2001 an investigation of capture effects in ieee 802.11
TRANSCRIPT
University of WollongongResearch Online
University of Wollongong Thesis Collection University of Wollongong Thesis Collections
2001
An investigation of capture effects in IEEE 802.11networksChristopher Graham WareUniversity of Wollongong
Research Online is the open access institutional repository for theUniversity of Wollongong. For further information contact the UOWLibrary: [email protected]
Recommended CitationWare, Christopher Graham, An investigation of capture effects in IEEE 802.11 networks, Doctor of Philosophy thesis, School ofElectrical, Computer and Telecommunications Engineering, University of Wollongong, 2001. http://ro.uow.edu.au/theses/1941
A n Investigation of Capture Effects in IEEE 802.11 Networks
A thesis submitted in fulfilment of the requirements for the award of the degree
Doctor of Philosophy
from
THE UNIVERSITY OF WOLLONGONG
by
Christopher Graham Ware
Bachelor of Engineering (Honours Class I) Bachelor of Science (Physics) University of Wollongong, 1997
SCHOOL OF ELECTRICAL, COMPUTER AND TELECOMMUNICATIONS ENGINEERING
AUGUST 2001
Abstract
Local area wireless networking is quickly becoming a preferred technology for
many networking applications. This is driven by the introduction of the IEEE
802.11 and ETSI HiperLAN standards, combined with the promise of explicit
quality of service mechanisms designed to support exciting new services over
wireless media. These developments have the potential to fundamentally alter
the way a user interacts with the network, as well as opening a raft of potential
new applications able to exploit the inherent benefits of wireless media.
Incorporating both the wireless local area and mobile ad hoc network paradigm,
this thesis presents a comprehensive investigation of the impact of capture ef
fects on the fairness properties of the IEEE 802.11 wireless Medium Access
Control ( M A C ) protocol in topologies involving hidden terminals. Through
empirical investigation, a strong relationship between the relative received sig
nal power of contending hidden connections and the fairness behaviour of the
network is identified. A signal power difference of greater than 5dB between
competing connections was observed to result in a channel capture state for
the stronger connection. This behaviour has a significant impact on the ability
of the M A C to provide fair service to all contending nodes, and in extreme
circumstances can result in extremely poor performance for the weaker hidden
nodes in the network. The signal strength dependent capture behaviour iden
tified in this thesis has been presented within the IEEE 802.1 Working Group,
having significant influence on the design of the Hybrid Co-ordination Function
centralised QoS M A C .
ii
Abstract iii
Analytical investigation of the impact of c o m m o n spreading code interference
confirms the empirical observation, with a received signal power difference of
greater than 2dB found to be sufficient to result in the observed bias. Simulation
tools such as ns-2 play a significant role in the development of new wireless pro
tocols and services. The ability of current receiver models to accurately match
empirical data is investigated. A new model based on the physical operation of
an IEEE 802.11 interface is introduced in response to inadequacies identified in
current receiver models. This model, termed Message Retraining, is shown to
provide a significant improvement over current receiver models in terms of the
ability to match the m o d e m capture characteristics of an IEEE 802.11 network
interface.
Finally, techniques designed to prevent unfair behaviour resulting from relative
signal power dependent capture effects are presented. These techniques are able
to operate within a distributed or centralised M A C . A n algorithm is developed
employing the relative observed signal power to determine a probability variable
for each identified hidden neighbour. This variable is then employed by one of
three separate techniques designed to provide additional transmission opportu
nities for hidden nodes at a relative disadvantage. Each scheme is shown to be
able to significantly improve the fairness characteristics for hidden connections,
preventing stronger hidden hosts from dominating the radio resource. However,
the three schemes are differentiated on the basis of the impact on aggregate
throughput, implementation complexity, and flexibility.
Statement of Originality
This is to certify that the work described in this thesis is entirely m y own,
except where due reference is made in the text.
No work in this thesis has been submitted for a degree to any other university
or institution.
Signed
Christopher Graham Ware
August, 2001
IV
Acknowledgments
Firstly, I would like to thank m y supervisors, Associate Professor Tadeusz
Wysocki and Professor Joe Chicharo. Their advice and support as proven in
valuable throughout this project.
I would like to thank my colleagues in the 'SNRC lab', Paul, Phil, Ben, Ricky,
Chun Tung, and Justin. The technical discussions, coffee, and endless distrac
tions have all contributed significantly to the development of this work. I must
also thank Eryk Dutkiewicz and John Judge for their assistance in the early
stages of this project.
I am also greatful for the support of my parents, Graham and Adele, my brother
Brad, and sister Julie through the years of study. This is it guys, I promise.
And finally, to my wonderful girlfriend Jane whose support and patience has
formed the solid ground that allowed m e to pursue this dream.
v
Contents
1 Introduction 1
1.1 Background 1
1.2 Thesis Outline 2
1.3 Contributions 4
1.4 Publications 6
2 Literature Review 8
2.1 Introduction 8
2.2 Medium Access Control Protocols for
Shared Wireless Media 9
2.2.1 Access Mechanisms 11
2.2.2 Signalling Mechanisms 16
2.2.3 Collision Resolution Algorithms 23
2.2.4 QoS Mechanisms 24
2.2.5 MAC Comparison Matrix 26
2.3 The IEEE 802.11 MAC/PHY Protocol 28
2.3.1 Medium Access Control Layer 29
2.3.2 Direct Sequence Spread Spectrum Physical Layer .... 33
2.3.3 Future Extensions to 802.11 PHY 36
vi
C O N T E N T S vii
2.4 Packet and Channel Capture Phenomena in Wireless Networks . 36
2.4.1 Packet Capture 37
2.4.2 Capture Probability Analysis 37
2.4.3 Channel Capture 39
2.5 Investigation of the Fairness Properties of Wireless LAN's ... 41
2.5.1 Fairness Definitions 42
2.5.2 Experimental Fairness Investigations 43
2.5.3 Mechanisms to Prevent Unfair Behaviour 46
2.5.4 Discussion 50
2.6 Summary 51
2.6.1 Summary of Open Research Issues Identified In Current Literature 52
3 Experimental Investigation of Capture Effects and Fairness Behaviour 54
3.1 Introduction 54
3.2 Experimental Motivation 55
3.3 Experimental Methodology 59
3.4 TCP Experiments 60
3.4.1 RTS Handshake - Hidden Terminals 62
3.4.2 Impact of Varying Signal Strength 65
3.5 UDP Experiments 68
3.5.1 Equal Signal Power 69
3.5.2 Unequal Signal Power 70
3.6 Conclusions 74
CONTENTS viii
4 Error Probability Analysis - Hidden Terminal Jamming 77
4.1 Introduction 77
4.2 Spread Spectrum Error Probability
Analysis 78
4.3 Error Probability of Captured Frame 81
4.3.1 DSSS Basic Rate Physical Layer 83
4.3.2 DSSS High Rate Physical Layer 84
4.4 Numerical Results 85
4.4.1 Single Interferer, K = 2 85
4.4.2 Multiple Interferers, K > 2 86
4.5 The Retraining Hypothesis 88
4.6 Conclusion 91
5 Modelling Packet Capture Behaviour 93
5.1 Introduction 93
5.2 Capture Models 94
5.2.1 Delay Capture 97
5.2.2 Power Capture 98
5.2.3 Hybrid Capture 98
5.3 Message Retraining Reception Model 99
5.4 Simulation Investigation 101
5.4.1 Methodology 101
5.4.2 Simulation Environment 103
5.4.3 UDP Results 104
5.4.4 TCP Results 109
C O N T E N T S ix
5.5 Fairness Study 116
5.5.1 Jain's Fairness Index 117
5.5.2 Kullback-Leibler Fairness Index 117
5.5.3 Results 118
5.5.4 Discussion 123
5.6 Conclusions 124
6 Prevention Of Signal Strength Dependent Unfairness 126
6.1 Introduction 126
6.2 Analysis of Topology Dependent
Unfairness Prevention Algorithms 127
6.3 Average Signal Strength Based
Probability 129
6.3.1 Identification of Hidden Nodes 131
6.4 Algorithms to Control Signal Strength
Dependent Unfairness in Hidden Node
Scenarios 132
6.4.1 Probabilistic Access at Backoff Countdown 136
6.4.2 Probabilistic Discard 137
6.4.3 Enhanced CTS Suppression 137
6.5 Performance Investigation and Comparison 140
6.5.1 Simulation Methodology 140
6.5.2 Comparison Criteria 141
6.5.3 Simple Case - Static Hidden Nodes 143
6.5.4 Static Scenario Discussion 148
6.5.5 General Dynamic Case - Hidden and
In-Range Nodes 150
CONTENTS x
6.5.6 Dynamic Scenario Discussion 155
6.6 Conclusions and Recommendations 157
7 Conclusions 161
7.1 Overview 161
7.2 Significant Results 161
7.3 Further Work 164
A Hidden Node Detection Mechanisms 177
B Additional Fairness Algorithm Results 183
B.l 3 Node TCP Results 183
B.l.l Static topology Results 183
B.l.2 Dynamic Topology Results 187
List of Abbreviations
ACK
AP
BB
BER
BPSK
BSS
CCK
CDMA
CFP
CP
CTS
CW
DBPSK
DCF
DIFS
DSSS
ECTS
EDCF
EIFS
ERTS
FAMA
FHSS
FTP
GAMA
Acknowledgement
Access Point
Black Burst
Bit Error Rate
Binary Phase Shift Keying
Basic Service Set
Complementary Code Keying
Code Division Multiple Access
Contention Free Period
Contention Period
Clear To Send
Contention Window
Differential Binary Phase Shift Keying
Distributed Co-ordinate Function
Distributed (co-ordinate function) Interframe Space
Direct Sequence Spread Spectrum
Enhanced C T S
Enhanced D C F
Extended Interframe Space
Enhanced RTS
Floor Acquisition Multiple Access
Frequency Hopping Spread Spectrum
File Transfer Protocol
Group Allocation Multiple Access
xi
List of Abbreviations xii
HCF
IEEE
IETF
IP
ISM
LAN
LRC
MAC
MACA
MACA-BI
MACAW
MANET
MIB
MMPDU
MPDU
MSDU
NAV
OFDM
PCF
PHY
PLCP
QPSK
QoS
RA
RF
RIMA
RTS
SA
SIFS
SNR
SRAM
Hybrid Co-ordination Function
Institution of Electrical and Electronic Engineers
Internet Engineering Task Force
Internet Protocol
Industrial, Scientific, and Medical
Local Area Network
Long Retry Count
Medium Access Control
Multiple Access Collision Avoidance
M A C A By Invitation
M A C A for Wireless
Mobile Ad Hoc Network
Management Information Base
M A C Management Protocol Data Unit
M A C Protocol Data Unit
M A C Service Data Unit
Network Allocation Vector
Orthogonal Frequency Division Multiplexing
Point Co-ordinate Function
Physical Layer
Physical Layer Convergence Protocol
Quadrature Phase Shift Keying
Quality of Service
Receiver Address
Radio Frequency
Receiver Initiated Multiple Access
Request To Send
Source Address
Short Interframe Space
Signal to Noise Ratio
Split channel Reservation Multiple Access
List of Abbreviations xiii
SSMA Spread Spectrum Multiple Access
SSRC Station Short Retry Count
STA 802.11 Network Station
TCP Transport Control Protocol
UDP User Datagram Protocol
W A N Wide Area Network
W G Working Group
List of Figures
2.1 Illustration of non-persistent C S M A Operation (Tobagi and Klein-
rock, 1976) 12
2.2 The Hidden terminal problem, Node A and C cannot sense each
other's carrier 13
2.3 The Exposed terminal problem, Node B prevents Node C from
transmitting 14
2.4 The Collision Avoidance handshake using RTS/CTS messages . 15
2.5 The FAMA CTS message dominates RTS messages. Host A will
detect the CTS message and backoff accordingly (Fullmer and
Garcia-Luna-Aceves, 1997b) 20
2.6 Hidden terminal collisions with the RTS/CTS exchange. Trans
mission of the RTS at t2 by node C collides with the C T S from
node B at t2, or the RTS from node C at t\ collides with the
D A T A frame from node A. F A M A attempts to prevent collisions
of this type by enforcing dominance of C T S messages over the
RTS (Fullmer and Garcia-Luna-Aceves, 1995) 21
2.7 Elimination Yield NMPA Channel Access Cycles (European Telecom
munications Standards Intsitute, 1998) 28
2.8 Interframe Space Relationships (Institution of Electrical and Elec
tronic Engineers, 1999a) 30
2.9 RTS/CTS/DATA/ACK and NAV Setting. 'Other' node is a node
within range of either Source or Destination. Receipt of RTS or
C T S M M P D U will set N A V accordingly (Institution of Electrical
and Electronic Engineers, 1999a) 31
2.10 PCF Frame Exchange Sequence (Institution of Electrical and
Electronic Engineers, 1999a) 32
xiv
LIST OF FIGURES xv
2.11 Square Experimental Topology (Gerla et al., 1999a) 41
2.12 Example wireless network topology including both hidden and
visible stations (Ozugur et al., 1998). Link access probabilities
are assigned to each logical network link in accordance with Equa
tions 2.6 and 2.7. Arrows indicate the logical links, and probabil
ity pij assigned to each link. Node A has a set of neighbours B.
The set of nodes C are hidden from A, having at least one con
nection with a neighbour of A. All other nodes in the network
belong to a separate set, D 47
3.1 Experimental Topology 56
3.2 Simulation Experiment - Hidden terminals, DATA/ACK only.
Connection A captures the resource until approximately 3 sec
onds when Connection B is able to access the channel. The abil
ity of either connection to capture the channel is random in this experiment 57
3.3 Simulation Experiment - Hidden terminals, RTS/CTS/DATA/ACK 58
3.4 Experiment 1: Lucent Barker PHY Equal SNR 25dB, No RTS/CTS
Handshake 63
3.5 Experiment 2a: Lucent Barker PHY - Equal SNR 25dB for both
connections, aRTSThreshold 500 bytes 64
3.6 Experiment 2b: Cisco CCK PHY - Equal SNR 25dB for both
connections, aRTSThreshold 500 bytes 64
3.7 Experiment 3a: Lucent Barker Code PHY - Unequal SNR Con
nection A 25dB and Connection B 20dB, aRTSThreshold 500
bytes 66
3.8 Experiment 3b: Cisco CCK PHY - Unequal SNR Connection A
20dB and Connection B 25dB, aRTSThreshold 500 bytes .... 66
3.9 Experiment 4: Lucent Barker Code PHY - Controlled SNR, aRT
SThreshold 500 bytes 67
3.10 Experiment 5a: Lucent Chipset CCK PHY - UDP Trace - both
connections 25dB 69
3.11 Experiment 5b: Cisco Chipset CCK PHY - UDP Trace - both
connections 25dB 70
LIST OF FIGURES xvi
3.12 Experiment 6a: Lucent CCK P H Y - UDP Trace - stronger host
(Connection A - 25dB) commencing prior to weaker host (Con
nection B - 20dB) 71
3.13 Experiment 6b: Cisco CCK PHY - UDP Trace - stronger host (Connection A - 25dB) commencing prior to weaker host (Con
nection B - 20dB) 71
3.14 Experiment 7a: Lucent CCK PHY UDP Trace - stronger host
(Connection A - 25dB) commencing after weaker host (Connec
tion B - 20dB) 72
3.15 Experiment 7b: Cisco CCK PHY - UDP Trace - stronger host
(Connection A - 25dB) commencing after weaker host (Connec
tion B - 20dB) 73
4.1 DSSS System Model (Pursley, 1977) 79
4.2 Autocorrelation function for 11-chip Barker sequence, +1,-1,+1,+1,-
1,+1,+1,+1,-1,-1,-1, employed in the 802.11 DSSS PHY 84
4.3 Correlator Output BER Experienced by ith Frame for 2 Mbit/s
Barker spreading code 87
4.4 Correlator Output BER Experienced by Initial Frame for 5.5
Mbit/s C C K Spreading Sequence Set 88
4.5 Correlator Output BER Experienced by Initial Frame for 11
Mbit/s CCK Spreading Sequence Set 89
4.6 Barker Code {K - 1) interferers, Ebk/N0 = 20dB 89
4.7 CCK codes, (K - 1) interferers, Ebk/N0 = 20dB 90
5.1 Potential Slot Time Error 97
5.2 Operation of the Message Retraining model 100
5.3 Trace Data UDP Transport: Lucent Chipset 105
5.4 No Capture Model UDP Transport 106
5.5 Delay Capture Model UDP Transport 107
5.6 Power Capture Model UDP Transport 107
LIST OF FIGURES xvii
5.7 Hybrid Capture Model UDP Transport 108
5.8 Message Retraining Capture Model UDP Transport 109
5.9 Trace Data TCP Transport: Lucent Chipset 110
5.10 No Capture Model TCP Transport Ill
5.11 Delay Model TCP Transport 112
5.12 Power Model TCP Transport 113
5.13 Hybrid Model TCP Transport 114
5.14 Message Retraining Model TCP Transport 114
5.15 UDP Experimental Data Trace obtained with Cisco chipset. Con
nection A commences with an SNR of 20dB 1 second later Con
nection B commences with an SNR of 25dB 119
5.16 Comparison of model fairness performance against experimental
UDP trace data. Top Figure is Jain's index, bottom Figure is
Kullback-Leibler index. Both indices illustrate the ability of the
Message Retraining model to provide an improved match with
empirical data with respect to the Power, Delay and Hybrid models 120
5.17 TCP Experimental Data Trace obtained with Cisco chipset. Con
nection B commences with an SNR of 20dB, 1 second later Con
nection A commences with an SNR of 25dB 121
5.18 Comparison of model fairness performance against experimental
TCP trace data. Top Figure is Jain's index, bottom Figure is
Kullback-Leibler index. Both indices illustrate the ability of the
Message Retraining model to provide an improved match with
empirical data with respect to other models for a fairness horizon
of less than 400 frames 122
6.1 Diagramatic representation of ECTS Suppression Scheme. Nodes
C and D are identified through group address as hidden from
Node A. Node B determines when an Enhanced CTS reply is re
quired to suppress Nodes C and D, allowing Node A fair channel
access. The duration set within the ECTS can be tuned to meet
the specific fairness objective 139
6.2 Three Node Hidden Terminal Topology 144
LIST OF FIGURES xviii
6.3 Three node topology, static scenario for p-Persistence on Backoff
Countdown algorithm using U D P - /3 — 0.25 146
6.4 Three node topology, static scenario for Probabilistic Discard
algorithm using U D P - 0 = 0.40 147
6.5 Three node topology, static scenario for enhanced CTS Suppres
sion algorithm using U D P - /3 = 0.75 149
6.6 Example received SNR trace during dynamic experiment .... 151
6.7 Three node topology, dynamic scenario for p-Persistence on Back
off Countdown algorithm using U D P /3 = 0.25 152
6.8 Three node topology, dynamic scenario for Probabilistic Discard
algorithm using U D P 0 = 0.30 154
6.9 Three node topology, dynamic scenario for Enhanced CTS Sup
pression algorithm using U D P j3 = 0.55 156
A.l Hidden Terminal Message Exchange Semantics. Node 3 observes
node 1 as hidden via C TS and A C K frames 178
A.2 STA 3 observes STA 1 as hidden through the timing constraints
places on C T S and A C K frames. In (a) STA 3 is not hidden from
STA 1, hence Duration and SA fields are consistent. In case (b)
where STA 3 is hidden from STA 1, the cached Duration and SA
fields will be inconsistent with the values observed in the C T S or A C K frames 180
A.3 STA 3 observes STA 1 as hidden as no RTS or DATA frames
have been received to update known neighbour list in STA 3. In
(a) STA 3 is not hidden from STA 1, hence STA 3 is able to
identify and maintain STA 1 in the known neighbours list. In
case (b) where STA 3 is hidden from STA 1, the R A within C TS
and A C K frames will not be found in the known neighbours list 181
B.l 3 node topology, static scenario for p-Persistence on Backoff Count
down algorithm using T C P - optimal /J = 0.25 184
B.2 3 node topology, static scenario for Probabilistic Discard algo
rithm using T C P - optimal j3 = 0.40 185
B.3 3 node topology, static scenario for Enhanced CTS Suppression
algorithm using T C P - optimal 0 = 0.55 186
LIST O F FIGURES xix
B.4 3 node topology, static scenario for p-Persistence on Backoff Count
down algorithm using T C P - optimal /? = 0.05 188
B.5 3 node topology, dynamic scenario for Probabilistic Discard al
gorithm using T C P - optimal /5 = 0.30 189
B.6 3 node topology, dynamic scenario for Enhanced CTS Suppres
sion algorithm using T C P - optimal /3 = 0.25 190
List of Tables
2.1 MAC protocol comparison matrix and summary 27
2.2 Modulation Techniques and Spreading Codes for 802.11 DSSS
PHY 34
2.3 802.11 DSSS PHY Parameters 35
2.4 QPSK Encoding Scheme 35
2.5 Comparison of Fairness Definitions 44
3.1 Summary of Experimental Trials 61
5.1 Modem Simulation Parameters 104
6.1 Comparison of improvement in fairness index and reduced aggre
gate normalised throughput for static and dynamic UDP scenar
ios with each fairness control technique 157
xx
Chapter 1
Introduction
1.1 Background
The recent explosion in popularity of local area wireless networks can be at
tributed to the availability of a standardised Medium Access Control (MAC)
and Physical Layer (PHY) protocol, the IEEE 802.11 standard for wireless Lo
cal Area Networks (LAN's). The standard initially defined a signalling rate of
2 Mbit/s, though several recent enhancements have resulted in the addition of
signalling rates up to 54 Mbit/s to the standard. Traditionally, wireless data
networks have been used in an access paradigm, in which terminals obtain access
to the fixed network via a centralised base station. In addition to the tradi
tional access paradigm, a new application has recently been developed based
on the Mobile A d Hoc Network ( M A N E T ) paradigm. A M A N E T allows nodes
to form a dynamic network configured to suit the application requirements of
participating nodes. Nodes within the network are able to act as both an end
terminal and a router, forwarding packets to the destination via neighbouring
nodes. The dynamic nature of a M A N E T is such that node mobility results
in an ever changing topology. In such circumstances, both hidden nodes and
varying propagation conditions are expected to present significant challenges
for the M A C protocol.
Further, wireless MAC protocols are currently being developed whose aim is
1
Introduction 2
to implement Quality of Service (QoS) mechanisms capable of supporting the
requirements of future demanding applications. Central to the provision of a
QoS guarantee, is the ability of the M A C protocol to provide fair channel access
for competing nodes across a range of scenarios. A significant cause of unfair
behaviour is the so called channel capture problem, in which a given host is able
to dominate the channel at the expense of other hosts in the network. In wireless
networks, this is further complicated by the intrinsic packet capture behaviour of
a radio receiver. Accordingly, a solid understanding of the relationship between
packet and channel capture behaviour, and the impact this has on the overall
fairness properties of the network is extremely important.
1.2 Thesis Outline
This thesis aims to provide a detailed investigation of the impact capture be
haviour has on the fairness properties exhibited in an IEEE 802.11 network.
Chapter 2 initially reviews the development of wireless M A C protocols, leading
to the IEEE 802.11 M A C / P H Y protocol. Following this, literature investigat
ing capture phenomena and related fairness issues is reviewed. In particular, a
lack of empirical work investigating the fairness properties of the IEEE 802.11
M A C in general topology scenarios is identified.
An experimental investigation of the fairness properties of a physical IEEE
802.11 network is presented in Chapter 3. The performance of the 802.11
M A C / P H Y protocol in general topology scenarios involving hidden terminals is
presented. Of specific interest is the ability of the M A C to provide fair channel
access for all competing connections through the R T S / C T S handshake designed
to set up clear transmission opportunities for competing hidden nodes. The im
pact of varying signal strength conditions on the performance of the protocol
are investigated in detail. Throughout this thesis, the terms node, host and
Station (STA) are used interchangeably. In general, the term S T A is employed
when discussing an issue directly related to the IEEE 802.11 standard.
Introduction 3
A significant observation from Chapter 3 is that the ability of the network to
avoid channel capture in hidden terminal scenarios is strongly dependent on
the relative received signal power between hidden connections. In response to
the observations in Chapter 3, Chapter 4 presents an analytic investigation of
the impact interfering transmissions have on the reception of an 802.11 frame.
The scenario under investigation is based on the topology employed in the
experimental investigation of Chapter 3. This analysis investigates the impact
a common code interfering signal has on the successful reception of a frame. The
results of this investigation confirm empirical measurements, indicating that a
difference of greater than 2dB is sufficient to afford preferential reception to a
stronger signal. This analysis has application in the development of appropriate
receiver models for accurate simulation of an IEEE 802.11 radio interface.
Chapter 5 investigates the necessary features required of a packet capture model
to match the empirical data when employed in a simulation environment. Fair
ness indices are employed as criteria to determine the ability of significant cap
ture models to match fairness characteristics of empirical trace data. A new
capture model, based on the design of a physical 802.11 R F front-end, is pro
posed to address shortcomings in current models. The outcome of Chapter 5 is
an accurate simulation technique with specific application in the development
of mechanisms to prevent the unfair behaviour observed in physical systems.
Combining the results of previous chapters, Chapter 6 develops techniques to
overcome the significant unfairness evident in hidden terminal scenarios. Given
the number of different P H Y protocols defined within the current IEEE 802.11
standard, the development of specific techniques to prevent such behaviour
within each P H Y is not practical. Any mechanism designed to prevent relative
signal power dependent unfairness must be able to operate with any P H Y ,
in the same manner as the M A C . Therefore, the techniques developed in this
chapter use information taken from the P H Y to provide the M A C with the
ability to identify and prevent relative signal power dependent unfair behaviour.
A mechanism to identify potential unfairness based on average relative signal
Introduction 4
power amongst neighbouring nodes is developed. This mechanism determines
a probability variable for each identified neighbour based on observations of
received signal power. This probability variable is then employed one of three
by techniques developed to provide additional transmission opportunities for
weaker nodes. The three techniques presented are:
• p-Persistence on Backoff Count down, in which the channel access prob
ability of offending nodes is controlled in proportion to their received
relative signal power
• Probabilistic Discard, in which a common node is able to discard RTS or
D A T A frames from offending nodes thereby forcing the offending node
into a backoff period
• Enhanced CTS Suppression, making use of a new interpretation of the
R T S / C T S exchange to enforce a suppression period on an offending node
The performance of each scheme is determined, along with analysis of imple
mentation issues within the IEEE 802.11 M A C framework.
Finally, Chapter 7 concludes the thesis with a summary of the major results
obtained in earlier chapters, as well as presenting a summary of related open
research issues in the area of wireless local area networks.
1.3 Contributions
Below is a list of the major contributions of this thesis, and the section in which
they appear. Relevant publications are also cited with each contribution.
1. Empirical investigation of fairness properties of a CSMA/CA MAC em
ploying an R T S / C T S / D A T A / A C K handshake (IEEE 802.11) using both
greedy U D P sources and T C P sources in hidden terminal topologies (Sec-
Introduction 5
tions 3.4 and 3.5) (Ware et al., 2000; Ware et al., 2001b; Ware and
Dutkiewicz, 2001)
2. Experimental analysis of fairness properties observed with hidden termi
nals in dynamic signal strength scenarios for an IEEE 802.11 wireless
local area network. This investigation identifies a 5dB threshold effect
in the fairness characteristics of the IEEE 802.11 M A C / P H Y protocol
(Sections 3.4 and 3.5) (Ware et al., 2000; Ware et al., 2001b; Ware and
Dutkiewicz, 2001)
3. Derivation of analytical expressions describing the characteristics of a hid
den terminal collision for the 802.11 1, 2, 5.5 and 11 Mbit/s physical layers,
assuming a B P S K modulated signal. This provides an analytical basis for
the threshold effect at 3-5 dB in the relative signal power based unfair
behaviour (Sections 4.3) (Ware et al., 2001a; Ware et al., 2001b)
4. Application of network layer fairness, measured through both Jain's Fair
ness Index (Jain et al., 1984) and the Kullback Leibler Index (Koksal et al.,
2000), in determining the suitability of packet capture models for accurate
802.11 network interface simulation. This work involves a detailed quanti
tative comparison of empirical data with simulation trace data generated
using Delay, Power, and Hybrid capture models (Sections 5.4 and 5.5).
