rec(old)

r − Si,r · λi · L · α−1

[P

(e)(i,BS) ·

(1− P

(e)(i,r)

)· Ebr · ξ + E

(Rx)b

](3.18)

where α is given in (3.15).

In addition, every sensor node i must reserve sufficient power budget for transmitting its

own data frames. At each step, the algorithm updates the remaining power budget at i using

2i = j is the case when the non cooperative ARQ protocol is used to transmit data frames from i to thebase-station. Si,i is derived using (3.17).

81

(3.12) as follows:

P rec(new)

i = P rec(old)

i − Si,rλi(1− P

(e)(i,r) · P (e)

(i,BS) − P(e)(r,BS) ·

(1− P

(e)(i,r)

)· P (e)

(i,BS)

) ·L ·Ebi· ξ (3.19)

where α is given in (3.15).

The algorithm stops after N steps, when one relay ∀i ∈ D is chosen. The complexity of

the algorithm is O(N2).

Table 3.1. Pseudo code of the greedy algorithm

begin algorithmfor(∀ sensor nodes i ∈ D){for(∀ sensor nodes j ∈ N ){

Compute Si,j

}Find ri | Si,j = maxj∈N (Si,j)Si,ri

= Si,j

Update energy available at sensor node rUpdate energy available at sensor node iif(i | di = mink∈D(dk))Sup−bound = Si,ri

}end algorithm

Table 3.1 provides a pseudo code description of the algorithm. The description also indi-

cates how to compute an upper bound for the one-relay ARQ protocol saturation throughput.

The upper bound is based on the observation that sensor node i with the smallest value of

di receives the least amount of MW recharge power and/or has the lowest probability of suc-

cessfully transmitting data frames to the base-station. Based on the saturation throughput

definition given in Section 3.3, S can not exceed the value of throughput achievable by i. S(i,ri)

represents then an upper bound for S, as explained in the next example. Consider sensor node

j 6= i ∈ D. There are two possible cases. 1) S(j,rj) ≥ S(i,ri): in this case i still constrains the

82

value of S. 2) S(j,rj) < S(i,ri): in this case j constrains the value of S. Therefore the value

S(i,ri) is an upper bound for the value of the saturation throughput (S).

3.5 Performance Evaluation

This section reports various saturation throughput results obtained for the two classes of ARQ

protocols in a number of GAP4S scenarios. The saturation throughput is used to determine

under what conditions the ARQ-C protocol outperforms the ARQ-NC protocol and quantifies

the potential gain. Results are presented after a short description of the assumptions made

on the wireless network.

3.5.1 Wireless Network Assumptions

Both path loss and fading are taken into account in the uplink channel. Only path loss is

taken into account in the MW downlink recharging channel.

The path loss is modeled as

Ebr = Ebt ·GT ·GR · λ2

(4π)2(d)n(3.20)

where:

• Ebr , Ebt : energy per bit at the receiver and transmitter, respectively,

• GT , GR: transmitter and receiver antenna gain, respectively,

• d: transmitter-receiver distance,

• λ: wavelength at the channel center frequency,

83

• n: path loss exponent, n = 2 in free space, typically 2 ≤ n ≤ 4 for environments with

structures and obstacles [21, 90].

Fading is assumed to be Rayleigh slow and flat, i.e., the fading coefficients are considered

constant over a single frame transmission (Section 2.1.1.3). The fading experienced by any

given frame transmission is statistically independent of the fading experienced by any other

frame transmission. The instantaneous signal to noise ratio of the uplink channel at receiver

j given a transmission from transmitter i is:

γ(i,j) =Ebr

N0

· α2(i,j) (3.21)

where:

• N0: noise spectral density of the Additive White Gaussian Noise (AWGN),

• α(i,j): a Rayleigh distributed random variable used to model the Rayleigh fading mag-

nitude between transmitter i and receiver j, E[α2(i,j)] = 1, ∀i, j.

It is assumed that the MW downlink channel is error free. On the uplink channel, sensor

nodes send data augmented with a cyclic redundancy (CRC) code. Each block contains B

bits (including the CRC bits). The probability of receiving a data frame incorrectly (error

probability) is a function of both γ(i,j) and the code (if any) used to add redundancy to the

transmitted data. The probability of detecting an erroneous codeword P(block)(i,j) when a coded

data frame, i.e., a codeword, is sent from transmitter i to receiver j is upper bounded by the

following expression [30,91]

ζ = min

1,

∞∑D=Df

a(D) · P (D|γ(i,j))

(3.22)

84

P(block)(i,j) ≤ 1−

∫ ∞

0

(1− ζ)B · p(γ(i,j))dγ(i,j) (3.23)

where:

• B: number of data bits in each block (data plus CRC bits), i.e., number of trellis

branches in the codeword,

• Df : free distance of the code [92],

• a(D): spectrum of the code [40], i.e., number of codewords of weight D,

• P (D|γ(i,j)): probability that a wrong path at distance D is selected,

• p(γ(i,j)): probability density function of the instantaneous SNR.

Binary PSK with soft decoding is employed, i.e.,

P (D|γ(i,j)) = Q(√

2 ·D · γ(i,j)) (3.24)

where Q(·) is the Marcum Q function [93] and D is the weight of the codeword.

It is assumed that the CRC can detect all erroneous codewords. Thus, the data frame

error probability, i.e., the probability that retransmission of a data frame is required is P(e)(i,j) =

P(block)(i,j) .

The following parameter values are used. Data frames have fixed length and carry B = 128

bits of combined data and CRC, which are encoded into 256 bit codewords using a rate-

compatible punctured convolutional code (RCPC) with rate 1/2, parent code rate of 1/4,

puncturing period of 8, and memory of 4 [40]. The frame error probability P (e) for this RCPC

85

−20 −10 0 10 20 30 4010

−5

10−4

10−3

10−2

10−1

100

SNRrec

[dB]

P(e

)

Figure 3.7. Error probability (P (e)) as a function of SNRrec

versus SNR at the detector stage of the receiver, SNRrec, is shown in Figure 3.7. SNRrec

accounts for both Ebr/N0 and the receiver noise figure F . It is assumed that F = 5 dB.

The values for the transmitting and receiving antenna gain at the base-station and at every

sensor node are assumed to be GT = GR = 1. The recharge power constantly radiated at the

base-station is PBS = 10 W. It is assumed that, at the sensor node, the energy received from

the MW channel by the antenna is fully transferred into its battery and circuitry losses are

negligible.

The effects of various energy consumption factors at the sensor node, i.e., analog-digital

conversion, processing, power management, receiver and coding are all taken into account and

combined using parameter ξ introduced in Section 3.3. Consistent with [94], ξ = 300. Traffic

is uniform, i.e., λi = 1, ∀i ∈ N .

The formulations presented in Section 3.3 are solved using LP Solve 5.5.0.3 [95].

86

The radiated energy per bit Eb, i.e., the value of energy that is radiated by a sensor node

to transmit one bit, is assumed to be the same at each sensor node. Unless otherwise specified,

the energy necessary to receive one bit when a sensor node acts as a relay is E(Rx)b = 30 nJ [96].

Multiple experiments are carried out by randomly placing 300 sensor nodes within a circular

footprint of radius R according to a uniform distribution. The base-station is at the footprint

center. The presented averages are obtained ensuring a confidence interval value of 8% or

better at 98% confidence level.

10−12

10−11

10−10

10−3

10−2

Eb [J]

S

ARQ−C (300)ARQ−C (1)ARQ−NC

Figure 3.8. Saturation throughput (S) versus energy per bit radiated by the sensor nodes

(Eb). R = 50 m, G = 1, ξ = 300, F = 5 dB, E(Rx)b = 30 nJ, n = 3

Figure 3.8 plots the saturation throughput (S) versus the radiated energy per bit (Eb).

ARQ−C(300) refers to the case of the cooperative ARQ protocol with no constraints on the

number of relays per source. ARQ− C(1) refers to the one-relay cooperative ARQ protocol.

All curves show a similar pattern, with a maximum at some intermediate value of Eb. High

87

10−12

10−11

10−3

10−2

Eb [J]

S

Sup−bound

, Eb(rx)=0.3n

ARQ−C (300), Eb(rx)=0.3n

ARQ−C(1), Eb(rx)=0.3n

ARQ−C (300), Eb(rx)=30n

Sup−bound

, Eb(rx)=30n

ARQ−C(1), Eb(rx)= 30n

Figure 3.9. Saturation throughput (S) versus energy per bit radiated by the sensor nodes(Eb). R = 50 m, G = 1, ξ = 300, F = 5 dB, n = 3

values of Eb increase the SNR and the probability of successful transmission but consume too

much energy. In this case, all ARQ protocols perform similarly, as retransmission attempts are

sporadic. Low values of Eb save energy but require too many retransmission attempts. In this

case, the ARQ protocols behave significantly different from one another. The ARQ-C protocols

achieve their maximum S at slightly lower Eb values than the ARQ-NC protocol. The reason

for this is the combined effect of energy borrowing and two-hop retransmission of the ARQ-C

protocols. Sensor nodes, whose energy consumption-to-recharge rate ratio is not favorable,

can borrow energy from other (richer in energy) sensor nodes by asking them to perform

data frame retransmissions. At the same time, the transmission energy level at the sensor

node may be lowered without compromising throughput due to the shorter transmission range

offered by the (two-hop) relay retransmission mechanism. Note that lowering the transmitted

energy level has the initial effect of increasing the amount of retransmissions performed by the

88

relay, i.e., more energy is borrowed from the relay. Further lowering the transmitted energy

level compromises the ability of the relay to help, causing a sharp throughput decay. Finally,

the plots indicate that when the energy amount available at the sensor node is the limiting

factor in the system, the cooperative ARQ protocol saturation throughput may be more than

twice the saturation throughput of the non-cooperative ARQ protocol. In addition, limiting

the number of relays per source to one is a good tradeoff between network throughput and

protocol complexity.

Figure 3.9 shows the effect of E(Rx)b on S. Two values are considered, i.e., E

(Rx)b = 30 nJ

and E(Rx)b = 0.3 nJ. As intuition suggests, S grows with decreasing E

(Rx)b . The figures also

shows the upper bound for ARQ− C(1) derived in Section 3.4.2. Note that the performance

gap between ARQ− C(300) and ARQ(1) decreases with decreasing E(Rx)b .

Table 3.2 documents the effect of the uplink frequency (fup), the footprint radius (R),

and the path loss exponent (n) on S for the two classes of ARQ protocols. The S values

reported in the table are obtained using the value for Eb that yields the maximum S in

each case. Results in the top part of the table are obtained using an uplink frequency fup=

2.4 GHz. Results in the middle part of the table are obtained using an uplink frequency fup=

916.5 MHz. Results in the bottom part of the table are obtained using an uplink frequency

fup= 433.92 MHz. Depending on the surrounding environment — which may affect the value

of n — the practical size of the footprint spans from tens to hundred meters. Even though the

ARQ-C protocol always yields higher saturation throughput values than the ones achieved by

the ARQ-NC protocol, noticeable performance differences are found only in those cases when

the direct transmission of data frames from the sensor node to the base-station is not favored

by the surrounding environment, e.g., for increasing values of fup, R and n.

89

Table 3.2. ARQ-C protocol: saturation throughput (S) as a function of the uplink frequency(fup), footprint radius (R) and path loss exponent (n)

fup = 2.4 GHz, GT = GR = 1, E(Rx)b = 30 nJ, ξ = 300, F = 5 dB

n = 3 R = 50 m

R = 10 m R = 50 m R = 100 m n = 2 n = 3 n = 4

ARQ-NC 64.297 4.1179E-3 6.4297E-5 10.208 4.1179E-3 1.6361E-6

ARQ-C(1) 111.66 1.1199E-2 2.4269E-4 11.733 1.1199E-2 1.1332E-5

ARQ-C(300) 123.72 1.5117E-2 2.5264E-4 15.121 1.5117E-2 1.1332E-5

fup = 916.5 MHz, GT = GR = 1, E(Rx)b = 30 nJ, ξ = 300, F = 5 dB

n = 3 R = 50 m

R = 10 m R = 50 m R = 100 m n = 2 n = 3 n = 4

ARQ-NC 440.05 2.7284E-2 4.4005E-4 69.688 2.7284E-2 1.1258E-5

ARQ-C(1) 753.91 5.5398E-2 1.2840E-3 70.912 5.5398E-2 6.5263E-5

ARQ-C(300) 769.34 7.4938E-2 1.6419E-3 82.343 7.4938E-2 7.8625E-5


n = 3 R = 50 m

R = 10 m R = 50 m R = 100 m n = 2 n = 3 n = 4

ARQ-NC 1948.2 1.2587E-1 1.9481E-3 314.62 1.2587E-1 4.8166E-5

ARQ-C(1) 2470.1 2.2278E-1 4.0794E-3 318.23 2.2278E-1 2.5434E-4

ARQ-C(300) 3204.1 2.7221E-1 5.4379E-3 342.66 2.7221E-1 3.2046E-4

90

Similar results are found under non uniform traffic patterns, which are studied extensively

in Chapter 4.

3.6 Conclusion

To ensure reliable sensor data delivery to the base-station in GAP4S, two classes of ARQ

protocols – one of which is based on cooperative communications — were considered and

compared.

It was found that when the energy available at the sensor node is the main limiting fac-

tor in the system, the cooperative ARQ protocol saturation throughput may be more than

twice the saturation throughput of the non-cooperative ARQ protocol. Cooperating sensor

nodes are able to effectively balance their energy consumption-to-recharge rate ratio by bor-

rowing energy from one another during the data frame retransmission phase. At the same

time, the transmission energy level at the sensor node may be lowered without compromising

throughput due to the shorter transmission range offered by the (two-hop) relay retransmis-

sion mechanism. The latter is a well-known advantage in multi-hop networking. It was also

found that limiting the number of relays per source to one constitutes a good tradeoff between

network performance and protocol complexity.

In summary, GAP4S is a maintenance-free wireless sensor network solution. While using

acceptable microwave signal levels radiated at the base-station, footprint sizes in the hundred

meter range are possible. Potential GAP4S application fields span from building, airport, and

monument monitoring and control to industrial and agricultural activities, personal safety

and monitoring and alerting systems.

CHAPTER 4

ARQ-CN : A CLASS OF CASCADE COOPERATIVE ARQ PROTOCOLS

4.1 Introduction

In Chapter 3, two classes of ARQ protocols were introduced and compared. The class of

ARQ protocols based on radio cooperation showed promising results in increasing the value

of the saturation throughput of the GAP4S. With the ARQ-C protocol, upon an unsuccessful

data frame transmission, the source is able to borrow energy from the relay by asking it to

retransmit the data frame to the base-station. This cooperative approach is effective as long

as the relay is close enough to the base-station to guarantee a low data frame error probability.

When the source does not have a (one-hop) relay in the proximity of the base-station, this

type of retransmission mechanism might not be effective in overcoming the distance from the

source to the base-station.