The Delay, Power, and Hybrid capture models are shown to be unable
to accurately match the fairness characteristics of the empirical data in
terms of either magnitude or timescale of fairness behaviour (Ware et al.,
2001c; Ware et al., 2001d; Ware et al., 2001e)
5. Development and performance characterisation of the Message Retrain
ing capture model based on operation of IEEE 802.11 radio interface de
sign. This is developed in response to identified inadequacies in the Delay,
Power, and Hybrid packet capture models in terms of ability to match rel
evant fairness characteristics. The message retraining model is shown to
more accurately match the fairness characteristics, in terms of both fair
ness index magnitude and timescale of the empirical data, than the Delay,
Introduction 6
Power, and Hybrid models (Section 5.3) (Ware et al., 2001c; Ware et al.,
2001d; Ware et al., 2001e)
6. Development of the relative signal strength based heuristic technique to
identify unfair network conditions due to relative received signal power
differences in hidden terminals scenarios (Section 6.3)
7. Development and detailed performance investigation of the p-Persistence
on Backoff Count down fairness control mechanism (Section 6.4.1)
8. Development and detailed performance investigation of the Probabilistic
Discard mechanism to control relative received signal power dependent
unfairness (Section 6.4.2)
9. Development and detailed performance investigation of the Enhanced
C T S Suppression mechanism to control relative received signal power de
pendent unfairness (Section 6.4.3)
1.4 Publications
Publications arising from work directly related to this thesis are listed below:
Journal Publications
Ware, C. G., Wysocki, T., and Chicharo, J. F. (2001c). Simulating Capture
Behaviour in IEEE 802.11 Radio Modems. Journal of Telecommunications and
Information Technology.
International Conferences
Ware, C. G., Judge, J., Chicharo, J. F., and Dutkiewicz, E. (2000). Unfairness
and Capture Behaviour in 802.11 Adhoc Networks. In International Conference
on Communications, ICC 2000, volume 1, pp 159-163 New Orleans. IEEE Press.
Ware, C. G., Wysocki, T., and Chicharo, J. F. (2001a). Hidden Terminal
Jamming Problems in IEEE 802.11 Mobile Ad Hoc Networks. In International
Introduction 7
Conference on Communications, ICC 2001, volume 1, Helsinki.
Ware, C. G., Wysocki, T., and Chicharo, J. F. (2001e). Modelling Capture
Behaviour In IEEE 802.11 Radio Modems. In International Conference on
Telecommunications, ICT 2001, Special Sessions Volume, Bucharest. IEE /
IEEE.
Ware, C. G., Wysocki, T., and Chicharo, J. F. (2001d). Simulation of Capture
Behaviour In IEEE 802.11 Radio Modems. Vehicular Technology Conference,
Fall 2001, New Jersey. IEEE V T C FALL.
IEEE 802.11 Working Group Contributions
Document IEEE 802.11-01/058, "Network Capture and DCF QoS", Monterey
January 2001.
Document IEEE 802.11-01/232, "Packet Capture UDP Experiments", Florida,
May 2001.
Patents
Provisional Patent has been filed regarding the fairness control mechanisms
presented in Chapter 6
Journal Papers currently under review
Ware, C. G., Wysocki, T., and Chicharo, J. F. (2001b). On The Hidden Termi
nal Jamming Problem in IEEE 802.11 Ad Hoc Networks, submitted to: IEEE
Transactions on Vehicular Technology.
Chapter 2
Literature Review
2.1 Introduction
In recent years there has been a significant increase in the popularity of Wire
less LAN's, driven by the development of the IEEE 802.11 standard. Initially,
wireless LAN's were significantly slower than wired counterparts, operating at 1
or 2 Mbit/s. However, recent improvements in the IEEE 802.11 standard have
increased channel bit rates to 11 Mbit/s and with the addition of the 5.2 G H z
band and Orthogonal Frequency Division Multiplexing ( O F D M ) signalling, up
to 54 Mbit/s will be possible in the near future. With this increase in bit
rate, multimedia applications will inevitably be possible, and therefore issues
regarding quality of service and network fairness will come to the fore.
A further driving force behind local area wireless networking is the development
of the M A N E T . A M A N E T is a network in which nodes are mobile, and able
to form dynamic connections with neighbours. Nodes may even act as routers,
forwarding traffic through the network on behalf of other nodes. The Inter
net Engineering Task Force (IETF) are currently developing network routing
protocols which will allow a group of nodes to form a M A N E T .
The removal of the need for physical connectivity with the network introduces
great flexibility in the way users are able to interact with and use the network.
8
Literature Review 9
Fundamental to each of these developments is the performance of the P H Y
and M A C protocols used in each network interface. The combination of the
M A C and P H Y must be able to provide fair access to the channel, within the
constraints of the broadcast medium, and also within the physical constraints
of an R F signal environment. Fair access is required to allow the development
of suitable QoS mechanisms. In this thesis, a local area wireless network is
considered to constitute any wireless network that covers a local geographic
area, either a traditional base station - client style wireless L A N , or a M A N E T .
In this chapter, we review the current state of wireless MAC protocol research,
with a specific aim of identifying key research issues which remain to be ad
dressed. Section 2.2 provides an overview of current wireless M A C protocols
proposed in the literature. As the IEEE 802.11 protocol is the most widely
available, Section 2.3 describes in detail the operation of the M A C and P H Y
layers of the IEEE 802.11 protocol. As capture effects may adversely effect the
fairness properties of the network, Sections 2.4 and 2.5 review the current state
of capture model development and analysis, and literature on fairness properties
in wireless M A C protocols respectively. Section 2.6 outlines key research issues
remaining to be resolved.
2.2 Medium Access Control Protocols for Shared Wireless Media
Given recent advances in VLSI technology, bringing down the cost of compo
nents enabling high speed wireless packet networks, it comes as no surprise the
significant number of wireless M A C protocols having been proposed in litera
ture. Packet radio networks were first introduced in the late 1960's with the
A L O H A system (Abramson, 1970). However, a revolution has taken place in
recent years with the introduction of the IEEE 802.11 wireless L A N standard
(Institution of Electrical and Electronic Engineers, 1999a), and the ETSI Hiper-
L A N I standard (European Telecommunications Standards Intsitute, 1998).
Literature Review 10
M A C protocols perform a number of vital functions. The basic task is to de
termine when a transmit opportunity exists for each node, thereby providing a
mechanism to provide access to a shared medium for a large number of nodes. In
the case where a node is unable to transmit immediately, a backoff mechanism
is required. The M A C must also provide reliable data transfer through retrans
mission of lost frames. The M A C also provides an interface to the P H Y , acting
as a buffer between the network layer and the timing constraints introduced by
the P H Y .
In this review, we describe and categorise the mechanisms employed by signifi
cant M A C protocols proposed in literature. This should be considered as an in
troduction to the historical development of wireless M A C protocols. This review
will lead to the combination of various features that have been standardised in
the IEEE 802.11 (Institution of Electrical and Electronic Engineers, 1999a) and
HiperLAN (European Telecommunications Standards Intsitute, 1998) Wireless
L A N M A C standards. In this review, we restrict the discussion to random ac
cess approaches, as fixed assignment time division approaches have been shown
to be very inefficient when the number of terminals is large and each terminal
is transmitting a bursty traffic stream (Tobagi and Kleinrock, 1976). Further,
M A C protocols implemented in wireless LAN's today (Institution of Electri
cal and Electronic Engineers, 1999a),(European Telecommunications Standards
Intsitute, 1998) are based on random access approaches. While polling based ap
proaches (Section 2.2.1.2) may also be considered as fixed assignment schemes
when employed in certain scenarios, we are considering applications in both
wireless LAN's and M A N E T ' s in which dynamic random access is achieved us
ing a contention free polling mechanism. The aim of this discussion is to group
M A C protocols into relevant categories, identifying the important features of
each before highlighting mechanisms that have been incorporated into the M A C
protocols employed in current wireless LAN's and MANET's.
Literature Review 11
2.2.1 Access Mechanisms
The mechanism employed to determine when a node is able to transmit is
a fundamental property of the M A C protocol, and can be used to categorise
M A C protocols into two main groups. The first are those that rely on a physical
or virtual carrier sense (Kleinrock and Tobagi, 1975; Tobagi and Kleinrock,
1975; Kara, 1990; Fullmer and Garcia-Luna-Aceves, 1995), and the second
are those that rely on the reception of a polling beacon indicating a current
transmit opportunity (Tobagi and Kleinrock, 1976; Tzamaloukas and Garcia-
Luna-Aceves, 1999; Talucci et al., 1997; Garcia-Luna-Aceves and Tzamaloukas,
1999). Whether the node senses the physical channel, a separate signalling
channel or a virtual channel, the basic technique remains common: listen to the
channel, if an ongoing transmission is detected defer and backoff, otherwise this
is a potential transmission opportunity. The alternative is to rely on another
controlling entity to poll each node, indicating the transmit opportunity for the
polled node. In this section we outline the mechanisms employed by the major
M A C protocols.
2.2.1.1 Carrier Sense Approaches
A carrier sense mechanism provides a terminal with sufficient information to
prevent collisions occurring in the immediate area. The basic approach was
initially proposed and analysed by Tobagi and Kleinrock as an improvement
on slotted A L O H A , (Kleinrock and Tobagi, 1975). Carrier Sense Multiple Ac
cess (CSMA) is a multiple access technique where each node senses the channel
thereby avoiding collisions before transmission. A significant number of M A C
protocols have extended this approach (Tobagi and Kleinrock, 1975; Tobagi and
Kleinrock, 1976; Karn, 1990; Bharghavan et al., 1994; Fullmer and Garcia-Luna-
Aceves, 1995; Institution of Electrical and Electronic Engineers, 1999a; Euro
pean Telecommunications Standards Intsitute, 1998; W u et al., 2000), though
the basic carrier sense access mechanism remains the same.
CSMA has been proposed in both a p-persistent form, where a node transmits
3 0009 03287429 4
Literature Review 12
Unsuccessful Transmission
Vulnerable Period Successful Transmission
I I Node D - Busy channel, defer 2nd Collision - Node C
1st Collision - Node B
Busy Period \ Channel sensed idle Node A Transmission Attempt
Idle Period Busy Period All Stations Backoff Station A returns, senses idle
channel and transmits
Figure 2.1 Illustration of non-persistent C S M A Operation (Tobagi and Kleinrock, 1976)
onto an idle channel with probability p or defers to the next timeslot with prob
ability (1 — p), and a non-persistent form where a node will undertake a binary
exponential backoff on sensing a busy channel (backoff approaches will be dis
cussed in Section 2.2.2.1). The non-persistent form of C S M A is illustrated in
Figure 2.1. Each station senses the channel, and if no carrier is detected, trans
mission commences. The vulnerable period at the start of each transmission
period is the interval during which a carrier sense operation may not detect a
transmission having just commenced from another node. In Figure 2.1, node
A commences transmission on detection of an idle channel. Nodes B and C
also detecting an idle channel during the vulnerable period, commence trans
mission. Node D however, senses the channel after the vulnerable period, and
detecting carrier, enters a backoff state. Node A is the first to sense the channel
again after an idle backoff period, and there being no transmissions during the
vulnerable period, successfully transmits the packet.
The most significant problem for common channel carrier sense wireless MAC
protocols is known as the hidden terminal problem. Both the hidden terminal
and exposed terminal problems, have a significant impact on the performance
Literature Review 13
Q Node A
Carrier Sense- Idle Channel
Collision
Q NodeC
Carrier Sense- Idle Channel
Figure 2.2 The Hidden terminal problem, Node A and C cannot sense each other's carrier
of the C S M A protocol, reducing performance to a level equal with the A L O H A
protocol (Tobagi and Kleinrock, 1975). The hidden node problem arises when
two nodes attempting to communicate with a common node, are unable to
sense a carrier from each other. In Figure 2.2, nodes A and C are attempting
to transmit a frame to node B. Node A senses a clear channel and commences
transmission. Node C's carrier sense also finds the channel clear, and therefore
commences transmission. The result is a collision at node B which is unde
tectable by either node A or C. This is a significant problem, as neither node
A or C has any ability to prevent such a collision. Also, as all collisions occur
at the receiver, node A and C are forced to wait for a timeout on each packet
before retransmitting.
The exposed terminal problem, as illustrated in Figure 2.3, arises when a trans
mission from node B to node A prevents node C from transmitting, even when
the indented recipient from node C (i.e. node D ) is out of range of node A. In
this case, node C would be able to transmit without colliding with the trans-
o NodeB
Literature Review 14
/
i i i i J . — » .
t Carrier Sense - Busy Channel Cannot transmit to Node D
Figure 2.3 The Exposed terminal problem, Node B prevents Node C from transmitting
mission from B to A, but is prevented as the carrier sense operation detects the
transmission from B to A.
To solve the hidden terminal problem, a signalling mechanism is required to
indicate when a node intends to transmit a data packet. This will then allow
potential interferers an opportunity to defer a transmission that would oth
erwise resulted in a receiver side collision. C S M A with Collision Avoidance
( C S M A / C A ) (Colvin, 1983) was one of the first protocols proposing a mech
anism to implement such a handshake. The collision avoidance mechanism is
based around a control message handshake between the intending transmitter
and receiver. As illustrated in Figure 2.4, a Request-To-Send (RTS) message
is sent by the intending transmitter. The intended receiver then responds with
a Clear-To-Send (CTS) message indicating the transmitter may now send the
impending data frame. Nodes in range of either the transmitter or receiver will
receive the R T S or C T S message addressed to another node and will be able to
determine the duration of the impending transmission and defer until after this
transmission concludes. C S M A / C A is the basic access mechanism employed in
o Node A
o NodeB
Node D O o NodeC
Literature Review 15
0 Node A Node A Receives CTS
RTS CTS DATA
Idle Channel Send RTS
Node A Commences DATA Tx
0 NodeB RTS CTS DATA
Node B receives RTS, Responds with CTS
Node C Detects CTS, defers Tx
0 NodeC CTS
Figure 2.4 The Collision Avoidance handshake using RTS/CTS messages
IEEE 802.11.
2.2.1.2 Polling Based Approaches
The main alternative to C S M A style approaches, are those based on a polling
mechanism. A node is unable to transmit until it receives a token, or polling
beacon indicating a transmit opportunity. The mechanism must also include a
mechanism to add a new node to the polling list which is maintained by a cen
tralised node, such as a base station. This technique results in a contention free
transmission mechanism generally not requiring carrier sense for normal data
transmission. The allocation (or election) of the node controlling the polling
process provides a distinction between proposals of this type. In a wireless L A N
paradigm, a base station acts as a centralised co-ordinator, alternatively in an
ad hoc paradigm, the polling may be receiver based (Garcia-Luna-Aceves and
Tzamaloukas, 1999).
Analysis of the basic polling approach following a simple roll call process is
provided by (Tobagi and Kleinrock, 1976). This idea is extended to include a
Literature Review 16
separate reservation channel on which a station is able to request bandwidth.
The polling station places each request in a queue. Stations are then polled
in accordance with individual requests. The reservation channel may use a
contention mechanism, allowing stations to place themselves on the polling list.
This technique, termed Split Channel Reservation Multiple Access (SRMA) is
very similar in nature to the Point Co-Ordinate Function (PCF) adopted within
the IEEE 802.11 M A C .
Many schemes have been presented which attempt to reverse the collision avoid
ance mechanisms. Proposals such as Multiple Access Collision Avoidance By In
vitation (MACA-BI) (Talucci et al., 1997; Talucci and Gerla, 1997) and Receiver
Initiated Multiple Access (RIMA) (Garcia-Luna-Aceves and Tzamaloukas, 1999;
Tzamaloukas and Garcia-Luna-Aceves, 2001; Tzamaloukas, 2000) require each
station to poll surrounding nodes indicating a readiness to receive. While such
schemes have been shown to provide good performance in the presence of hid
den terminals, they require a potential receiver to have excellent knowledge of
a potential neighbours traffic pattern in the form of either a statistical history
or detailed traffic model. For this reason, receiver polling schemes are still in
the relatively early stages of development.
2.2.2 Signalling Mechanisms
As discussed in Section 2.2.1.1, solving the hidden node problem requires a
signalling mechanism to provide information about impending transmissions
for all nodes that may potentially cause interference. In this section, we out
line the two basic approaches used to solve this problem; mechanisms using a
common channel approach (Karn, 1990; Bharghavan et al., 1994; Fullmer and
Garcia-Luna-Aceves, 1995; Fullmer and Garcia-Luna-Aceves, 1997b; Fullmer
and Garcia-Luna-Aceves, 1997a), and those which split the available radio chan
nel and create a separate signalling channel. As outlined in Section 2.2.1.2, there
are also a group of schemes which attempt to reverse the collision avoidance
handshake, relying on the receiver to initiate a data transfer from an intending
Literature Review 17
transmitter (Tzamaloukas, 2000; Garcia-Luna-Aceves and Tzamaloukas, 1999;
Tzamaloukas and Garcia-Luna-Aceves, 2001; Talucci et al., 1997; Talucci and
Gerla, 1997). With the exception of the M A C A - B I scheme (Talucci et al., 1997;
Talucci and Gerla, 1997), schemes such as Receiver Initiated Collision Avoid
ance (RICA) are designed for slow frequency hopped channels (Tzamaloukas,
2000; Garcia-Luna-Aceves and Tzamaloukas, 1999) and will not be considered
further. Finally, we highlight likely performance issues for each scheme in real
network environments where transmission is unreliable. A significant shortcom
ing of work in this area is a lack of rigorous investigation in realistic propagation
environments.
2.2.2.1 Common Channel Signalling
A popular technique to overcome hidden terminal collisions is to use a common
channel signalling mechanism. The majority of schemes are based around the
RTS/CTS handshake, with C S M A / C A being the most simple protocol of this
type. As mentioned earlier, many protocols have extended the basic C S M A ac
cess technique to handle hidden node problems through a variety of signalling
handshakes. Another key feature of each single channel approach is the backoff
algorithm invoked when a busy channel is detected. The following discussion
highlights significant features of M A C protocols proposed in the literature em
ploying a common channel signalling technique.
(a) Multiple Access with Collision Avoidance - MACA
MACA was first proposed by Karn (Karn, 1990) as a new channel access tech
nique for packet radio, inspired by the C S M A / C A method which was used in
Apple Localtalk networks. The mechanism removes the carrier sense compo
nent of C S M A / C A . The author (Karn, 1990) argues that in cases where hidden
and exposed terminals are present, the carrier sense mechanism will provide an
inaccurate result at the receiver. A n indication that transmission is possible
when this may in fact result in a collision at the receiver, or vice versa, is highly
probable. A n R T S message is transmitted before every data frame. If a C T S
Literature Review 18
is not received, the sender will undergo binary exponential backoff where the
backoff counter is doubled every collision and reset to the minimum value on
successful transmission.
MACA was extended by Bharghavan (Bharghavan et al., 1994), with the ad
dition of a link layer positive acknowledgement (ACK) packet, and the inclu
sion of a mechanism to distribute backoff counter values amongst competing
nodes. The resulting protocol, M A C A W , is designed for a pico-cellular envi
ronment where hidden and exposed terminals are very common. The additions
to M A C A are in response to a channel capture state which may arise when
two stations compete for the same shared resource using standard M A C A . One
station is able to maintain a continuously lower average backoff window, and
therefore achieve access to the channel more frequently. The distribution of the
backoff counter value with each data packet forces each node to have the same
backoff window value at the start of the next contention period. The authors
(Bharghavan et al., 1994) claim this mechanism allows the M A C protocol to
distribute bandwidth fairly amongst nodes, though does so at the expense of
overall throughput.
In addition to this, the backoff mechanism was adjusted to implement a mul
tiplicative increase linear decrease algorithm rather than the binary exponen
tial backoff algorithm which can lead to significant oscillation in the backoff
window values. O n each collision, the backoff window is increased by a mul
tiplicative factor of 1.5, and decreased by 1 for each successful transmission.
Results presented in (Bharghavan et al., 1994) illustrate that the combination
of mechanisms included in M A C A W are able to overcome the channel capture
problem.
However, potential problems exist in exposed terminal configurations. The Data
Send (DS) packet was therefore introduced by the authors (Bharghavan et al.,
1994) to indicate that the R T S / C T S exchange was successful. This is designed
to alleviate cases where an exposed terminal cannot receive a C T S when the
common middle node is transmitting, effectively preventing the exposed node
Literature Review 19
from transmitting. The additional D S signalling packet is designed to increase
the robustness of the RTS/CTS exchange. Interestingly, there has been no
additional work on the inclusion of the additional signalling packet, and interest
in standardisation groups never materialised.
(b) Floor Acquisition Multiple Access - FAMA
FAMA, initially proposed by Garcia-Luna-Aceves, (Fullmer and Garcia-Luna-
Aceves, 1995; Fullmer and Garcia-Luna-Aceves, 1997b; Fullmer and Garcia-
Luna-Aceves, 1997a), is an attempt to unify many of the single channel RTS/CTS
based M A C protocols. The design of F A M A aims to overcome scenarios where
RTS, C T S and D A T A frames may collide, by forcing correct acquisition of the
'floor' prior to transmission (equivalent to the geographic area defined by the
transmission radius of a specified node). F A M A has been presented with a
number of variations:
1. FAMA employing RTS/CTS without carrier sense (Fullmer and Garcia-
Luna-Aceves, 1995). This is identical to the M A C A protocol.
2. F A M A - N C S employing RTS/CTS with a non-persistent carrier sense
mechanism (Fullmer and Garcia-Luna-Aceves, 1995; Fullmer and Garcia-
Luna-Aceves, 1997b). This is similar to the IEEE 802.11 C S M A / C A
approach, however, C T S packets are lengthened to be larger than the
aggregate of an RTS message, one maximum round trip time, transmit
to receive turnaround time, and any processing time. As illustrated in
Figure 2.5, this gives the C T S dominance over RTS messages. In the two
cases illustrated, station A either transmits an RTS frame just after or
before station B commences a C T S transmission. The length of the C T S
message has the effect of an in-band busy tone, jamming station A as a
potential interferer who will backoff after the unsuccessful RTS.
3. F A M A employing non-persistent packet sensing designed for a peer re
ceiver (Fullmer and Garcia-Luna-Aceves, 1997b; Fullmer and Garcia-Luna-
Aceves, 1997a). In this variant all stations are assumed to have the same
Literature Review 20
A
B
, *—*
RTS
i '
i
/ ' i
i ,
/
i
CTS
h *-
e - * •
receiver
e
4 i
noise/jamming at A
RTS| e
\ i
\ 1 \ 1 \ 1
\ 1
\ 1
» 1 I t > 1 1 i
1 /
>
< ••' >
/
CTS ;
t
i
e RX/TX turnaround time t Propagation delay
Figure 2.5 The F A M A CTS message dominates RTS messages. Host A will detect
the CTS message and backoff accordingly (Fullmer and Garcia-Luna-Aceves, 1997b)
functionality, following the M A N E T paradigm. Timing constraints are
quite tight in order to force correct 'acquisition' of the floor.
4. FAMA employing non-persistent packet sensing with a base station re
ceiver (Fullmer and Garcia-Luna-Aceves, 1997a). This scheme is similar
to the peer receiver scheme, though the timing constraints are relaxed on
the base station node when an RTS message is received. In this variant,
the base station is assumed to be able to sense carrier from all nodes, and
is the intended recipient of all frames from surrounding terminals.
Analytical performance studies (Fullmer and Garcia-Luna-Aceves, 1998) ap
plied to F A M A - N C S in an ad hoc network scenario illustrate that F A M A - N C S
outperforms A L O H A , C S M A , C S M A / C A , M A C A , and M A C A W when hid
den terminals are present in the network. In cases where no hidden terminals
are present, F A M A - N C S suffers with the extra overhead introduced by the ex
tended duration CTS. These results are not surprising, as F A M A - N C S is able
Literature Review 21
RTS DATA t2 tl
A and C hidden A and C hidden
Figure 2.6 Hidden terminal collisions with the RTS/CTS exchange. Transmission of the RTS at t2 by node C collides with the CTS from node B at t2, or the RTS from node C at t\ collides with the D A T A frame from node A. F A M A attempts to prevent collisions of this type by enforcing dominance of CTS messages over the RTS (Fullmer and Garcia-Luna-Aceves, 1995)
to prevent potential hidden interferers from transmitting, rather than reacting
to hidden terminal collisions as in the basic R T S / C T S approaches. This is il
lustrated in Figure 2.5 where F A M A attempts to prevent collisions of this type
through the extended duration of the C T S message. F A M A trades off increased
overhead against a reduced number of uncontrolled collisions. IEEE 802.11 out
lined in Section 2.3 does not use the lengthened C T S message, as the additional
overhead would significantly reduce effective network capacity.
2.2.2.2 Multi Channel Signalling
A second approach to the prevention of hidden terminal collisions is to split the
radio channel into separate data and control channels, using the control channel
exclusively to prevent potentially interfering transmissions.
Busy Tone Multiple Access (BTMA) was first proposed in the seminal paper
identifying the hidden terminal problem by Tobagi and Kleinrock (Tobagi and
Kleinrock, 1975). The basic operation of the protocol requires that the available
resource is split into a messaging channel and a busy tone channel. W h e n a
station senses carrier on the messaging channel, a tone is broadcast on the busy
Literature Review 22
tone channel. This is a receiver based scheme where each node that is receiving
a transmission is able to block any other node from colliding with the ongoing
transmission. Prior to transmission, a node will sense the busy tone channel,
and when carrier is detected, transmission will be deferred.
An extension to the BTMA concept has since been proposed, termed Dual
Busy Tone Multiple Access ( D B T M A ) (Deng and Haas, 1998; Deng and Haas,
1999). D B T M A includes a busy transmit tone, as well as a busy receive tone.
A separate control channel is also introduced, on which nodes will employ an
RTS/CTS exchange prior to an attempted transmission on the data channel. In
a similar manner to B T M A , a node will transmit a tone on the busy transmit
channel while transmitting, or on the busy receive tone while receiving. Before
transmitting an R T S message, a node will sense the busy receive channel. A
clear channel indicates that no nodes are receiving and the transmission can go
ahead. W h e n a node receives an RTS, it must sense the busy transmit channel
to ensure it is able to reply with a CTS. This basic mechanism is followed
for all transmissions. Combined with the RTS/CTS exchange, the mechanism
decouples the transmission directions allowing an otherwise exposed node to
transmit. Analysis of this mechanism (Deng and Haas, 1999) illustrates that
the benefit of continual channel sensing provided by the additional tones leads
to a throughput performance improvement over M A C A of up to 60-70% in
scenarios where hidden and exposed nodes are present.
Extension of the DBTMA protocol to include power control (Wu et al., 2000)
again shows further improvement in ad hoc network scenarios. Power control is
employed to manage the topology, thereby reducing the number of interfering
nodes for a given transmission pair. The RTS/CTS handshake is employed to
gauge the appropriate power level for the data transmission. A number of other
multi-channel M A C protocols (Nasipuri et al., 1999; Chandra et al., 2000; Tang
and Garcia-Luna-Aceves, 2000) have been proposed in the literature, however;
in terms of implementation, multi-channel M A C protocols are not popular.
Multi-channel mechanisms introduce significant complexity in the receiver, for
Literature Review 23
example D B T M A requires 3 separate channels, and also requires the receiver
to be capable of transmitting a busy tone at the same time as receiving a data
frame without causing interference to the data frame. This has been a significant
hurdle for R F interface designs to overcome. However, the recent development
of O F D M systems will potentially open the way for the implementation of
multichannel M A C protocols.
2.2.3 Collision Resolution Algorithms
The mechanism employed to resolve collisions can be broken into two ap
proaches; those employing a backoff mechanism, and those which employ per
sistence. Backoff mechanisms have long been the preferred collision resolution
mechanism in both wired and wireless M A C protocols. The most common
technique, originally applied in Ethernet C S M A / C D networks was a binary
exponential backoff, where the backoff period was successively doubled each
time a busy medium or collision is detected. Binary exponential backoff has
been shown to be unfair in many cases (Bharghavan et al., 1994) and may lead
to a channel capture state where one station is able to maintain a statistically
lower average backoff counter, thereby gaining preferential access to the channel.
Improvements have been made to this basic algorithm, for example the Multi
plicative Increase, Linear Decrease (MILD) approach presented by Bharghavan
(Bharghavan et al., 1994). A backoff distribution algorithm is also presented by
the authors (Bharghavan et al, 1994) to prevent a station from maintaining a
continuously lower backoff counter than surrounding neighbours. The approach
taken in the IEEE 802.11 M A C is to use a backoff period dictated by a uniform
random variable drawn on [0,CWmax\. CWmax increases exponentially with
each successive backoff.
Persistence approaches can considered to be a statistical technique of imple
menting a backoff counter (Cali and Gregori, 2000; Geraniotis and Soroushne-
jad, 1987). Each station attempts to access the channel, and on finding a clear
channel will transmit with a probability of p, or defers for a slot time with prob-
Literature Review 24
ability (1 — p). W h e n a busy channel is detected, the station delays transition
by a single slot time. If the channel is detected free after the deferred period,
the same process is repeated. If the channel is busy, the station will continue
waiting until the channel is free, then engage in the channel access procedure.
This technique combines the contention resolution algorithm with the channel
access mechanism.
The application of a persistence technique has the disadvantage of requiring sta
tions to generate a pseudo-random variable prior to each packet transmission.
However, persistence techniques are considered to have superior fairness proper
ties and resilience to channel capture behaviour often observed with exponential
backoff mechanisms (Bharghavan et al., 1994).
2.2.4 QoS Mechanisms
Recently, the inclusion of mechanisms to support quality of service differenti
ation at the M A C layer has become an important feature of a wireless M A C
protocol. From the preceding discussion, it is evident that most M A C protocols
proposed in recent years have been designed with best effort traffic in mind.
This is partially a result of the difficulty involved in building a physical wireless
packet network, and the lack of a need historically for quality of service differ
entiation. This is no longer the case, and a number of M A C protocols and QoS
mechanisms have been developed recently to address this need.
Transmission scheduling approaches have been presented in (Lu et al., 1999;
Luo et al., 2000). These schemes rely on nodes to schedule the transmission
of data packets and are generally considered as a network wide optimisation
problem that, given the level of complexity, must be solved using approximate
methods. The most common approaches are based on the use of utility func
tions. Scheduling mechanisms of this type typically attempt to provide bounds
on throughput, delay, delay jitter, and loss for individual traffic stream. The
main drawback of such approaches is the requirement for accurate distributed
knowledge of the network state. This limits the ability of the scheduling ap-
Literature Review 25
proach to maintain a steady state, particularly in a dynamic network.
Tiered contention or virtual MAC schemes are currently being considered for
inclusion in the IEEE 802.11 M A C protocol (Chesson et al., 2001). In schemes
of this type, each node maintains a number of M A C queues, with a virtual
C S M A / C A M A C applied on each queue. Differentiation between streams is
achieved by weighting the deferral period between detection of an idle channel
and eventual transmission (the DIFS in the case of 802.11). A higher priority
queue has a shorter interval, and will generally be served before a lower priority
queue. This results in a statistically higher throughput for classes with a lower
deferral period. Disadvantages with a tiered contention approach include an
inability to provide bounds or guarantees on delay, throughput, or delay jitter.
The quality of service provided is in the form of differentiation between service
classes.
Group Allocation Multiple Access (GAMA) has been proposed in (Muir and
Garcia-Luna-Aceves, 1997) as a technique to provide guarantees to real time
traffic through the allocation of groups. Nodes are allocated positions in a
transmission group. The channel is divided into a contention period, and a
group transmission period. Each node uses the RTS/CTS exchange during the
contention period to assert a position within a group, corresponding to a trans
mit opportunity within the group transmission period. Each node maintains a
group position while it has data to transmit.
A similar, more rigorous scheme termed Black Burst Contention is presented
by the authors of (Sobrinho and Krishnakumar, 1999). In this scheme, nodes
contend for the channel using a Black Burst (BB) jamming tone of a length pro
portional to the period of time the impending packet has been queued. Each
node senses the channel at the end of a B B to determine if it was the longest of
the contending BB's. The winning node then transmits the impending packet
without delay, while other nodes wait for the channel to become idle again, be
fore commencing another round of BB's. Nodes requiring time bounded service
employ this method, while nodes carrying best effort traffic revert to standard
Literature Review 26
C S M A / C A . This scheme is shown to provide an upper bound on delay for real
time traffic, without significantly increasing average delay for best effort traffic.
The performance of B B Contention and G A M A with hidden nodes has not been
established. The analysis presented thus far in the literature assumes all nodes
have perfect carrier sense information.
Ideas from both GAMA and the BB Contention MAC have been incorporated
in Elimination Yield Non-pre-emptive Priority Multiple Access (EY-NPMA),
which has been included as the primary M A C in the HiperLAN I standard (Eu
ropean Telecommunications Standards Intsitute, 1998). This approach can be
based on either a synchronised channel access cycle which includes a number of
distinct phases to establish priority amongst contending nodes prior to trans
mission, or a channel free access cycle containing only the transmission phase
as illustrated in Figure 2.7. In the synchronised access cycle, the role of the
Prioritisation phase is to allow each node to establish the priority of impending
transmissions. Nodes surviving this phase all have an equal priority transmis
sion pending. The Elimination phase then eliminates as many as possible of
the remaining nodes. The Yield phase is the final stage in this process, com
plementing the elimination phase by resolving any further contention. Finally,
the transmission phase allows transmission for the successful node.
This is quite a complex approach to a distributed MAC QoS mechanism, though
several researchers have illustrated that the E Y - N P M A M A C is able to pro
vide reasonable, stable performance for higher priority streams (Anastasi et al.,
2000).
2.2.5 MAC Comparison Matrix
Table 2.1 provides a summary and a useful comparison of the MAC protocols
presented in this section in terms of the salient features identified in the preced
ing discussion. That is, each M A C protocol is classified in terms of the access
technique, signalling mechanism, and the QoS mechanism employed.
Literature Review 27
03
=3 CO
X • -H SH
c3
(=1 O CO i-i
o3 ft
a o CJ
"o u o +3
o FH
ft
U m
i — i
F Q
0
a • FH
a CO
.a o
CO
o
or
a • F H
F-H
a a bX)
• FH
CO
CD
a • FH
a Fa u
8 CO CO CD
u u < o o 0 FH
OH
O
CO
a o I=I
a CI 03
.a o a o
a a o CJ
co
a o a
co i=l
o a
a a a CJ
a o
a a o
CO
a o a
1
co a o a
hO
a CO =3
a <$ a a «3 M
.FH "-1
H-=
13 & FO T—1
a
co a eg a Si ^ CO _FH
CO *§ CO >> • H CO
£ a 03 -Q
o
H PQ
CO
a o a
faC
a , --H
CO !=H
a «§ a a • -H «
-a * a ce
H
CO CO
a CO CO F H
CO • F H F H
FH
03 CJ
a -a
H PQ Q
CO
a o a
a a o3
.a o a o
a a o CJ
a • H
ft
l-H
CQ i
<
CO
a o a
r—H
CO
a a 03
•6 a o
a a o CO
FH co a ci
o § ., - F H
si FH a CO 03 03 F-,
" 13
<
time bounded
transmission
"co
a a 03 Fa CJ
a o
a a o o
a o
• F-H
03 H-3
co a co S
a « M Is FH CO
CO 03 •C FO
«3 & CJ a
o FH
bO
<
time bounded
transmission
"co
a a 03
-a CJ
a o
a a o CJ
Fa a
•S -2
< ° co a O FQ
a o
• F H
a CO
-s o O PQ PQ
possible EDCF
HCF combines EDCF and scheduler
a
•s a o
a a o CJ
-a a o3
O O
CO
O
i-H H
CN O 00
2 •FH
a o
• F H
-F^
03
a
a 3
"co
a a o3 Fa CJ
a o
a a o CJ <
Pu
F—1
F5
< F-3
OH h—1
Literature Review 28
1
«= 9»
Prioritisation Phase
3hase < >
Phase ^ >
< ;»<
Contention Phase
Transmission Phase
Channel Free Channel Access Cycle
—=»
|
Synchronised Channel Access Cycle
Figure 2.7 Elimination Yield N M P A Channel Access Cycles (European Telecommunications Standards Intsitute, 1998)
2.3 The IEEE 802.11 MAC/PHY Protocol
The IEEE 802.11 standard (Institution of Electrical and Electronic Engineers,
1999a), combined with subsequent enhancements, 802.11b (Institution of Elec
trical and Electronic Engineers, 1999b) and 802.11a (Institution of Electrical
and Electronic Engineers, 1999c), defines a M A C Protocol, and four distinct
P H Y protocols. This includes; an Infra-Red (IR) P H Y , a Frequency Hopping
Spread Spectrum (FHSS) P H Y operating in the 2.4 G H z Industrial Scientific
and Medical (ISM) band, a Direct Sequence Spread Spectrum (DSSS) P H Y
in the same 2.4 G H z ISM band, and a P H Y operating in the 5.2 G H z ISM
band employing Orthogonal Frequency Division Multiplexing ( O F D M ) . In the
following sections, the basic operation of the M A C protocol and DSSS P H Y are
outlined. Future planned enhancements are discussed in Sections 2.3.1.3 and
2.3.3.
Literature Review 29
2.3.1 Medium Access Control Layer
The MAC layer within 802.11 employs one of two distinct techniques to solve
the multiple access problem. The first, a distributed contention based approach
employing C S M A / C A , is termed the Distributed Co-ordinate Function (DCF)
and is described in Section 2.3.1.1. The second is a contention free polling
based mechanism, termed the Point Co-ordinate Function (PCF), and will be
described in Section 2.3.1.2.
2.3.1.1 Distributed Co-ordinate Function
The Distributed Co-ordinate Function (DCF) is a contention based approach
employing C S M A / C A with a random backoff time following the detection of
a busy channel. All D A T A frame transmissions are immediately positively ac
knowledged by an A C K frame. The sender is then able to schedule retransmis
sion of the D A T A frame if an A C K is not received within a specified time. The
basic access mechanism is described as follows:
1. Prior to a Station (STA) transmitting, the medium is sensed to determine
if another STA is transmitting
2. If the medium is determined to be busy, the STA will defer until the end
of the current transmission then undergo a backoff period governed by
equation (2.1) before attempting retransmission
3. If the medium is determined to be free, the STA will defer for a D C F Inter-
Frame Space (DIFS) period, ensuring the media has been idle throughout
this period before transmitting the frame
Timing relationships are illustrated in Figure 2.8. Two additional guard times
are shown in this Figure; the Short InterFrame Space (SIFS) used to separate a
D A T A and A C K frame, and the Point Coordination Function InterFrame Space
(PIFS) used by the Point Co-ordinator in the Point Co-ordination Function
mode. Several enhancements to the C S M A / C A mechanism are also included.
Literature Review 30
Immediate access when medium is free > DIFS
DCS
Busy Medium
DIFS Contention Window
Backoff Window Next Frame
Defer Access
Slot time
Select Slot and Decrement Backoff as long as medium is idle
Figure 2.8 Interframe Space Relationships (Institution of Electrical and Electronic Engineers, 1999a)
A Request-to-Send (RTS) and Clear-to-Send (CTS) exchange may be employed
prior to the transmission of a M A C Protocol Data Unit ( M P D U ) or M A C
Management Protocol Data Unit ( M M P D U ) . As described in Section 2.2.2, the
RTS/CTS exchange is a mechanism to prevent hidden terminal collisions, by
way of reservation of the Wireless Medium ( W M ) for the duration of the planned
D A T A transmission. Several simulation studies (Gerla et al., 1999b; Fullmer
and Garcia-Luna-Aceves, 1997a; Bharghavan et al., 1994) have illustrated that
the RTS/CTS exchange is able to provide good immunity to hidden terminal
collisions in cases, whilst maintaining good fairness properties. W e investigate
the performance of this mechanism in later sections of this thesis.
Virtual carrier sense is employed through the Network Allocation Vector (NAV).
The N A V is a counter which counts down to zero, representing the time at
which the wireless medium will become idle. RTS, CTS, and D A T A frames
may set the N A V of an STA with the Duration field, indicating the period of
time the impending transmission will occupy the W M . Figure 2.9 illustrates the
RTS/CTS handshake, and setting of the N A V for neighbouring stations. When
a station senses the W M , the N A V is also checked. If the counter is zero the
W M is considered idle, and when non-zero the W M is considered busy. The
carrier sense operation combines the physical detection and virtual carrier sense
indication to determine the state of the W M .
Literature Review 31
Source
DIFS
RTS
Destination
Other
SIFS SIFS
CTS
DATA
SIFS
ACK
N A V (RTS)
N A V (CTS)
Defer Access
DIFS Contention Window
/ / /
Backoff After
Defer
Figure 2.9 RTS/CTS/DATA/ACK and N A V Setting. 'Other' node is a node within range of either Source or Destination. Receipt of RTS or C T S M M P D U will set N A V accordingly (Institution of Electrical and Electronic Engineers, 1999a)
A random backoff mechanism is used for all PDU's (with the exception of
A C K frames) when carrier sense indicates a busy channel. The backoff pe
riod commences once an S T A has deferred through an impending transmis
sion, then determined the medium to be free for a DIFS period. A random
value, RandomQ, is drawn from a uniform distribution on the interval [0, C W ]
where C W represents the current Contention Window, an integer in the range
[CWmin, CWmax]. aSlotTime is a Management Information Base (MIB) vari
able defined independently for each P H Y . With each successive backoff, CW is
increased exponentially. The backoff interval is then calculated as:
Backoff Time = RandomQ x aSlotTime (2.1)
The D C F can operate in either an Access Point (AP) mode, where each station
is associated with an AP, or in an ad hoc mode where peer-to-peer connections
are made possible with each neighbour by the distributed nature of the D C F .
In this case, the D C F M A C is reasonably well suited to the M A N E T multiple
access application.
Literature Review 32
Contention Free Repetition Interval
Contention Free Period
SIFS SIFS SIFS PIFS SIFS
Beacon Dl + Poll D2 + ACK
+ poll
D3+ACK
+poIl D 4 + poll
D x = Frames sent by Point Co-oidinator Ux = Frames sent by polled stations
— Contention Period
PIFS
NAV
Ul+ACK + poll U2 + ACK
U4 + ACK CF-END
Reset NAV SIFS SIFS SIFS
Figure 2.10 PCF Frame Exchange Sequence (Institution of Electrical and Electronic Engineers, 1999a)
2.3.1.2 Point Co-ordinate Function
The Point Co-ordinate Function (PCF) is a mechanism to provide contention
free M P D U transfer. In this mode, the A P assumes the role of a Point Co
ordinator (PC), which by implication prohibits ad hoc mode operation. Fig
ure 2.10 illustrates the polling technique, and the co-existence of the Contention
Free Period (CFP) with the Contention Period (CP). The P C maintains a
polling list of STA's to poll during the CFP. STA's are able to request ad
dition to the polling list during the CP. A beacon frame is used to signify the
start of the CFP. During the C F P the P C sequentially sends any D A T A frames
along with a polling beacon to each STA on the polling list. STA's respond
with a D A T A frame (if there is D A T A to send). ACK's are piggy backed onto
D A T A frames, along with Polling frames from the AP.
During the CFP, the NAV is set to the duration of the CFP by the point
co-ordinator. This effectively prevents any station transmitting during this
period without the permission of the point co-ordinator. Non CF-Pollable STA's
are able to operate as normal, with the C F P appearing as a (large) single
transmission. This mechanism, while being complex, is very stable under high
Literature Review 33
load.
2.3.1.3 Future QoS Extensions
Currently, the IEEE 802.11 Working Group (WG) is making a significant effort
to add QoS mechanisms to the 802.11 M A C protocols (TGe, 2001). Extensions
are currently under development for the D C F (Chesson et al., 2001), along with
a new M A C termed the Hybrid Co-Ordinate Function (HCF) (Fischer, 2001).
The H C F is designed operate in place of the P C F to provide a centralised QoS
enabled M A C . The tiered contention, or virtual D C F scheme (Chesson et al.,
2001) discussed in Section 2.2.4 is currently quite advanced through the IEEE
802.11 W G , though a final QoS enhancement for 802.11 is still in early stages
of development. A final QoS standard is anticipated for early 2002.
2.3.2 Direct Sequence Spread Spectrum Physical Layer
The DSSS PHY in 802.11 currently provides 4 different bit rates. As illustrated
in Table 2.2, each of the 4 data rates employ a unique combination of modulation
technique and spreading code to achieve the desired symbol rate, and number of
bits per symbol. The Basic Rate (BR) comprises the 1 and 2 Mbit/s data rates,
and employs a Barker spreading code with D B P S K or D Q P S K respectively.
The common 11 chip code used by all stations for both the 1 and 2 Mbit/s
physical layers is:
-r-1,-1,+1,+1,-1,+1,+1,+1,-1,-1,-1
The High Rate DSSS (HR-DSSS) physical layer, comprising the 5.5 and 11
Mbit/s rates, employs Complementary Code Keying (CCK) with a spreading
code of length 8, generated by a generalised Hadamard transform, Equation
(2.2), where 4>i is added to all code chips, <j)2 to all odd code chips, 03 to all odd
pairs, and 04 to all odd quads of code chips. In each case, the chipping rate is
Literature Review 34
Table 2.2 Modulation Techniques and Spreading Codes for 802.11 DSSS PHY
Bit Rate (Mbit/s)
1 2 5.5 11
Coding Scheme
Barker Sequence (11 Chip) Barker Sequence (11 Chip) CCK or optional BCC CCK or optional BCC
Modulation Technique
DBPSK DQPSK DQPSK QPSK
Bits per Symbol
1 2 4 8
11 Mchip/sec.
c _ eJ(<t>l+<p2+(t>3+(p4) eJ(01+03+04) eJ{<t>l+<t>2 + <t>i)
_eJ(</»l+04) eJ(<t>l+<t>2 + <!>3) eJ{<Pl+<l>3) _ g j ( 0 1 + 0 2 ) eJ4>l (2.2)
Equation (2.2) is used to create 8 complex chips (c0 to CT) with c0 transmitted
first in time. For CCK 5.5 Mbit/s modulation at 4 bits/symbol, 0i is encoded
by data bits d0 and d\ based on DQPSK. Data bits d2 and d3 CCK encode the
basic symbol by:
02 = d2 X 7T + | (2.3)
03 = 0 (2.4)
04 = di X 7T (2.5)
This leads to a family of 16 distinct spreading sequences which are used to
indicate the symbol transferred.
Table 2.3 lists several key parameter values for the DSSS PHY.
For CCK 11Mbit/sec modulation (at 8 bits/symbol), 0i is again encoded by
d0 and dl using DQPSK. Data bits (d2,dz), (d4,d5),and (d6,d7) are used to
QPSK encode 02, 03, and 04 respectively, as shown in Table 2.4 This leads to
a matrix of 256 spreading sequences, which support the transmission of 8 bits
per symbol.
Literature Review 35
Table 2.3 802.11 DSSS PHY Parameters
Parameter
aSlotTime aSIFSTime aCCATime aCWmin aCWmax DIFS PIFS EIFS
Value
20 /is 10 lis < 15/zs
31 1023
aSIFSTime + 2 x aSlotTime aSIFSTime + aSlotTime aSIFS + (8xACKSize)
Table 2.4 QPSK Encoding Scheme
Bit Pattem[di, di+i]
00 01 10 11
Phase
0 TT/2
7T
3TT/2
Literature Review 36
2.3.3 Future Extensions to 802.11 P H Y
Work is already underway within the IEEE 802.11 WG to extend bit rates
supported by the P H Y protocol (Webster and Halford, 2000; Institution of
Electrical and Electronic Engineers, 1999c). The first of these is the recently
standardised 802.11a (Institution of Electrical and Electronic Engineers, 1999c)
employing an O F D M physical layer at 5.2 GHz. The O F D M P H Y employs
52 sub-carriers, spaced at 0.3125 M H z intervals within the allocated frequency
band, comprising 48 data sub-carriers and 4 pilot sub-carriers. Prior to trans
mission, the data is encoded, interleaved, then mapped to a series of complex
numbers according to the BPSK, Q P S K , 16-QAM, or 64-QAM encoding ta
ble. A n inverse Fourier transform is then performed on the data stream, and
the result mapped to the 48 data sub-carriers to form the O F D M symbol for
transmission. The receiver essentially reverses this operation. The choice of
modulation technique is based on the bit rate required.
There are also a number of additional extensions to the 802.11 standard under
development. In addition to the QoS mechanisms mentioned in Section 2.3.1.3,
there is a proposal before the 802.11 W G aiming to provide bit rates beyond
11 Mbit/s in the 2.4 G H z band (Webster and Halford, 2000). A power control
mechanism is also under development, with two competing proposals (Hansen
et al., 2001; Choi et al., 2001) currently before the working group.
2.4 Packet and Channel Capture Phenomena in Wireless Networks
Capture behaviour is a crucial aspect of the performance characteristic of any
practical wireless network. T w o distinct capture phenomena can be identified:
Packet Capture, and Protocol Capture. In the following discussion, we review
both capture phenomena, describing the significance of each in the context of a
wireless M A C protocol.
Literature Review 37
2.4.1 Packet Capture
Packet, or modem capture is a result of the ability of a modem to 'lock' onto a
signal in the presence of received interference caused by either an external noise
source or another transmission within the same region. This behaviour is a re
sult of the combination of both m o d e m design and the P H Y protocol employed.
In particular, both Frequency Modulation and Spread Spectrum receivers which
use a phase locked loop or correlation detection circuit display this ability (Rap-
paport, 1996). A number of authors have illustrated the improved throughput
performance of a network with some form of capture capability, with (Lau and
Leung, 1992; Arnbak, 1987; Goodman and Saleh, 1987) representing the most
significant work in this area. In each case, the approach has been to determine
the improvement in channel throughput a specific capture model may provide.
This is outlined in the following section.
2.4.2 Capture Probability Analysis
Over the past 25 years, a significant body of literature has developed investi
gating the impact packet capture behaviour has on network throughput across
a range of scenarios. The original wireless M A C protocol, A L O H A , was first
analysed in the late 70's by Abramson (Abramson, 1977). In this seminal paper,
Abramson determined the throughput performance of both a pure and slotted
A L O H A system employing a very basic capture model. The model dictates the
successful reception of a packet transmitted from a distance r from the receiver
unless overlapped by another packet broadcast from another user at a distance
slightly greater than r. In the latter case, both packets would be lost at the re
ceiver. The analysis presented in (Abramson, 1977) illustrates how the capture
effect increases the throughput obtained for users inside a critical radius, the so
called Sisyphus distance, outside which a user would expect to receive little or
no access to the slotted A L O H A channel.
Goodman and Saleh also identified the Near-Far effect in slotted ALOHA
(Goodman and Saleh, 1987), in which stations closer to the receiver are able
Literature Review 38
to achieve higher throughput than stations at a greater distance from the re
ceiver. The authors illustrate that under a Poisson traffic assumption for all
nodes in the network, capture has a beneficial impact on system throughput,
helping all nodes including those further from the base station to achieve higher
throughput than would otherwise be the case.
These fundamental results (Abramson, 1977; Goodman and Saleh, 1987) have
since formed the basis of almost all published work investigating the perfor
mance of wireless M A C protocols, for example (Davis and Gronemeyer, 1980;
Arnbak, 1987; Lau and Leung, 1992; Cheun and Kim, 1998). The general
technique employed throughout the literature is to determine the probability
with which a packet will be captured by the receiver, then determine channel
throughput as the product of offered load and capture probability. This involves
assuming a model for the packet arrival process, node distribution, and signal
fading or variation. Three main capture models have since been assumed:
Delay Capture in which the arrival time of each packet is randomised
over a defined interval, corresponding to the different propagation delays
experienced by packets emanating from nodes at different distances from
the receiver (Davis and Gronemeyer, 1980). The first arriving packet is
assumed to be captured by the receiver if no other packet arrives within
the acquisition time required by the receiver to successfully detect and
commence reception of the first packet.
Power Capture in which the strongest packet arriving in a timeslot is
received, provided that the strongest packet has a power that is greater
than the sum of all other packets by at least the capture ratio (Arnbak,
1987). This model requires the assumption that each signal varies quickly
enough to allow incoherent addition of the phasors of each signal. This is
the most common model used in simulation of wireless receivers.
Hybrid Capture combining the power and delay capture models (Cheun
and Kim, 1998). This model attempts to describe the behaviour of a re-
Literature Review 39
ceiver more accurately, by allowing the first frame to be received provided
that the total power received during the acquisition time is less than the
first arriving packet by at least the capture threshold. This approach at
tempts to model the operation of a spread spectrum receiver having good
noise immunity during the initial detection and synchronisation stages of
reception.
Capture probability analysis is based on modelling the operation of an ideal
receiver. In each case, only the initial capture probability is analysed, making no
guarantees with regard to the retention of the packet once capture has occurred.
A n opportunity exists to compare the performance of simulation models with
empirical measurements from a physical network, and subsequently investigate
the fairness properties of a network with packet capture capabilities. A further
opportunity exists to identify appropriate characteristics of a simulation model
for an IEEE 802.11 DSSS receiver.
2.4.3 Channel Capture
Protocol, or channel, capture is a state that may arise when a given host is
able to monopolise the channel at the expense of contending hosts. This is typ
ically the result of interactions between timers at various protocol layers (Gerla
et al., 1999b; Gerla et al., 1999a; Tang and Gerla, 1999; Holland and Vaidya,
1999). In particular, the Transport Control Protocol (TCP) is quite vulnerable
to the interaction between transport layer timers and the M A C layer. This
sensitivity is multiplied in a multihop wireless network, where repeated channel
contention severely affects transport layer timers. Delayed acknowledgements
are often treated as a sign of congestion, when in fact the channel may have
been in a capture state. The resulting backoff period and transmission window
reductions will, in effect, contribute to the capture state by preventing the host
from transmitting when it may have been able to do so. This is a separate
problem from that of acknowledgement or data frames that are affected by an
error burst and incorrectly interpreted as congestion by the sender.
Literature Review 40
The authors of (Gerla et al., 1999b; Gerla et al., 1999a) have performed a sim
ulation study of T C P performance over various wireless network architectures,
focusing on interactions with the M A C layer. Three different M A C protocols
are investigated, C S M A / C A , Multiple Access Collision Avoidance for Wireless
( M A C A W ) (Bharghavan et al., 1994), and Floor Acquisition Multiple Access
(FAMA) (Fullmer and Garcia-Luna-Aceves, 1995). Their results indicate that
in many circumstances, T C P requires a window size of 1 packet (effectively
becoming a stop and wait protocol) in order to achieve any throughput across
a multiple number of hops. Further experimental investigation has illustrated
that T C P does not alleviate, and may even complicate, channel capture when
the hidden terminal problem arises while using a non reservation based M A C
protocol.
Simulation experiments performed by the authors of (Gerla et al., 1999b), fo
cus on interactions between the M A C and transport layer protocols in multihop
scenarios. This work also includes an experimental component, in which pro
prietary (non-802.11) C S M A / C A wireless L A N equipment is used to examine
the protocol channel capture problem. Nodes were arranged in a square topol
ogy, with single hop connections along each side of the square as illustrated
in Figure 2.11. This topology combines the hidden and exposed terminals.
The results indicate how protocol timers interact and result in protocol capture
states. The M A C protocol employed did not include an RTS/CTS signalling
mechanism, allowing the T C P retransmission timers to interact with the M A C
backoff mechanism. In each trial, a single connection was able to gain signifi
cantly greater channel access than the other three. As this behaviour is a result
of protocol interaction between two connections competing for a common re
ceiver, a simpler topology with a single pair of hidden terminals will provide the
same result when no signalling mechanism is employed to minimise the impact
of hidden terminal collisions.
Tang (Tang and Gerla, 1999) presented simulation results which illustrate that
the C S M A / C A with an R T S / C T S / D A T A / A C K 4 way handshake is able to
Literature Review 41
o ••a
GO
V
TCP Stream A
Hidden s/
.•' Terminals '•.
TCP Stream C
H
n on
u w
Figure 2.11 Square Experimental Topology (Gerla et al., 1999a)
provide fair channel access in hidden terminal scenarios. The simulation envi
ronment used in (Gerla et al., 1999b; Gerla et al., 1999a; Tang and Gerla, 1999)
is based on an ideal channel, in which each host receives all intended packets
without error, and with a fixed propagation delay. Using an ideal channel does
not allow the authors to investigate the impact that varying radio conditions
may have on the performance of the protocols. A n accurate indication of per
formance in a true mobile environment, in which signal detection and reception
are unreliable, and propagation delays are constantly changing is required.
Both packet and channel capture effects can be expected to have an impact on
the fairness properties of the network. Accordingly, we present a discussion of
work investigating the fairness properties of wireless networks.
2.5 Investigation of the Fairness Properties of Wireless LAN's
Unfair behaviour in wireless networks can result from of a number of distinct
physical effects and protocol interactions. The combination of protocol induced
channel capture, hidden terminal collisions, and m o d e m capture can result in
significant unfairness for hosts in a general topology wireless network. Recently,
Literature Review 42
a number of publications have appeared in the literature examining fairness
problems in wireless networks from both an experimental and analytical view
point. This literature can be divided into two groups. The first, discussed
in Section 2.5.2, investigates the performance of a variety of M A C protocols
but do not propose mechanisms to overcome the identified unfairness. The
second group, discussed in Section 2.5.3, propose algorithms to prevent unfair
behaviour caused by exposed nodes in disconnected topologies.