The objective of this chapter is to investigate a technique capable of further increasing

the sensor node to base-station saturation throughput in the power-constrained GAP4S. The

goal is accomplished by introducing a new class of cooperative ARQ protocols able to trade

performance for implementation complexity at the base-station. This class of protocols is

based on an iterative version of the ARQ-C protocol. With the ARQ-C protocol, when the

relay retransmission attempt is unsuccessful, the base-station begins a new transmission cycle

by requesting the source retransmit the data frame. This could be avoided by inviting a second

relay — within the earshot of the first relay — to actively participate in the transmission

process. With this mechanism, when retransmission is required, it is as if the data frame

91

92

migrates to sensor nodes that are more likely to produce a successful transmission. Data

frames are thus delivered to the base-station through a chain of relays, hence the name cascade

cooperative ARQ protocols.

First the class of cascade cooperative ARQ protocols is formally described. Then a flow

model to asses the value of the saturation throughput achievable by this new class of protocols

is introduced. The value of the average number of times a data frame needs to be transmitted

in order to be successfully delivered to the base-station is computed for each one of the three

classes of retransmission protocols. The two classes of ARQ protocols, introduced in Chapter 3,

and the class of cascade cooperative ARQ protocols are, then, compared in terms of both

saturation throughput and average number of required transmissions. A number of network

scenarios with different sensor node density, footprint size, and circuit energy consumption

are considered. Results show that, via its “multi-hop” retransmission mechanism, the cascade

cooperative ARQ protocol is most indicated in working scenarios where the footprint size

reaches hundreds of meters.

4.2 ARQ-CN : a Cascade Cooperative Protocol

This section describes a cascade cooperative ARQ protocol that may be used in the GAP4S

to overcome transmission errors. It is assumed that transmission errors may occur only on

the wireless uplink channel, as the sensor node power budget limits the value of the SNR.

Transmission errors on the downlink MW channel are negligible due to the relatively high

power of the MW signal and can be easily overcome with a timeout-triggered retransmission

at the base-station.

This cooperative ARQ protocol takes into account the unique nature of the GAP4S, where-

93

jRelay

iRelay

Base−station

Source

Figure 4.1. ARQ-CN protocol: cooperation among three sensor nodes

by the base-station — characterized by non-stringent power constraints — remotely recharges

the sensor node, broadcasts slot synchronization, polls sensor nodes for collision free uplink

transmissions, requests uplink data frame transmissions and retransmissions, and sends ac-

knowledgments. All decisions are made at the base-station. Sensor nodes obey the control

frames received from the base-station. The base-station is responsible for the choice of the

cooperating sensor nodes and for scheduling collision-free transmissions and retransmissions

of the sensor nodes.

The ARQ-CN protocol is a recursive version of the ARQ-C protocol. Figure 4.1 sketches

how the ARQ-CN protocol works. The source transmits to the base-station. Relay i overhears1

and stores the data frame. To accomplish this task, the optional RX section in the radio

1The base-station chooses sensor node i as the relay.

94

transceiver module shown in Figure 1.3 is required at each sensor nodes. If the source data

frame is not correctly received, the base-station requires relay i to retransmit the data frame.

(If relay i does not transmit, the base-station can safely assume that relay i did not correctly

receive the data frame from the source and requires the source to retransmit.) During the

retransmissions at relay i, relay j, chosen by the base-station, overhears and stores the data

frame. If the data frame retransmission from sensor node i is not correctly received, the

base-station requires relay j to retransmits the data frame. (If relay j does not transmit, the

base-station can safely assume that relay j did not correctly receive the data frame from relay

i and requires relay i to retransmit.) This procedure continues recursively, possibly requiring

the cooperation of additional relays, until the frame is correctly received at the base-station.

The rationale behind this recursive protocol is the assumption that the relay is chosen

to have (a) a higher probability of successful (re)transmission than the one of the source

(or previous relay) and/or (b) to require a lower power to transmit the data frame to the

base-station. Thus, when retransmission is required, the data frame migrates to sensor nodes

that are more likely to produce a successful transmission. For each data frame transmission,

the sequence of the relay is chosen by the base-station. The base-station may choose that

in a probabilistic way, according to some predefined distribution values. The complexity at

the base-station in this case is slightly higher than the one of the ARQ-C protocol, as the

migration of each data frame must be tracked. If necessary, in order reception of data frames

may be enforced by the base-station.

A final note on the similarities and differences between the ARQ-CN protocol and the

store and forward: the similar aspect is that the data frame may reach the base-station via a

multi-hop route. The difference is that the ARQ-CN protocol remains a layer 2 protocol, in

95

which the base-station acts as a central controller, while store and forward is a layer 3 solution

based on routing tables stored at the sensor nodes.

4.3 Assessing Saturation Throughput of the ARQ-CN Protocol

This section attempts to assess the saturation throughput that is achieved by the retransmis-

sion protocol discussed in the previous section. The saturation throughput (S) defines the

throughput values (≤ S) that can be sustained by the system under two constraints: the av-

erage power consumption at each sensor node cannot exceed the power recharge rate, and fair

access is granted to all sensor nodes in the area surrounding the base-station. It is assumed

that other system parameters are not limiting the system throughput, e.g., transmission rate,

channel and electronics bandwidth, buffer capacity, network latency, QoS, etc.

The sensor nodes and the base-station are stationary. The battery recharge characteristic

at each sensor node is ideal, i.e., linear and indefinitely rechargeable.

The base-station determines which are the cooperating nodes for each source. The coop-

erating sensor nodes are chosen to maximize the saturation throughput, as explained next.

The problem of maximizing the throughput for the presented protocol can be formulated

using a flow model that relies on the following input and variables.

Input:

• N : the set of sensor nodes,

• N = |N |: the number of sensor nodes,

• PBS: recharge power radiated on the MW channel at the base-station,

96

• P reci : amount of recharge power radiated at the base-station that reaches the battery of

sensor node i,

• di,j: distance between sensor node i and j,

• di,BS: distance between sensor node i and the base-station,

• Ebi: energy radiated per bit at sensor node i,

• E(Rx)b : energy necessary to receive one bit of data frame at the relay,

• L: number of bits per data frame,

• ξ: the transmission energy overhead at the sensor node, defined as the ratio between

the energy consumption and the radiated energy Ebi,

• P(e)(i,j): probability of unsuccessful reception at sensor node j of a data frame sent by

sensor node i,

• P(e)(i,BS): probability of unsuccessful reception at the base-station of a data frame sent by

sensor node i,

• λi: normalized offered load at sensor node i.

Variables:

• S: network saturation throughput,

• S · λi: offered load2 at sensor node i,

2This definition of the offered load permits to model the traffic intensity at every sensor node using a singlevariable, i.e., S.

97

S1

λ

S2

λ

P(e)

(2,BS)

P(1,BS)

(e)

P(1,2)

(e)

(2,1)λ

λ(2,2)

λ(1,1)

(1,2)λ

λ∗

λ∗1

2(1−P )

(2,1)

(e)

Figure 4.2. ARQ-CN protocol: flow model

• p(i,j): probability that the base-station chooses sensor node j to be the relay for sensor

node i,

• λ(i,j): normalized rate of data frames transmitted at source i while having relay j active,

• λ∗i : total number of data frames transmitted per unit of time at sensor node i.

Figure 4.2 shows the flow model used for the ARQ-CN protocol. Since at each retransmis-

sion the relay can take further advantage of another relay, only one queue at each sensor node

is needed in this model.

max : S

subject to:( ∑

j∈N ,j 6=i

λ(j,i)

P(e)(j,BS)

)· L · E(Rx)

b + (λ∗i · L · Ebi· ξ) ≤ P rec

i ∀i ∈ N (4.1)

λ∗i = S · λi +∑j∈N

(P

(e)(i,j) · λ(i,j)

)+

∑j∈N

((1− P

(e)(j,i)

)· λ(j,i)

)∀i ∈ N (4.2)

98

∑j∈N

λ(i,j) = P(e)(i,BS) · λ∗i ∀i ∈ N . (4.3)

After a transmission error at sensor node i, flow λ∗i is divided into all the possible relays.

The probability p(i,j) is then equal to λ(i,j)

λ∗i P(e)(i,BS)

. Notice that the solution of the presented LP

model can, therefore, be used to set the values of the probabilities p(i,j) at the base-station.

Obviously, sensor node j can be the relay only if the transmission from sensor node i

to sensor node j is error free. λ(i,j) ·(1− P

(e)(i,j)

)is the flow of data frames — from λ(i,j) —

successfully reaching sensor node j from sensor node i. When calculating the power dissipated

to overhear data frame transmissions at sensor node j, i.e., when sensor node j is acting as

the relay of sensor node i, it is not possible to know a-priori whether data frames from source

i are affected by a transmission error or not. When sensor node j is selected as a relay, it

has, therefore, to overhear all data frames transmitted per unit of time at sensor node i.

The number of data frames per unit of time relay j has to overhear from sensor node i is

p(i,j) · λ∗i =λ(i,j)

P(e)(i,BS)

. The total number of data frames per unit of time sensor node j has to

overhear from all the sensor nodes is therefore∑

i∈N ,i6=j

λ(i,j)

P(e)(i,BS)

.

Equation (4.1) balances the total energy used to transmit and overhear data frames and

the energy received from the base-station. Equation (4.2) expresses the total number of data

frames that sensor node i has to transmit/retransmit. This is the sum of three terms: new

data frames, data frames that the base-station designated to be retransmitted by relay j but

were not successfully received at sensor node j, and data frames that i — when acting as

relay for sensor node j — has to retransmit because the retransmission operated by j was not

successful. Equation (4.3) ensures that λ(i,j) ≤ P(e)(i,BS) · λ∗i , therefore, it is always p(i,j) ∈ [0, 1].

99

4.4 Average Number of Transmissions for the Three Classes of ARQ Protocols

This session attempts to asses the value of the average number of times a data frame needs to

be transmitted in order to be successfully received at the base-station. When computing this

value, the number of transmissions of the source and the relay(s) — if any — are considered.

The average number of transmissions is computed for the three classes of ARQ protocols

presented so far in the dissertation, i.e., ARQ-NC, ARQ-C and ARQ-CN .

4.4.1 ARQ-NC

For the non cooperative ARQ protocol, the average number of times a data frame — originat-

ing at sensor node i — needs to be transmitted to be successfully received at the base-station

is defined as [24]:

E[Txi](ARQ−NC) =1

1− P(e)(i,BS)

∀i ∈ N (4.4)

where P(e)(i,BS) represents the data frame error probability from sensor node i to the base-station.

4.4.2 ARQ-C

For the ARQ-C protocol, the average number of times a data frame — originating at sensor

node i — needs to be transmitted to be successfully received at the base-station is defined as:

E[Txi](ARQ−C) =λ∗i +

∑j∈N λ(i,j)

S · λi

∀i ∈ N (4.5)

where λ∗i is the relative total flow of data frames transmitted by sensor node i, i.e., both data

frame transmissions and retransmissions. λ(i,j) represents the flow of data frames successfully

reaching sensor node j when sensor node j acts as the relay for sensor node i. The numerator in

100

(4.5) provides the total number of times a data frame — transmitted the first time by sensor

node i — is transmitted by the source and by the relay(s) involved in the retransmission

process before being successfully received at the base-station.

4.4.3 ARQ-CN

For the ARQ-CN protocol, a different approach is needed. The average number of times a data

frame — originating at sensor node i — needs to be transmitted to be successfully received

at the base-station is defined as:

E[Txi](ARQ−CN ) =

∑i∈N λi

S · λi

∀i ∈ N (4.6)

where the value of λi is obtained by solving the system of linear equations reported below:

C · Λ + S ·Λ = 0 (4.7)

where:

C =

c11 c12 . . . c1N

c21

.... . .

cN1 cNN

, Λ =

λ1

...

...

λN

, Λ =

λ1

...

...

λN

.

When computing the value of λi, the value of the normalized offered load at sensor node

i (λi) is set to be equal to 1, and the value of the normalized offered load at all other sensor

nodes is forced to zero, i.e., λj = 0, ∀j ∈ N , j 6= i.

The system of linear equation is derived from the (4.1), (4.2) and (4.3). N is the total

number of sensor nodes in the network. Each coefficient in Λ represents the total number of

101

data frames transmitted per unit of time at sensor node i, S is the value of the saturation

throughput, and Λ is the matrix of the unknown variables. Coefficients in C are defined as:

ci,j = P(e)(j,BS) · p(j,i)

(1− P

(e)(j,i)

)∀i, j ∈ N , i 6= j (4.8)

ci,j = P(e)(i,BS)

(p(i,i) +

∑j∈N

P(e)(i,j) · p(i,j) − 1

)∀i, j ∈ N , i = j. (4.9)

p(i,j) represents the probability that the base-station chooses sensor node j to be the relay for

sensor node i as defined in Section 4.3:

p(i,j) =λ(i,j)∑

j∈N λ(i,j)

∀i, j ∈ N . (4.10)


This section presents saturation throughput values (S) that are obtained for the ARQ− CN

protocol. Saturation throughput values are measured in packets per second.

The following assumptions are used. Both path loss and fading are taken into account in

the uplink transmission. Only path loss is taken into account in the MW downlink recharging

signal. A path loss exponent of n = 3 is used. Fading is assumed to be Rayleigh slow and

flat; i.e., the fading coefficients are considered constant over a single frame transmission (Sec-

tion 2.1.1.3). The fading experienced by each frame transmission is statistically independent

of the fading experienced by any other frame transmission. More details on the models used

for path loss and fading can be found in Chapter 3.

It is assumed that the MW downlink channel is error free. On the wireless uplink channel,

data frames are augmented with a cyclic redundancy code (CRC). Each block contains B

bits (including the CRC bits). The probability of receiving a data frame incorrectly (error

probability) is a function of both the instantaneous SNR and the CRC. The probability of

102

detecting an erroneous codeword is upper bounded in [30]. The CRC is used to detect the

case of an erroneous codeword decoding, in which case retransmission is required. We assume

that the CRC is able to detect all erroneous codewords. It is assumed that binary PSK with

soft decoding is employed.

The following parameter values are used. Data frames have fixed length and carry B = 128

bits of combined data and CRC, which are encoded into 256 bit codewords using a rate-

compatible punctured convolutional code (RCPC) with rate 1/2, parent code rate of 1/4,

puncturing period of 8, and memory of 4 [40]. The data frame error probability P (e) for

this RCPC versus SNR at the detector stage of the receiver, SNRrec, is shown in Figure 3.7.

SNRrec accounts for both Ebr/N0 and the receiver noise figure F . It is assumed that F = 5 dB.

The values for the transmitting and receiving antenna gain at the base-station and at every

sensor node are assumed to be GT = GR = 1. The recharge power constantly radiated at the

base-station is PBS = 10 W. It is assumed that, at the sensor node, the energy received from

the MW channel by the antenna is fully transferred into its battery and circuitry losses are

negligible.

The effects of various energy consumption factors at the sensor node, i.e., analog-digital

conversion, processing, power management, receiver, and coding are all taken into account

and combined using parameter ξ introduced in Section 4.3. Consistent with [94], ξ = 300.