The first requirement of an examination of the fairness properties of a wireless
network or protocol is to define behavioural characteristics which are considered
indicative of fair behaviour. W e present a discussion of the many possible
fairness definitions and how each may be applied to local area wireless networks.
This discussion is followed by a review of relevant work in this area.
2.5.1 Fairness Definitions
The view of which characteristics constitute fairness in a distributed wireless
network will vary in accordance with the applications and services the network is
supporting. For example, a real time voice connection would consider a network
unfair if one or a group of nodes was able to gain immediate access to the
channel at the expense of other connections of the same type. Or, alternatively
a data transfer m a y consider a network unfair if throughput was not shared
proportionally amoung similar connections. These simple examples illustrate
how the definition of fairness will depend on the application, data type, or
service being supported by the wireless network. Accordingly, a number of
fairness definitions are possible in a general topology wireless network, including
for example:
• the ability of the network to allow each node access to the channel on
demand, with a transmission rate that is proportional to the required or
requested data rate
• the ability of a node to access the channel within a defined interval of the
Literature Review 43
initial attempt
• the ability of the network to prevent a single user from monopolising the
channel
• the ability of the network to allow equal sharing of capacity amongst nodes
attempting to access the network
• the ability of the network to provide equal performance in terms of spec
ified metrics such as loss, delay, or throughput for contending nodes
Each of the above fairness definitions can be categorised in terms of their scope,
timescale, potential application and other characteristics as illustrated in Ta
ble 2.5. In this comparison we group the above definitions into one of 4 broad
areas, though as mentioned previously, a significant range of definitions is possi
ble depending on the specific circumstances involved. The type category (qual
itative or quantitative) represents whether the fairness behaviour can be mea
sured against a specified metric. The potential application category gives an
indication of how the performance achieved with a specific traffic type may be
assessed.
A suitable exact fairness definition will be necessary in the later stages of this
thesis, and as such it will draw on appropriate elements of this discussion.
2.5.2 Experimental Fairness Investigations
Investigation of wireless network fairness properties presented to date has typ
ically been simulation based, with a small number of publications presenting
empirical results. Simulation investigations in (Tang and Gerla, 1999) and
(Tang et al., 2001) attempt to quantify the interactions between the transport
protocol and M A C backoff timing mechanisms. The performance of several
M A C protocols supporting T C P streams are investigated in a number of differ
ent topologies including hidden terminal scenarios. Results for the C S M A / C A
M A C with an R T S / C T S / D A T A / A C K handshake, illustrate that this variant of
Literature Review 44
en
a o a mi o
Q CO CO O)
PI • i — I
PI O co • H fH 03
ft O
O
CN
i—i
Potential
Application
CP
>>
H
CP
0
o CO
CP >
s
CP >
O
CP
CP
o
S CP
CO
0
CP
P CO CO
CP
a u
real time
data streams
(voice/video)
>
X
X
>
>
Bounded
Access
Delay
transport protocols
including re
transmission timers
X
>
X
X
>
Prevention
of Capture
State
co o
O* • i-H
fH
O
PI
>
X
>
X
X
Throughput
Proportional to
Population
CO
o
CD CO 03
a 03 O) fH -t-a CO
>
X
>
X
>
Equal
Throughput
to Demand ratio
Literature Review 45
the C S M A / C A protocol is able to provide fair access to the resource for both
connections. This is the mechanism employed in 802.11, and as such, an imple
mentation of 802.11 would be expected to perform in a similar manner when
deployed in such a topology.
However, the simulation environment used in (Tang and Gerla, 1999) and (Tang
et al., 2001) includes an ideal channel in which all packets are received by
all nodes in range without error. This approach, while providing a means of
investigating the protocol interactions in isolation through the initial stages of
an investigation, does not allow for a detailed investigation in a similar manner
for a physical implementation.
An empirical investigation is presented by (Koksal et al., 2000), and (Swan
and Raman, 2000). While not considering fairness as a primary objective, the
authors of (Gerla et al., 1999b) present simple empirical results from a wireless
L A N testbed employing a basic C S M A M A C without any signalling mechanism
to prevent hidden terminal losses.
The authors of (Pagtzis et al., 2001) present an empirical investigation of the
fairness properties of the 802.11 M A C / P H Y in a W L A N configuration carrying
a U D P traffic stream. The results indicate that the rate selection algorithm
may result in unfairness for nodes further from the base station, who have a
lower received signal power as a consequence of their relative position. As the
802.11 standard requires that all packets headers be transmitted at the lowest
possible rate, IMbit/s, the authors claim this unfairly penalises stations that
are capable of transmitting at higher rates. This is due to the time taken to
transmit the header compared with the time taken to transmit the payload at a
higher rate. The authors then propose an extension to the M A C which includes
proportional bandwidth allocation within the P C F as a means of preventing
this unfairness.
The requirement for a common header rate is included in the 802.11b standard
to ensure backward compatibility across all DSSS PHY's. Further, the authors
Literature Review 46
(Pagtzis et al., 2001) neglect the short P H Y header option described in the
802.11b standard, which is employed between all stations capable of transmit
ting at either the 5.5 Mbit/s or 11 Mbit/s rates. This reduces the P H Y header
overhead significantly, thus ameliorating the negative impact of the requirement
for a common header transmission rate.
The performance of the MAC protocol using a UDP transport layer requires
further investigation. The majority of work presented to date employs T C P as a
higher layer transport, and while this provides useful information regarding the
interactions between protocol timers and retransmission mechanisms, isolation
of M A C performance in these scenarios becomes difficult.
2.5.3 Mechanisms to Prevent Unfair Behaviour
Schemes which aim to prevent unfair behaviour maintain a precise definition
of fairness and are able to move the measured behaviour closer to the required
operating point by adjusting parameters either individually within each node,
or in concert within a larger groups of nodes across the network. Three distinct
approaches have appeared in the literature aiming to provide network wide fair
medium access, and in the following sections we provide an overview of each
approach.
• Link Access Probability Techniques
Techniques based on the calculation of a link access probability have been pre
sented by the authors of (Ozugur et al., 1998; Ozugur et al., 1999). The tech
nique is based on a mechanism which assigns a predefined access probability for
each link with a neighbour. This is illustrated in Figure 2.12. W h e n the backoff
counter expires, node i will contend for the link with node j with probability
Pij. In this case the normal backoff mechanism is retained. Each node has a
visible set of neighbours, denoted by VJ. Each neighbour in the set then broad
casts the number of logical connections available, Sj. If the number of logical
connections for node i, Si = Y^j&v SJ> then node *is a central node and nas n0
Literature Review 47
Figure 2.12 Example wireless network topology including both hidden and visible stations (Ozugur et al., 1998). Link access probabilities are assigned to each logical network link in accordance with Equations 2.6 and 2.7. Arrows indicate the logical links, and probability pij assigned to each link. Node A has a set of neighbours B. The set of nodes C are hidden from A, having at least one connection with a neighbour of A. All other nodes in the network belong to a separate set, D.
hidden terminals. In this case, Pij = 1, V? G V;.
If Si < Yljev Sj ^nen n°de i will have hidden stations in the vicinity and must
choose the link access probabilities more judiciously. The maximally connected
neighbouring node for node i has a number of connections, 5™°*. In this case,
the link access probability is selected based on the link node i is attempting to
access. If node i is attempting to access the maximally connected node, then
Pij is chosen as:
ftj = mtn|l,-^b| (2-6)
Literature Review 48
And in the case where node i attempts to access any other node A;:
p^ = mm |l,-^M (2.7)
where Sk is the number of logical connections for node k, the intended recipient.
This can be performed as a dynamic process, provided that each node is able to
maintain an accurate account of the variables Si and Sj each time the network
topology changes.
A second method is presented (Ozugur et al., 1998), which uses a time based
approach to determine the appropriate link access probability. Each node is
required to periodically broadcast a traffic descriptor Lij used to indicate the
presence of traffic destined on the link from i to j. A measure of the contention
period duration is also required at each node to determine the link access prob
ability as the average contention period by the mean value of the contention
periods of all neighbouring links.
The connection based scheme provides a relatively simple mechanism to prevent
unfair behaviour on a traffic flow basis (where a flow is defined as the data
flowing along a logical link between two nodes). However, this scheme is based
only on topology information, or the measured contention period experienced
by each node. Accordingly, these mechanisms require detailed knowledge of the
congestion state of the network, or the topology and flow information within
the extended local region. The performance of these mechanisms in dynamic
environments is unknown, particularly in cases where the topology is rapidly
changing (in a M A N E T for example), or the congestion state of the network is
difficult to accurately assess for the M A C layer. Further, the requirement for
this additional knowledge introduces significant additional complexity to the
M A C layer, greatly expanding the scope of a distributed M A C protocol.
• Backoff Window Adjustment
A technique which adjusts the backoff window has been proposed in (Bensaou
et al., 2000). In this technique, a node requires a pre-defined indication of the
Literature Review 49
fractional throughput fa it may expect to achieve. Combined with the actual
throughput achieved W j , and the offered load Li, a fairness index is defined as
a function of the ratio of the achieved load to the fractional throughput:
{max (^rS^-J | V i,j : ) * *j( } (2.8)
• (Wi Wj\ | v '
mm U' t) J The task is then to minimise this fairness index for each node in the network.
This is achieved by each node estimating the throughput neighbouring nodes
obtain W0, through observation of all visible packets. A dynamic fairness index
variable is maintained, Fid = 0ji)/(^s-), and used to either double or half the
maximum contention window when compared with another constant C. This
approach requires two variables which must be tuned to operate effectively. It
also applies the contention window change to all logical connections.
Mechanisms to distribute the backoff counter value amoung neighbouring nodes
have also been presented in (Bharghavan et al., 1994). This technique attempts
to ensure each node starts each contention period on the same footing after
periods of extended backoff. The authors then illustrate that this is able to
prevent a node from suffering unfair excessive backoff periods. In the case
of both (Bensaou et al., 2000) and (Bharghavan et al., 1994), ideal channel
characteristics have been assumed.
• Generalised Persistence Techniques
A generalised analytic framework to determine fair rate adaptation algorithms
is presented in (Nandagopal et al., 2000). Identifying the spatial dependence
of contention in a wireless network, the authors propose a general framework
for the derivation of a fairness algorithm to suit a specified fairness objective.
Using a graph theory approach, combined with a utility function defining the
fairness objective, a rate adaptation algorithm can be derived which is shown
to converge on a fair service for all nodes in the network.
Literature Review 50
The example given is a proportional fair contention resolution mechanism de
rived using this framework. Proportional fairness is defined by the utility func
tion U(r) = log(r), where r is the rate allocation vector for a given node. The
channel allocation algorithm derived using this technique is:
fi = a- PPM (2.9)
where a and /? are parameters used to control the efficiency (increasing a, de
creasing 0) and fairness (decreasing a, increasing /?). Each node in the network
implements this rate adaptation algorithm, resulting in an approximation of
the network wide fairness objective. Through comparison of the proportional
rate control scheme with 802.11 and other M A C protocols in a number of static
topologies, the authors illustrate (Nandagopal et al., 2000) that the proportional
rate control mechanism provides an improved measure of fairness throughout
the network.
The stability and convergence properties of this method has not been analysed
or tested, particularly in a mobile network. Further, the selection of the two pa
rameters, a and /? is a not trivial concern for a general network. The suitability
of this scheme for a network in which the topology is dynamically changing is
unknown.
2.5.4 Discussion
Each of the schemes presented in Section 2.5.3 are designed to solve the fairness
problems resulting from adverse protocol interactions arising in both discon
nected and hidden terminal topologies. The impact of variable propagation
environments, unreliable signal detection and reception, and packet capture ef
fects have not been considered in any of these mechanism. Further issues that
must be addressed with each scheme include:
1. How each algorithm operates within the MAC/PHY relationship
2. The processing and messaging overhead associated with each scheme
Literature Review 51
3. Implementation complexity
4. Practical incorporation into relevant standards (802.11, HiperLAN I)
5. The sensitivity of each scheme to realistic propagation conditions and
radio m o d e m behaviours
From the above discussion, we contend that while each scheme may operate well
when considering protocol based unfairness problems, suitable performance in
a realistic network environment can only be obtained when signal strength vari
ability and realistic packet capture mechanisms are taken into account. The
motivation for this is that both packet capture and variable propagation condi
tions may result in unfair behaviour outside the scope of the schemes outlined
above. Based on the above discussion, it is evident that fairness algorithms re
quire the following attributes when applied in a dynamic network environment:
• Minimal 'protocol' overhead, in terms of additional messaging and infor
mation gathering processes
• Minimal additional resource requirements on each node, including pro
cessing overhead and implementation complexity
• Adaptive to changes in network topology
• Robust in the presence of variable propagation environments
• Ability to operate within current and future standards
2.6 Summary
Throughout this chapter, we have concentrated on the development of wire
less M A C protocols contributing to the development of the IEEE 802.11 M A C
protocol, and tried to focus on areas in which this thesis provides new contri
butions, namely the impact of capture effects on the fairness properties of a
Literature Review 52
wireless L A N . Whilst this review has shown wireless M A C protocol research
to be quite advanced in several aspects, there are clearly many areas in which
significant work is required. Our approach is to build on the existing state of
wireless M A C performance studies from the specific viewpoint of fairness in
the presence of capture effects. Issues identified in the literature which require
further investigation are summarised in the following section.
2.6.1 Summary of Open Research Issues Identified In
Current Literature
• The RTS/CTS handshake relies on all potentially interfering nodes being
able to successfully receive either the R T S or C T S message. Practical
wireless networks are characterised by noisy, lossy environments where
messages may or may not be correctly received. The performance of the
R T S / C T S mechanism in a real environment when supporting T C P and
U D P traffic streams is yet to be established.
• A n investigation of the sensitivity of the R T S / C T S handshake to node
movement and other signal degrading effects has not been presented.
• Detailed investigation of the performance of the IEEE 802.11 M A C pro
tocol in hidden terminal topologies is required.
• Investigation of the relative fairness properties of the significant capture
models presented in literature is required. A comparison of simulation
data generated by each capture model with empirical data has not been
undertaken.
• The salient features required from a capture model suitable for accurate
simulation of an IEEE 802.11 m o d e m need to be established.
• N o clear understanding of the empirical fairness properties of an IEEE
802.11 network in hidden terminal topologies exists.
• N o analysis of the detailed requirements of a fairness control scheme suit
able for the IEEE 802.11 M A C protocol has been undertaken.
Literature Review 53
The IEEE 802.11 M A C protocol has combined many of the features described
in earlier M A C protocols proposed in literature. The remainder of this thesis
will be concerned with an investigation of the fairness properties exhibited by
the IEEE 802.11 protocol.
Chapter 3
Experimental Investigation of Capture Effects and Fairness Behaviour
3.1 Introduction
A significant body of literature exists investigating the performance of various
M A C protocols, retransmission schemes, and hidden terminal avoidance mech
anisms across a range of circumstances. However, as discussed in Chapter 2, an
experimental investigation of these mechanisms will allow us to evaluate interre
lationships between protocol layers, as well as the impact variable propagation
conditions have on signalling mechanisms within the M A C .
Further, the tendency of communications systems to exhibit some form of cap
ture behaviour has been an issue for many years (Jain et al., 1984). In partic
ular, Chapter 2 illustrated how capture behaviour is observed at two separate
levels: the packet level due to physical layer mechanisms, and the protocol level
due to interactions between transmission timers. The combination of these two
capture effects can greatly impact on the fairness behaviour of a shared me
dia wireless network. Therefore, the remainder of this thesis focuses on the
investigation, modelling and analysis, and prevention of such capture effects in
physical wireless networks.
54
Experimental Investigation of Capture Effects and Fairness Behaviour 55
In this chapter, we present an experimental investigation of the fairness prop
erties of a C S M A / C A network employing the RTS/CTS common channel sig
nalling technique. W e concentrate on scenarios where hidden terminals are
present, or where carrier sense mechanisms may be unreliable. Section 3.2 out
lines our motivation for the experiments performed, while Section 3.3 outlines
the experimental methodology employed throughout the chapter. Section 3.4
presents experimental results obtained using T C P as the higher layer transport
protocol. Section 3.5 then presents results obtained using U D P as a higher layer
transport. Conclusions are drawn in Section 3.6, which provide direction for the
following chapters.
3.2 Experimental Motivation
Given the mobile, dynamic nature of an ad hoc network, it is anticipated that
hidden terminals are likely to be a common problem in typical network topolo
gies. Also, in a typical indoor wireless L A N , the characteristics of an indoor
office environment imply that absolute carrier sense information cannot be guar
anteed for all nodes throughout the network. Walls and other obstacles result in
significant signal attenuation and multipath components in the signal. There
fore, the performance of data streams competing for a common radio resource
in the presence of hidden terminals is an important issue.
The poor performance displayed by TCP over the MAC protocol (Gerla et al.,
1999b; Holland and Vaidya, 1999) is also a problem for shared media wireless
networks. The performance of a M A C protocol in terms of fairness and delay
is critical to the performance of the transport protocol. In this context, there
are two major interactions of interest: one between the M A C retransmission
timers and the T C P timers, and another between the routing protocol and T C P
retransmission timers. Investigations into both areas form the basis of much of
the published work in the area of wireless data transport (Tang and Gerla, 1999;
Gerla et al., 1999b; Holland and Vaidya, 1999; Chandran et al., 1998; Gerla
et al., 1999a). In our investigation, we consider single hop connections only,
Experimental Investigation of Capture Effects and Fairness Behaviour 56
Connection A
Transmission and C O Carrier Sense Range
Hidden Nodes
Figure 3.1 Experimental Topology
and therefore will be concerned with the M A C and transport layer interaction.
Tang (Tang et al., 2001; Tang and Gerla, 1999) presents simulation results il
lustrating that the IEEE 802.11 M A C protocol is able to provide fair channel
access in hidden terminal scenarios when the RTS/CTS/DATA/ACK 4-way
handshake is employed. To confirm this result, we have performed experiments
simulating two competing F T P file transfers over hidden terminal connections,
with the topology illustrated in Figure 3.1. The simulation environment is a
modified version of the Berkeley ns-2 network simulator (UCB/LBNL/VINT,
1999). ns is a detailed protocol simulation tool implementing the IEEE 802.11
M A C and T C P protocols. A two-ray ground propagation model with an Addi
tive White Gaussian Noise ( A W G N ) channel has been employed for these tests.
A power modem capture model (Arnbak, 1987) has been employed. In this
trial, each connection has an equal average Signal-to-Noise Ratio (SNR), as the
nodes are equally distant from the common host.
The results of these basic trials in Figures 3.2 and 3.3 confirm the result pre
sented by Tang (Tang et al., 2001), that the RTS/CTS handshake is able to
prevent adverse timing interactions between the M A C and TCP. Figure 3.2
provides evidence of the protocol capture problem without an RTS/CTS hand
shake. This behaviour is explained by the interaction between T C P timers and
the M A C retransmission and backoff counters. Connection B is started 0.1
seconds after Connection A, and is unable to gain access to the channel until
approximately 3 seconds. During this period, Connection A has captured the
Experimental Investigation of Capture Effects and Fairness Behaviour 57
1200
1000
- 800
I , 1 600 '8 tr M ^ J A M
Q 400
200
Connection A Connection B
3
Time (seel
Figure 3.2 Simulation Experiment - Hidden terminals, D A T A / A C K only. Connection A captures the resource until approximately 3 seconds when Connection B is able to access the channel. The ability of either connection to capture the channel is random in this experiment
radio resource. Figure 3.3 illustrates how the RTS/CTS signalling mechanism
attains a fair distribution of channel access for the competing connections.
The simulation environment used by the authors of (Tang et al., 2001; Gerla
et al., 1999a; Tang and Gerla, 1999) is based on an ideal channel, in which
each host receives all intended packets without error, and no m o d e m capture
ability is present. This allows investigation of the protocol mechanisms and
interactions in isolation, but will provide limited insight into the behaviour and
fairness properties that may be expected from a physical M A C / P H Y protocol.
In this section, we have illustrated that simulation with an A W G N channel and a
simple power capture model indicates behaviour consistent with that presented
by Tang (Tang et al., 2001; Gerla et al., 1999a; Tang and Gerla, 1999).
The critical questions is whether the RTS/CTS/DATA/ACK mechanism ex
hibits the same robust behaviour in real propagation environments with physical
Experimental Investigation of Capture Effects and Fairness Behaviour 58
1200
1000 -
~ 800 eft
f CO § 600
DC
2 S 400
200
Connection A o Connection B x
Figure 3.3 Simulation Experiment - Hidden terminals, RTS/CTS/DATA/ACK
signal detection and demodulation hardware, as opposed to the ideal receivers
modelled in previous simulation. H o w well this mechanism operates when signal
propagation conditions are not ideal (multipath, fading etc) remains an open
question. It is also unlikely that the channel conditions will be equal on each
of the contending connections as is the case with simulation results presented
thus far. These factors, combined with a lack of experimental results examining
the fairness properties of a wireless L A N involving hidden terminals, motivate
an investigation of the impact varying S N R conditions have on the fairness
performance of the R T S / C T S mechanism.
Through investigation of the dynamic behaviour of competing hidden connec
tions over an IEEE 802.11 network, this experimental study will address the
following issues:
1. the ability of the R T S / C T S / D A T A / A C K handshake to prevent adverse
protocol interactions, and associated protocol capture states when a higher
layer backoff and retransmission mechanism is employed.
Experimental Investigation of Capture Effects and Fairness Behaviour 59
2. the ability of the RTS/CTS/DATA/ACK handshake able to alleviate hid
den terminal collisions in a real propagation environment.
3. the resistance of the R T S / C T S / D A T A / A C K mechanism to protocol cap
ture induced by synchronisation of backoff timers at the M A C layer.
4. the fairness performance of the R T S / C T S / D A T A / A C K mechanism in a
practical propagation environment.
We are interested in instantaneous, dynamic behaviour of each connection rather
than longer term average behaviour, as this gives an accurate description of the
short term fairness properties of the system. This is important from the view
point of the QoS experienced by an individual user. If a system was asymptot
ically fair, but significantly unfair at shorter timescales, then the QoS experi
enced by an end user, in terms of channel access delay, instantaneous through
put, and bit error rate will be less than ideal. For the purposes of the exper
iments of this chapter, we use a relatively simple definition of fairness, one in
which both nodes should gain relatively equal access to the radio resource, and
no connection should be able to prevent another from gaining appropriate ac
cess to the channel for an extended period. This is a reasonable definition for
the current experiments, given that both data transfers utilise the same M A C
protocol, whose aim is to provide equivalent access for all nodes. More detailed
quantitative fairness studies are undertaken in Chapter 5.
3.3 Experimental Methodology
The aim of this investigation is not to determine absolute throughput perfor
mance, nor benchmark the equipment used to perform the experiments. W e are
interested in the ability of the M A C to provide fair channel access in hidden
terminal scenarios, and the susceptibility of the 802.11 M A C / P H Y to protocol
and other capture effects.
The experimental topology, illustrated in Figure 3.1, has hosts 1 and 3 mutually
Experimental Investigation of Capture Effects and Fairness Behaviour 60
out of range, attempting to communicate with host 2. Each experiment consists
of a simultaneous data transfer from hosts 1 and 3 to host 2. The network
sniffing program, tcpdump (Lawrence Berkeley Laboratories, 1999), is used at
host 2 to trace the progress of each data transfer. Each host has an 802.11b
wireless network interface, and is used in either the ad hoc mode or standard
access point mode, and employs a collision avoidance R T S / C T S handshake
governed by the aRTSThreshold parameter. The results are presented as traces
illustrating the time evolution of each transfer (data received vs time) to provide
a clear indication of the timescale of unfairness that may be present. The signal
strength, noise power, and S N R on each connection is measured using a software
package which sends test packets each second and records the signal strength
parameters reported by the R F interface. The results are averaged over 2 minute
intervals to provide an indication of the average link quality. The individual
experiments presented in the following sections (Section 3.4 and Section 3.5)
are summarised in Table 3.1.
Following the results presented in Section 3.2 combined with previously pub
lished simulation results (Tang et al., 2001; Tang and Gerla, 1999), it was antic
ipated that the reservation mechanism should enable reasonable sharing of the
radio resource. It was also anticipated that T C P connections should suffer no
serious ill effects (in terms of excessive retransmission or timeouts), given that
each connection is only a single hop, and that the M A C protocol employs im
mediate positive acknowledgement with retransmission. The experiments were
performed using three laptop PCs. Each P C was equipped with IEEE 802.11
network interface cards, and the experiments were performed using the Linux
operating system, and repeated using Windows 98 operating systems to test for
any operating system or driver dependent artifacts.
3.4 TCP Experiments
A number of separate experiments were undertaken to investigate the perfor
mance of T C P over competing hidden connections. This series of experiments
Experimental Investigation of Capture Effects and Fairness Behaviour 61
Table 3.1 Summary of Experimental Trials
Experiment
1
2a
2b
3a
3b
4
5a
5b
6a
6b
7a
7b
Figure
3.4
3.5
3.6
3.7
3.8
3.9
3.10
3.11
3.12
3.13
3.14
3.15
Scenario
no-RTS/CTS equal SNR
RTS/CTS equal SNR
RTS/CTS
equal SNR
RTS/CTS
unequal SNR
RTS/CTS unequal SNR
controlled
controlled SNR
RTS/CTS equal SNR
RTS/CTS
equal SNR
RTS/CTS unequal
stronger first
RTS/CTS unequal stronger first
RTS/CTS unequal
weaker first
RTS/CTS unequal
weaker first
Transport
TCP
TCP
TCP
TCP
TCP
TCP
UDP
UDP
UDP
UDP
UDP
UDP
P H Y Mbit/s
Barker
Barker
CCK
Barker
CCK
Barker
CCK
CCK
CCK
CCK
CCK
CCK
Vendor
Lucent
Lucent
Cisco
Lucent
Cisco
Lucent
Lucent
Cisco
Lucent
Cisco
Lucent
Cisco
Experimental Investigation of Capture Effects and Fairness Behaviour 62
examines both the performance of TCP under these conditions, and the abil
ity of the M A C protocol to avoid protocol channel capture under heavy load
conditions. These experiments were performed using both Orinoco WaveLAN
Lucent and Aironet 340 Cisco IEEE 802.11 cards. This is a deliberate choice, as
the Lucent and Cisco cards employ different R F front end designs. The Lucent
R F front end was developed by Lucent, while the Cisco cards employ a front
end designed by Intersil (formerly Harris Semiconductor). The overwhelming
majority of 802.11 interfaces available today employ one of these two R F front
ends.
3.4.1 RTS Handshake - Hidden Terminals
Initially, the performance without the RTS/CTS handshake and an equal SNR
on each connection is investigated. The resulting trace, illustrated in Figure 3.4,
show that even though connections A and B have an equal S N R as measured
at host 2, Connection A is able to capture the channel for the duration of
the transfer. The results in this simple case, mirroring those of Figure 3.2,
illustrate the impact timing mechanisms can have on contending connection.
Connection B has commenced transferring data when Connection A commences.
This leads to a period of receiver side collisions, won by Connection A, which
eventually manages to capture the channel. Host 3 (Connection B) now invokes
T C P congestion control measures, and undergoes periods of exponential backoff.
During this period, host 3 is unable to receive an acknowledgement for any data
frame it has attempted to transmit as host 1 has monopolised the channel.
The second experiment is a simple case where the SNR of each connection is
equal and the aRTSThreshold is set to 500 bytes. A typical example of the
resulting traces are shown in Figures 3.5 and 3.6. In this example, Connection
A is commenced a few seconds after Connection B. Even though T C P connec
tion setup (SYN) messages of 40 bytes are exchanged without an RTS/CTS
handshake, the channel is effectively shared until connection B finishes. This
experiment was run numerous times with a range of aRTSThreshold parameter
Experimental Investigation of Capture Effects and Fairness Behaviour 63
500
450
400
350
2- 300
250
200
150
100
50
0 M 6 8
Time (sec)
Connection A Connection B
10 12 14
Figure 3.4 Experiment 1: Lucent Barker P H Y Equal SNR 25dB, No RTS/CTS Handshake
values from 0 bytes to the maximum T C P segment size of 512 bytes, with little
impact on the relative fairness provided by the M A C .
The delayed start of the Connection A data transfer in Figures 3.5 and 3.6 is
due to the lack of precise control over the exact time at which a connection
will commence. This is not significant from a fairness point of view, and may
be considered to represent a case where an ongoing connection is joined by a
competing hidden terminal connection part way through. It can be seen that
the R T S / C T S handshake has an impact on the fairness of the channel access
achieved by each connection. Each host maintains a roughly equal share of the
channel capacity throughout the common period of each transfer, observed as
the parallel gradient of the individual traces between 4 seconds where connection
A commences, and 14 seconds where connection B finishes. The most interesting
result is the sensitivity of the capture behaviour. A very subtle change in
physical orientation of a terminal was able to sufficiently alter the received
signal strength, preventing fair access for both connections to the channel.
Experimental Investigation of Capture Effects and Fairness Behaviour 64
1200
1000 -
« 800
I CO | 600 -
'8 CD OC M
co a 400
200 -
I
-
-
/
/ ,
1 — 1 1
y-
X
i i
S"
I I
1 1 1
Connection A Connection B
• >
-
-
-
0 X
8 10 12
Time (sec)
14 16 18 20
Figure 3.5 Experiment 2a: Lucent Barker P H Y - Equal SNR 25dB for both connec
tions, aRTSThreshold 500 bytes
3500
3000
2500
2000
» 1500 a co Q 1000
500
Connection A o Connection B x
6
Time (sec)
10 12
Figure 3.6 Experiment 2b: Cisco C C K P H Y - Equal SNR 25dB for both connections,
aRTSThreshold 500 bytes
Experimental Investigation of Capture Effects and Fairness Behaviour 65
From these experiments, it is apparent that the RTS/CTS mechanism is able to
provide reasonable sharing of the channel when T C P is used as a higher layer
transport protocol and the signal strength on each connection is equal. This
is in accordance with the simulation results of Section 3.2. In the following
section, we investigate the sensitivity of this behaviour to the signal strength
conditions on each connection.
3.4.2 Impact of Varying Signal Strength
The next experiment, in which Connection A has an SNR 5dB higher than
Connection B, again uses an aRTSThreshold of 500 bytes. The scenario inves
tigates the performance under a 'near-far' hidden terminal scenario. The trial
results in behaviour illustrated in the examples shown in Figures 3.7 and 3.8
where Connection A is able to dominate, capturing the channel. Here, Con
nection A starts marginally after Connection B, yet manages to dominate the
contending host. None of the randomness of the previous two experiments was
evident. Over multiple trials the connection associated with the higher S N R
always captured the channel.
A 5dB difference between connections is quite minor and in practice can be
simply due to subtle variations in multipath propagation as the surrounding
environment changes. W e expect the scenario presented in the first experiment
(equal S N R ) will rarely arise with a hidden terminal topology in a multihop
wireless network, particularly given the number of factors affecting the S N R
observed on each connection. These results demonstrate the sensitivity of the
R T S / C T S mechanism within 802.11 to the S N R seen on competing hidden
connections.
To test the dependence on relative signal strength of which connection is able
to capture the channel, the fourth experiment involves reducing the S N R on
the stronger connection, Connection A, to a point below the weaker Connec
tion B midway through the file transfer. It is anticipated that Connection B
should be able to capture the channel at the expense of Connection A. This
Experimental Investigation of Capture Effects and Fairness Behaviour 66
500
450
400
350
2r 3 0° 250
OC co co Q
200
150
100
Connection A Connection B
10 12
Figure 3.7 Experiment 3a: Lucent Barker Code P H Y - Unequal SNR Connection A
25dB and Connection B 20dB, aRTSThreshold 500 bytes
2500
2000
ft 15 0°
1000
500 -
Connection A o Connection B x
12
Figure 3.8 Experiment 3b: Cisco C C K P H Y - Unequal SNR Connection A 20dB
and Connection B 25dB, aRTSThreshold 500 bytes
Experimental Investigation of Capture Effects and Fairness Behaviour 67
1200
15 20
Time (sec) 25
Connection A Connection B
30 35 40
Figure 3.9 Experiment 4: Lucent Barker Code P H Y - Controlled SNR, aRTSThresh
old 500 bytes
experiment provided a concrete test of the S N R dependence observed in previ
ous trials. Connections A and B commence the test with a S N R of 25dB and
20dB respectively. Five seconds into the trial the S N R of Connection A was
reduced using R F absorbing foam to approximately 17dB through to the end of
the experiment. A n example of the resulting trace is shown in Figure 3.9. Here,
the sensitivity to the signal strength is clearly illustrated. The new stronger
host, Connection B, manages to 're-capture' the channel once the S N R of Con
nection A is sufficiently reduced. Once Connection B has finished Connection
A is able to regain access to the channel, after a significant timeout period be
tween 5 seconds and 33 seconds. In the above experiments, the M A C is unable
to provide fair access amoung the contending hidden terminals. In each case,
the connection which is able to capture the channel suffers relatively few T C P
timeouts, and transmission errors are simply handled by the M A C and T C P
retransmission mechanisms. Conversely, the contending connection undergoes
continual timeout and exponential backoff at both the M A C and T C P levels.
In all cases, the connection with the stronger signal strength measured at the
Experimental Investigation of Capture Effects and Fairness Behaviour 68
common receiver is able to capture the channel. The experiments have been
performed in an indoor office environment subject to multipath and other sig
nal degrading effects. The propagation delay over the links employed in the
experiment is 50 nsec, significantly less than the 10 //sec SIFS, and the 50 / sec
DIFS. Multipath reflections were eliminated as a potential factor through sub
sequent experiments in a controlled multipath environment illustrating identical
behaviour.
In each of the static and controlled SNR experiments, it is obvious that TCP
is exacerbating the channel capture problem. The result is a series of signifi
cant periods of protocol induced capture for either connection whilst the other
connection is forced to repeatedly backoff and eventually timeout. This raises
a question regarding how significant a component protocol capture is in this
behaviour. If a more suitable transport layer protocol were employed then the
protocol capture state may be avoided. To test this, we undertake a number of
experiments using the User Datagram Protcol (UDP) as a transport protocol
in the following section.
3.5 UDP Experiments
In order to quantify the impact TCP timing interactions have on the fairness
on each connection, we perform experiments in this section in which we remove
TCP. In this series of experiments, a transport layer is used which has been
written to flood each connection with U D P packets of a specified size instead of
F T P / T C P used in the previous experiments. A n application sends a specified
number of packets to a remote machine on a given port number. The applica
tion will send packets as quickly as the network interface allows, without any
retransmission or timeout mechanisms. The network connection will not accept
packets from the transport layer (through the network socket) while the socket
and M A C buffers are full. The receiver runs a similar application which listens
to the identified port, and accepts the data from each host. The resulting sys
tem is one with two greedy traffic sources transmitting to a common traffic sink
Experimental Investigation of Capture Effects and Fairness Behaviour 69
Connection A Connection B
20 25
Time (sec) 50
Figure 3.10 Experiment 5a: Lucent Chipset C C K P H Y - U D P Trace - both con
nections 25dB
over contending hidden connections.
To test for artifacts that may be specific to a particular 802.11 interface imple
mentation, we continue experiments with both Orinoco 802.11 Lucent interfaces
and Aironet 340 Cisco interfaces. Each connection consists of either 5000 or
10000 512 byte U D P frames to introduce variability into the experiment.
3.5.1 Equal Signal Power
Experiment five was performed with an equal signal power on each connection,
as measured at the common receiver. The topology is again the hidden terminal
topology illustrated in Figure 3.1. In this scenario, only the 802.11b UMbit/s
P H Y is employed as the previous results have indicated there to be little dif
ference between the 2Mbit/s Barker code P H Y and the UMbit/s C C K P H Y .
In each experiment, hosts 1 and 3 send a total of 10,000 U D P packets (512
bytes) to the receiver. The results of the equal signal power experiments are
illustrated in Figures 3.10 and 3.11. In these traces, fair sharing of the channel
Experimental Investigation of Capture Effects and Fairness Behaviour 70
2500
6
Time (sec)
Connection A O Connection B x
12
Figure 3.11 Experiment 5b: Cisco Chipset C C K P H Y - U D P Trace - both connec
tions 25dB
is obvious. Neither host gains preferential access to the channel, and there is
no visual evidence of any unfairness due to M A C backoff timer synchronisation.
As mentioned earlier, more detailed investigation of the quantitative fairness
properties of the experimental traces is undertaken in Chapter 5.
3.5.2 Unequal Signal Power
The sixth experiment was performed with one of the connections at a relative
difference of 5dB. In this case, the stronger host commences transmission prior
to the weaker connection. The resulting traces are shown in Figures 3.12 and
3.13. In both cases Connection A is the stronger connection with an average
measured S N R of 25dB and Connection B the weaker host, 5dB weaker with
an average measured S N R of 20dB. Both Figures 3.12 and 3.13 illustrate the
relative difference between the stronger and weaker connections. During the life
time of the stronger connection, the weaker connection suffers continual timeout
and retransmission of M A C frames, resulting in the much lower throughput ob-
Experimental Investigation of Capture Effects and Fairness Behaviour 71
Connection A Connection B
40 45
Figure 3.12 Experiment 6a: Lucent C C K P H Y - U D P Trace - stronger host (Con
nection A - 25dB) commencing prior to weaker host (Connection B - 20dB)
2500
2000
£ 1500
1000
500 -
Connection A Connection B
o X
12
Figure 3.13 Experiment 6b: Cisco C C K P H Y - U D P Trace - stronger host (Con
nection A - 25dB) commencing prior to weaker host (Connection B - 20dB)
Experimental Investigation of Capture Effects and Fairness Behaviour 72
5000
4500
4000
3500
§• 3000
1 2500 co
8 01 2000 CO CO Q
1500 1000
500
0 0 5 10 15 20 25 30 35 40 45
Time (sec)
Figure 3.14 Experiment 7a: Lucent CCK PHY UDP Trace - stronger host (Con
nection A - 25dB) commencing after weaker host (Connection B - 20dB)
served during this period. Each frame suffers a significantly higher delay for
this reason. The weaker connection suffers a significantly higher average backoff
period, and will require a higher number of retransmission attempts than the
stronger host. Combined with the fact that the greedy source is unable to send
packets while the socket and M A C buffers are fully occupied and will therefore
pause until room is made available in the local socket and M A C buffer. The
latter explains the small gaps in time present in the weaker trace, corresponding
to periods where the M A C cannot successfully access the radio channel.
Experiment seven is identical to experiment six, with the exception that the
starting order of the data transfers is reversed, and the stronger connection
commences after the weaker connection. This experiment, combined with the
previous experiment, provides a mechanism to investigate the impact a stronger
connection has on a weaker connection. The resulting traces are shown in
Figures 3.14 and 3.15. In both traces, the stronger host again prevents the
weaker host from gaining equal access to the channel. Once the stronger host
J i_
Connection A o Connection B *
Experimental Investigation of Capture Effects and Fairness Behaviour 73
2500
2000
| 1500
1000
500
Connection A Connection B
12
Figure 3.15 Experiment 7b: Cisco C C K P H Y - U D P Trace - stronger host (Con
nection A - 25dB) commencing after weaker host (Connection B - 20dB)
has commenced, the weaker host is again prevented from gaining equal access to
the channel. The weaker host suffers significantly higher loss and delay during
the period of the stronger connection. Each of the traces from experiments
using U D P (Figures 3.10 to 3.15) display the same behaviour as the T C P based
results of the previous section. The stronger of the hidden connections will gain
preferential access to the channel at the expense of weaker connections.
Mitigation of physical artifacts or other factors in the above experiments was
an important aspect of the experimental process. Experiments were repeated
with updated versions of drivers, card firmware, and operating systems. M A C
parameters were also carefully controlled, including fixing the transmission rate
to prevent auto rate selection reducing the transmit rate when an increased
B E R is observed.
In each experiment described in this chapter, a consistent result is evident.
W h e n a signal power difference of greater than 5 dB exists between two hidden
connections, the stronger connection will gain preferential access to the channel
Experimental Investigation of Capture Effects and Fairness Behaviour 74
at the expense of the weaker connection.
3.6 Conclusions
In this chapter, we have presented a detailed empirical investigation of the per
formance of a C S M A / C A M A C protocol using an common channel RTS/CTS
signalling technique to guard against hidden terminal collisions. The experi
ments have examined the performance of the M A C protocol from the viewpoint
of the higher layer protocols, in a real propagation environment. Through in
vestigation of hidden terminal topologies, the experiments performed in this
chapter have illustrated a strong signal power dependence of channel capture
behaviour with the IEEE 802.11 M A C / P H Y protocol. The various scenarios
investigated have illustrated that the RTS/CTS handshake, is unable to pre
vent unfair behaviour in the form of channel capture in near-far signal power
scenarios. A S N R differential as small as 5dB was shown to result in capture
for the stronger connection. A log-distance path loss model with exponent 4
(corresponding to an indoor nlos environment) and an experimentally measured
reference of 25dB at 15m results in a 5dB path loss corresponds at an additional
spacial separation of only 4.74m.
Under all but the most ideal of conditions, channel capture is evident during
periods of high traffic load. The mechanisms behind this require detailed inves
tigation.
In terms of the aims outlined in Section 3.2, the experiments performed in this
chapter have illustrated the following:
• The RTS/CTS/DATA/ACK handshake is able to support fair channel
access for both T C P and U D P connections in hidden terminal scenarios
where the average signal power measured at the receiver is equal for both
connections. There is no evidence of adverse interactions between the
transport protocol and the M A C in the case where the S N R is equal on
Experimental Investigation of Capture Effects and Fairness Behaviour 75
each competing connection.
• The RTS/CTS/DATA/ACK handshake is able to successfully alleviate
hidden terminal collisions only in scenarios where the signal power mea
sured at the receiver is equal amongst all hidden nodes. The results of the
equal S N R U D P and T C P experiments provide strong evidence of this.
In scenarios where the received signal power is not equal on each connec
tion, the R T S / C T S / D A T A / A C K handshake appears unable to support
fair access for all hidden connections. The reasons why this is the case
require further investigation.
• The RTS/CTS/DATA/ACK handshake does not illustrate the propensity
to suffer adverse synchronisation of M A C backoff timers in scenarios where
the received signal power is equal on each connection. In scenarios where
the received signal power is not equal, the U D P experiments illustrate that
the weaker host suffers significantly longer average backoff period, and
requires a greater number of transmission attempts. This is not direct
evidence of synchronisation of backoff timers, rather an indication that
the congestion state of the network observed by the stronger and weaker
connections differ in accordance with their relative received signal power.
The stronger host does not observe significant congestion, while the weaker
host observes significant congestion through continual timeout, backoff
and retransmission cycles.
• The RTS/CTS/DATA/ACK handshake is unable to provide reasonable
fairness in scenarios where the received signal power is not equal on all
hidden connections. In such cases, the fairness properties are heavily
skewed towards the stronger connection. With TCP, the stronger connec
tion suffers no serious ill effects, whilst the weaker connection will suffer
significant loss and timeout at the M A C layer. This causes the T C P
congestion control mechanisms to undergo continual backoff and retrans
mission, preventing the weaker connection from gaining equal access to
the channel. With a U D P based greedy source, the stronger connection is
Experimental Investigation of Capture Effects and Fairness Behaviour 76
able to occupy the significant majority of the radio resource at the expense
of the weaker connection.
The most fundamental contribution arising from the experiments presented in
this chapter, is the sensitive dependence on signal strength of the fairness prop
erties observed empirically. This indicates that the M A C protocol operates
effectively in cases where the signal power observed at the receiver is equal for
all competing hidden nodes. The M A C is unable to provide fair access in cases
where a relative received signal strength of greater than 5dB is present between
two competing hidden connections. Relative received power has a strong in
fluence on the fairness properties of the M A C , indicating that m o d e m capture
behaviour may have a significant influence on the performance of the M A C pro
tocol. In the following chapter we undertake an investigation of the physical
layer signal reception process in order to investigate the impact this has on the
ability of the M A C protocol to provide equal access for all hidden connections
across a suitable range of signal conditions.
Chapter 4
Error Probability Analysis -Hidden Terminal Jamming
4.1 Introduction
Detection and subsequent reception of a spread spectrum signal is a fundamental
aspect of the performance of the P H Y protocol. One of the fundamental reasons
behind the selection of a direct sequence spread spectrum physical layer in the
IEEE 802.11 standard is the inherent immunity to noise and multipath achieved
with a spread spectrum signalling technique. In an ideal system, a receiver
should be able to receive a given signal in the presence of a reasonable level of
background interference
However, in the previous chapter, it was shown that the relative received sig
nal power is a significant factor in determining the performance of competing
traffic flows in network scenarios involving hidden terminals. In cases with a
small relative difference in received signal strength, the stronger connection re
ceives a favourable throughput over weaker connections, despite the physical
and virtual carrier sense mechanisms within the 802.11 C S M A / C A multiple
access technique. This result adds an unexpected dimension to the protocol
capture problems already known. The theory behind this behaviour will be
investigated in this chapter.
77
Error Probability Analysis - Hidden Terminal Jamming 78
Through investigation of the reception mechanism within an 802.11 spread spec
trum receiver, we aim to determine the impact interfering signals have on the
probability of successful reception of a given frame. The remainder of this
chapter is organised as follows: Section 4.2 provides background on the analysis
techniques employed, outlining the analytical results used in following sections.
In Section 4.3 we derive closed form expressions describing the bit error rate for
a received frame in the presence of hidden terminal interference for each of the
802.11 DSSS P H Y protocols under the assumption of a B P S K modulated signal.
Section 4.4 presents numerical results obtained with these expressions which il
lustrate how a m o d e m will be unable to receive a transmission in the presence
of an interferer with a relative signal power between the signals of greater than
2dB. Finally, Section 4.6 draws conclusions, describing the key points from this
analysis that must be considered in the development of a model to accurately
reflect empirical behaviour.
4.2 Spread Spectrum Error Probability
Analysis
The initial analysis of the multiple access interference problem was undertaken
in the late 70's by Pursley (Pursley, 1977) with later extension by Geraniotis
(Geraniotis and Pursley, 1982). (Pursley, 1977) presents a method to deter
mine the level of multiple access interference caused by other users in a spread
spectrum multiple access system. W e briefly review this result here.
Phase coded spread spectrum systems allow multiple access by assigning each
user a unique spreading code. The receiver is then able to detect and receive
a signal from a given user at the same time as a number of other users are
transmitting. In this model, we consider an asynchronous system, being an
accurate description of the IEEE 802.11 system. Spreading sequences are of
length N, with a total of K active users. The signal is assumed to be B P S K
modulated.
Error Probability Analysis - Hidden Terminal Jamming 79
b(t)
»2W
b(t)
\^a(t)cos(<ot+e,) Delay
V2Paa)cos(o)t+e,) 2 C L
\/Sa^t)cos((»,t + eK)
KO Receiver
n(t)
Figure 4.1 DSSS System Model (Pursley, 1977)
A direct sequence spread spectrum signal transmitted by the kth user under
these assumptions can be represented as:
sk(t) = V2Pak(t)bk(t) cos (ujct + 0k), k = l,...,K (4.1)
where ak(t) is the spreading code, bk(t) the data signal, 9k the phase of the
kth. carrier, UJC the centre frequency, and P the signal power. A n asynchronous
DSSS system model is illustrated in Figure 4.1.
After passing through the channel and incurring both channel noise and delay,
the received signal, r(t), is then described as:
K
r{t) = n{t) + ^2 V2Pak(t - rk)b(t - rk) cos(ujct + 0k - ucTk) (4.2)
fc=i
where n(t) is a two sided gaussian noise process of spectral density N0/2, and rk
is the delay of the /cth signal. The signal is then passed to a correlation receiver.
The output of the receiver at time t = T, written in terms of the component
data signals bk(t), is given by:
K
Zt = y/P/2hi)0T+ ^2 [h,-iRk,i{Tk) + bk,oRk,i{n)]-cos9k
+ / n(t)ai(t) Jo
k=l,k^i
cos ujctdt (4.3)
Error Probability Analysis - Hidden Terminal Jamming 80
Where Rkji and Rkii are the continuous time partial cross correlation functions
for sequences (af^), (of). Rk)i and Rkji are expressed in terms of the aperi
odic cross correlation functions for the spreading sequences, defined in Equa
tion (4.4): ( N-l-l
J2 °*0'KO' + 0 o</<iv-i 3=0
N-l+l
£ ak(j-l)a*(j) 1-N<1<0 3=0
0 elsewhere
ck,S) = { (4.4)
To determine the average signal to noise ratio at the output of the ith. correlation
receiver, the Root Mean Square (RMS) noise component in the correlator output
is required. This is given by Equation (4.5):
VarZi = PT2
12N3 £ rkA+\N0T (4.5)
To simplify the S N R calculation, Pursley defined (Pursley, 1977) an average
interference parameter, rk)i in terms of the aperiodic cross correlation for the
two sequences defined in Equation (4.4). Using the cross-correlation parameters:
N-l
»kAn)= 1 3 Ck,i{l)Ck>i(l + n) (4.6) /=1-JV
the average interference parameter is written as:
r*,t = 2/4^(0)+//*,< (1) (4.7)
The signal to noise ratio is then expressed as y/P/2T divided by the R M S noise
component in the correlator output as described by Equation (4.5). The final
signal to noise ratio is given by Equation (4.8):
-1/2
SNRi = 1% , _1_ A 9 B? R A^3 2-^i 2E 6N3
Th,i
fc=l,/t
(4.8)
Equation (4.8) describes the signal-to-noise ratio experienced by a single user
in spread spectrum multiple access systems, as a result of multiple access inter
ference due to other active users in the system. Given that the channel model
Error Probability Analysis - Hidden Terminal Jamming 81
assumes no spatial isolation of transmitters, path loss need not be considered.
In the following sections, we apply this result to the IEEE 802.11 P H Y , as a
special case of an S S M A system, in which each user employs identical spread
ing codes, in an effort to describe the impact an interfering signal has on the
reception of a specified frame.
4.3 Error Probability of Captured Frame
The experimental results presented in Chapter 3 illustrate a distinct relation
ship between the ability of a host to capture the radio channel, and the relative
received signal power of each contending frame measured at the receiver. This
is a problem specific to topologies where the standard C S M A / C A access mech
anism is unable to sense a transmission that may result in a collision at the
intended receiver. A successful transmission relies on the reception of an R T S
frame by the intended receiver and the subsequent successful reception of a C T S
message by all potential interfering nodes.
When two or more hidden terminals are attempting to communicate with a
common receiver, we consider two possible collisions which may occur at the
receiver:
1. an RTS frame from connection A collides with a DATA frame from con
nection B
2. an RTS frame from connection A collides with an RTS currently under
reception from connection B
In each case the eventual behaviour will be dependent on many additional fac
tors, including the timing of the interfering frame arrival, and the relative signal
power of both transmissions. In case 1, the contention will be handled by the
M A C protocol. However, the measurements in Chapter 3 show that the stronger
host will be able to capture the channel after a number of backoff periods. Even
Error Probability Analysis - Hidden Terminal Jamming 82
though the R T S frame is relatively small, 40 bytes compared to several hun
dred for the D A T A frame, there is a high probability that the data frame will
be corrupted by the collision if the signal power is sufficiently high. This then
provides an opportunity for the stronger host to prevent a weaker host from
gaining access to the channel through a number of timeout and retransmission
cycles. This case is further complicated by the fact that all control messages
(RTS/CTS etc.) are transmitted at the 1 or 2 Mbit/sec transmission rate, re
sulting in the potential for a signal spread using the Barker sequence to collide
with a signal spread using the C C K codes generated with (2.2).
In case 2, the receiver will either retain capture of the original RTS frame and
return a valid CTS, or will loose both of the frames, unable to respond with
a C T S until an R T S has been received correctly. The experimental results in
Chapter 3 suggest that the stronger host will win this contention period, and
be able to capture the channel.
To examine the impact of an interfering transmission on the reception of a
previously acquired frame, we investigate the resulting B E R obtained at the
output of a correlation spread spectrum receiver. The model assumes that the
initial signal, i, is currently being received, at T > Ta, where Ta is the time
required by the correlation receiver to acquire and achieve synchronisation with
the signal, the acquisition time. A n asynchronous interfering signal, k, arrives
at a time T2 > Ta. For both the BR-DSSS and HR-DSSS physical layers,
we determine the impact this has on the correlation receiver output B E R as
a function of the relative power difference between the signals, and hence the
ability of the receiver to maintain capture of the initial signal.
Our analysis for the DSSS physical layer is based on the results described in
Section 4.2. W e consider the IEEE 802.11 as a restricted case of a spread
spectrum multiple access system in which all users employ a common sequence
or set of sequences (in the case of C C K ) . This has not been examined previously
in literature.
Error Probability Analysis - Hidden Terminal Jamming 83
W e define the relative signal strength between the contending signals as
ft = §* (4-9)
where Ebk and E^ are the respective bit energies for each signal.
The SNR experienced by the original (ith) signal, at the correlator output of a
B P S K asynchronous D S - C D M A receiver due to the presence of the interfering
(kth) signal is given by Equation (4.8). In the case under investigation, K is the
total number of concurrent signals received (including the ith signal whose B E R
we are investigating), iV0 the one sided noise power spectral density, Ebi the bit
energy of the zth signal, N the sequence length, and rkj is the Average Inter
ference Parameter (AIP). It has been shown that the AIP can be approximated
(Karkkainen, 1992) as: N-l
rk,i~2 J2 \Ck,M2 (4-10) 1=1-N
where Ckji is the aperiodic cross correlation between the two sequences a,j and ic \^k i s.a tiic opcuuuii- uuaa I^VJIICICILUJII uciwccu LU G IJWU D C ^ U C U L C O 0
Oj , as defined in Equation (4.4). If we assume that the interfering signal A;
arrives with a power 5k times greater than the current frame i, (4.8) can be
written as:
SNRi = N0 1 K N-1
+ i E « E iwIs 2EM 3N3 bl k=l,k^i 1=1-N
(4.11)
The B E R is then expressed as:
BERi = Q (SNRi) (4.12)
where Q is the complementary error function.
4.3.1 DSSS Basic Rate Physical Layer
The use of a single spreading sequence for the both basic rates allows us to
simplify (4.11). The aperiodic cross correlation Ck>i is replaced by the autocor
relation function, Ck>k for the specific Barker sequence employed. The autocor
relation sequence for the 11-chip Barker sequence employed in an IEEE 802.11
Error Probability Analysis - Hidden Terminal Jamming 84
• 4 - 2 0 2
Delay (chips)
Figure 4.2 Autocorrelation function for 11-chip Barker sequence, +1,-1,+1,-1-1, 1,+1,+1,+1,-1,-1,-1, employed in the 802.11 DSSS P H Y
DSSS receiver is illustrated in Figure 4.2. Combined with the approximation
derived in (Karkkainen, 1992), the final S N R expression reduces to:
SNRi = INo 2Ebi
K
k=l,k^y
(4.13)
where a — 480.3. The value of a is determined using the autocorrelation
function for the Barker sequence illustrated in Figure 4.2, and the length of the
sequence, iV = 11. This analysis assumes B P S K modulation. The BR-DSSS
P H Y employs D B P S K for the 1 Mbit/s rate, and D Q P S K for the 2 Mbit/s rate.
A practical system employing differential modulation will require even higher
S N R at the receiver to achieve equivalent B E R performance (Proakis, 1995).
4.3.2 DSSS High Rate Physical Layer
Spreading codes for the high rate physical layer are generated using Equation
(2.2) resulting in 16 complex codes for the 5.5 Mbit/s rate at to 4 bits per
symbol, and 256 distinct complex spreading codes for the 11 Mbit/s rate at 8
Error Probability Analysis - Hidden Terminal Jamming 85
bits per symbol. For each rate, we use Equation (4.11) to generate an expression
for the output B ER.
If we again use the ratio of bit energies for the current and interfering signals,
Sk, and the approximations of the previous section, we can use Equation (4.11)
to determine the S N R experienced by the ith frame, averaging this result across
all sequences in the set to determine the average probability of error.
4.4 Numerical Results
4.4.1 Single Interferer, K = 2
This scenario corresponds to Host 1 in Figure 3.1 attempting to send an RTS
or D A T A frame to Host 2, which is currently involved in the reception of a
frame from Host 3. The B E R given by (4.12), using (4.11) and (4.13) has
been calculated for a range of Eu/N0 values, as a function of 5k. In this case,
the i sequence corresponds to the signal currently being received, and the k
sequence the interfering signal. In each of Figures 4.3, 4.4, and 4.5 the presence
of the interfering signal from Host 1 (the A;th signal) significantly effects the
received B E R of the signal from Host 3, ith signal. This is observed through
the increased B E R as the value of 6k increases (i.e. the relative power of the
kth signal from Host 1 increases).