Unless otherwise specified, traffic is uniform, i.e., λi = 1, ∀i ∈ N .

The formulations presented in Section 4.3 are solved using LP Solve 5.5.0.3 [95].

The radiated energy per bit Eb, i.e., the value of energy that is radiated by a sensor node

to transmit one bit, is assumed to be the same at each sensor node. Unless otherwise specified,

the energy necessary to receive one bit when a sensor node acts as a relay is E(Rx)b = 30 nJ [96].

103

Results are averaged over 5 distinct instances of sensor node distribution. Each instance

is obtained by randomly distributing the sensor nodes within a circular footprint of radius R.

The base-station is at the center of the footprint. The polar coordinates of each sensor node

with respect to the base-station are randomly chosen using a uniform distribution of the angle

in the [0, 2π) interval, and a triangular distribution of the magnitude in the (0, R] interval,

i.e., the density of sensor nodes is constant over the circular footprint.

4.5.1 The Effect of Sensor Density on S

The three ARQ protocol classes (ARQ-NC, ARQ-C, and ARQ-CN) are compared using two

functions, i.e., S versus Eb and E[Tx] versus the distance between the source and the base-

station. The value of E[Tx] is computed in the following way. The interval [0, R] is divided

into K intervals. For each one of the K intervals:

E[Tx] =1∑

m∈N Xm

·∑

i∈(j−1,j]

E[Txi], ∀j, j = 1, 2, · · · , K (4.11)

where

Xm =

1 if m ∈ (j − 1, j]

0 otherwise

and E[Txi], depending on the retransmission protocol used, is defined as in Section 4.4.

Unless otherwise specified, E[Tx] is plotted for the value of Eb that yields the maximum S

for the ARQ-CN protocol. Three different values of the footprint radius are chosen. For each

value of the footprint radius (R), three different values of the number of sensor nodes (N) are

considered. The objective is to create different instances of sensor node density so that the

working scenario in which each ARQ protocol is most suitable can be identified.

104

Figures 4.3, 4.4, and 4.5 depict a working scenario where R = 10 m, while N is equal

to 100, 300, and 500, respectively. Both cooperative ARQ protocols yield values of S that

are greater than the value of S for the ARQ-NC protocol. The difference between the two

cooperative ARQ protocols is marginal, as sensor nodes are all relatively close to the base-

station, i.e., using a chain of relays to successfully transmit a data frame does not yield any gain

when compared to the two-hop approach used by the ARQ-C protocol. Figures 4.3(b), 4.4(b)

and 4.5(b) show the value of E[Tx] when Eb = 6e−14 J. The two cooperative ARQ protocols

have almost the same value of E[Tx]. Varying the density of the sensor nodes in the footprint

does not affect the value of S.

10−14

10−13

10−12

101

102

Eb [J]

S

ARQ−NCARQ−C(1)ARQ−C(100)

ARQ−CN

(a) Saturation throughput

0 2 4 6 8 101

1.5

2

2.5

3

Distance from the base−station [m]

E[T

x]ARQ−NCARQ−C(1)ARQ−C(100)

ARQ−CN

(b) Average number of transmissions

Figure 4.3. R = 10 m, N = 100: (a) saturation throughput (S) versus energy per bit radiatedby the sensor nodes (Eb), (b) Eb = 6e−14: average number of transmissions (E[Tx]) versusthe distance of a sensor node from the base-station

105

10−14

10−13

10−12

101

102

Eb [J]

S


ARQ−CN


0 2 4 6 8 101

1.5

2

2.5

3


E[T

x]


ARQ−CN



10−14

10−13

10−12

101

102

Eb [J]

S


ARQ−CN


0 2 4 6 8 101

1.5

2

2.5

3


E[T

x]


ARQ−CN



106

Figures 4.6, 4.7, and 4.8 depict a working scenario where R = 50 m, while N is equal to 100,

300, and 500, respectively. The gap between the ARQ-NC protocol and the two cooperative

ARQ protocols has increased when compared to the previous scenario. The figures also show

the slight improvement in terms of S achieved by the ARQ-CN protocol when compared to

the ARQ-C protocol. Using a chain of relays to successfully deliver a data frame to the base-

station is somehow relatively beneficial. Sensor nodes that are at the edge of the footprint

may count on relays that are close to the base-station and whose data frame error probability

is low. Considering the increased complexity at the base-station that comes with the ARQ-

CN protocol, the ARQ-C protocol can still guarantee a good tradeoff between performances

and protocol complexity. Figures 4.6(b), 4.7(b) and 4.8(b) shows the value of E[Tx] when

Eb = 4e−12 J. For low values of Eb, cooperative communication is able to limit to a few

retransmission attempts the exponential growth typical of E[Tx]ARQ−NC . Intuitively, the

ARQ-CN protocol slightly benefits from an increased value of N .

10−12

10−11

10−10

10−3

10−2

Eb [J]

S


ARQ−CN


0 10 20 30 40 501

1.5

2

2.5

3


E[T

x]


ARQ−CN



107

10−12

10−11

10−10

10−3

10−2

Eb [J]

S


ARQ−CN


0 10 20 30 40 501

1.5

2

2.5

3


E[T

x]


ARQ−CN



10−12

10−11

10−10

10−3

10−2

Eb [J]

S


ARQ−CN


0 10 20 30 40 501

1.5

2

2.5

3


E[T

x]


ARQ−CN



108

Figures 4.9, 4.10, and 4.11 depict a working scenario where the value of R = 100 m, while

N is equal to 100, 300, and 500, respectively. This is the working scenario where the ARQ-CN

protocol appears to be the most beneficial in terms of S. For this value of R, the two-hop

approach used by the ARQ-C protocol might not be sufficient to overcome the distance from

the source to the base-station. Sensor nodes at the edge of the footprint do not have one-hop

relays close enough to the base-station to guarantee a low value of data frame error probability.

On the contrary, the ARQ-CN protocol is able to reach the base-station through a chain of

relays that, at each hop, have a higher probability of successful (re)transmission than that

of the source or the previous relay. Figures 4.9(b), 4.10(b) and 4.11(b) show the value of

E[Tx] when Eb = 1.5e−11 J. These figures show that using a chain of relays may reduce up

to 25% the average number of transmissions necessary to successfully deliver a data frame to

the base-station. The value of S for the ARQ-CN protocol increases with N .

10−11

10−10

10−9

10−4

Eb [J]

S


ARQ−CN


0 20 40 60 80 1001

1.5

2

2.5

3

3.5

4

4.5


E[T

x]


ARQ−CN


Figure 4.9. R = 100 m, N = 100: (a) saturation throughput (S) versus energy per bit radiatedby the sensor nodes (Eb), (b) Eb = 1.5e−11: average number of transmissions (E[Tx]) versusthe distance of a sensor node from the base-station

109

10−11

10−10

10−9

10−4

Eb [J]

S


ARQ−CN


0 20 40 60 80 1001

1.5

2

2.5

3

3.5

4

4.5


E[T

x]


ARQ−CN


Figure 4.10. R = 100 m, N = 300: (a) saturation throughput (S) versus energy per bitradiated by the sensor nodes (Eb), (b) Eb = 1.5e−11: average number of transmissions (E[Tx])versus the distance of a sensor node from the base-station

10−11

10−10

10−9

10−4

Eb [J]

S


ARQ−CN


0 20 40 60 80 1001

1.5

2

2.5

3

3.5

4

4.5


E[T

x]


ARQ−CN


Figure 4.11. R = 100 m, N = 500: (a) saturation throughput (S) versus energy per bitradiated by the sensor nodes (Eb), (b) Eb = 1.5e−11: average number of transmissions (E[Tx])versus the distance of a sensor node from the base-station

110

4.5.2 The Effect of E(Rx)b on S

Figure 4.12 shows the effect of E(Rx)b on S for two values of the footprint radius, R = 50 m

and R = 100 m, respectively, while N = 300. Intuitively, the value of S grows with decreasing

values of E(Rx)b . The gain is approximatively close to one order of magnitude for the ARQ-CN

protocol when compared to the values in Figure 4.7(a) and Figure 4.10(a).

10−14

10−13

10−12

10−11

10−2

10−1

Eb [J]

S


ARQ−CN

(a) R = 50 m

10−13

10−12

10−11

10−10

10−4

10−3

10−2

Eb [J]

S


ARQ−CN

(b) R = 100 m

Figure 4.12. E(Rx)b = 0.3 nJ, N = 300: saturation throughput (S) versus energy per bit

radiated by the sensor nodes (Eb)

Figure 4.13 shows the value of the average number of transmissions (E[Tx]) versus the

distance of a sensor node from the base-station. The E[Tx] values reported in the figure are

obtained using the value for Eb that yields the maximum S for each one of the three ARQ

protocols. The plots confirm the earlier claim that higher values of throughput might be

achieved by finding a good tradeoff between the value of energy used to transmit each bit and

the number of retransmissions necessary to ensure a reliable delivery of data frames.

Figure 4.12(a) shows that the maximum achievable saturation throughput (S) for the ARQ-

CN protocol is approximately equal to 10−1 packets per second. Figure 4.13(a) shows that with

111

0 10 20 30 40 501

2

3

4

5

6

7

8


E[T

x]ARQ−NCARQ−C(1)ARQ−C(300)

ARQ−CN

(a) R = 50 m

0 20 40 60 80 100

2

4

6

8

10

12

14


E[T

x]


ARQ−CN

(b) R = 100 m

Figure 4.13. E(Rx)b = 0.3 nJ, N = 300: average number of transmissions (E[Tx]) versus the

distance of a sensor node from the base-station

this protocol the maximum number of times a data frame needs to be transmitted in order to

be successfully received at the base-station is slightly greater than 7. Each transmitted data

frame has L = 256 bits, and the total number of sensor nodes in the network is equal to 300.

The uplink radio channel has to sustain a total bit rate that is in excess of 0.1 · 7 · 256 · 300 =

53760 bits per second. If this constraint is satisfied, the assumption made in Section 3.3

and Section 4.3 that the transmission rate is not a limiting factor in the evaluation of the

saturation throughput is accurate.

Table 4.1 documents the effect of E(Rx)b , R, and N on S for the three classes of ARQ

protocols. The S values reported in the table are obtained using the value for Eb that yields

the maximum S in each case. The values in the table confirm the trend shown in Figure 4.12

and the fact that ARQ-CN protocol benefits from a higher density of sensor nodes.

112

Table 4.1. ARQ-CN protocol: saturation throughput (S) as a function of the footprint radius(R) and the number of sensor nodes (N)

fup = 2.4 GHz, GT = GR = 1, E(Rx)b = 0.3 nJ, ξ = 300, F = 5 dB, n = 3

R = 50 m R = 100 m

N = 100 N = 300 N = 500 N = 100 N = 300 N = 500

ARQ-NC 4.1191E-3 4.1191E-3 4.1191E-3 6.4316E-5 6.4316E-5 6.4316E-5

ARQ-C(1) 1.6324E-2 1.6324E-2 1.6324E-2 2.2902E-4 2.5581E-4 2.5643E-4

ARQ-C(all) 1.6324E-2 1.6324E-2 1.6324E-2 2.2902E-4 2.5581E-4 2.5643E-4

ARQ-CN 1.1724E-1 1.1724E-1 1.3214E-1 2.2509E-3 4.2975E-3 5.5741E-3

4.5.3 The Effect of Uplink Frequency and Path Loss Exponent on S

Table 4.2 documents the effect of the uplink frequency (fup), the footprint radius (R), and the

path loss exponent (n) on S for the two classes of ARQ protocols. The S values reported in the

table are obtained using the value for Eb that yields the maximum S in each case. Results in

the top part of the table are obtained using an uplink frequency fup= 2.4 GHz. Results in the

middle part of the table are obtained using an uplink frequency fup= 916.5 MHz. Results in the

bottom part of the table are obtained using an uplink frequency fup= 433.92 MHz. Depending

on the surrounding environment — which may affect the value of n — the practical size of

the footprint spans from tens to hundred meters.

The value in the table confirm that the ARQ-CN protocol yields higher saturation through-

put values than the ones achieved by the ARQ-C protocol, only in those working scenarios

where the two-hop retransmission of data frames from the sensor node to the base-station is

not favored by the surrounding environment, e.g., for increasing values of fup, R and n.

113

Table 4.2. ARQ-CN protocol: saturation throughput (S) as a function of the uplink frequency(fup), footprint radius (R) and path loss exponenet (n)

fup = 2.4 GHz, GT = GR = 1, E(Rx)b = 30 nJ, ξ = 300, F = 5 dB

n = 3 R = 50 m

R = 10 m R = 50 m R = 100 m n = 2 n = 3 n = 4

ARQ-NC 64.297 4.1179E-3 6.4297E-5 10.208 4.1179E-3 1.6361E-6

ARQ-C(1) 111.66 1.1199E-2 2.4269E-4 11.733 1.1199E-2 1.1332E-5

ARQ-C(300) 123.72 1.5117E-2 2.5264E-4 15.121 1.5117E-2 1.1332E-5

ARQ-CN 112.49 1.9115E-2 5.7409E-4 13.158 1.9115E-2 7.8987E-5


n = 3 R = 50 m

R = 10 m R = 50 m R = 100 m n = 2 n = 3 n = 4

ARQ-NC 440.05 2.7284E-2 4.4005E-4 69.688 2.7284E-2 1.1258E-5

ARQ-C(1) 753.91 5.5398E-2 1.2840E-3 70.912 5.5398E-2 6.5263E-5

ARQ-C(300) 769.34 7.4938E-2 1.6419E-3 82.343 7.4938E-2 7.8625E-5

ARQ-CN 544.71 8.1060E-2 2.2739E-3 76.206 8.1060E-2 2.3279E-4


n = 3 R = 50 m

R = 10 m R = 50 m R = 100 m n = 2 n = 3 n = 4

ARQ-NC 1948.2 1.2587E-1 1.9481E-3 314.62 1.2587E-1 4.8166E-5

ARQ-C(1) 2470.1 2.2278E-1 4.0794E-3 318.23 2.2278E-1 2.5434E-4

ARQ-C(300) 3204.1 2.7221E-1 5.4379E-3 342.66 2.7221E-1 3.2046E-4

ARQ-CN 2173.1 2.6587E-1 6.3287E-3 328.54 2.6587E-1 5.2274E-4

114

4.5.4 Non Uniform Traffic Scenario

This section is intended to compare the performance, in term of S, of the three classes of ARQ

protocols in a working scenario with non uniform traffic.

Two instances of non uniform traffic patterns are reproduced, as explained next.

• Hot-spot-like traffic increase: this instance of non uniform traffic mimics a scenario that

is common in wireless sensor networks. In the sensor field, a set of neighboring sensor

nodes detect the same phenomenon at the same time. As a consequence, sensor nodes

positioned in the surroundings of the phenomenon will generate traffic at an higher rate

when compared to the other sensor nodes in the network.

• Two rates traffic increase: this instance of non uniform traffic mimics a scenario where

two different types of sensor nodes, each one generating traffic with a different rate, are

deployed in the sensor field.