The results for the BR-DSSS 1 and 2 Mbit/s rates are shown in Figure 4.3.
With Sk - 0 dB, the interfering signal (A;) arrives with a power equal to the
current signal (i). At higher Eu/N0 the presence of the interfering signal will
increase the B E R of the initial signal, but will still allow a high probability of
successful reception of the initial frame. In this case, both connections will have
an equal impact on the other, providing the M A C protocol with a relatively fair
scenario to operate. This provides a strong basis for the fair channel access
observed with the equal S N R T C P and U D P experiments of Figures 3.5, 3.6,
3.10, and 3.11.
Error Probability Analysis - Hidden Terminal Jamming 86
With Sk = 5 dB, the presence of the interfering frame raises the B E R to ~ 10 _ L 5 ,
significantly reducing the probability of successful reception of the initial frame.
If the A;th signal arrives with Sk < 0 dB, then the current ith signal will suffer
little increase in B E R and retain a high probability of successful reception.
The calculations for the HR-DSSS were performed by averaging the SNR as
given by (4.11) across the entire number of sequences in the set. This requires
the calculation of the interference parameter, rkj for each sequence in the set.
The number of codes in the set represents the number of interfering transmis
sions across which the result must be averaged.
For the 5.5 Mbit high rate sequence set shown in Figure 4.4 the BER follows very
closely that of the single Barker sequence employed by the BR-DSSS. In the case
of the 11 Mbit/s rate (Figure 4.5) the B E R impact is marginally worse, being
approximately 10~0-5 higher than for the BR-DSSS at Sk = 0 dB. This difference
is relatively insignificant, as in either case, the presence of an interfering frame
with Sk > 0 dB will, with a high probability, corrupt the current transmission.
4.4.2 Multiple Interferers, K > 2
Figure 4.6 illustrates the impact multiple interferers have on the average BER
for the BR-DSSS 1 and 2 Mbit/s rates. As the number of interfering frames is
increased from 1 to 4, the average B E R is increased from 10 - 7 to 10~4, 10~3,
and 10-2-5 respectively.
Figure 4.7 illustrates this for 11 Mbit/s with Ebi/N0 = 20 dB. Again, as the
number of interferers is increased, the B E R is significantly greater. In practice,
a single interferer with 6k > 2 d B will be sufficient to jam a competing hidden
connection. With these results we have assumed each interfering signal arrives
with equal <5fc. This result indicates that a host may be unable to success
fully access the radio channel when competing with a hidden terminal having
a marginally higher received signal power. Transmissions from any terminal
with a weaker received signal power are potentially jammed by the stronger
Error Probability Analysis - Hidden Terminal Jamming 87
Figure 4.3 Correlator Output B E R Experienced by ith Frame for 2 Mbit/s Barker spreading code
connection, making the R T S / C T S handshake effective for the strongest host
only. As stated earlier, this analysis is based on the assumption of a B P S K
modulated signal. More complex modulation schemes require a higher signal
to noise ratio at the receiver to achieve equivalent B E R performance (Proakis,
1995). Analysis of differential and quadrature modulation schemes is considered
unnecessary for two reasons. Firstly, this analysis based on B P S K shows quite
conclusively that an interfering signal arriving with a relative signal power of
greater than 2dB will result in the corruption of the original signal. Secondly,
unless the short header option is employed (as part of the IEEE 802.11b ex
tension (Institution of Electrical and Electronic Engineers, 1999b)), the P H Y
header and frame preamble are transmitted using BPSK. Collisions are most
likely to involve this section of the frame.
Error Probability Analysis - Hidden Terminal Jamming 88
10°
10"1
a: UJ
2io"2
2
6 § 1 0 J
a
DC
O ,„-4 s 10 & S ° n-5 E 10 2 £
R10"6
& <
10"7
10"*#
i;;:}|i;;;;;:|;;;i;il};;|:;;;;;;;;iS|
• ' • ' * * ^ ^ ^ ^ ^ T T T T I : | i |
'.'.•'.'•'•'.'.'.','.'.'.'.'.'.'.'.'. .,V 1 ; i • V •
H •::::::::::: ::::::::::'/::: ::::!;*
' ;> ; /,' r: :: : : : :,M : ::::::•:::::::: ^f; : : : ; : i • •
l-i-iill:!!!!!!!"!!!!!!*!!!!?!!!!!! ^ — # / >
.' / sijiiMji: ;•:••• i:Mfi::!<t::: ::::•!::
/ '
- / /
/ / '"
I:::*::::::::?::;::::::::::::::::;::
! 0 2
....*''.. >:!!-v* >::::/>:: •••••••
*::::::::
:::•!:::::'
• ;'''':«» ^ -
4 SdB
its** • i • •
; I ; ;
! /!'.
: > : :
; H :
.', . .
I 6
\&$
:::::::*::::::
it:::::-:::::::
: :
:::::::::::::: -
:::::: :-. : : ; : : ; I : : : : : : : : : : i 1:
-
:::::::•:::::::
Eb/No = Eb/No = Eb/No =
- - Eb/No = 1 8
! = :::!••-
::::::: Tz
= 5dB . 10dB :
= 15dB : = 20dB
10
Figure 4.4 Correlator Output B E R Experienced by Initial Frame for 5.5 Mbit/s C C K Spreading Sequence Set
4.5 The Retraining Hypothesis
The analysis presented in previous sections investigates a scenario where two
signals arrive at a common receiver, separated in time. W e investigate the
impact the later arriving signal has on the reception of the first arriving signal.
As each signal will arrive with an independent signal power, the relative power
between the signals is employed as the control parameter in the investigation.
IEEE 802.11 is different from a typical C D M A system, as all users employ the
same spreading code rather than rely on separate orthogonal codes to provide
multiple access. The assumption in this analysis is that the earlier arriving
frame has been detected, the receiver is synchronised and reception is underway.
This will involve detection of the peak in the correlator output signal.
The results in Section 4.4 illustrate that if the later arriving frame has a received
signal power greater than 2dB stronger than the earlier arriving frame, the B E R
Error Probability Analysis - Hidden Terminal Jamming 89
10"
10- -
DC UJ CD 1 0 -
S
a
§ 1 0 r
o O 10"
s _ '5 10
10
10
:::::
1 1 1 !
f>:*!
:::::
" : : : ::::: '.'.'.'.'
iji<^
- • : •
~ : : :
:::..:...!::,.:.::....1 . , : . • ' • T - - - V V 1 ,:::
I:::;:;::::::::::::::::::::::;:: \\\\\\\\\\\\\\±\*\*Uj&^.V.\\\\\ :•::::::::::::::::::::;:::::r.±;.«>.r.U^jijj#*\\::::;::::::::
; ^ • * < * ' ; ; • * / & * •
;u<~'*\ ;"'V<-;—« « ! « ? |t =' "='= M = = = = = = = = = ! = = = i i • - ^ i * ^ - = i= = = = = = . = = = = ; ::::::::::::•:::::::::::::::: ;.:•:: : • > ( ' * : : : : : : :::::: I::::::::;::: I ••::
..•'"• x V v-. . . . !>««*> ;.. . .
!!!!! M M ! i!:!! = i«fn!! = = /!!M!M:::!M:=3 = iM = :::i: = = = : = ! = : = = : = = : •••:::::::::::. ;!! I!!;:/; !/!;•;: •:::::::::!:::::::::::!: ::::>:!::::::
::::;••' . *. .. ,.•'...: /r.. >...: ;
/ / it!! = !!!n!!:!i!!!l/!^!!;i!i!!!!!!!i!!i!!!!!:!!!!!!!!!!!!!!!!!!!!!!!!! ::::::::::::::::: l/l :/:::::::::::::::::::: :::•:•::::::::::::: •/•*•
: ' • / • / : •
= = = = = = = = = = = =><;=#= = = = = = = = = = = ! = = = = = = = = = = = = = = = = = = = = = = : = = = : = : ====== = f ' '/••I
/ : / /• T ;=•;;=:
/ f ••'•/• ir
/ = = i = = = t\ '.'.'•'. \\ = = i i = = I : = = :• = 1 = M = = = = = = = = = : : =:= : = = = = = = = = = = = i = = : = = : =
t '
' i i i i
._.
— —
"s • ••
: : • : : : : : : :
:'; : : : : : :
Eb/No=
Eb/No= Eb/No= Eb/No= i
i::i:::::i
:! = = = = = = -::::;
_ ::::::::::
: : : -::::::::::
1:: = :::: "
: : : : : : : : n
::::::::::
::::::: :-
5dB 10 dB i 15 dB : 20 dB
4
SdB 10
Figure 4.5 Correlator Output BER Experienced by Initial Frame for 11 Mbit/s C C K
Spreading Sequence Set
10"
10"
210" Q.
| 10"3
§ 1 0 Jo
o O 10
3 10
S
10"
10-
! !"::*::,: ' ••••-•n
* v- -" ' *». .^r:.. ..,<'. * V ^ .•' ..,i* *.. yr.;,.-.'. .<:-. ••*• jT .„"•=••'•' -.•#• :'
• " / .,=-*•' « < *
::::::::::::::::*::^:::;;-::;:::::::>::::::::::. z'y'-y :-y
- = = = = :M = =*::=>n = = = =>?=M = = :: = ;:=J«: = = = = = = i = = M:= = * / 1 /
/ / >
/ / / • • /
::/:::/:::::-::::::::::::/::::::::::::::::::::::::: > y .-...: / .../„„o*=.. =:= = = = = = !/: = . = = = = = : : / ,-? ' / y .-•• /
{ ~ ' - ; ' • • ! ' • • •
:::::•::::: : :::::;:*:::::::::::::.!::::::::::::::::::: '•":•'' /
n!r!;i!!!!!ii!!<:!!!i!!!!!!!!!!!!i!!!!i!!!i!MM!!:i! /
•-" / • • : • : • •
T^nTH ' • : • : • $ * :
.<.
<•
:: >
•:;
• ; .;
• •
: :
1
TTiTi »
::••:•*
?2*^iV~~
•
• : I : : :3 : : : : : :
::::::'::::::; !!!!!!"!!!!!!
::::::;::::::
r
•Sli w
....
::::::::::
:•:•:::: fl
-
::;::::;-!
K = 1 .
K = 2 :
K = 3 : K = 4 •
4
SdB
10
Figure 4.6 Barker Code (K - 1) interferers, Ehk/N0 = 2MB
Error Probability Analysis - Hidden Terminal Jamming 90
10°
10"'
Receiver Output BE
o
o
Aquired Fr
ame Correlator
o
o
o
4,
i,
i.
10"7
10 ,
.*.~£j*^r^'"'':'~ >•>'"*'"•
::>::::y :i :::£::::: ::::::#:!::::: / / 0;.i *...
_...; :./
/ ' " • * " : ' '
'•
i!!ll!!!!!!i!!i!!!!r
!iH = ;;!!!;i!i = !i = r=
:::*:::::::::::::::::
_
• . . . i i i i
0 2 4 6 SdB
K=1 . K=2 ! K=3 :
- - K=4
8 10
Figure 4.7 C C K codes, (K - 1) interferers, Ehk/N0 = 20dB
at the output of the correlation receiver will be intolerably high, thus destroying
the earlier frame. This result in itself is not unexpected, as the later arriving
frame is simply a 'loud' noise source from the perspective of the the earlier frame.
However, if the later frame is sufficiently strong, the signal detection mechanism
can be effectively reset when this later frame arrives. The correlator output will
include both frames, offset in time, and m a y result in the re-synchronisation of
the receiver with the new stronger correlator output.
Therefore, in certain scenarios the receiver m a y be able to re-train onto the
new signal (Kim et al., 1995). IEEE 802.11 interfaces have been designed to
exploit this, and attempt to successfully receive both signals (Mud et al., 1999).
However, the results of this investigation suggest that the interfering signal will
have an intolerable impact on the received B E R of the initial signal. It is also
possible to reverse this scenario such that the interfering signal has a lower re
ceived power than the original signal. In this case, the B E R experienced by
the original signal will not be significantly increased by the interfering signal.
Error Probability Analysis - Hidden Terminal Jamming 91
Further, if the receiver is retrained onto the interfering signal, the scenario is
effectively reversed with the weaker signal becoming the new source of inter
ference for the stronger signal. The stronger signal will be received will little
impact on the B E R . The impact of this behaviour will be examined in greater
detail in the following chapter.
4.6 Conclusion
Understanding the impact an interfering signal will have on the reception of a
previously arriving signal is important for the development of a suitable model
to accurately describe the behaviour of a radio modem. In this chapter, we
have extended previous analysis to determine closed form expressions for the
bit error probability observed at the correlator output for a given signal in the
presence of a c o m m o n code interferer. Numerical results illustrate that the
strong relative received signal power dependence of channel capture behaviour
with the IEEE 802.11 M A C protocol described in Chapter 3 can be attributed
to the impact of an interfering signal.
Analytical expressions describing the BER of a received signal in the presence of
multiple access interference for a number of IEEE 802.11 DSSS physical layers
have been developed. Numerical results indicating that a received signal power
difference of greater than 2dB is sufficient for the stronger signal to effectively
jam a weaker signal, closely match the experiments presented in Chapter 3. This
renders the R T S / C T S handshake ineffective for all but the connection with the
highest received signal power.
The major contribution arising from this analysis is the qualification that rel
ative received signal power between contending hidden connections is the sig
nificant parameter in determining which connection is able to gain preferential
access to the radio resource. A relative signal power of greater than 2dB is suf
ficient to result in a channel capture state for the stronger connection. In the
following chapter, we investigate modelling capture behaviour, with respect to
Error Probability Analysis - Hidden Terminal Jamming 92
the ability of a specified model to match properties of empirical trace data. W e
employ the quantitative results derived in this chapter to develop an improved
model which allows for accurate simulation of the behaviour of an IEEE 802.11
radio modem.
Chapter 5
Modelling Packet Capture Behaviour
5.1 Introduction
In this chapter, we present an investigation of techniques used to model the
packet capture behaviour of an IEEE 802.11 wireless network interface. Chap
ter 2 outlined a number of capture models which have appeared in the literature
(Cheun and Kim, 1998; Davis and Gronemeyer, 1980; Arnbak, 1987). Capture
models are generally employed to either determine the channel throughput as
a function of offered load, or alternatively as a technique to model the radio
interface in a simulation environment. The ability of a capture model to ac
curately reflect empirically observed behaviour when employed in a simulation
application is an important aspect of capture model performance. The aim of
this chapter is to investigate the features a capture model should exhibit in
order to support improved simulation of an IEEE 802.11 radio interface. Meth
ods required to provide a reliable simulation tool will also be investigated. This
is performed through both a qualitative and quantitative comparison of sim
ulation results with empirical trace data. A second aim of this chapter is to
address the shortcomings of existing models through a new model, designed to
meet the requirements for improved simulation of an 802.11 radio interface.
93
Modelling Packet Capture Behaviour 94
Section 5.2 presents an overview of the significant capture models presented in
the literature. The results presented in Chapters 3 and 4 motivate the introduc
tion of a new model designed to represent the salient features of an IEEE 802.11
radio modem. Therefore, in Section 5.3 we introduce the Message Retraining
model based on the design of an IEEE 802.11 radio m o d e m (Mud et al., 1999)
and the results presented in Chapter 4, combined with previous work on parallel
receiver structures (Kim et al., 1995), and multiple access interference (Ware
et al., 2001a; Pursley, 1977). Improved simulation models are necessary for the
future development of mechanisms to overcome or prevent the unfair behaviour
observed in Chapter 3.
Section 5.4 presents the results of a qualitative analysis in which we compare
trace data obtained via simulation with empirical trace data. Given that T C P
timers were shown to have a significant impact in Chapter 3, we undertake initial
experiments with U D P as the primary transport. This will remove the poten
tial for adverse interaction between the T C P timers, and M A C retransmission
timers, and allow a study of the capture effect in isolation. T C P results are
included for completeness. As Chapter 3 has illustrated that fairness proper
ties are a key element of the performance characteristic in a physical network,
Section 5.5 presents a quantitative analysis of simulation and empirical trace
data using two fairness indices. The analysis presented in these two sections
will provide insight into the ability of each capture model match the fairness
behaviour of a physical system, and in doing so re-create salient features of
actual system performance. This will provide an indication of which model is
most applicable for a given scenario. Finally, Section 5.6 concludes the chapter.
5.2 Capture Models
A significant body of literature (Cheun and Kim, 1998; Davis and Gronemeyer,
1980; Arnbak, 1987) exits investigating the development of models describing
the initial capture of a frame by a radio modem. The common goal of each model
is to determine the probability with which a given frame may be captured by
Modelling Packet Capture Behaviour 95
the receiver as a function of the number of active stations. This probability
is then used to determine the channel throughput achieved, as the product of
capture probability and offered load.
As discussed in Chapter 2, capture can be considered to occur at two levels:
• Packet Capture, also termed Modem Capture is a property of both
the radio m o d e m and modulation technique employed (Soroushnejad and
Geraniotis, 1991). Packet capture results in a given transmission being
'captured' by the receiver while rejecting interfering frames as noise. Sev
eral models based on either power (Arnbak, 1987), time of arrival (Davis
and Gronemeyer, 1980), or both, (Cheun and Kim, 1998) have been pro
posed to evaluate the probability of a frame being captured by a receiver
as a function of the number of interfering frames.
• Protocol Capture, also termed Channel Capture is induced by protocol
timing, and results in a channel being monopolised by a single node, or
subset of nodes in a given geographic region. Protocol capture has been
identified as a significant problem for multihop packet networks in many
scenarios where disconnected topologies exist (Nandagopal et al., 2000;
Bensaou et al., 2000), or higher layer retransmission and backoff timers
are employed (Gerla et al., 1999a; Tang and Gerla, 1999).
Two significant stages are present in the successful reception of a frame by a
radio modem. Initially, the frame must be successfully detected and captured by
the receiver. Following this, successful reception of the frame must be achieved
in the presence of interference, from other transmissions and external noise
sources. Most literature (Cheun and Kim, 1998; Davis and Gronemeyer, 1980)
has considered only the probability with which successful detection and capture
of a frame at the start of a transmission slot occurs. The second aspect requires
an understanding of the impact multiple access interference will have on the
captured frame (Soroushnejad and Geraniotis, 1991; Pursley, 1977; Ware et al.,
Modelling Packet Capture Behaviour 96
2001a) and depends significantly on the modulation technique and spreading
codes employed.
Capture models are often used when simulating the performance of wireless
networks. The results presented in Chapters 3 and 4 suggest a more complex
capture behaviour is present in the case of an IEEE 802.11 radio interface,
resulting in the significant unfairness evident in experimental data. Further
complications arise in cases where hidden nodes are likely (e.g. a mobile ad hoc
network). Specifically, there is a strong probability of late starting transmissions
colliding with other signals at the common receiver. In a scenario where all
nodes are able to sense carrier, slot boundaries are easily identified and defined,
thereby reducing significantly the probability of a new transmission interfering
with an ongoing reception.
In scenarios where carrier sense mechanisms are unreliable, it is possible for
a node to have little knowledge of an ongoing hidden transmission. This in
troduces the potential for an interfering transmission to arrive at a common
receiver at any time during a slot. As illustrated in Figure 5.1, this can be due
to differences in the slot time boundaries observed by both hidden nodes. This
is further complicated by the slot timing mechanisms within 802.11. Rigid slot
boundaries are not maintained, requiring nodes to infer slot boundaries from the
beginning and end of surrounding transmissions. Data transmissions are able
to occupy multiple 'slot times'. Guard times are inserted between sensing an
idle channel and transmitting (DIFS), or returning management frames (SIFS)
to maintain the semi-slotted channel. However, the lack of carrier from an op
posing hidden node increases the possibility that a hidden node will transmit at
what appear random times to the central node at which the collision occurs. In
the example shown in Figure 5.1, Host 3 has commenced a data transfer prior
to Host 1 (being hidden from Host 3) initiating a carrier sense operation. O n
sensing a clear channel, Host 1 defers for a DIFS then transmits an R T S mes
sage. This collides with the data frame from Host 3, illustrating the potential
for a late starting transmission to interfere with an ongoing transmission.
Modelling Packet Capture Behaviour 97
HOST1
HOST 2
HOST 3
Sense Clear Channel
i i Slot Time/ / /
Error / DIFS / RTS /
\
1
/ReceiveDATA / Collision
4 / /
/ Transmit DATA / 1 'II / Time
Figure 5.1 Potential Slot Time Error
In the following sections we briefly review the significant capture models pre
sented in the literature.
5.2.1 Delay Capture
Delay capture originally described by Davis and Gronemeyer (Davis and Grone
meyer, 1980), enables the capture of a frame in a given timeslot, provided no
other frame arrives within a given capture time, Tc of the initial frame. Only
the initial frame is able to be received. Frame arrivals are assumed uniformly
distributed on the interval [0,TU], where Tu represents the maximum variation
in packet arrival times at the receiver. The initial frame arrives at time T\, and
may be captured by the receiver provided that Tt > Tx + Tc, where Tt is the
arrival time of the ith frame. This model is chiefly controlled by the parameter
Tc, governing the period of time required by a receiver to detect, correlate with,
and lock onto the received signal. The larger the Tc/Tu ratio, the less effective
the m o d e m is at capturing a frame.
Modelling Packet Capture Behaviour 98
5.2.2 Power Capture
Power capture, originally described with Rayleigh fading, and constant trans
mitter power (Arnbak, 1987), is described by the following inequality over the
interval [0,TC]:
N
Pmax > 7 ^ Pi (5.1) i=l
where Pmax is the power of the strongest of N arriving signals, each with power
Pi, within the capture time Tc. The model allows a frame to be captured pro
vided Pmax is greater than the sum of the power of all other received signals,
Pi, times the capture ratio, 7. The received signals are assumed to have phase
terms varying quickly enough to allow incoherent addition of the received power
of each frame. This model is the most commonly employed in the simulation
of radio modems, allowing the first arriving frame in a slot to be received pro
vided no other frame arrives within the capture time, Tc having a signal power
violating Equation (5.1). In the case where Equation (5.1) is violated, no frame
is captured.
5.2.3 Hybrid Capture
The hybrid model was originally proposed by Cheun and Kim (Cheun and Kim,
1998). The power capture effect is used to increase the capture probability of
the first arriving frame in a given timeslot, even though the delay model would
otherwise indicate capture has not occurred. Capture occurs when the following
inequality holds:
N
1Y,pi\Ti+Tc-Ti]<TcPi (5-2)
i=2
where T\ and Pi are the time of arrival and power of the initial signal respec
tively, Tc the capture time required by the receiver to synchronise with and
Modelling Packet Capture Behaviour 99
prepare to receive the initial signal, T; and Pi the time of arrival and power of
the zth arriving signal respectively, and 7 the capture ratio. The total accu
mulated energy must be less than the energy received from the first packet, Pi
over the capture interval Tc. This model results in a greater capture probability,
reflecting the ability of a direct sequence spread spectrum receiver to correlate
with the initially detected signal and reject other signals as noise.
5.3 Message Retraining Reception Model
As discussed in Section 4.5, contrary to each of the models presented above,
(Mud et al., 1999) describes an enhanced capture technique which allows a mo
dem to successfully receive a signal that would otherwise be considered lost by
the previous models. The m o d e m implements a Message In Message process,
whose function is to monitor the energy received on either antenna during re
ception of a frame. If an increase in energy beyond a given threshold, jmr is
observed, the m o d e m attempts to synchronise with and demodulate the new
energy as a potential new signal. If this is achieved, a retraining process allows
the m o d e m to prepare to receive this new signal once the prior transmission has
finished.
Another factor which motivates the introduction of a new model is the behaviour
of the correlation detector when a new stronger signal arrives. The new signal
is a source of interference for an ongoing reception process. In cases where the
power of the new signal is sufficiently higher than the initial signal, then the
potential exists for the correlation detector to be 'reset' by this increase in signal
power. This is due to the use of common spreading codes for all users in the
network, as the output of the correlator appears identical for all signals, simply
offset in time. The result may be the subsequent loss of the initial signal, and
successful reception of the new signal. In cases where the detection circuit does
not employ multiple reception paths discriminating between users through time
separation of the arriving signals (Kim et al., 1995), the receiver will be unable
to receive the initial signal in the case where an interfering signal has sufficient
Modelling Packet Capture Behaviour 100
HOST1
HOST 2/ AP
Rx Message
Rx Signal
HOST 3
/ Signal 2 /
/ Rx Signal 1 / Rx Signal 2 /
, J r 1 Y
1
/ - //////////////
Signal 1 Starts > Signal 2 Starts Time ' Rx signal level increases
If Increase in Rx Signal >y ml
Signal 1 corrupted, Signal 2 received Else, standard power capture
Figure 5.2 Operation of the Message Retraining model
power. In the alternate case, a stronger signal would suffer little interference as
a result of the weaker signal.
In either case, each of the capture models previously described will result in a
pessimistic capture probability for a frame over a given duration. The message
retraining ability of the m o d e m also extends the time scale over which capture
must be considered. Retraining may take place at any time during frame recep
tion, as opposed to the delay, power and hybrid capture models which consider a
short duration at the start of a frame or slot. W e therefore propose an extended
capture model, termed Message Retraining which incorporates this enhanced
capture behaviour.
The model allows the modem to receive a new signal (Signal 2 in Figure 5.2),
which may arrive at a random time during the reception of a previous signal
(Signal 1 in Figure 5.2), provided the new signal has sufficient relative power
to enable successful synchronisation and demodulation of the frame preamble.
Modelling Packet Capture Behaviour 101
In Chapter 4, we illustrated that the Eh/N0 associated with the new signal
will have a significant impact on the B E R observed at the correlator output
for the original signal. Results indicate the previous signal will be corrupted
if the power difference between the new and existing signals is greater than a
threshold of greater than 2dB. The Message Retraining model accounts for this
by dropping the initial frame if a new frame is detected with a signal power
greater than the current by the Message Retraining threshold, jmr. Successful
reception of a frame, Fj will occur provided that over the duration of this
transmission: N
Imr J2 P i < P J (5-3)
This model allows for the successful reception of the frame received with the
strongest power throughout its own duration, i.e. Fj will be successfully received
provided no other frame arrives over the duration of Fj with a signal power
greater than Pj + jmr (measured in d B m ) . Furthermore, the initial frame may
be successfully received provided that power capture described by Equation
(5.1) holds.
5.4 Simulation Investigation
5.4.1 Methodology
The aim of this experiment is to perform a qualitative comparison of the trace
output of simulation trials obtained with each capture model against the em
pirical trace data. The scenario under investigation involves controlling the
received signal power on each connection throughout the experiment, using the
topology illustrated in Figure 3.1. Each node initiates a data transfer into the
common node.
In the UDP experiments, both connections commence with an equal received
S N R of 25dB. 10 seconds into the experiment, the signal power of Connection
A is reduced by approximately 8dB through to the end of the experiment,
Modelling Packet Capture Behaviour 102
thus making Connection B the stronger connection from this time on. In each
experiment, Connection B commences 0.5 seconds after Connection A.
With the TCP experiments, Connection A commences the experiment with a
received S N R of 25dB, Connection B with a received S N R of 20dB. 5 seconds
after the start of the Connection A trace, the signal power scenario is reversed
with Connection B made the stronger of the two connections.
Both signal power scenarios are examined as a means of introducing variability
into the results. Through this method, it also becomes possible to examine
the impact a change in received signal power has on the performance of both
the M A C protocol and the transport protocol with each capture model. In
the case of U D P , a single change is employed to examine the ability of each
capture model to cause M A C behaviour matching the empirical data. With the
T C P experiment, a different scenario is employed in which the signal power on
each connection is reversed. This is to investigate the ability of the capture
model to accurately induce the interactions between transport and M A C layer
retransmission timers observed in the empirical data. The assists with the
qualitative analysis in which the ability of the model to match the characteristic
shape of the empirical trace.
The series of results presented in this section involve the transfer of a constant
number of bytes. The differences between each model are illustrated in the char
acteristics of each curve, including the different duration of each trial. Therefore
when presenting these results it is more beneficial to be able to examine the
progress of each trace in a manner which allows the instantaneous behaviour
of each connection to be observed, rather than comparing the aggregate perfor
mance over a fixed timescale. This provides the reader the ability to visually
determine which connection is gaining preferential access to the channel and
how the related capture effect is evolving throughout the experiment.
Simulation trials have been undertaken with the environment described in the
following section. Results for the U D P and T C P experiments are presented in
Modelling Packet Capture Behaviour 103
Sections 5.4.3 and 5.4.4.