4.5.4.1 Hot-Spot-Like Traffic Increase

The three ARQ protocol classes (ARQ-NC, ARQ-C, and ARQ-CN) are compared in terms of

saturation throughput (S) versus the energy radiated at the sensor node (Eb).

An hot-spot is created in two steps. First, one sensor node in the network is randomly

selected to be the center of the hot-spot. Then, all sensor nodes whose distance from the

center of the hot-spot is less than or equal to a fixed value — called Rhot−spot — have their

traffic rate increased by ten times when compared to the other sensor nodes in the network.

Figure 4.14, 4.15, and 4.16 depict a working scenario where the center of the hot-spot is

placed in the center of the footprint.

115

−50 0 50−50

−40

−30

−20

−10

0

10

20

30

40

50

X [m]

Y [m

]Sensor with increased traffic rate

(a) Sensor node positions

10−12

10−11

10−10

10−3

10−2

Eb [J]

S

ARQARQ−C(1)ARQ−C(300)

ARQ−CN

(b) Saturation throughput

Figure 4.14. Hot-spot in the center of the footprint: R = 50 m, N = 300, Rhot−spot = R/8


placed at the edge of the footprint.


placed in the middle of the footprint.

Each plot represents one instance of position of the sensor nodes and one instance of

position of the center of the hot-spot. The set of figures depict the case when Rhot−spot =

R/8, Rhot−spot = R/4 and Rhot−spot = R/2, respectively. For each hot-spot scenario R =

50 m and N = 300. Results for this type of non uniform traffic confirm the trend already

shown in Section 4.5.1. Irrespective of the hot-spot position and dimension, the two classes

of cooperative ARQ protocols show better performance in terms of S when compared to the

non cooperative ARQ protocol. In addition, as already shown in Figures 4.6(a) to 4.11(a),

the plots confirm the improvement in terms of S achieved by the ARQ-CN protocol when

compared to the ARQ-C protocol.

116

−50 0 50−50

−40

−30

−20

−10

0

10

20

30

40

50

X [m]

Y [m

]

Sensor with increased traffic rate


10−12

10−11

10−10

10−3

10−2

Eb [J]

S


ARQ−CN



−50 0 50−50

−40

−30

−20

−10

0

10

20

30

40

50

X [m]

Y [m

]



10−12

10−11

10−10

10−3

10−2

Eb [J]

S


ARQ−CN



117

−50 0 50−50

−40

−30

−20

−10

0

10

20

30

40

50

X [m]

Y [m

]



10−12

10−11

10−10

10−4

10−3

Eb [J]

S


ARQ−CN


Figure 4.17. Hot-spot at the edge of the footprint: R = 50 m, N = 300, Rhot−spot = R/8

−50 0 50−50

−40

−30

−20

−10

0

10

20

30

40

50

X [m]

Y [m

]



10−12

10−11

10−10

10−4

10−3

Eb [J]

S


ARQ−CN



118

−50 0 50−50

−40

−30

−20

−10

0

10

20

30

40

50

X [m]

Y [m

]



10−12

10−11

10−10

10−4

10−3

Eb [J]

S


ARQ−CN



−50 0 50−50

−40

−30

−20

−10

0

10

20

30

40

50

X [m]

Y [m

]



10−12

10−11

10−10

10−3

10−2

Eb [J]

S


ARQ−CN


Figure 4.20. Hot-spot in the middle of the footprint: R = 50 m, N = 300, Rhot−spot = R/8

119

−50 0 50−50

−40

−30

−20

−10

0

10

20

30

40

50

X [m]

Y [m

]



10−12

10−11

10−10

10−3

10−2

Eb [J]

S


ARQ−CN



−50 0 50−50

−40

−30

−20

−10

0

10

20

30

40

50

X [m]

Y [m

]



10−12

10−11

10−10

10−4

10−3

Eb [J]

S


ARQ−CN



120

4.5.4.2 Two Rates Traffic Increase Scenario

The three ARQ protocol classes (ARQ-NC, ARQ-C, and ARQ-CN) are compared in terms of

saturation throughput (S) versus the energy radiated at the sensor node (Eb).

Figures 4.23, 4.24, and 4.25 depict a scenario where 25%, 50%, and 75% of the total

number of sensor nodes are randomly selected to generate a traffic rate ten times higher than

the other sensor nodes in the field. For each scenario, R = 50 m and N = 300. Sensor

nodes with the increased traffic rate are chosen according to a uniform distribution. Each

plot represents one instance of position of the sensor nodes and one instance of position of

the sensor nodes generating traffic with an higher rate. Figures confirm that the use of radio

cooperation is also beneficial in achieving higher values of saturation throughput with this

type of non uniform traffic.

−50 0 50−50

−40

−30

−20

−10

0

10

20

30

40

50

X [m]

Y [m

]



10−12

10−11

10−10

10−4

10−3

Eb [J]

S


ARQ−CN


Figure 4.23. Two rates traffic increase: R = 50 m, N = 300, 25% of all sensor nodes have anincreased value of generated traffic rate

121

−50 0 50−50

−40

−30

−20

−10

0

10

20

30

40

50

X [m]

Y [m

]



10−12

10−11

10−10

10−4

10−3

Eb [J]

S


ARQ−CN



−50 0 50−50

−40

−30

−20

−10

0

10

20

30

40

50

X [m]

Y [m

]



10−12

10−11

10−10

10−4

10−3

Eb [J]

S


ARQ−CN



122

4.6 Conclusion

This chapter presented a cascade cooperative ARQ protocol that may be used in the GAP4S

to overcome transmission errors.

When compared to the cooperative ARQ protocol introduced in Chapter 3, it was found

that the cascade cooperative ARQ protocol is most indicated in those scenarios where the two-

hop retransmission mechanism of the ARQ-C protocol is not sufficient to overcome the distance

from the source to the base-station. When retransmission is required, the ARQ-CN protocol

is able to reach the base-station through a chain of relays that have a higher probability of

successful (re)transmission at each hop than the one of the source or the previous relay. For

this reason, the presented protocol is more indicated for scenarios where the footprint size

reaches hundreds of meters.

CHAPTER 5

SPP-DiR PROTECTION SWITCHING SCHEME

5.1 Introduction

Wavelength Division Multiplexed (WDM) networks are evolving to respond quickly and eco-

nomically to dynamic traffic demands. A WDM network consists of a number of optical

switches interconnected by fiber-optic links to form, in general, an arbitrary topology. The

basic services provided by WDM networks are high speed, all-optical end-to-end channels,

also referred to as lightpaths [52]. Lightpaths are dynamically created between node pairs to

both provide the desired network connectivity and accommodate arriving traffic demands.

The unexpected failure of a network element may have severe consequences due to the large

amount of traffic carried by the WDM channels. WDM networks can be made more reliable

by means of protection switching schemes that are implemented at the WDM layer [97].

A protection scheme requires the allocation of spare (or stand by) resources, that can be

used in the event of a fault occurrence. For a lightpath, the protection scheme consists

of assigning a working and a protection path between the source and the destination. The

working path carries the offered traffic during normal network operations. When the working

path is disrupted by a fault, the interrupted traffic is re-routed over the protection path until

the fault is repaired.

Each working (and protection) path that needs to be created in the WDM network is

assigned both a route and a wavelength — this is the so called Routing and Wavelength

123

124

Assignment (RWA) problem. When traffic demands dynamically enter and depart from the

network, the problem is referred to as the online RWA problem. One of the online RWA

problem objectives is to reserve minimum number of network resources (wavelengths) for each

arriving traffic demand. It is expected that, by minimizing the amount of reserved resources

per arriving demand, the blocking probability is reduced — where a demand is blocked when it

cannot be created due to the lack of available wavelengths in the network. In general, finding

the optimum solution for the RWA problem is a challenging combinatorial problem, whose

complexity — i.e., the size of the solution space — grows with both the network size and the

number of demands.

In this chapter, two open problems are addressed: how to contain the amount of network

resources reserved for the arriving demand and how to solve the online RWA problem swiftly.

In simple terms, the first problem is how to guarantee the desired level of reliability for

arriving traffic demands (lightpaths), while avoiding unnecessary over-reservation of network

resources. Conventional protection schemes [98] are capable of providing full protection in the

presence of a single network fault. These solutions are simple, and provide valid approaches in

many network scenarios [99–101]. However, when over-reservation of network resources is not

acceptable, some of these solutions may not be adequate. For example, in the Dedicated Path

Protection (DPP) scheme the wavelengths reserved for the protection path of a demand are

dedicated to that demand only [47]. The Shared Path Protection (SPP) scheme may then be

used to reduce the amount of resources required by allowing multiple working paths to share

some wavelengths that are reserved for protection. For static networks, it is possible to show

that under certain circumstances, the same minimum degree of reliability can be guaranteed

to the demands by both DPP and SPP, with SPP requiring a significantly smaller amount of

125

network resources [102]. The SPP resource saving is achieved at a cost of increased complexity

of the protection scheme. Further reduction of the required resources can be achieved in some

instances, by using the concept of Differentiated Reliability (DiR). The DiR concept — when

applied to networks with static traffic (offline RWA problem) — yields significant reduction of

the total network resources that are required to accommodate a given set of demands [83,84].

In this chapter, the SPP scheme, combined with the DiR concept, is applied to WDM

networks with dynamic traffic. The resulting scheme is referred to as SPP-DiR. In the sim-

plest DiR formulation, each arriving demand is assigned a reliability degree, defined as the

probability that the established demand is still available after the occurrence of a single fault

in the network. The reliability degree is chosen to match each traffic demand requirement and

must be met by the protection scheme independently of the actual network topology, design

constraint, device technology, and demand span. This assumption makes it possible to reserve

the minimum amount of network resources that are required to achieve the level of reliability

requested by the arriving demand. The origin of this DiR advantage — that conventional

protection schemes, e.g., SPP, do not offer — can be clarified as follows. The former scheme

(DiR’s) focus is on the reliability degree offered to each individual demand. Conversely, the

focus of the latter schemes is on the network reliability offered against any single network fault.

Consequently, with the latter schemes the actual reliability degree offered to a demand may

vary significantly as a function of the path span and Mean Time Between Failure (MTBF) of

the network elements. Besides creating unfair handling of demands, the latter schemes may

also over-reserve spare resources in the network, which in turn produces an unnecessarily high

reliability degrees to some demands.

The second problem addressed in this chapter is how to provide — in reliable WDM net-

126

works — a satisfactory (sub-optimal) solution to the online RWA problem, when operating

under constrained computational time. One approach which is widely used to select the (work-

ing) route for each demand is based on a variation of the multicommodity flow problem [76].

Some examples can be found in [103–105]. This approach is based on the intuitive reasoning

that the careful pruning of the set of possible candidate paths [106] leads to a (sub-optimal)

solution of the multicommodity flow problem that may be satisfactory from both the com-

plexity and the performance standpoint. A well-known pruning technique consists of choosing

only the k-shortest paths found in the graph that represents the network topology [107]. It can

be shown that, for unprotected networks, a relatively small value of k may already produce

results that are close to the optimum. On the contrary, when dealing with reliable networks,

the use of the k-shortest paths may require a much larger value of k. The reason is twofold.

First, at least one route disjoint path-pair must be found for each source-destination pair.

(This is a necessary condition to yield a feasible RWA solution in single-fault reliable net-

works.) Second, a sufficiently large number of distinct path-pair candidates must be available

between each source-destination pair. This latter condition is needed to allow some degree of

flexibility in choosing the best path-pair for the arriving demand. (As shown in Section 5.4,

the approach based on the single shortest disjoint path-pair [76,108] may not yield satisfactory

performance.) When k is large, however, the set of candidate paths remaining after pruning

may be too large to provide fast and satisfactory solutions to the RWA problem.

For this second problem, this chapter proposes an alternative pruning technique to the k-

shortest paths based on the Disjoint Path-Pair Matrix (DPM). The objective of the proposed

pruning technique is to control and limit (1) the number of route disjoint candidate path-pairs,

(2) the number of hops of the working paths, (3) the number of hops of the protection paths,

and (4) the hop difference between the working and the protection paths. These objectives can

127

be accomplished by the DPM while maintaining a solution performance that is comparable

to the — less controllable — solution obtained by the k-shortest path pruning technique. In

addition, the DPM technique requires a smaller search space than the one obtained by the

k-shortest path. This fact may yield an advantage to DPM when the computational time

available to find a solution is constrained.

The DPM is applied to solve the RWA problem for both the conventional SPP and SPP-

DiR schemes based on a centralized network status database. Numerical results are shown

using a Pan-European topology as a benchmark. When compared to the conventional SPP,

the SPP-DiR scheme requires less network resources and yields improved blocking probability,

already with a small and controlled reduction of the demand reliability degree. It is also shown

that, when compared to a path pruning technique based on the k-shortest path algorithm,

the DPM technique yields slightly better solutions when the computational time allowed to

solve the RWA problem is constrained to few milliseconds.

5.2 The SPP-DiR Model for WDM Networks with Dynamic Traffic

This section describes the assumptions made and defines the SPP-DiR scheme and the related

RWA problem.

It is assumed that the WDM network has an arbitrary physical topology (mesh), wave-

length conversion is not available in the network, only link failures are possible, and any link

failure disrupts demands in both directions of propagation. The widely used single link failure

assumption [97,109] is adopted, i.e., the probability that two or more links are down concur-

rently is considered to be negligible. Rerouting of working lightpaths that are not affected by

the fault is not allowed.

128

The WDM mesh is modeled as a graph G(N ,L), where N represents the set of network

nodes and L the set of network links. It is assumed that, for each direction of propagation,

every network link consists of a set of fibers, F . Each fiber carries a set of wavelengths, W .

Each link (m,n) ∈ L is characterized by three parameters: |F |: the number of available

fibers; |W |: the number of available wavelengths in each fiber; and Pf (m,n): the value of the

conditional link failure probability. Based on the single failure assumption, the conditional

link failure probability is the conditional failure probability given that a single link failure has

occurred in the network. By assuming the single link failure scenario, the link failure proba-

bility is given by the product of the conditional link failure probability and the probability of

having a single failure. For example, assuming a uniform distribution of faults among all the

links, the conditional link failure probability is:

Pf (m,n) =1

|L| ∀(m,n) ∈ L. (5.1)

It is assumed that the demand arrivals cannot be predicted. Thus, they are modeled as

a random process. Demands must be served in the same order as they are generated. Each

demand requires one working lightpath to be created between two nodes. Each lightpath is

created using one single wavelength. Each arriving demand is characterized by a Maximum

Conditional Failure Probability (MCFP ). The MCFP represents the maximum acceptable

probability that, given the occurrence of a network link failure, the demand data flow will be

permanently disrupted.

When using the conventional SPP scheme, each working path is assigned a route-disjoint

protection path ready to be used if the working path is affected by a link failure. Working

and protection paths of the same demand need not have the same wavelength assigned. Only

distinct protection paths whose corresponding working paths are route-disjoint can share the

129

same link and wavelength. Each demand is thus 100% survivable against any single fault, i.e.,

the SPP supports MCFP = 0 only.

To offer a wider range of MCFP values, the SPP-DiR scheme is derived from the SPP

scheme as follows. For a demand with a less stringent MCFP > 0, the protection path does

not need to be always available for every possible link failure scenario. Thus, it is possible

to select a set of links H(d)u of the working path for which arriving demand d will not require

to resort to the protection path. Set H(d)u must be selected to satisfy the demand required

reliability degree, formally expressed by the demand’s MCFP . Notice that, with SPP-DiR,

two (or more) demands whose working paths have a common link may also share a link and

a wavelength for their respective protection paths. This option is available when at least one

of the two demands can afford to be permanently disrupted upon the failure of the link that

is shared by the working paths. By the same reasoning, it is also possible to have a working

path completely unprotected if the working path failure probability still satisfies the reliability

requirement indicated by the demand’s MCFP .

The SPP-DiR scheme has the potential to yield a more efficient resource utilization when

compared to the conventional SPP scheme, while still guaranteeing each demand sufficient

resources to satisfy its reliability requirement. The example shown in Figure 5.1 illustrates this

possibility. All links in the network are bidirectional and can accommodate two wavelengths for

each direction of propagation. Assuming uniform link failure distribution, the link conditional

failure probability is Pf (m,n) = 17∀(m,n) ∈ L. Three demands are shown. Demand d1 arrives

first and requires MCFP (d1) = 0. The chosen working path is C −B. The protection path is

C−E−B. Demand d2 arrives second and requires MCFP (d2) = 0. The chosen working path

is D − E −A. The protection path is therefore D − C −B −A. Demand d3 arrives last and

130

D1D2D3

A

E B

CD

Figure 5.1. SPP-DiR example

requires MCFP (d3) = 17. This reliability requirement permits to have demand d3 protected

against any single fault but one. Taking advantage of this possibility, it is possible to route

the working path along D − E − B and have link D − E unprotected. The protection path

for d3 is D − C − B and is used only in the case of a failure on link (E, B). As shown in

the example, protection resources along link (C, B) can be shared between demands d2 and

d3 even though their working paths are not route disjoint. Notice that by requiring a higher

reliability degree, i.e., MCFP (d3) < 17, demand d3 is then blocked due to the lack of available

wavelengths in the network.

5.2.1 The Online RWA Problem for SPP-DiR

The online RWA problem for the SPP-DiR scheme consists of choosing both the working

and protection path-pair and the wavelength(s) to be assigned to each arriving demand.

The choice must be made so that both the amount of available resources that is reserved to

accommodate the arriving demand is minimized, and the MCFP required by the demand is

131

satisfied. It is expected that such optimization has a favorable impact on the overall network

blocking probability. The SPP-DiR RWA problem is formally defined next. The formulation

is provided assuming |F | = 1 for all links. Its extensions to the case of multiple fibers per link

is straightforward.

Let λ(d)w , λ

(d)p ∈ W be the wavelengths that are chosen for the working and protection paths

of demand d, respectively, i.e., the working and protection wavelength λ(d)w and λ

(d)p need not

be the same. Let H(d)w be the set of links that are in the working path assigned to demand

d, i.e., the set of working links for d. Let H(d)p be the set of links that are in the protection

path assigned to demand d, i.e., the set of protection links for d. Let H(d)u ⊆ H

(d)w be the set

of working links of d that are unprotected, i.e., upon the failure of a link in H(d)u demand d is

permanently disrupted. Let MCFP (d) be the minimum reliability degree requested by d.

Let D be the set of demands that are already established in the network. Initially, D = ∅.

Let d be the arriving demand. Demand d is accepted (and added to set D) if all the following

conditions can be satisfied:

H(d)w ∩H(d)

p = ∅ (5.2)

i.e., working and protection paths must be route-disjoint,

∀d ∈ D, d 6= d, H(d)w ∩H(d)

w 6= ∅ ⇒ λ(d)w 6= λ(d)

w (5.3)

i.e., in any link a working wavelength can be assigned to only one (demand) working path,

∀d ∈ D, d 6= d, (H(d)w \H(d)

u )∩ (H(d)w \H(d)

u ) 6= ∅ ⇒ H(d)p ∩H(d)

p = ∅ ∨ λ(p)w 6= λ(d)

p (5.4)

i.e., a protection wavelength can not be shared by multiple demands if they share the same

(protected) working link,

P(d)f =

∑

(i,j)∈H(d)u

Pf (i, j) ≤ MCFP (d) (5.5)

132

i.e., the conditional failure probability guaranteed to demand d does not exceed the MCFP

required by d. If any one of the above four conditions cannot be satisfied, demand d is blocked

(and not added to set D).

Notice that the protection paths of demands d and d ∈ D are allowed to share wavelength

on a common link, i.e.,

λ(d)p = λ(d)

p ∧ (H(d)p ∩H(d)

p ) 6= ∅ (5.6)

only if condition

(H(d)w ∩H(d)

w ) ⊆ (H(d)u ∪H(d)

u ) (5.7)

is satisfied. Let H(d)s ⊆ H

(d)p be the set of protection links of demand d in which the spare

wavelength is shared by at least one other protection path already reserved in the network,

i.e.,

H(d)s = {(m,n) : ∃d ∈ D : (m,n) ∈ (H(d)

p ∩H(d)p ) ∧ (λ(d)

p = λ(d)p )}. (5.8)

A cost function measuring the goodness of the RWA chosen for both the working and protec-

tion paths of demand d is

C(d) = |H(d)w |+ |H(d)

p | − |H(d)s |+ (MCFP (d) − P

(d)f ). (5.9)

The optimal solution of the RWA problem for demand d is the one that minimizes C(d), while

satisfying (5.2), (5.3), (5.4), and (5.5). The cost function C(d) quantifies both the amount of

resources that must be reserved to accommodate demand d and the excess of reliability that is

guaranteed to demand d — defined as (MCFP (d)−P(d)f ) ≥ 0. The reason for choosing such a

cost function is twofold. First, each demand is guaranteed the working and protection path-

pair that requires the least amount of newly reserved resources. Second, over-provisioning of

wavelengths is avoided by matching the arriving demand’s MCFP as closely as possible.

133

Notice that if MCFP (d) = 0 for all arriving demands d, then H(d)u = ∅, ∀ d ∈ D. In this

case C(d) becomes the cost function that must be minimized to find the optimum solution of

the RWA problem for the conventional SPP.

5.3 Solving the Online RWA Problem for both the Working and Protection Paths

In this section, a two-step approach is presented to find a good sub-optimal solution to the

RWA problem defined in Section 5.2. In step A, the DPM (Disjoint Path-Pair Matrix) is

built for each source-destination pair using only selected disjoint path-pairs, i.e., the path-

pair candidates. In step B, the RWA problem of the SPP-DiR scheme is solved by running

a Simulated Annealing (SA) [77, 110] algorithm that searches for the best candidate in the

DPM. In general, any optimization algorithm can be used for the latter step to replace SA.

The SA approach is chosen here as it was found to provide satisfactory results.

The two steps are described next.

5.3.1 Step A: Construction of the DPM

One Disjoint Path-Pair Matrix is built for each source-destination pair. The DPM is computed

beforehand and is then used to route all the arriving connection requests. The candidates of

the DPM are computed based on the observation that the space of possible solutions contains

only route disjoint path-pairs. Let k1 be the desired number of candidate working paths.

The candidate working paths are the first k1 paths that are found by the k-shortest loopless

paths algorithm [107] applied to graph G(N ,L). Let k2 be the desired number of candidate

protection paths for each candidate working path. The candidate protection paths for working

path i are the first k2 paths that are found by the k-shortest loopless paths algorithm applied

134

Table 5.1. Pseudo code of the algorithm to construct DPM

begin Construction of the DPMfor(∀ node pairs (s, d), s 6= d, s, d ∈ N ){

Compute Ws,d on G(N ,L)for(i = 0, 1, . . . , k1 − 1, i ∈ Ws,d){L(i) = L − i

Compute Ps,d,i on G(i)(N ,L(i))for(j = 0, 1, . . . , k2 − 1, j ∈ Ps,d,i){

DPMs,d(i, j) = (Ws,d(i),Ps,d,i(j))}}}

end Construction of the DPM

to graph G(i)(N ,L(i)), where L(i) is the set of links in L that are not in path i. A k1 × k2

DPM of route disjoint path-pair candidates is now available for each source-destination (s, d)

pair, i.e.,

DPMs,d(i, j) : i = 0, 1, . . . , k1 − 1, j = 0, 1, . . . , k2 − 1, ∀ s, d ∈ N , s 6= d (5.10)

where i identifies the working path candidate and j identifies the associated protection path

candidate. Due to the arbitrary topology of the WDM network, it is possible that specific

node pairs may have fewer working path candidates than k1, and/or fewer protection path

candidates than k2.

Let Ws,d (|Ws,d| = k1) and Ps,d,i (|Ps,d,i| = k2) be the set of k1 candidate working path and

k2 candidate protection paths for each candidate working path i ∈ Ws,d between source node

s and destination node d, respectively. Paths are sorted in each set based on their length,

i.e, from the shortest to the longest. A pseudo code that summarizes the algorithm used to

construct the DPM for each source-destination pair (s, d) is shown in Table 5.1.

The computational complexity of building the DPM is related to the computational com-

135

plexity of the k-shortest path algorithm. In the worst case, the computational complexity

of the k-shortest path algorithm in [107] is O(K · |N | · (|L| + |N | · log |N |)), where K, |L|,

and |N | represent the number of computed loopless shortest paths between any given source-

destination pair, the number of links and the number of nodes in the network, respectively.

Let l be the average number of links that belong to each k-shortest path found. The worst

case complexity of the DPM approach is: O(|N |2 · {k1 · |N | · (|L|+ |N | · log |N |) + k1 · (2l +

k2 · |N | · [(|L| − l) + |N | · log |N |]}) = O(|N |3 · (k1 · k2)(|L|+ |N | · log |N |) + |N |2 · k1 · l). By

properly choosing the values of both k1 and k2, it is possible to arbitrarily prune down the

solutions that are available to the optimization process described next.

5.3.2 Step B: the RWA Algorithm

The objective of the RWA algorithm described in this section is to search for the best path-pair

candidate that can be found in matrix DPMs,d, where s and d are the source and destination

of the arriving demand, d. The best path-pair is the one that minimizes the cost function

in (5.9), while satisfying the four conditions in (5.2), (5.3), (5.4), and (5.5). Those solutions

that do not satisfy all conditions in (5.2), (5.3), (5.4), and (5.5) are called unfeasible.

The RWA algorithm consists of two sub-steps. In the first sub-step (Step B.1), the al-

gorithm determines the reliability degree of d with coarse granularity. Depending on the

reliability degree requested by d, i.e., MCFP (d), the chosen working path is either entirely

protected, i.e., H(d)u = ∅, or entirely unprotected, i.e., H

(d)u = H

(d)w . In the second sub-step

(Step B.2), the algorithm attempts to modify set H(d)u to closely match MCFP (d).

136

5.3.2.1 Step B.1: First Fit Algorithm

Upon arrival of demand d, both the working path and wavelength are chosen using First Fit

(FF). The first working path i = 0, 1, . . . , k1 − 1 ∈ DPMs,d(i, ∗) that is found to be able to

accommodate d is chosen. Let i(d) be such a path. Set H(d)w contains all links in path i(d).

The first wavelength λ = 1, 2, . . . , |W | that is found to be available along path i(d) is selected

to be the working wavelength λ(d)w . (If a working path cannot be found in DPMs,d, or no

wavelength is found to be available along path i(d), d is blocked.) If path i(d) does not need to

be protected — i.e., condition in (5.5) is satisfied when H(d)u = H

(d)w — Step B.1 terminates,

and the algorithm continues to Step B.2.

Conversely, if i(d) path needs to be protected, set H(d)u is set to ∅ and the algorithm chooses

the first path j = 0, 1, . . . , k2 − 1 ∈ DPMs,d(i(d), j) that is found to be able to provide a

protection path to d. Let j(d) be such path. Set H(d)p contains all the links in path j(d). All

wavelengths λ = 1, 2, . . . , |W | are, in turn, considered as candidate protection wavelengths.

The wavelength λ that is found able to maximize the value of |H(d)s | — i.e., the number

of protection links of demand d in which λ is shared by at least one other protection path

already routed — is set to be the protection wavelength. (Notice that sharing of protection

wavelengths with the demands already in D is permitted when condition (5.7) is satisfied

given, H(d)u = ∅.) If a protection path j(d) that satisfies the above condition can not be found,

the solution is set to be equal to path-pair DPMs,d(i(d), 0) and the protection wavelength λp

is set to be equal to 0. In this case, the solution found is said to be unfeasible. Regardless of

the feasibility of the found solution, the algorithm continues to Step B.2.

The (worst case) computational complexity of Step B.1 is O(k1 · k2 · |W |2 · l2).

137

5.3.2.2 Step B.2: SA Algorithm

The objective of this step is to reduce the resources (wavelengths) that must be reserved to

satisfy MCFP (d), if possible at all. For this purpose, a SA algorithm is designed to identify

which links that must be in the final sets H(d)w , H

(d)p , and H

(d)u . The cost function to be

minimized by the SA algorithm is the one given in (5.9) for all feasible solutions. Unfeasible

solutions are assigned an arbitrary high cost.

The path-pair found in Step B.1, i.e., i(d) and j(d), is used as the initial solution for running

the SA algorithm. The initial sets H(d)w , H

(d)p , and H

(d)u , and the initial wavelengths λ

(d)w and

λ(d)p are those obtained in Step B.1. At each SA iteration, a neighboring solution is obtained

by randomly choosing one of the following three moves.

1. Randomly select another working path i′(d) 6= i(d) from those in the DPM. If the new

path-pair (i′(d), j(d)) satisfies all conditions in (5.2), (5.3), (5.4), and (5.5), the solution

is said to be feasible, and the first wavelength λ = 1, 2, . . . , |W | that is found to be

available along path i′(d) is selected to be the working wavelength λ(d)w . Conversely, if

the new path-pair does not satisfy all conditions in (5.2), (5.3), (5.4), and (5.5), or no

available working wavelength is found along path i′(d), the new path-pair solution is said

to be unfeasible and another move is randomly selected.

The (worst case) computational complexity of move 1 is O(|W | · l).

2. Randomly select a new protection path j′(d) 6= j(d) from those in the DPM. All wave-

lengths λ = 1, 2, . . . , |W | are, in turn, considered as candidate protection wavelengths for

the new path-pair (i(d), j′(d)) . The wavelength that is found able to maximize the value

of |H(d)s | is set to be the protection wavelength. If no available protection wavelength is

138

found along path j′(d), the new path-pair solution is said to be unfeasible and another

move is randomly selected.

The computational complexity of move 2 is O(|W | · l).

3. Randomly select link (m,n) ∈ H(d)w and

• if (m,n) ∈ H(d)u , (m, n) is removed from H

(d)u and the working wavelength λ

(d)w is

left unchanged,

• if (m,n) 6∈ H(d)u , (i, j) is added to H

(d)u under the condition that the resulting

P(d)f ≤ MCFP (d). The working wavelength λ

(d)w is not changed. If the resulting

P(d)f > MCFP (d), another move is randomly selected.