5.4.2 Simulation Environment
Each of the capture models described previously has been implemented within
the ns-2 simulation package (version 2.1b3) (UCB/LBNL/VINT, 1999). This
package contains a detailed 802.11 P H Y / M A C layer model, as well as provid
ing excellent implementations of higher layer protocols such TCP/IP, UDP,
F T P etc. The channel model employed is an Additive White Gaussian Noise
( A W G N ) Two-Ray Ground model. Capture decisions are made within each
modem based on the received signal strength, capture threshold, and other rel
evant parameters for each model. Each node receives a copy of the transmitted
packet and based on the received power, determines whether the transmission
was observable or not. If the frame is received with sufficient signal power, a
capture decision in accordance with each model is made prior to passing the
frame up to the M A C protocol. The network model considered is one involv
ing hidden terminals over a semi-slotted 802.11 M A C / P H Y layer, illustrated in
Figure 3.1. All nodes employ a common spreading code with no power control.
Parameters for the radio interface are listed in Table 5.1. The capture thresh
old, 7, is selected based on measurements presented in Chapter 3, results of
analysis performed in Chapter 4, and design parameters of the message retrain
ing process in an 802.11 modem (Mud et al., 1999). Pt represents the nominal
transmitter power of the radio modem, Rb the channel bit rate (determined by
the combination of spreading sequence and modulation technique employed),
/ the operating frequency, and Tc the capture interval which corresponds to
the duration of the preamble and sync bits in the 802.11 P H Y header. As the
802.11 standard requires that the P H Y preamble and header are transmitted at
IMbit/s with an 11 chip Barker code using D B P S K modulation, or possibly 2
Mbit/s with the 11 chip Barker code using D Q P S K modulation where the short
P L C P preamble/header option is available, we use a value of R B at 2Mbit/s.
Simulation trials are also performed over shorter time scales than the experi-
Modelling Packet Capture Behaviour 104
Table 5.1 Modem Simulation Parameters
Parameter
7 Pt (Nominal)
Rb Sensitivty
/
Tc
Value
2dB 15dBm
2 MBit/sec -95 dBm 2.412 GHz 120/zs
mental trace as we are investigating the ability of the simulation model to react
to changes in the received signal power scenario. Accordingly, we are not using
comparative aggregate throughput as a performance metric.
5.4.3 UDP Results
The following results are from experimental trials employing a UDP transport
layer. Each source is modelled as a greedy source which transmits packets from
the network layer as quickly as the M A C layer will allow. In each case, it is
important to acknowledge that we are examining the ability of the simulation
model to match the dynamic behaviour of the real system, therefore we are
examining instantaneous results, rather than comparing longer term averages.
The behaviour we are attempting to match through simulation is in itself an
instantaneous behaviour. Therefore, in an effort to provide an equivalent basis
for comparison, each simulation trial is performed with the same random num
ber generation seed for all capture models. Complete sets of experiments were
repeated with a number of different random seeds. The results presented in this
and the following section represent typical results from this process.
Trace Data
The empirical trace data presented in Figure 5.3 illustrates similar unfairness to
the trace data presented in Section 3.5. The reduction in received signal power
Modelling Packet Capture Behaviour 105
Connection A Connection B
30
Time (sec) 40 50 60
Figure 5.3 Trace Data U D P Transport: Lucent Chipset
of Connection A just before the 10 second point is obvious in the trace. Again,
the horizontal axis represents time, the vertical axis data successfully received
at the common node. The received signal power is reduced using R F absorbing
foam at the transmitting node.
No Capture
Figure 5.4 illustrates a simulation trial without a capture model. Any colliding
frame at the receiver will result in both frames being corrupted. Accordingly,
there is no indication of any change in the behaviour when the signal power
of connection A is reduced. This is to be expected as signal power plays no
role in determining the reception of a frame in the event of a collision with
this model. The trace exhibits short periods where either connection is able to
dominate the channel. This behaviour is random, and due to the ability of a
node to maintain a lower average backoff window over the period in which it has
captured the channel. Obviously, this trace does not exhibit similar behaviour
to the empirical trace of Figure 5.3.
Modelling Packet Capture Behaviour 106
1400
1200
1000
s m £. 800
DC co
a
600
400
200
_
-
1 1
/ /
— * — ^ - 1
1 1 1
/ / /' /
i i i
i i • I
/ X ,
f -y
Connection A Connection B
i i i
-
+ X
8 9 Time (sec)
10 11 12 13
Figure 5.4 No Capture Model U D P Transport
Delay Capture
Figure 5.5 illustrates a simulation trial with the delay capture model described
in Section 5.2.1. Again, this trace exhibits alternate periods of channel capture,
where either connection gains access to the channel at the expense of the other
host. Between 7 and 9 seconds, Connection A has captured the channel, then at
9 seconds, Connection B manages to reverse this scenario. There is no indication
that the relative signal power change has had any impact at 10 seconds, as is
evident in the empirical trace after the signal power of connection A is reduced.
Power Capture
The power capture trace illustrated in Figure 5.6 again illustrates periods of
random capture through the period where signal power is equal. After Connec
tion A is reduced at 10 seconds, random periods of capture appear to remain as
both connections are still able to capture the channel for short periods. Between
approximately 10 and 11 seconds, Connection A is able to maintain preferential
access to the channel despite being the weaker connection. This behaviour is
Modelling Packet Capture Behaviour 107
1400
1200
1000
m * 800
| 600 eg eo a 400
200
Connection A Connection B
8 9
Time (sec) 10 11 12 13
Figure 5.5 Delay Capture Model U D P Transport
1600
1400
1200
£ m vr
T& > 0>
ai CC ra a a
1000
800
son
400
200 •
_
-
-
•
-
-
— 1
/
1 1
..J
1 1 i
/
<
1 1
, . /
/
Connection A Connection B
/
X
-
-
-
-
-
-
8 9
Time (sec) 10 11 12 13
Figure 5.6 Power Capture Model U D P Transport
Modelling Packet Capture Behaviour 108
1600
1400
1200
5, 1000 m
> 800 o a d>
IT 2 600 CO
a 400
200
I •
-
-
-4
/
<- * — t —
1 1
ry ' >
/ / /
• 1 1
/
i i i
i 1 —
Connection A Connection B
> •
/
-
•
_
-t-
X
8 9
Time (sec) 10 11 12 13
Figure 5.7 Hybrid Capture Model U D P Transport
not in concert with the empirical trace in Figure 5.3.
Hybrid Capture
The Hybrid model results illustrated in Figure 5.7 illustrates much less exag
gerated periods of capture for each host, with improved sharing throughout
much of the period, though Connection B does achieve a higher throughput
than Connection A. W h e n the signal power is reduced on Connection A at 10
seconds, Connection B is able to achieve greatly improved access to the channel,
matching the empirical data more closely than the previous models.
Message Retraining Capture
The Message Retraining model results illustrated in Figure 5.8 exhibit approx
imate fair sharing through until the signal power change at 10 seconds, when
Connection B is able to capture the radio channel. Short periods where either
connection is able to slightly dominate the channel are present until this point
(as with the Hybrid model). The resulting trace illustrates an improved ability
Modelling Packet Capture Behaviour 109
1600
1400
1200 •
<D
co
i > <D O <D
tr (0 CO
D
1000
/
Connection A + Connection B *
8 9 10
Time (sec)
13 14
Figure 5.8 Message Retraining Capture Model U D P Transport
to match the empirical data compared with the Delay and Power models.
Both the Hybrid and Message Retraining models illustrate improved ability
to match the experimental data. In particular, both are able to match the
behaviour after the change in relative signal strength occurs 10 seconds into
the experiment. This result indicates that from a qualitative viewpoint, the
Delay, and Power models do not model the necessary features of an 802.11 radio
interface in order to support accurate simulation. Additional investigation of the
performance of the Hybrid and Message Retraining model supporting T C P data
streams will further establish the ability of each model to reflect the empirical
data.
5.4.4 TCP Results
Trace Data
Figure 5.9 illustrates the time evolution of the T C P experiment used for com
parison with simulation. Connection A commences as the stronger connection
Modelling Packet Capture Behaviour 110
1200
1000
•5T 800
£ m 5 600 -
tr
Q 400
200 i
f
10 15 20
Time (sec)
25
Connection A Connection B
30 35 40
Figure 5.9 Trace Data T C P Transport: Lucent Chipset
with an S N R of 25dB. Connection B remains at 20dB throughout. 5 seconds
into the experiment, the scenario is reversed with the received signal power
of Connection A reduced by 8dBm using R F absorbent foam. Connection B
then manages to capture the channel, preventing Connection A from accessing
the channel. In this example, Connection A suffers a significant T C P time
out between approximately 5 and 33 seconds though the experiment. In this
case we would expect the simulation traces to match the changes in behaviour
at approximately 5 seconds, through the common period of each connection.
The stronger host should gain preferential access to the channel, with the trace
having the same characteristic shape.
Modelling Packet Capture Behaviour 111
2500
2000
& m 1500
i > O « 1000 To Q
500
0 4 6 8 10 12 14 16 18 20 22 24
Time (sec)
Figure 5.10 No Capture Model T C P Transport
No Capture
In the case where no modem capture is implemented, any colliding signal at the
common receiver will result in both frames being lost. Backoff and retransmis
sion then results in approximate sharing of the radio channel. In Figure 5.10,
alternating periods where either connection is able to dominate the radio re
source are due to protocol timing interactions between the M A C timers in each
node, the M A C and T C P retransmission timers, and the T C P timers in each
node (Gerla et al., 1999b). These periods are similar in nature to those caused
by the M A C layer with the U D P experiment. In this case, it is not possible to
identify which interaction results in the alternating channel capture periods.
Delay Capture
The Delay capture model makes no account of signal strength characteristics.
Figure 5.11 exhibits random periods during which one of the connections is
able to capture the majority of the channel resource. This is again due to the
interaction between M A C backoff timers and the T C P timers at the transport
1 1
-
• y //:' ,
i i i
/ /
/
1 1 1
1 " t 1
• <
i i
r
Connection A + Connection B x
< •
Modelling Packet Capture Behaviour 112
y / , ,+ Connection A +
/ Connection B x * rL i i i i i i i i
4 6 8 10 12 14 16 18 20 22 Time (sec)
Figure 5.11 Delay Model T C P Transport
layer. Connection B also gains a slightly higher transfer rate than Connection
A, showing no evidence of the changed signal power at 10 seconds. This is due to
Connection B starting before Connection A, and therefore having a larger T C P
window at the time Connection A commences. Connection B is able to expand
it's T C P window without contention for the channel, whereas Connection A
must contend from the establishment of the T C P connection. The connection
start times were staggered in this manner to match the experimental data.
Power Capture
The power capture model trace in Figure 5.12 displays similar behaviour to
the Delay capture model. Neither connection is able to dominate. There is
no evidence of the sustained channel capture exhibited in Figure 5.9, nor any
evidence of the transmission power change at 10 seconds.
As with the UDP experiments, the Power and Delay models are again unable
to accurately reflect the performance of an actual 802.11 radio interface. As
both these models tend to be reasonably pessimistic, destroying both inter-
ilJUU
2000
m 1500
£ co Q
1000
500
Modelling Packet Capture Behaviour 113
2500
2000
m 1500
1000
500
0 <* 4
Connection A Connection B
10 12 14
Time (sec) 16 18 20 22
Figure 5.12 Power Model TCP Transport
fering frames in many circumstances, is would be reasonable to expect that a
more suitable model should behave in a more robust manner when considering
collisions. Both the Hybrid and Message Retraining models fit this criteria.
Hybrid Capture
Connection B again gains the advantage of a larger TCP window at the time
Connection A commences. Unfortunately, as with the Power model, there is
no evidence of behaviour approaching that observed in the empirical data of
Figure 5.9. This is relatively unexpected, as the Hybrid model was able to
reflect empirically observed behaviour with U D P quite well. However, this result
indicates that the Hybrid model does not reflect the appropriate behaviour
required to accurately model an IEEE 802.11 receiver.
Message Retraining
The trace for the Message Retraining model in Figure 5.14 appears to match the
measured data of Figure 5.9 in terms of characteristic shape, and response to
Modelling Packet Capture Behaviour 114
2500
2000
m 1500
0)
8 OC CO CO Q
1000
500
0 <>• 4
1 I
-
/y " ,
i
s
/
i
y y
i
i • 1 — i 1
, • / /
-x /
/
/
/
Connection A + Connection B x
i i i i
6 8 10 12 14
Time (sec)
16 18 20 22
Figure 5.13 Hybrid Model TCP Transport
2500
2000
m 1500
1000
500 -
n 1 1 1 r -i r
_i i
Connection A Connection B
j i 12 14 16
Time (sec)
24
Figure 5.14 Message Retraining Model TCP Transport
Modelling Packet Capture Behaviour 115
the changes in signal strength. Once Connection A commences as the stronger
connection, Connection B is prevented from gaining reliable channel access.
10 seconds into the trace, Connection B, the new stronger connection, is able
to capture the channel from Connection A, which is in turn prevented from
gaining fair access until Connection B finishes. In this experiment, neither
connection suffers a timeout of a similar magnitude to that of Connection A in
the empirical trace, however, both connections do suffer timeouts in excess of 2
seconds throughout the experiment.
The changes in received signal power throughout the experiment which have
such a dramatic impact in the empirical trace data, are not reflected in Fig
ures 5.11, 5.12, and 5.13. This represents a significant shortcoming for the
Delay, Power and Hybrid capture models, which this investigation shows are
unable to accurately reflect the impact of changes in relative signal strength on
the fairness performance of each connection. This can be attributed to the be
haviour of each model in cases where a collision of signals with a relative power
difference above a specified threshold occurs. In each model, once a collision
is determined to have occurred (as opposed to the capture of a frame), both
frames are assumed lost. One conclusion from this investigation is that this is
a pessimistic assumption, resulting in a model which displays a greater level of
fairness than is present in a physical system.
The results presented in both this and the previous section indicate that from
a qualitative viewpoint, the Message Retraining capture model is capable of
providing the basis of an improved simulation technique for an 802.11 radio
interface. However, this comparison is on the basis of a qualitative visual com
parison of trace data only. In the following section, we employ two fairness
indices in a more detailed investigation which will quantify the relative perfor
mance of each model against empirical data using fairness indices.
Modelling Packet Capture Behaviour 116
5.5 Fairness Study
Fairness in wireless networks can be a difficult quantity to define. In the cur
rent context we require that each node is able to access the channel without
sustained delay, and that no node is able to monopolise the radio channel at
the expense of other nodes. This should be independent of the physical network
topology. In cases where the M A C does aim to provide a guarantee on delay
bound or throughput, a more detailed definition of the fairness properties would
be necessary.
Previous experiments in Chapters 3 and 4 have illustrated the significance of
relative signal power in determining the distribution of channel access. There
fore we have designed a hidden terminal experiment for this investigation in
corporating signal power changes throughout the data transfer. Connection B
commences the data transfer with a received S N R of 20dB. Connection A then
commences a data transfer with a received S N R of 25dB, 1 second after Connec
tion B. This experiment examines the ability of each model to reflect the impact
of a change in the relative received signal power at the common receiver. This
experiment is performed with greedy U D P sources. The experiment is then
repeated with with greedy T C P sources, and both the received signal power
and starting order of each connection reversed.
To make a quantitative comparison of the results obtained with each capture
model, a fairness metric is required. Following (Koksal et al., 2000), we employ
two fairness indices : Jain's Fairness Index (Jain et al., 1984), and a new index
proposed in (Koksal et al., 2000), the Kullback-Leibler Fairness Index. In each
case, a sliding window method is used to calculate the fairness over a specified
horizon. The window slides along a packet sequence which indicates which
node has successfully gained access to the channel. A n instantaneous value is
determined for each index, with the average calculated across the entire trace.
Results are presented as curves illustrating the fairness index as a function of
window size. This provides an indication of the fairness horizon, or time scale
Modelling Packet Capture Behaviour 117
over which a user may expect a specified level of fairness measured with an
appropriate index.
As the TCP trace records successfully acknowledged data, this investigation will
provide an indication of the fairness associated with the data transfer at the
transport layer, including effects from the M A C and P H Y layers. W e include the
transport protocol in this manner, as T C P is the most common transport pro
tocol in use today, and any wireless M A C / P H Y protocol should be expected to
support competing T C P streams without imposing additional fairness charac
teristics. Further, comparison with the U D P results also provides useful insight
into the impact protocol timing interactions have on the observed performance
of the network.
5.5.1 Jain's Fairness Index
This index has been used widely in the literature to describe the fairness char
acteristics in both congestion control (Jain et al., 1984) and wireless M A C pro
tocols (Koksal et al., 2000). A n ideal fair distribution of channel access would
result in a value of 1 for this index, though values above 0.95 are typically con
sidered to indicate excellent fairness properties. The index , Fja, is defined in
Equation (5.4) below:
(5>) Fja = - = £ (5.4)
1=1
where pi is the fractional share achieved by the ith connection, and N is the
number of active connections. A value of 0.7 would imply that 3 0 % of nodes
were suffering significant unfairness.
5.5.2 Kullback-Leibler Fairness Index
The Kullback-Leibler Fairness Index was first proposed in (Koksal et al., 2000).
The technique considers the distribution of channel access for each node as a
Modelling Packet Capture Behaviour 118
probability distribution, f. The Kullback-Leibler distance D (T\\T\ , an entropy
measure of the 'distance' between two probability distributions, is calculated
between the desired distribution T, and the measured distribution, f. This
index is defined below in Equation (5.6):
" 1 1 1 D(r||f) = D ([Pl,p2...pn] N' N'" N
(5.5)
N
= [Y^Pi l0S2 Pi + loS2 N
J=l
where N is the number of nodes, and pi the fractional share achieved by the
ith node. A value of 0 corresponds to a perfectly fair system, with values below
0.05 typically indicating a system with excellent fairness properties.
5.5.3 Results
Simulation trials of the UDP and TCP experiments were undertaken, and both
fairness indices calculated as a function of the sliding window size. Figures 5.15,
and 5.17 illustrate the experimental data used for comparison with the simu
lation models. Figures 5.16, and 5.18, present both fairness indices for each
capture model described in previous sections, the empirical data, and a simula
tion trial employing no capture. The window size in each case does not extend
beyond 1000 frames, as this represents a very large fairness horizon of several
seconds.
UDP
Commencing with the UDP experiment, the Power, Delay, and Hybrid models
over estimate the measured fairness as the window increases in size. At very
small window sizes, all models illustrate significant unfairness. The Power,
Delay, and Hybrid models quickly display increased fairness as the window
increases. Figure 5.16 illustrates the significant difference between the Power,
Delay, and Hybrid capture models and experiment. Both Jain's index and the
Kullback-Leibler index indicate the Message Retraining model is able to provide
an accurate estimate of the fairness properties. Jain's index indicates that
Modelling Packet Capture Behaviour 119
2500
2000
s m 1500
1 O
« 1000 co Q
500
0 0 2 4 6 8 10 12
Time (sec)
Figure 5.15 U D P Experimental Data Trace obtained with Cisco chipset. Connection A commences with an SNR of 20dB 1 second later Connection B commences with an SNR of 25dB
the Message Retraining model matches the trace data within 2% for windows
between 50 and 500 frames, and within 5 % for fairness windows greater than
500 frames. Alternatively, the Kullback-Leibler index matches within 1 % for
windows greater than 50 frames. Conversely, for windows between 100 and 500
frames, the Power, Delay and Hybrid models over estimate fairness with Jain's
index by an average of 30%, and an average of 4 2 % with the Kullback Leibler
index.
With respect to the trace data, over a fairness horizon of less than 500 frames,
the Message Retraining model represents an average improvement over the
Power, Delay and Hybrid models of 2 8 % with Jain's index, and greater than
4 0 % with the Kullback-Leibler index.
TCP
The TCP experiment illustrated in Figure 5.18 also demonstrates how Message
Retraining results in improved accuracy in the fairness properties observed with
Connection A Connection B
i
Modelling Packet Capture Behaviour 120
CD •o
CD
0.85
0.8
0.75
0.7
0.65
0.6
0.55 -
CD
9
0.5
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
^.& .© © or'
Message Retraining + Hybrid * Power » Delay °
No Capture • Trace Data ° 0 100 200 300 400 500 600 700 800 900
Window (frames)
1000
V* 5>X-;
TV \ h *
- r\ k\
1 1 1 1 1 1 1 1 1
Message Retraining Hybrid Power Delay
No Capture Trace Data
" • — • • » - • *
*—---:•*.,;-.;.. ' * • • - . - • • •
V *-•-—»^, \
""" fc —.~ y*-~
^s^-^t. 1 1 1 1 1 1 1 I 1
+
X
u a •
o
~ " -H
100 200 300 400 500 600
Window (frames) 700 800 900 1000
Figure 5.16 Comparison of model fairness performance against experimental U D P
trace data. Top Figure is Jain's index, bottom Figure is Kullback-Leibler index. Both
indices illustrate the ability of the Message Retraining model to provide an improved
match with empirical data with respect to the Power, Delay and Hybrid models
Modelling Packet Capture Behaviour 121
2500
2000
m 1500
1000
500
/
/
/
/
/
/
/ /
Connection A Connection B
6
Time (sec) 10 12
Figure 5.17 T C P Experimental Data Trace obtained with Cisco chipset. Connection B commences with an SNR of 20dB, 1 second later Connection A commences with an SNR of 25dB
the simulation trace. For fairness windows less than 400 frames, the Message
Retraining matches trace data within 4 % on Jain's index, and within 1% on the
Kullback-Leibler index. The Power, Delay, and Hybrid models over estimate
the fairness on the Kullback Leibler index by approximately 6 5 % at 200 frames,
and approximately 4 8 % at 100 frames. Using Jain's index, the Power, Delay,
and Hybrid models overestimate fairness by approximately 2 8 % and 2 5 % at 200
and 100 frame windows respectively.
As with the UDP experiment, with respect to the trace data the Message re
training model proves a minimum improvement of 2 1 % on Jain's index and 4 7 %
on the Kullback-Leibler index over the Power, Delay, and Hybrid models for a
fairness horizon of less than 400 frames.
Modelling Packet Capture Behaviour 122
0.95
0.9 -
0.85
£ 0.8 -
« 0.75 h CD
CD
3 0.7
0.65
0.6
0.55
1 1
" A**
/ y*' y s
// r a/ p •r / y
/ /y •'/ fy
7 * 7 / W // ?// - Ml
T -•**"
" >F ."a"'-
5 r + ,7 •*a •••'*
(Srl-I-F
1
.a'
y
.
^ - U l
I
1 1
^ — i n w r f ^ " " * *"'"'
--»*"* ^ «.-
. - • " * '
i '
1
• • - & —
..+ "'
„,
* • » '
1
i r i "
D -Q "
«- -»
..+ ""
,--°" --•°'~
„--'
Message Retraining Hybrid Power Delay
No Capture Trace Data
i i i
+
X
* D
•
0
...--
y
-
-
_
0 100 200 300 400 500 600 700 800 900 1000 Window (frames)
0.8
0.7
0.6
0.5
S 0.4 -CO
0.3
0.2
0.1
Message Retraining + Hybrid » Power « Delay °
No Capture • Trace Data °
0 100 200 300 400 500 600 700 800 900 100Q Window (frames)
Figure 5.18 Comparison of model fairness performance against experimental TCP
trace data. Top Figure is Jain's index, bottom Figure is Kullback-Leibler index. Both
indices illustrate the ability of the Message Retraining model to provide an improved
match with empirical data with respect to other models for a fairness horizon of less
than 400 frames
Modelling Packet Capture Behaviour 123
5.5.4 Discussion
The experiments described in Section 5.5.3 illustrate that the Delay, Power and
Hybrid capture models provide a significant overestimate of the fairness proper
ties observed in the experimental trace data. However, the Message Retraining
model provides an improved indication of the fairness properties present in the
experimental data over a horizon of less than 400 frames.
Matching the fairness characteristics of an empirical trace represents a chal
lenging task for each capture model. The introduction of a competing hidden
connection has a significant impact on the fairness properties of the empirical
data, which should be reflected in the simulation traces also. As illustrated
qualitatively in Sections 5.4.3 and 5.4.4, this was not the case. W e would there
fore expect the Delay, Power and Hybrid models to significantly overestimate
the fairness achieved for each connection. The results in the previous section
confirm this, with the Delay, Power and Hybrid models all significantly overesti
mating the fairness present in the experimental data. The Message Retraining
model is shown to provide an improved indication of the fairness properties
present in the empirical trace data.
The two fairness indices employed in this experiment exhibit similar behaviour.
However, due to the logarithmic nature of the Kullback-Leibler index, there
is a tendency for this index to be more stable than Jain's index across the
range of windows investigated. Conversely, Jain's index is more variable over
shorter timescales. Both indices demonstrate agreement between the Message
Retraining model and the empirical data for a fairness horizon less than 400
frames. The results presented in this chapter will assist with the selection of a
fairness index, and appropriate fairness horizon in the following chapter.
Modelling Packet Capture Behaviour 124
5.6 Conclusions
Simulation plays an important role in performance evaluation of wireless MAC
protocols. Therefore, accurate simulation models are vital if simulation tech
niques are to provide a reliable indication of the performance of a protocol in
a specific scenario. Previous m o d e m capture models have not been designed
with a specific receiver in mind, and with the development of the IEEE 802.11
M A C / P H Y protocol, this situation needed to be addressed. In particular, rela
tive fairness has not traditionally been considered a significant aspect of capture
model performance. However, assessment of the features a capture model should
exhibit in order to accurately model an IEEE 802.11 receiver indicates that fair
ness should be considered a significant factor. Our results in Chapters 3 and
4 indicate that in cases where fairness is an important component of network
performance, a more detailed capture model is required to reflect the impact
varying signal strength characteristics have on the fairness characteristics ob
served between competing traffic streams.
Building on the empirical and analytic evidence presented in Chapters 3 and 4,
this chapter has proposed a new m o d e m capture model, termed Message Re
training. A detailed qualitative investigation of the performance of a number of
common m o d e m capture models presented in literature, in terms of their abil
ity to accurately reflect fairness properties of empirical trace data, illustrates
that the Message Retraining model is able to model the dynamic fairness prop
erties of the IEEE 802.11 M A C / P H Y protocol under varying signal strength
conditions more accurately than either the Delay, Power, or Hybrid capture
models.
Quantitative comparison between experimental trace data and simulation traces
for each capture model using both Jain's Fairness index and the Kullback-
Leibler fairness index, illustrates than the Delay, Power, and Hybrid capture
models provide an overly optimistic estimate of the fairness afforded to the
contending hidden connections. The Hybrid, Power, and Delay models were
Modelling Packet Capture Behaviour 125
shown to overestimate fairness indices by as much as 3 0 % on Jain's index and
6 5 % on the Kullback-Leibler index in certain cases. The Message Retraining
model is shown to match the experimental data within 4 % on Jain's index, and
1 % on the Kullback-Leibler index for a fairness horizon of less than 400 frames.
In terms of the ability to exhibit fairness characteristics, the Message Retraining
model is shown to provide an improved technique for IEEE 802.11 M A C / P H Y
interface simulation.
Understanding the fairness horizon associated with a MAC protocol is impor
tant in achieving good performance for real time multimedia traffic flows, and
smoothing the flow of T C P acknowledgements. The Message Retraining model
can be employed in situations where varying signal strength is expected to im
pact on system performance. This has specific relevance in cases where nodes
in a given topology are unable to sense carrier from near neighbours. The
Message Retraining model will provide a solid basis on which to develop and
test mechanisms to prevent the unfair behaviour observed in empirical traces.
The Message Retraining model may also have application in the development
of quality of service mechanisms for the IEEE 802.11 wireless M A C protocols.
Chapter 6
Prevention Of Signal Strength Dependent Unfairness
6.1 Introduction
In previous chapters we have investigated the nature of packet capture behaviour
observed with IEEE 802.11 radio modems, and the impact this has on the
ability of the M A C to provide adequate performance for higher layer protocols.
In particular, the results in Chapter 3 have illustrated the distinct unfairness
suffered by weaker connections in hidden terminal topologies. In this chapter
we introduce techniques to restore the fairness characteristics of the network to
a state equivalent to a scenario without hidden nodes. The desired outcome is
a state in which connections are able to gain fair access to the radio channel
independent of relative signal strength. Therefore, in this chapter, we make use
of the analysis and modelling presented in Chapters 4 and 5 to develop and
analyse techniques to correct the unfairness present in near-far signal strength
conditions when hidden terminals are present in an 802.11 network.