The computational complexity of move 3 is O(1).

Each of the three moves is equally likely to be chosen. Sets H(d)w , H

(d)p , and H

(d)u , and

wavelengths λ(d)w and λ

(d)p are updated at the end of each move accepted by the SA algorithm.

If a feasible solution is found by the SA algorithm, d is added to set D. Otherwise, d is

blocked.

Let itermax be the number of iterations performed by the SA algorithm each time Step B.2

is executed. Since each of the three moves is equally likely to be chosen, the computational

complexity of Step B.2 in a worst case analysis is itermax

3O(|W | · l) + itermax

3O(|W | · l) +

itermax

3O(1) = O(|W | · l). The overall (Step B.1 and Step B.2) computational complexity for

the RWA algorithm is O(k1 · k2 · |W |2 · l2) + O(|W | · l) = O(k1 · k2 · |W |2 · l2).

139


This section presents a collection of results that are obtained by means of the RWA algorithm

and the DPM pruning technique that are presented in Section 5.3. Both SPP and SPP-DiR

schemes are considered.

To provide a comparison benchmark for the DPM technique, results that are obtained

using the path pruning technique based on the k-shortest loopless paths are also shown. This

benchmark pruning technique is referred to as Linear Based (LB). For LB, candidate path-

pairs are computed as follows. For any possible node pair, only the first k-shortest loopless

paths are considered. All the possible route-disjoint path-pairs that can be generated from

the considered k candidate paths are then used to create the LB matrix. The LB matrix is

then used by the RWA algorithm described in Section 5.3.2. The computational complexity

of the LB solution in a worst case analysis is O(|N |3 ·K · (|L|+ |N | · log |N |) + |N |2 ·K · l2).

Solutions are found for the topology of the European optical network, that is shown in

Figure 5.2(a). This network comprises |N | = 19 nodes and |L| = 39 bidirectional links.

It is assumed that each link accommodates |F | = 1 fiber for each direction of propagation.

Each fiber carries |W | = 32 wavelengths. The conditional link failure probability is obtained

assuming a uniform distribution of failures over all links in L. Hence, Pf (i, j) = 139∀(i, j) ∈ L.

The demand arrivals form a Poisson process with rate λ. Source and destination nodes

of each demand are randomly chosen using a uniform distribution over all possible node

pairs. Unless otherwise specified, each demand is assigned a reliability degree requirement

of MCFP = 0.03. With this value and in the network topology under consideration, each

demand may be able to have up to one working link that is unprotected. Once established,

a demand remains in the system for a time that is exponentially distributed with parameter

140

1µ

= 1. It is assumed that the signaling latency in the network is negligible, and the correct

network status information is available at all nodes.

17

0

18

1

2

3

4

5

6

7

8

9

10

11

12

13

14

15

16

(a) European optical network topology

WDM NETWORK

Input Slot

(b) Virtual single-slot input buffer

Figure 5.2. The European optical network topology and the virtual single-slot input buffer

To provide results that are not dependent upon any specific call admission control, all

arriving demands are first stored in a virtual centralized buffer, as shown in Figure 5.2(b). At

most one demand can be stored in the buffer at once. A demand that upon arrival cannot be

established in the network due to lack of available resources is stored in the buffer until it can

be established. Demands that arrive while the buffer is busy are blocked and dropped. Let

Pb be the probability of blocking and dropping a demand.

For all results, the simulation time is set to achieve a confidence interval value of 5% or

better, at 98% confidence level.

141

Table 5.2. Statistics on candidate pathsScheme k1 k2 NW NP Npp Hcw Hcp

LB k = 20 20 6.512 130.24 4.157 3.778LB k = 60 60 15.868 952.1 5.008 4.646LB k = 100 100 22.594 2259.4 5.536 5.063

DPM 30 10 30 9.282 278.5 4.507 4.435DPM 30 5 30 4.685 140.5 4.507 3.803DPM 20 10 20 9.289 185.8 4.157 4.438DPM 20 5 20 4.688 93.8 4.157 3.819DPM 10 10 10 9.338 93.4 3.604 4.459DPM 10 5 10 4.709 47.1 3.604 3.875

5.4.1 Comparison Between Pruning Techniques

Table 5.2 shows some statistics that are collected on the route-disjoint pair-paths obtained by

both the DPM and LB pruning techniques. From left to right, the table reports the pruning

technique used, the value of k1 and k2 used for building the DPM, NW defined as the average

number of candidate working paths per source-destination pair, NP defined as the average

number of candidate protection paths associated with each working path, Npp defined as the

average number of candidate route disjoint path-pairs per source-destination pair, Hcw defined

as the average hop length of the candidate working paths, and Hcp defined as the average hop

length of the candidate protection paths.

The values reported in Table 5.2 support the earlier claim that by using the DPM pruning

technique the size of the solution space may be reduced when compared to the LB solution

space. In some instances, i.e., when comparing LB with k = 60 and DPM with k1 = 20 and

k2 = 5, the DPM approach is able to reduce the solution space by one order of magnitude.

Table 5.2 also shows that with the DPM pruning it is possible to better control the hop length

of both the working and protection paths.

The top part of Table 5.3 shows results that are collected for the LB technique, using

142

Table 5.3. LB solutions found by the SA algorithm

Rep = 100, λ = 300, MCFP (d) = 0.03k T0 Tf a Pb |Hw| |Hp| |Hs| RCT RCTFF RCTSA

20 6 1 0.9 4.38E-3 2.315 4.019 3.801 7.41E-3 2.09E-4 7.19E-360 6 1 0.9 1.94E-3 2.370 4.597 4.359 1.12E-2 5.69E-4 1.06E-2100 6 1 0.9 3.13E-3 2.412 4.801 4.501 1.18E-2 8.92E-4 1.09E-2

k = 60, λ = 300, MCFP (d) = 0.03Rep T0 Tf a Pb |Hw| |Hp| |Hs| RCT RCTFF RCTSA

100 3 1 0.9 4.03E-3 2.407 4.539 4.237 6.56E-3 5.81E-4 5.97E-325 6 1 0.9 5.73E-3 2.446 4.387 3.944 2.90E-3 4.73E-4 2.42E-350 6 1 0.9 3.41E-3 2.412 4.513 4.192 5.34E-3 5.02E-4 4.82E-3100 6 1 0.9 1.94E-3 2.370 4.597 4.359 1.12E-2 5.69E-4 1.06E-21000 6 1 0.9 9.89E-4 2.290 4.634 4.510 9.56E-2 5.84E-4 9.50E-2100 25 1 0.9 1.50E-3 2.341 4.633 4.438 1.75E-2 5.40E-4 1.69E-2100 100 1 0.9 1.18E-3 2.327 4.640 4.467 2.52E-2 6.13E-4 2.46E-2100 300 1 0.9 1.03E-3 2.318 4.641 4.475 3.18E-2 5.65E-4 3.12E-2500 50 1 0.9 1.10E-3 2.289 4.629 4.508 9.60E-2 5.47E-4 9.54E-2100 6 1 0.99 1.15E-3 2.292 4.634 4.508 1.07E-1 6.27E-4 1.06E-1100 25 1 0.99 7.46E-4 2.282 4.620 4.507 1.98E-1 6.54E-4 1.98E-1100 6 1 0.999 7.42E-4 2.259 4.605 4.508 9.75E-1 5.18E-4 9.75E-1

k = 20, 60 and 100. The best blocking probability is obtained when k = 60. This value is

chosen to obtain all the subsequent results. Statistics collected from various solutions found by

the SA algorithm are reported in the bottom part of Table 5.3 (LB) and in Table 5.4 (DPM).

Simulations are run using Linux boxes with Athlon XP 2200 processors. The complier used is

g++, ver. 3.2.2. Simulation time is measured in seconds. Statistics refer to arriving demand

d with MCFP (d) = 0.03. For DPM k1 = 20 and k2 = 10. From left to right, both tables

report Rep defined as the number of iterations performed by SA at any given temperature,

T0 defined as the starting temperature, Tf defined as the final temperature, a defined as the

cooling factor, Pb, |Hw| defined as the average hop length of the chosen working path, |Hp|

defined as the average hop length of the chosen protection path, |Hs| defined as the average

number of shared links, and RCT defined as the average running time of the RWA algorithm,

RCTFF defined as the average running time of First Fit algorithm, and RCTSA defined as

143

the average running time of the SA algorithm. The cooling function is geometric. The DPM

technique outperforms the LB technique in terms of |Hw|, |Hp|, and ||Hp| − |Hw|| with any

set of SA parameter values shown. Table 5.3 also shows that when the computational time is

limited, i.e., RCT in the order of few milliseconds, the DMP technique is better than the LB

technique in terms of Pb.

For the rest of the chapter, the following SA parameter values are chosen: Rep = 100,

T0 = 6, Tf = 1, a = 0.9.

Table 5.4. DPM solutions found by the SA algorithm

k1 = 20, k2 = 10, λ = 300, MCFP (d) = 0.03Rep T0 Tf a Pb |Hw| |Hp| |Hs| RCT RCTFF RCTSA

100 3 1 0.9 2.11E-3 2.323 4.271 4.058 5.84E-3 2.03E-4 5.64E-325 6 1 0.9 3.43E-3 2.366 4.237 3.934 2.65E-3 1.88E-4 2.46E-350 6 1 0.9 2.03E-3 2.332 4.267 4.038 4.76E-3 1.72E-4 4.58E-3100 6 1 0.9 1.72E-3 2.307 4.271 4.089 1.19E-2 2.33E-4 1.16E-21000 6 1 0.9 1.12E-3 2.270 4.246 4.125 1.12E-1 2.91E-4 1.11E-1100 25 1 0.9 1.45E-3 2.295 4.262 4.102 1.66E-2 2.22E-4 1.64E-2100 100 1 0.9 1.21E-3 2.287 4.258 4.113 2.31E-2 1.88E-4 2.29E-2100 300 1 0.9 1.26E-3 2.284 4.249 4.106 2.91E-2 2.59E-4 2.89E-2500 50 1 0.9 9.14E-4 2.268 4.228 4.100 9.21E-2 1.82E-4 9.19E-2100 6 1 0.99 1.20E-3 2.270 4.249 4.128 9.09E-2 2.25E-4 9.06E-2100 25 1 0.99 9.17E-4 2.261 4.245 4.129 1.63E-1 2.01E-4 1.63E-1100 6 1 0.999 1.16E-3 2.243 4.242 4.142 8.94E-1 2.30E-4 8.94E-1

5.4.2 Comparison of SPP and SPP-DiR Schemes

The results shown in this section provide a performance comparison between the SPP-DiR

and the conventional SPP schemes. As already mentioned, the SPP scheme can offer only

MCFP (d) = 0.

Figures 5.3 and 5.4 show Pb (blocking probability) versus λ (arrival rate) for the SPP-DiR

and the SPP protection scheme, respectively. The plots show that with a mild reduction of

144

250 300 350 400 450 50010

−5

10−4

10−3

10−2

10−1

100

λ

Blo

ckin

g pr

obab

ility

DPM (k1 = 1, k2 = 1)DPM (k1 = 10, k2 = 1)DPM (k1 = 20, k2 = 1)LB (k = 60)DPM (k1 = 20, k2 = 10)DPM (k1 = 20, k2 = 5)

Figure 5.3. Blocking probability (Pb) versus arrival rate (λ): SPP-DiR

the offered reliability degree (MCFP (d) = 0.03), the SPP-DiR scheme may strongly reduce Pb

when compared to the SPP scheme. Moreover, the plots show that the DPM technique better

solves the RWA problem when compared to the LB technique, due to the reduced size of the

solution space in both the SPP-DiR and SPP schemes. The figure also highlights the impor-

tance of making use of multiple candidate path-pairs in obtaining satisfactory performances.

If the values of k1 and/or k2 are too small, Pb is negatively and significantly affected.

Figures. 5.5 and 5.6 plot |Hw| (the average hop length of the working path) versus λ.

Figures. 5.5 and 5.6 plot |Hp| (the average hop length of the protection path) versus λ.

Results obtained for both the SPP-DiR and SPP schemes are shown. The DPM technique is

effective in reducing both |Hw| and |Hp| under any traffic load.

Figures. 5.9 and 5.10 plot |Hs| (the average number of shared protection links) versus λ.

145

102.4

102.5

102.6

10−2

10−1

100

λ

Blo

ckin

g pr

obab

ility


Figure 5.4. Blocking probability (Pb) versus arrival rate (λ): SPP

Results obtained for both the SPP-DiR and SPP schemes are shown. In the case under study,

it is found that by closely matching the demand’s reliability requirement, the SPP-DiR scheme

improves the number of shared protection links by 49% when compared to SPP.

Figure 5.11 shows the normalized average excess of reliability versus λ. The excess of

reliability, defined in 5.9, is averaged over all the serviced traffic requests and normalized to

MCFP = 0.03. The obtained excess of reliability is below 20%. The DPM solution appears

to yield slightly smaller values of excess of reliability when compared to the LB solution.

Simulation results show that the excess of reliability obtained by the DPM solutions with

k1 < 20 and k2 = 1 is equal to the excess of reliability obtained by the DPM solution with

k1 = 20 and k2 = 1.

Figure 5.12 shows Pb versus MCFP (d). Clearly, the plots indicate the existing trade-

146

102.4

102.5

102.6

2.2

2.25

2.3

2.35

2.4

2.45

λ

Ave

rage

leng

th −

wor

king

pat

h


Figure 5.5. Average hop length of the working path (|Hw|) versus arrival rate (λ): SPP-DiR

off between the demand’s guaranteed reliability degree and the blocking probability. Values

shown at MCFP (d) = 0 represent the blocking probability of the SPP scheme. These results

confirm that by attempting to closely match the demand’s reliability requirement, the SPP-

DiR scheme is successful in reducing the average amount of network resources that must be

reserved to establish a newly arrived demand. In turn, this fact may reduce Pb significantly.

5.5 Conclusion

The chapter proposed an approach to dynamically create reliable demands in WDM networks

keeping in mind two objectives: (1) to guarantee the desired demand reliability level while

minimizing the required network resources and (2) to produce satisfactory solutions under

constrained computational time.

147

102.4

102.5

102.6

2.2

2.22

2.24

2.26

2.28

2.3

2.32

2.34

2.36

2.38

λ

Ave

rage

leng

th −

wor

king

pat

h


Figure 5.6. Average hop length of the working path (|Hw|) versus arrival rate (λ): SPP

The first objective was pursued by generalizing the SPP scheme to the SPP-DiR scheme.

The SPP-DiR scheme is applied for the first time to create demands dynamically with the

desired reliability level. The main advantage of this scheme is the ability to guarantee the

demand reliability level, independently of the network topology and size, source-destination

distance, and MTBF of the network elements. In some circumstances, the use of an SPP-

DiR scheme was found to significantly reduce the amount of network resources that must

be reserved for the incoming demand. In turn, this fact was shown to yield a remarkable

reduction of the demand’s blocking probability.

The second objective was pursued by proposing the use of the Disjoint Path-Pair Matrix,

which contains a number of preselected candidate path-pairs for both working and protec-

tion routes. The solution produced by the DPM approach was compared with the solution

148

102.4

102.5

102.6

3.2

3.4

3.6

3.8

4

4.2

4.4

4.6

4.8

5

λ

Ave

rage

leng

th −

pro

tect

ion

path


Figure 5.7. Average hop length of the protection path (|Hp|) versus arrival rate (λ): SPP-DiR

produced by the widely used k-shortest paths approach. In order to provide satisfactory re-

sults, the DPM approach was found to require up to one order of magnitude fewer candidate

path-pairs than the k-shortest paths approach does. For this reason, the DPM approach

is best suitable when the computational time available for choosing each demand routing is

constrained. The DPM solution was also found to require reduced average hop length for

both the working and protection paths (up to 3% and 14% respectively) when compared to

the k-shortest paths solution. This chapter shows only the SPP and SPP-DiR schemes; it

is expected that similar advantages of the DPM approach will be found when other path

protection switching schemes are used.

149

102.4

102.5

102.6

2.2

2.22

2.24

2.26

2.28

2.3

2.32

2.34

2.36

2.38

λ

Ave

rage

leng

th −

wor

king

pat

h


Figure 5.8. Average hop length of the protection path (|Hp|) versus arrival rate (λ): SPP

102.4

102.5

102.6

1.5

2

2.5

3

3.5

4

4.5

λ

Pro

tect

ion

path

− A

vera

ge n

umbe

r of

sha

red

links


Figure 5.9. Average number of shared protection links (|Hs|) versus arrival rate (λ): SPP-DiR

150

102.4

102.5

102.6

1.4

1.6

1.8

2

2.2

2.4

2.6

2.8

3

λ

Pro

tect

ion

path

− A

vera

ge n

umbe

r of

sha

red

links


Figure 5.10. Average number of shared protection links (|Hs|) versus arrival rate (λ): SPP

250 300 350 400 450 5000.14

0.15

0.16

0.17

0.18

0.19

0.2

λ

Nor

mal

ized

ave

rage

exc

ess

of r

elia

bilit

y


Figure 5.11. Normalized average excess of reliability versus arrival rate (λ)

151

0 0.02 0.04 0.06 0.08 0.110

−5

10−4

10−3

10−2

10−1

100

MCFP

Blo

ckin

g pr

obab

ility


Figure 5.12. Blocking probability (Pb) versus MCFP (d)

CHAPTER 6

CONCLUSIONS

This dissertation presented ways to achieve reliable networking for GAP4S over both the

wireless and the wired segments.

In the wireless segment, the dissertation investigated the problem of providing reliable and

fair data transmission from the sensor node to the base-station.

Chapter 3 considered and compared two classes of ARQ protocols. The first is the con-

ventional class of ARQ protocols while the second takes advantage of cooperative communi-

cations. In this second class of ARQ protocols, one or more relays assist the source during

the retransmission process. It has been shown that the advantage of using a cooperative

retransmission mechanism is twofold. First — by asking relays to perform retransmissions

— it is as if sensor nodes could borrow energy from one another and balance their energy

consumption to match their own battery recharging rate. The second advantage of using this

cooperative ARQ protocol derives from its inherent two-hop retransmission mechanism. If the

relay is located between the source and the base-station, data frames are (re)transmitted over

a shorter range, thus reducing the required amount of transmitted energy. However, when

the source does not have a (one-hop) relay in the proximity of the base-station, this type of

retransmission mechanism might not be effective in overcoming the distance from the source

to the base-station.

To cope with the latter problem, Chapter 4 extended the class of the two-hop cooperative

ARQ protocols to a new class of cooperative ARQ protocols based on a multi-hop retransmis-

152

153

sion mechanism. With this new class of protocols, sensor nodes’ data frame delivery to the

base-station is accomplished through a chain of relays, each one within earshot of the other

and each one with a lower data frame error probability.

In a variety of anticipated scenarios, it was found that when the energy available at the

sensor node is the main limiting factor in the system, the two ARQ protocols based on

cooperative communication may more than double the saturation throughput of the non-

cooperative ARQ protocol. Equivalently, it can be said that the energy required to operate

the system may be reduced by half. With acceptable microwave signal levels, it has been

shown that is possible to reach footprint sizes in the hundreds of meters range.

Based on these encouraging results, further study is going to be carried out on cooperative

ARQ protocols applied to wireless sensor networks. For instance, it will be interesting to

investigate the transmission scheduling strategies at the base-station and the medium access

control protocols that are best suited for GAP4S.

In the wired segment, the dissertation investigated the problem of providing end-to-end

survivable optical circuits under the assumption of dynamic traffic conditions. For the first

time the SPP-DiR scheme was proposed to dynamically reserve network resources to incoming

connection requests, with the objective of providing Differentiated Reliability (DiR) by means

of Shared Path Protection (SPP). It was shown that, when compared to the conventional SPP

scheme, the proposed SPP-DiR scheme reduces the overall blocking probability by efficiently

making use of reusable protection wavelengths, while guaranteeing the required reliability

degree of each connection.

The concept of DiR has a wider range of applicability. DiR can be applied in a straightfor-

ward way to any circuit switched network architecture. Additionally, the idea can find very

154

interesting applications in the wireless segment of the GAP4S where different types of reliable

delivery can be required by the sensor nodes, depending upon the nature of the message and

the scope of the delivery.

REFERENCES

[1] P. Gupta. Coded Cooperation Automatic Repeat Request Protocols for Slotted Radio

Networks. Master’s thesis, University of Texas at Dallas, December 2004.

[2] D. Estrin, R. Govindan, J. Heidemann, and S. Kumar. Next Century Challenges: Scal-

able Coordination in Sensor Networks. In Proceedings of Mobicom ’99, pages 263–270,

August 1999.

[3] Standard Wireless Protocols Create New Sensor Networks. Technical Report 8, National

Aeronautics and Space Administration (NASA), August 2000.

[4] I.F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci. Wireless Sensor Net-

works: A Survey. Elsevier Computer Networks, 38(4):393–442, March 2002.

[5] D. Katsaros, A. Nanopoulos, and Y. Manolopoulos, editors. Wireless Information High-

ways. IRM Press, 2005.

[6] S. Mahlknecht. Energy-Self-Sufficient Wireless Sensor Networks for the Home and

Building Environment. PhD thesis, Vienna University of Technology, September 2004.

[7] Picoradio Project. http://bwrc.eecs.berkeley.edu/Research/Pico Radio/.

[8] R. Shad. Energy Scavenging for Wireless Sensor Networks : with Special Focus on

Vibrations . Kluwer Academic, 2004.

[9] µAMPS Project. http://www-mtl.mit.edu/researchgroups/icsystems/uamps/.

155

156

[10] W.R. Heinzelman, A. Chandrakasan, and H. Balakrishnan. Energy Efficient Commu-

nication Protocol for Wireless Micro-Sensor Networks. In Proceedings of the Second

European Workshop on Sensor Networks (EWSN), February 2005.

[11] WIseNET Project. http://www.csem.ch/wisenet.

[12] WINS Project. http://www.janet.ucla.edu/WINS.

[13] WSSN Project. http://www.ict.tuwien.ac.at/wireless/.

[14] EnOcean GmbH. http://www.enocean.de.

[15] GAP4S Project. http://opnear.utdallas.edu/projects/gap4s/gap4s/gap4s home.htm.

[16] A. Sartori, F. Maloberti, A. Simoni, and G. Torelli. A 2D Photosensor Array with

Integrated Charge Amplifier. In Proceedings of Sensors and Actuators, pages 247–250,

1995.

[17] T. Bolthauser and H. Baltes. Capacitive Humidity Sensors in SACMOS Technology with

Moisture Absorbing Photosensitive Poyimide. In Proceedings of Sensors and Actuators,

pages 509–512, 1990.

[18] Micro-Battery Development. http://www.afrlhorizons.com/Briefs/Dec01/PR0104.html.

[19] Micropower for Remote, Authonomous Microsystems. Brigham University. NSF Grant.

No. CMS-9978737.

[20] J.M. Rabaey, M.J. Ammer, J.L. da Silva Jr., D. Patel, and S. Roundy. PicoRadio

supports ad hoc ultra-low power wireless networking. Computer, 33(7):42–48, 2000.

157

[21] T. S. Rappaport. Wireless Communications: Principles and Practice (2nd Edition).

Prentice Hall PTR, 2001.

[22] A. Nosratinia, T.E. Hunter, and A. Hedayat. Cooperative Communication in Wireless

Networks. IEEE Communications Magazine, 42(10):74 – 80, October 2004.

[23] P. Herhold. Cooperative Relaying Protocols and Performance. PhD thesis, Technischen

Universit at Dresden, July 2005.

[24] D. Bertsekas and R. Gallager. Data Networks (2nd Edition). Prentice-Hall, Inc, 1992.

[25] I. Cerutti, A. Fumagalli, and P. Gupta. Delay Models of Single-Source Single-Relay

Cooperative ARQ Protocols in Slotted Radio Networks with Poisson Frame Arrivals.

Technical Report UTD/EE-12/2005, The University of Texas at Dallas, 2005.

[26] T. M. Cover and A. A. El Gamal. Capacity Theorems for the Relay Channel. IEEE

Transactions on Information Theory, 25(5):572–584, September 1979.

[27] M. Gastpar and M. Vetterli. On the Capacity of Wireless Networks: The Relay Case.

In Proceedings of IEEE INFOCOM ’02, pages 1577–1586, New York, NY, 2002.

[28] M. Gastpar, G. Kramer, and P. Gupta. The Multiple-Relay Channel: Coding and

Antenna-Clustering Capacity. In Proceedings of IEEE ISIT, Lausanne, Switzerland,

2002.

[29] J. N. Laneman, G. W. Wornell, and D. N. C. Tse. An Efficient Protocol for Realizing

Cooperative Diversity in Wireless Networks. In Proceedings of IEEE ISIT, page 294,

Washington, 2001.

158

[30] M. Janani, A. Hedyat, T. E. Hunter, and A. Nosratinia. Coded Cooperation in Wireless

Communications: Space-Time Transmission and Iterative Decoding. IEEE Transactions

on Signal Processing, 52(2):362–371, February 2004.

[31] A. Sendonaris, E. Erkip, and B. Aazhang. User Cooperation Diversity–Part I: System

Description. IEEE Transactions on Communications, 51(11):1927–1938, 2003.

[32] A. Sendonaris, E. Erkip, and B. Aazhang. User Cooperation Diversity–Part II: Imple-

mentation Aspects and Performance Analysis. IEEE Transactions on Communications,

51(11):1939–1948, 2003.

[33] E. Zimmermann, P. Herhold, and G. Fettweis. The Impact of Cooperation on Diversity-

Exploiting Protocols. In Proceedings of 59th IEEE Vehicular Technology Conference,

May 2004.

[34] T. E. Hunter and A. Nosratinia. Cooperative Diversity through Coding. In Proceedings

of IEEE ISIT, page 220, Laussane, Switzerland, 2002.

[35] P. Herhold, E. Zimmermann, and G. Fettweis. Cooperative Multi-Hop Transmission in

Wireless Networks. Computer Networks, 49:299–324, June 2005.

[36] J. N. Laneman, D. N. C. Tse, and G. W. Wornell. Cooperative Diversity in Wireless

Networks: Efficient Protocols and Outage Behavior. IEEE Transactions on Information

Theory, 50(12):3062–3080, December 2004.

[37] P. A. Anghel, G. Leus, and M. Kaveh. Distributed Space-Time Coding in Cooperative

Networks. In Proceedings of the Nordic Signal Processing Symposium, Norway, October

2002.

159

[38] P. A. Anghel, G. Leus, and M. Kaveh. Multi-User Space-Time Coding in Cooperative

Networks. In Proceedings of the International Conference on Acoustics, Speech and

Signal Processing (ICASSP), Hong Kong, April 2003.

[39] Y. Xue and T. Kaiser. Cooperated TDMA Uplink Based on Hierarchically Modulated

Alamouti Code. In Proceedings of International Zurich Seminar on Communications

(IZS), pages 200–203, Zurich, February 2004.

[40] J. Hagenauer. Rate-Compatible Punctured Convolutional Codes (RCPC Codes) and

their Applications. IEEE Transactions on Communications, 36(4):389–400, April 1988.

[41] M. C. Valenti and B. Zhao. Distributed Turbo Codes: Towards the Capacity of the

Relay Channel. In Proceeding of IEEE Vehicular Technology Conference, October 2003.

[42] B. Zhao and M. C. Valenti. Distributed Turbo Coded Diversity for the Relay Channel.

IEE Electronics Letters, 39(10):786–787, May 2003.

[43] B. Zhao and M. C. Valenti. Cooperative Diversity Using Distributed Turbo Codes.

In Proceeding of Virginia Tech Symposium on Wireless Personal Commununication,

Blacksburg, VA, USA, June 2003.

[44] E. Zimmermann, P. Herhold, and G. Fettweis. On the Performance of Cooperative

Ralying Protocols in Wireless Networks. European Transactions on Telecommunications

(ETT), 16(1):17–35, 2005.

[45] B. Zhao and M. C. Valenti. Practical Relay Networks: a Generalization of Hybrid-ARQ.

IEEE Journal on Selected Areas of Communications, 23(1):7–18, January 2005.

[46] P. Gupta, I. Cerutti, and A. Fumagalli. Three Transmission Scheduling Policies for a

160

Cooperative ARQ Protocol in Radio Networks. In Proceedings of WNCG Conference,

October 2004.

[47] R. Ramaswami and K. N. Sivarajan. Optical Networks: a Pratical Perspective. Morgan

Kaufmann Publishers, Inc., 1998.

[48] Telecommunication Standardization Sector of ITU. Series G: Transmission Systems and

Media, Digital Systems and Networks: Digital transmission Systems - Digital Networks

- Optical Networks.

[49] M. W. Chabt, E. Grard, and L. Berthelon others. Toward Wide-Scale All-Optical

Transparent Networking: the ACTS Optical Pan-Eurpean Network (OPEN) Project.

IEEE Journal on Selected Areas in Communications, 16(7):1226–1244, September 1998.

[50] A. Jourdan, F. Masetti, G. Soulange, M. Garnot, and M. Sotom. Design and Imple-

mentation of a Fully Reconfigurable All-Optical Crossconnect for High Capacity Multi-

wavelength Trasport Network. IEEE Journal of Lightwave Technology, 14(6):1198–1206,

June 1996.

[51] T. Shiragaki, N. Henmi, T. Kato, M. Fujiwara, M. Misono, T. Shiozawa, and S. Suzuki.

Optical Cross-Connect System Incorporated with Newly Developed Operation and Man-

agement System. IEEE Journal on Selected Areas in Communications, 16(7):1179–1189,

September 1998.

[52] I. Chlamtac, A. Ganz, and G. Karmi. Lightpath Communications: a Novel Approach to

High Bandwidth Optical WAN-s. IEEE Transactions on Communication, 40(7), July

1992.

161

[53] H. Lee, S.-B. Kim, and Y. Go. Modeling and Algorithms for Physical Layer Network

Survivability. In Proceedings of IEEE ICC ’95, volume 1, pages 161–166, June 1995.

[54] O. Gerstel, R. Ramaswami, and G. H. Sasaki. Fault Tolerant Multiwavelength Optical

Rings with Limited Wavelength Conversion. In Proceedings of IEEE INFOCOM ’97,

volume 1, pages 506–515, April 1997.

[55] O. Gerstel, R. Ramaswami, and G. H. Sasaki. Fault Tolerant Multiwavelength Opti-

cal Rings with Limited Wavelength Conversion. IEEE Journal on Selected Areas in

Communications, 16(7):1166–1178, September 1998.

[56] B. Van Caenegem, N. Wauters, and P. Demeester. Spare Capacity Assignment for

Different Restoration Strategies in Mesh Survivable Networks. In Proceedings of IEEE

ICC ’97, volume 1, pages 288–292, June 1997.

[57] B. V. Caenegem, W. V. Parys, F. De Turck, and P. M. Demeester. Dimensioning

of Survivable WDM Networks. IEEE Journal on Selected Areas in Communications,

16(7):1146–1157, September 1998.

[58] M. Gryseels, K. Struyve, M. Pickavet, and P. Demeester. Common Pool Survivability

for Meshed SDH-Based ATM Networks. In Proceeddings of SYBEN ’98, 1998.

[59] P. Demeester, M. Gryseels, A. Autenrieth, C. Brianza, L. Castagna, G. Signorelli,

R. Clemente, M. Ravera, A. Lajszczyk, D. Janukowicz, K. Van Doorselaere, and Y. Hara-

da. Resilience in Multilayer Networks. IEEE Communications Magazine, 37, No 8,

August 1999.

[60] D. Johnson. Survivability Strategies for Broadband Networks. In Proceedings of IEEE

GLOBECOM ’96, volume 1, pages 452–456, November 1996.

162

[61] J. Armitage, O. Crochat, and J.-Y. Le Boudec. Design of a Survivable WDM Photonic

Network. In Proceedings of IEEE INFOCOM ’97, volume 1, pages 244–252, April 1997.

[62] T.-H. Wu. Fiber Optic Survivability. Artech House, 1992.

[63] R. R. Iraschko, M. H. MacGregor, and W. D. Grover. Optimal Capacity Placement

for Path Restoration in Mesh Survivable Network. In Proceedings of IEEE ICC ’96,

volume 3, pages 1568–1574, June 1996.

[64] L. Nederlof, M. Garnot, and F. Masetti. A Comparison of Survivability Methods for

the Optical Layer. In Proceedings of NOC ’97, pages 103–110, June 1997.

[65] L. Wuttisittikulkij and M. J. O’Mahony. Use of Spare Wavelengths for Traffic Restora-

tion in Multi-Wavelenth Transport Network. In Proceedings of IEEE ICC ’96, volume 3,

pages 1778–1782, June 1996.

[66] Y. Miyao and H. Saito. Optimal Design and Evalutation of Survivable WDM Transport

Networks. IEEE Journal on Selected Areas in Communications, 16, No 7, September

1998.

[67] S. Ramamurthy and B. Mukherjee. Survivable WDM Mesh Networks, Part I - Protec-

tion. In Proceedings of IEEE INFOCOM’99, New York, USA, March 21-25 1999.

[68] S. Baroni, P. Bayvel, R. J. Gibbens, and S. K. Korotky. Analysis and Design of Resilient

Multifiber Wavelength-Routed Optical Transport Networks. IEEE Journal of Lightwave

Technology, 17, No 5, May 1999.

[69] M. Garnot, F. Masetti, and M. Sotom. Benefits of Wavelength Conversion in the Di-

mensioning and Configuraion of WDM Optical Transport Networks. In Proceedings of

163

ICC ’96, May 1996.

[70] A. Fumagalli, I. Cerutti, M. Tacca, F. Masetti, R. Jagannathan, and S. Alagar. Surviv-

able Networks Based on Optimal Routing and WDM Self-Healing Rings. In Proceeding

of INFOCOM ’99, 1999.

[71] A. Fumagalli and L. Valcarenghi. Fast Optimization of Survivable WDM Mesh Networks

Based on Multiple Self-Healing Rings. In All-Optical Networking 1999: Architecture,

Control, and Management Issues, volume 3843, pages 44–55, Boston, MA, 19-21 Sep-

tember 1999.

[72] A. Fumagalli, M. Tacca, and I. Cerutti. Minimizing the Number of Optical Crossconnect

Ports in Mesh Networks Based on Line-Switched WDM Self-Healing Ring Protection.

In Proceedings of IC3N ’99, Boston, MA, October 1999.

[73] A. Jukan, A. Monitzer, and H. R. Van As. Service-Specific Recovery of Wavelength

Connections in WDM Networks. In Proceedings of OFC/IOOC ’99, volume 1, pages

164–166, 1999.

[74] A. Jukan, A. Monitzer, and H. R. Van As. QoS-Restorability in Optical Networks.

In Proceedings of 24th European Conference on Optical Communication (ECOC’98),

volume 1, pages 711–712, 1998.

[75] A. Jukan and A. Monitzer. Restoration Strategies with QoS-guarantee for Optical Net-

work Services. In Proceeedings of the EUROPTO Symposium on Broadband European

Networks, volume 3408, pages 228–234. SPIE, May 1998.

[76] M. Gondran and M. Minoux. Graphs and Algorithms. Wiley Interscience, 1979.

164

[77] S. Kirkpatrick, C. D. Jr. Gelatt, and M. P. Vecchi. Optimization by Simulated Anneal-

ing. Science, 220, No 4598, 13 May 1983.

[78] L. Wuttisittikulkij and M. J. Mahony. Design of a WDM Network Using a Multiple

Ring Approach. In Proceedings of IEEE GLOBECOM ’97, volume 1, pages 551–555,

November 1997.

[79] A. Jukan and H.R. Van As. Service-Specific Wavelength Allocation in QoS-Routed

Optical Networks. In Proceedings of IEEE GLOBECOM ’98, volume 4, pages 2270–

2275, 1998.

[80] K. Chi-Hung, M. Mezhoudi, and Y Hu. Comprehensive End-to-End Reliability Asses-

ment of Optical Network Transport. In Proceedings of the Optical Fiber Communications

(OFC) 2002 Conference, March 2002.

[81] M. Ali, G. Chiruvolu, D. Elie-Dit-Cosaque, and L. Tancevsky. QoS Routing in Optical

Networks. In Proceedings of OFC ’02, March 2002.

[82] A. Fumagalli and M. Tacca. Optimal Design of Optical Ring Networks with Differ-

entiated Reliability (DiR). In A. Bianco M. Ajmone Marsan, editor, Proceedings of

the International Workshop on QoS in Multiservice IP Networks, number 1989/2001 in

Lecture Notes in Computer Science, pages 299–313, January 2001.

[83] A. Fumagalli and M. Tacca. Differentiated Reliability (DiR) in WDM Ring without

Wavelength Converters. In Proceedings of IEEE ICC ’01, volume 9, pages 2887–2891,

June 2001.

[84] A. Fumagalli, M. Tacca, F. Unghvary, and A. Farago. Shared Path Protection with

165

Differentiated Reliability. In Proceedings of IEEE ICC ’02, volume 4, pages 2157–2161,

April/May 2002.

[85] K. Wu, L. Valcarenghi, and A. Fumagalli. Restoration Schemes with Differentiated

Reliability. In Proceedings of IEEE ICC ’03, May 2003.

[86] M. Tacca. Differentiated Reliability in WDM Networks. PhD thesis, University of Texas

at Dallas, May 2002.

[87] M. Tacca, A. Fumagalli, A. Paradisi, S. Lakshmanan, S.M. Rossi, A. de Campos Sachs,

D.S. Shah, and F. Unghvary. Differentiated Reliability in Optical Networks: Theoretical

and Practical Results. OSA/IEEE JLT, special issue on Optical Networks, 21(11):2576–

2586, November 2003.

[88] W. R. Heinzelman, A. Chandrakasan, and H. Balakrishnan. Energy Efficient Protocol

for Wireless Microsensor Networks. In Proceedings of Hawaii International Conference

on System Sciences, pages 4–7, January 2000.

[89] SensoNet Project. http://users.ece.gatech.edu/∼weilian/Sensor.

[90] J. Wall and J. Y. Khan. An Advanced ARQ Mechanism for the 802.11 MAC Protocol.

In Proceedings of Australian Telecommunications, Networks and Applications Confer-

ence(ATNAC), 2003.

[91] E. Malkamaki and H. Leib. Evaluating the Performance of Convolutional Codes Over

Block Fading Channels. IEEE Transactions on Information Theory, 45(5):1643–1646,

July 1999.

166

[92] S. B. Wicker. Error Control Systems for Digital Communication and Storage. Prentice-

Hall, 1995.

[93] J. G. Proakis. Digital Communications (Fourth Edition). McGraw-Hill International

Edition, 2001.

[94] Y. Sankarasubramaniam, I. F. Akyildiz, and M. S. W. Energy Efficiency Based Packet

Size Optimization in Wireless Sensor Networks. In Proceedings of First IEEE Interna-

tional Workshop on Sensor Network Protocols and Applications,, pages 1–8, 2003.

[95] Lp solve 5.5.0.3. http://www.geocities.com/lpsolve/.

[96] Company Chipcon: Datasheet CC2400 (Rev 1.2), February 2004.

[97] Optical Network Magazine, Special Issue on Protection and Survivability in Optical

Network. volume 2. SPIE, Kluwer, July/August 2001.

[98] A. Fumagalli and L. Valcarenghi. IP Restoration vs. WDM Protection: Is There an

Optimal Choice? IEEE Network, 14(6):34–41, November/December 2000.

[99] S. Sengupta and R. Ramamurthy. From Network Design to Dynamic Provisioning and

Restoration in Optical Cross-Connect Mesh Networks: An Architectural and Algorith-

mic Overview. IEEE Network Magazine, 15(4):46–54, July-August 2001.

[100] V. Anand and C. Qiao. Dynamic Establishment of Protection Paths in WDM Networks,

Part I. In Proceedings of the 9th International Conference on Computer Communications

(ICCCN), October 2000.

[101] Pin-Han Ho and H. T. Mouftah. Allocation of Protection Domains in Dynamic WDM

Mesh Networks. In Proceedings of the 10th IEEE International Conference on Network

167

Protocols (ICNP), pages 1–2, 2002.

[102] M. Tacca, A. Fumagalli, and F. Unghvary. Double-Fault Shared Path Protection Scheme

with Constrained Connection Downtime. In Proceedings of Design of Reliable Commu-

nication Networks (DRCN), 2003.

[103] G. Mohan, C. Siva Ram Murthy, and A.K. Somani. Efficient Algorithms for Rout-

ing Dependable Connections in WDM Optical Networks. IEEE/ACM Transactions on

Networking, 9(5):553–566, October 2001.

[104] H. Zang, J.Jue, and B. Mukherjee. Review of Routing and Wavelength Assignment Ap-

proaches for Wavelenght-Routed Optical WDM Networks. Optical Network Magazine,

2000.

[105] C. Ou, J. Zhang, H. Zang, L.H. Sahasrabuddhe, and B. Mukherjee. Online Algorithms

for Shared-Path Protection in WDM Mesh Networks. Technical Report CSE-2002-6,

Department of Computer Science, University of California, Davis, March 2002.

[106] S. Dixit C. Xin, Y. Ye and C. Qiao. A Joint Working and Protection Path Selection Ap-

proach in WDM Optical Networks. In Proceedings of IEEE GLOBECOM ’01, volume 4,

pages 2165–2168, November 2001.

[107] J. Y. Yen. Finding the K Shortest Loopless Paths in a Network. Management Science,

17:712–716, 1971.

[108] J.W. Suurballe. Disjoint Paths in a Network. Networks, 4:125–145, 1974.

[109] S. Ramamurthy and B. Mukherjee. Survivable WDM Mesh Networks, Part I - Protec-

tion. In Proceedings of INFOCOM ’99, volume 2, pages 21–25, March 1999.

168

[110] I. Chlamtac, A. Fumagalli, and L. Valcarenghi. Computational Intelligence in Telecom-

munications Networks. CPedrycz and Vasilakos eds, RC Press LLC, 2001.

VITA

Paolo Monti was born in Novara, Italy, on May 12, 1973, the son of Ileana Bogogna and Mario

Monti. After completing his work at the Liceo Scientifico “G. Galilei”, Borgomanero, Italy, he

entered the Politecnico di Torino at Torino, Italy, in 1992. In March 1999 he began working

for Boston Communication Networks as a research engineer until March 2001. He received the

Laurea Degree (Dott. of Engineering) in May 2001. In September 2001 he entered the Grad-

uate School of Engineering of the University of Texas at Dallas, Richardson, Texas. During

that time he was a Research Assistant and a member of the OpNeAR (Optical Networking

Advance Research) laboratory. He contributed in the following publications:

1. “Cooperative and Reliable ARQ Protocols for Microwave Recharged Sensor Nodes”,

submitted for publication to IEEE Transactions on Wireless Communications.

2. “Optimized Transmission Power Levels in a Cooperative ARQ Protocol for Microwave

Recharged Wireless Sensors”, in Proceedings of IEEE ICC ’05, Seoul, Korea, May 2005.

3. “Cooperative and Non-Cooperative ARQ Protocols for Microwave Recharged Sensor

Nodes”, in Proceedings 2nd European Workshop on Wireless Sensor Networks (EWSN),

Istanbul, Turkey, January-February 2005.

4. “The Disjoint Path-Pair Matrix Approach for Online Routing in Reliable WDM Net-

works”, in Proceedings of IEEE ICC ’04, Paris, France, June 2004.

5. “An Energy-Efficient Method for Nodes Assignment in Cluster-Based Ad Hoc Net-

works”, ACM/Kluwer Wireless Networks Journal (WINET), Vol.10, No.3, pp.223-231,

May 2004.

6. “Resource-Efficient Path-Protection Schemes and Online Selection of Routes in Reli-

able WDM Networks”, The OSA Journal of Optical Networking, special issue on Next-

Generation WDM Network Design and Routing, Vol. 3, pp. 188-203, April 2004.

7. “A Differentiated Reliability (DiR) Approach for Dynamic Provisioning in WDM Net-

works”, invited Paper, in Proceedings of the 40th Annual Allerton Conference on Com-

munication, Control and Computing, Monticello, Illinois, October 2002.

8. “Energy Efficient Design of Wireless Ad Hoc Networks”, in Proceedings of IFIP-TC6

Networking 2002, LNCS, Vol. 2345, Springer, May 2002.

Technical Report UTD/EE/13/2005 November 2005

Documents

Transcript of Technical Report UTD/EE/13/2005 November 2005