Several techniques have been presented in the literature which attempt to cor
rect for protocol based unfair behaviour in disconnected topologies where hid
den nodes are present (Nandagopal et al., 2000; Bensaou et al., 2000; Ozugur
et al., 1998; Ozugur et al., 1999). These are outlined in Section 6.2. None
126
Prevention Of Signal Strength Dependent Unfairness 127
of these schemes take the relative signal power dependent nature of the unfair
behaviour identified in Chapter 3 into account. Therefore, in Section 6.3, we
present a mechanism which uses the average relative signal strength to deter
mine a probability variable. This is then employed in one of three techniques
outlined in Section 6.4 to prevent unfair behaviour as a result of relative signal
strength. In Section 6.5 we undertake an investigation of the performance of
each scheme in a number of distinct topologies and scenarios, with conclusions
and recommendations presented in Section 6.6.
While the primary focus of this thesis has been the MAC and PHY protocols
defined in the IEEE 802.11 standard, the mechanisms presented in this chap
ter are more general in nature, and are equally applicable within any general
contention based M A C protocol.
6.2 Analysis of Topology Dependent Unfairness Prevention Algorithms
Several algorithms have been presented in recent literature which aim to pro
vide fair access to the radio channel in cases where specific topologies result
in poor performance of the C S M A / C A mechanism (Nandagopal et al., 2000;
Bensaou et al., 2000; Ozugur et al., 1998; Ozugur et al., 1999). Each of these
schemes or protocol enhancements aim to meet a different fairness objective
through control of either the channel access persistence or backoff window. The
problem identified in this thesis requires consideration of an additional factor -
the relative signal strength between competing hidden connections.
The backoff window adjustment scheme presented in (Bensaou et al., 2000)
requires each node to estimate the amount of traffic other nodes in the network
are transmitting. It then uses a pre-defined fair share to control the transmission
rate of each individual node. Detailed operation of this scheme is outlined in
Section 2.5.3. T w o significant problems exist with this approach. The first is
that each node is required to estimate the quantity of traffic due to surrounding
Prevention Of Signal Strength Dependent Unfairness 128
nodes. As we have illustrated, a frame transmitted over a weaker connection
will incur a significantly higher loss probability if a stronger signal collides at the
receiver. In this scenario, a node will not be able to gain an accurate measure of
the traffic being carried through the network, leading to an inaccurate estimate
of the network fairness metric. Secondly, each node is pre-programmed with
an identified 'fair share' of network capacity. A share of 0.5 indicates that this
node should obtain 5 0 % of the available bandwidth. The manner in which the
node adjusts the contention window it tied to this fair share parameter. Setting
such a parameter is a complex problem in a dynamic network, particularly a
multihop topology employing a M A N E T routing protocol in which nodes may
be acting as both end stations and routers.
The generalised persistence control approach presented in (Nandagopal et al.,
2000) is derived through a graph theoretic technique. The result is a rate control
mechanism applied by each node in the network. The operation of this scheme
is outlined in Section 2.5.3. This scheme requires a detailed indication of the
topology with which to derive the eventual rate control algorithm. Further,
in a dynamic network topology, convergence of node transmission rates will be
an issue. One advantage of the approach is the ability to define any fairness
model in terms of a utility function which then determines the appropriate
rate control algorithm. However, this algorithm is based on the assumption of
perfect knowledge of the contention state in the network. Contention is used as
implicit feedback to control the transmission rate of each node. Disparity exists
in a case when two hidden terminals compete at different signal levels. The
stronger node observes no significant contention or loss, while the weaker node
observes significant contention, loss and delay. In a manner similar to T C P
congestion control, the rate control algorithm will reduce the transmission rate
for the weaker node in response to the observed congestion, while the stronger
node will increase the transmission rate as no congestion is observed. This will
lead to an imbalance in the fairness properties observed in the network with
both nodes assuming fairness is being achieved.
Prevention Of Signal Strength Dependent Unfairness 129
Similar problems also exist with the contention approach presented in (Ozugur
et al., 1998; Ozugur et al., 1999) where each node calculates a link access prob
ability based on the surrounding topology. This probability is used to control
the rate at which a node is able to transmit. In cases where two hidden nodes
have individually defined access probabilities, there is a potential for significant
disparity to occur. The weaker of the hidden nodes will be restricted in the
number of packets it is able to transmit by the given link access probability.
As packets from the weaker nodes suffer a greater loss probability, the node is
forced to retransmit lost packets yet receives no additional link access to account
for the lost frames. The stronger node suffers no restriction and the majority
of packets are successfully received by the common node.
Each of these schemes aim to solve topology and protocol dependent unfairness
problems. However, as the examples discussed above illustrate, additional con
sideration of the relative signal strength amoung nodes will be required. In the
following section we introduce a technique to determine a probability variable
proportional to the relative signal strength of a given connection. Algorithms to
correct for signal strength dependent unfairness are introduced in Section 6.4.
6.3 Average Signal Strength Based Probability
The aim of this chapter is to develop mechanisms to control unfair behaviour
resulting from signal power differences in the presence of hidden terminals. Sec
tion 6.4 below outlines three proposed mechanisms. In this section we outline
a mechanism which maintains a probability variable based on the average re
ceived signal strength for each neighbour. The probability determined with this
mechanism is then employed by each of the fairness control techniques outlined
in the following section.
The proposed mechanism to determine the relevant probability variable is based
on an average signal strength metric for each node. As each connection will
Prevention Of Signal Strength Dependent Unfairness 130
have a dynamic signal strength due to fading, shadowing, and mobility, signal
strength averages should be filtered through a standard 1st order delay filter of
the form:
Pi = a- Pi(n) + (1 - a) • P^n - 1) (6.1)
where a, often termed the 'forgetting factor', is typically within the range 0.5
to 0.9, and n is the sample number index. Each node records a historic average
signal strength for each neighbour using Equation (6.1). The average signal
power relative to the all other neighbours is determined for each logical con
nection. Using the algorithm outlined in Equation (6.2) below, a probability
variable for each of the N total neighbours is maintained:
5i = max(Pi(m) - Pj(m), me[l,N], ie V)dB (6.2)
where V is the set of neighbours who have hidden nodes within range of the
node performing this calculation, m is an index variable, and the max function
selects the largest value in the set. The relative power is then used to calculate
a probability value, pi for each neighbour i € V according to:
, 1 , Si<y Pi={ (6.3)
U(Si), otherwise
where 7 is the observed unfairness threshold, and may be tuned to a specific
P H Y . U(8i) represents a utility function, continuous over the range [0,1]. In
Chapter 4 we illustrated that a relative difference of greater than 2dB will
result in the corruption of the weaker signal. Therefore, a threshold of 7 = ZdB
is selected. A utility function is employed which reduces the probability in
proportion to 7:
U(6i) = 1 - ^ (6.4) 7
where # is the control parameter. This algorithm is maintained by each re
ceiver in the network for all identified neighbours. Neighbours are identified by
the Source Address (SA) field of observed RTS, D A T A and other M M P D U ' s ,
Prevention Of Signal Strength Dependent Unfairness 131
and maintained in a cache of MAC addresses. The probability variable for the
ith neighbour, pt, is updated on reception of each valid frame. Once a neigh
bour is identified as having a hidden neighbour (Section 6.3.1), the current
probability for that neighbour is employed with one of the schemes outlined in
Sections 6.4.1, 6.4.2, and 6.4.3.
To prevent excessive restriction of a stronger node in cases where the traffic from
a weaker node is bursty, or a weaker node has little impending traffic, a lifetime
parameter Tj is associated with the average signal strength measurements, Pi.
Reception of a new frame from the ith node resets the lifetime of the average
signal strength variable Pj. O n expiration of the lifetime for Pi, the value is
set to zero, and the neighbour is removed from the set of identified neighbours,
V. Once the neighbour has been removed from V the relative probabilities will
readjust for all other neighbours within V accordingly.
Further, the control parameter j3 must be chosen to control the aggressiveness
of the correction mechanism. A higher /3 value will result in a more aggressive
control scheme. The selection of an appropriate /? for each scheme in various
scenarios is discussed in Section 6.5.
6.3.1 Identification of Hidden Nodes
Identification of hidden nodes is a key issue for the performance of any tech
nique designed to control unfair behaviour as a result of hidden terminals in
a practical implementation. The mechanism to identify potential hidden node
pairs should operate within the 802.11 framework. This can then be employed
to instigate the fairness control algorithms under development. Within the
context of a C S M A / C A M A C , a potential mechanism to identify hidden nodes
exists through observation of message exchange semantics within either the
R T S / C T S / D A T A / A C K or D A T A / A C K handshake. The reader is referred to
Appendix A for details of the proposed mechanism based on message exchange
semantics.
Prevention Of Signal Strength Dependent Unfairness 132
The aim of this chapter is to investigate the performance of the fairness con
trol schemes outlined in Section 6.4, rather than the performance of a hidden
node detection algorithm. Therefore the initial performance investigation is
Section 6.5 will assume that all hidden nodes are identified to the common
node. This assumption allows investigation of the fairness control mechanism
in isolation, free from potential interactions with the hidden node detection
mechanism.
6.4 Algorithms to Control Signal Strength Dependent Unfairness in Hidden Node Scenarios
The analysis in Chapter 4 illustrates that a relative signal power between two
interfering signals of greater than 2dB is sufficient to result in the unfairness as
measured experimentally. In this chapter, we focus on mechanisms to correct
unfairness due to relative signal strength variation in general hidden terminal
scenarios.
The effect we are attempting to correct for exhibits threshold behaviour. Once
the relative signal strength is greater than a given value, unfairness becomes evi
dent. Once the average relative signal strength is below the threshold identified,
we have observed in Chapter 3 that reasonable fairness results. Accordingly we
can restrict ourselves to a mechanism aiming to correct signal strength imbal
ances above the threshold to a point where the fairness performance matches
that observed with a relative signal strength below the threshold on each con
nection.
A significant issue is the positioning of a scheme designed to prevent unfair be
haviour due to relative signal power within the layered protocol structure. The
capture effect responsible for the unfair behaviour is an intrinsic feature of both
the P H Y and the receiver. To this extent, it m a y seem appropriate to improve
the P H Y to increase the robustness in the presence of multiple access interfer-
Prevention Of Signal Strength Dependent Unfairness 133
ence. However, given the number of P H Y protocols currently defined within
the IEEE 802.11 framework, a significant number of different solutions would
be required. A more generic solution would provide significant advantages, al
lowing the introduction of new P H Y protocols requiring tuning of the generic
solution rather than the development of a new control mechanism. Therefore,
the mechanisms proposed in this chapter are designed to reside between the
P H Y and M A C layer. This allows the mechanism to use information obtained
from the P H Y to correct the contention state observed by the M A C .
Closed loop power control also represents a potential mechanism to prevent
an unfair state due to relative received signal power differences arising in the
network. In this case, very fine control within a few dB would be required. In a
dynamic, general topology M A N E T , this is a non-trivial problem. Controlling
the transmission power of a competing hidden node will potentially result in
significant 'knock-on' effects through the network. For example, increasing or
decreasing transmission power will have a significant impact on the routing
protocol and the reliability of links throughout the network.
Another important issue requiring investigation is whether the scheme should be
receiver or sender based. A receiver based scheme will require the transmission
of information to each sender at appropriate intervals, however, if this can be
incorporated within existing M A C frame exchanges (in a field within an A C K
or another M M P D U frame for example) the impact on channel efficiency will
be negligible. Also, mechanisms such as the probabilistic discard or enhanced
C T S suppression (outlined in Section 6.4.3) are only possible in a receiver based
context. A sender based scheme in which an S T A was able to identify when
it is causing problems for another S T A in the network follows the paradigm
presented in (Bensaou et al., 2000; Nandagopal et al., 2000). However, we are
attempting to solve a problem that exists at a receiver where a transmitter
has no prior knowledge of the signal strength the receiver observes, nor any
knowledge of the relative strength of other hidden transmissions a receiver may
be able to observe. This requires information from the receiver before any
Prevention Of Signal Strength Dependent Unfairness 134
action can be taken to correct an unfair state. This separation implies that a
transmitter based scheme will be unable to take signal strength information into
account. In this context, a transmitter based scheme which includes received
signal strength as a parameter is not possible without significant dissemination
of information throughout the network. Therefore, we will focus on schemes in
which the receiver undertakes the majority of the required processing and is able
to minimise additional transmission, or where possible, incorporate information
transfer within the current 802.11 M A C frame exchange sequences and formats.
The fairness goal we aim to achieve through a control scheme of this type is
not one based on differentiation between nodes or traffic classes. As we are
correcting for an inherent bias in the system towards stronger connections, we
can only aim to remove this bias, and return the network to a state where each
connection achieves equal opportunity to access the channel over a specified
time. This also corresponds to maximising the fairness index. This implicit
fairness goal is the only goal we can attempt to achieve in a best effort network
where no differentiation between services is provided. In cases where service
differentiation were applied, a mechanism to correct for the signal strength
dependent unfair behaviour would provide a fair basis on which to apply service
differentiation.
Finally, complexity is a significant issue in terms of future adoption within the
802.11 standard. A preferred scheme will not introduce significant additional
complexity in terms of random number generation, or signal strength sampling
for example. However, while reduction of undue complexity should be a con
sideration, complexity itself should not be an overriding factor in determining
the suitability of a particular scheme.
From the above discussion, we are able to identify features a suitable fairness
control scheme should exhibit:
• the scheme should be as simple as possible
• the scheme should be able to be integrated into the existing standard
Prevention Of Signal Strength Dependent Unfairness 135
with minimal effort, or alternatively operate within the constraints of the
(future) standard in terms of available messaging syntax
• the scheme should take the results presented in Chapter 4 into account
when considering signal thresholds
• the scheme should be dynamic, adjusting to changes in signal strength,
and the introduction and departure of STA's from the network
In the light of the above feature requirements, we propose three potential options
to correct for signal strength dependent unfairness:
p-Persistence at Backoff Countdown: Each STA retains the nor
mal backoff process. O n detection of a clear channel, the node will defer
again with probability 1 — p determined in accordance with the relative
signal strength at a common receiver. This method is outlined in Sec
tion 6.4.1.
Probabilistic Discard: A receiver probabilistically discards data frames
from stronger STA's, forcing the offender to backoff and retransmit. This
method is outlined in Section 6.4.2.
Enhanced CTS Suppression: A receiver is able to direct a C T S mes
sage to a given STA, forcing suppression for the indicated interval. This
method is outlined in Section 6.4.3.
For each of the proposed mechanisms, distinction must be made between traffic
destined for an S T A that has no hidden terminals, and traffic that will compete
over a hidden connection for a common receiver. In random network topologies,
STA's may have traffic destined for multiple local STA's which raises the issue
of whether each of the above schemes should be applied on a per STA basis
or on a per traffic stream basis. The correction techniques listed below adjust
different parameters in an offending STA, which may or may not be capable
of distinguishing traffic streams. This will be discussed in greater detail in the
following sections.
Prevention Of Signal Strength Dependent Unfairness 136
6.4.1 Probabilistic Access at Backoff Countdown
This technique builds on the basic mechanism presented by Ozugur (Ozugur
et al., 1998; Ozugur et al., 1999). The idea is to apply a channel access proba
bility (or persistence) whilst retaining the standard backoff mechanism. W h e n
an S T A senses an idle channel, the transmission proceeds with probability p, or
defers with probability 1 — p. The normal backoff mechanism is retained.
Once a pair of hidden STA's have been identified through mechanisms outlined
in Appendix A, the algorithm outlined in Section 6.3 is employed to determine a
probability value in proportion to the relative received signal power between the
competing hidden STA's. O n reception of a valid D A T A frame, the common
STA includes the probability value in an 'enhanced' A C K reply (or possibly
another M M P D U ) . The station receiving the A C K then records the persistence
value and interprets this as a channel access probability when attempting to
transmit to the host sending the A C K .
The introduction of an additional field into an ACK frame will result in a
lack of compatibility with original versions of the 802.11 standard, and as such
represents a potential drawback for this approach. However, STA's supporting
the probabilistic access method will consider the A C K a valid frame and given
that the A C K is a directed frame (i.e. not a broadcast or multicast frame), such
an S T A will receive this frame. This mechanism relies on the inclusion of a single
additional field in an A C K frame which represents a very small overhead. It
may also become possible to distribute the persistence variable within another
M M P D U , as the current 802.11e W G (TGe, 2001) will introduce several new
M M P D U frames into the standard.
This scheme can be applied on a per-flow or per-node basis, as the probability
variable is only employed when attempting to transmit to the common STA.
Traffic not destined for this STA need not use this access control probability.
The detailed performance of this approach is investigated in Section 6.5 below.
Prevention Of Signal Strength Dependent Unfairness 137
6.4.2 Probabilistic Discard
In the probabilistic discard approach, the receiving STA interprets the prob
ability calculation of (6.2) as a drop probability for incoming D A T A or RTS
frames from identified stronger hidden STA's.
The common STA having identified a stronger hidden STA, will drop received
D A T A or R T S frames with probability 1 — p and deliberately fail to respond
with an A C K or C T S frame as would normally be required. This forces the
transmitting S T A to backoff more often than would otherwise be the case. This
approach makes use of the native backoff and timeout mechanisms within the
M A C protocol, and does not require information to be sent to the offending
STA's. Feedback is in the form of an apparently higher loss probability which
the offending S T A will (correctly) assume is due to congestion. This allows
the weaker S T A an opportunity to transmit. In the uncorrected situation, the
stronger S T A receives an incorrect indication of the actual congestion state of
the network, as it's transmissions are preferentially received. The probabilistic
dropping of frames from a stronger hidden STA corrects this by introducing
additional congestion specific to this connection.
This is a simple scheme which places very little additional requirement on each
STA, and is controlled by a single parameter. Information is implicitly dis
seminated through the network in the form of 'forced congestion' an offending
STA would otherwise not observe. No messaging is required, and each STA can
choose when to apply the scheme individually. This scheme could be applied to
all D A T A and R T S frames from an offending STA. The detailed performance
of this scheme is investigated in Section 6.5.
6.4.3 Enhanced CTS Suppression
This approach makes use of an enhancement to the interpretation of RTS and
C T S control messages proposed in (Sherman, 2001). W e propose the applica
tion of the Enhanced R T S (ERTS) and Enhanced C T S (ECTS) messages in
Prevention Of Signal Strength Dependent Unfairness 138
suppressing a stronger STA through the virtual carrier sense mechanism out
lined in Section 2.3.1.1. The common STA uses the probability variable to
determine when to force a suppression period upon a stronger hidden STA.
The enhanced RTS/CTS frame rules currently before the IEEE 802.11 WG
(Sherman, 2001) allow an STA to intentionally suppress an STA, or group of
STA's, identified by a special range of group or multicast addresses. In the
current context, an STA will maintain a list of group or multicast addresses,
and use these to identify hidden STA's. O n receipt of an RTS or E R T S from
an offending hidden STA, the common STA has the ability to set the duration
field in an E C T S reply to a value which provides a weaker STA an opportunity
to attempt a transmission. O n receipt of the E C T S (addressed to the offending
STA via a specified multicast address) the STA will then update the N A V to
the value returned via the ECTS. The stronger STA requires suppression for a
period of at least several hundred microseconds, which allows a competing hid
den STA to perform a successful RTS/CTS exchange. This will extend the N A V
of the stronger STA through the normal RTS/CTS N A V update mechanism,
and allow the weaker STA an opportunity to transmit. This mechanism is only
applicable on a per-STA basis, as it makes use of the NAV, which will return
a busy indication regardless of the destination address of any impending data
frame. Figure 6.1 illustrates this mechanism in a scenario where the stronger
STA group (Nodes C and D) is suppressed for a period allowing the weaker
STA (Node A) to transmit several data frames.
This alternative is attractive from the point of view that the technique may
easily be implemented using the ERTS/ECTS mechanism within an 802.lie
network. This is inter-operable with legacy stations, which are not affected
by the E C T S mechanism. While there are still many issues to be addressed
within the IEEE 802.11 W G , in particular the allocation of group and multi
cast addresses, given that the E R T S / E C T S mechanism allows the potential for
a directed R T S or C T S message, there is a potential application for this mech
anism with the probability determination algorithm as a method of controlling
Prevention Of Signal Strength Dependent Unfairness 139
Node A Sends RTS, Receives ECTS Node A Commences DATA Tx
0 Node A RTS ECTS
0 NodeB RTS ECTS
DATA DATA ACK
Multiple Frame Exchanges Possible While Nodes C and D Suppressed
DATA DATA ACK
Node B receives RTS, Responds with ECTS
0 0 ECTS
Node D Node C Nodes C & D identified through Group Address as hidden from A
Node C and Node D Detect Group Addressed ECTS Defer For Duration specified in ECTS Allows Node A additional access to channel
Figure 6.1 Diagramatic representation of ECTS Suppression Scheme. Nodes C and
D are identified through group address as hidden from Node A. Node B determines
when an Enhanced CTS reply is required to suppress Nodes C and D, allowing Node
A fair channel access. The duration set within the ECTS can be tuned to meet the
specific fairness objective.
Prevention Of Signal Strength Dependent Unfairness 140
signal strength dependent unfairness.
6.5 Performance Investigation and Comparison
Sections 6.4.1, 6.4.2, and 6.4.3 have described three proposed mechanisms to
correct for relative signal strength dependent unfairness when hidden terminals
are present in a general topology network. In this section, we undertake a
detailed performance analysis and comparison of each scheme. Our aim is to
determine the optimal performance range for each scheme, and identify scenarios
in which each scheme may be more appropriate. This will be achieved through
a number of performance and comparison criteria.
6.5.1 Simulation Methodology
The simulation techniques employed are identical to those of the previous chap
ter, though we present a brief review here. Each technique has been imple
mented within the ns-2 simulation package (version 2.1b3) (UCB/LBNL/VINT,
1999), incorporating the Message Retraining capture model developed in the
previous chapter, ns contains a detailed 802.11 P H Y / M A C layer model, as well
as providing excellent implementations of higher layer protocols such TCP/IP,
U D P , F T P etc. The channel model employed is an Additive White Gaussian
Noise ( A W G N ) Two-Ray Ground model. Capture decisions are made within
each m o d e m based on the received signal power, channel noise power, and cap
ture threshold. Each node receives a copy of the transmitted packet and based
on the received power, determines whether the signal is observable. If the signal
is received with sufficient power, the message retraining capture model deter
mines the appropriate course of action prior to passing the completed frame
up to the M A C protocol. In order to examine the performance of the fairness
control mechanisms in isolation, it is also assumed that each hidden node has
been identified to the common node.
Prevention Of Signal Strength Dependent Unfairness 141
6.5.2 Comparison Criteria
In the previous chapter we employed two fairness metrics to compare the per
formance of the capture models under investigation. Chapter 5 illustrated how
both the Kullback-Leibler and Jain's index show good agreement between ex
perimental trace data and simulation using the Message Retraining model for
fairness horizon's less than 400 frames. In this chapter, we employ Jain's fair
ness index over a fairness horizon of 100 frames as the main aim of each scheme
is to improve the relative fairness over short timescales.
The basis on which we compare the performance of the three mechanisms will
include the following:
• Qualitative improvement in trace characteristics, (plots a and c)
• Qualitative improvement in channel access times (plots b and d)
• Per connection and aggregate normalised throughput (plots f and h)
• Instantaneous and average fairness index throughout the experiment (plots
e and g)
Results are presented as a combination of 8 individual graphs presented for
each scenario. Figures 6.3, 6.4, 6.5, 6.7, 6.8, and 6.9 combine the 8 individual
graphs outlined below. Each graph highlights a specific aspect of the behaviour
of the fairness control mechanism. The 8 graphs presented, and the relevant
information each presents is as follows:
Plot a Presents Received Data vs time with no fairness control (/3 = 0).
This illustrates the relative fairness between contending connections at
the network layer. The relative slope of the graph indicates the differing
channel access obtained by the competing nodes in the network.
Plot b Presents Relative channel access times for each node without fair
ness control (/3 = 0). Again, this illustrates the relative access between
Prevention Of Signal Strength Dependent Unfairness 142
contending nodes.
Plot c Presents Received Data vs time with the optimal ft value, deter
mined in plots (g) and (h). This illustrates improvement in relative per
formance observed at the network layer with each mechanism operating
at the optimal point.
Plot d Presents Relative channel access times for each node with pre
ferred /3 value, illustrating the effect the fairness control algorithm has on
the ability of each connection to access the channel.
Plot e Presents Jain's fairness index over the length of the preferred
f3 trace, illustrating how the fairness properties varies throughout the
experiment. The average value over the entire experiment is included. A
value over 0.95 is considered to represent excellent fair behaviour.
Plot f Presents Normalised throughput for each connection over the length
of the preferred fi trace, averaged over 2 second intervals. This indi
cates the ability of each fairness control algorithm to prevent a stronger
connection from gaining appreciably greater channel access than weaker
connections.
Plot g Presents Jain's fairness index over the range of /3 values exam
ined. This illustrates the preferred /3 value as the peak in this curve.
Again, the fairness horizon is 100 frames.
Plot h Presents Normalised average throughput for each connection over
the range of /3 values examined. A preferred j3 value can be observed with
this trace as the point at which each connection achieves equal average
throughput. This corresponds to a long term fair allocation of bandwidth.
The combination of these criteria allow both a qualitative and quantitative com
parison between the fairness control mechanisms. As outlined in previously, the
reasons for implementing a particular scheme are based on additional factors
including implementation and standards inter-operability issues. The primary
Prevention Of Signal Strength Dependent Unfairness 143
aims of this investigation are to establish the performance of each scheme on
both a qualitative and quantitative basis, as well as a range of appropriate /3
values for each mechanism. The preferred /3 values for each mechanism are de
termined as the range of j3 combining the highest fairness measured with Jain's
index, with the range over which the average throughput for each connection
does not illustrate the distinct unfairness evident without fairness control. This
may have future application in an adaptive mechanism to control ft.
Using the criteria presented in this section, each of the schemes will be com
pared, and the range of /3 values for optimum performance identified.
6.5.3 Simple Case - Static Hidden Nodes
Initially, we investigate the performance of each mechanism in a simple case
where two hidden nodes compete for a common receiver using both U D P and
T C P traffic streams. Results for the three node topology using U D P traffic
streams, illustrated in Figure 6.2 are included in this chapter, while the results
obtained with T C P are included in Appendix B . U D P is used exclusively here
as it removes potential interactions between M A C and higher layer protocol
timers, illustrated in Chapters 2 and 3 to have an additional effect on network
layer fairness. This allows examination of the ability of each mechanism to
remove bias due to relative signal strength.
In this static scenario, distance from the central node is fixed and equal for
both nodes. The stronger Connection A has a transmit power adjusted to
achieve a received signal power of -87 d B m , corresponding to an average S N R
of approximately 27dB. The transmit power of the weaker Connection B is
adjusted to achieve a received average signal power of-90dBm, corresponding to
an average S N R of approximately 24dB. Transmit power and distance from the
central node remain constant throughout the experiment. Dynamic scenarios
are considered in later sections. This scenario, whilst potentially representing a
limited physical scenario, is investigated primarily as a means of verifying the
ability of each mechanism to effectively control unfair behaviour.
Prevention Of Signal Strength Dependent Unfairness 144
Connection A y^ >v Connection B
,0 (r®0 j © >-- \ y •*
'^•Hidden Nodes •' '
Figure 6.2 Three Node Hidden Terminal Topology
The results for each algorithm with the 3 node topology are presented in Fig
ures 6.3 through 6.5.
Prevention Of Signal Strength Dependent Unfairness 145
p-Persistence on Backoff C o u n t d o w n
Figures 6.3(a) and (b) clearly illustrate the advantage Connection A obtains over
Connection B without a fairness control mechanism. Employing a value of f3 =
0.25, Figures 6.3(c) and (d) illustrate the ability of the p-Persistence mechanism
to provide increased access opportunities for Connection B at the expense of
Connection A. Figures 6.3(e) and (f) illustrate the variation of Jain's index and
the per connection throughput throughout the experiment respectively. In this
case, the fairness index is increased from 0.52 with 0 = 0 to 0.83 with {3 = 0.25.
Figures 6.3(g) and (h) indicate that the preferred (3 value lies between 0.25 and
0.35, corresponding to the range over which Jain's fairness index is maximised
and the throughput on each connection is equal.
Probabilistic Discard
Figures 6.4(a) and (b) again illustrate the advantage Connection A obtains
over Connection B without a control mechanism. Figures 6.4(c) and (d) il
lustrate the ability of this mechanism to increase channel access opportunities
for Connection B at the expense of Connection A. However, compared to the
p-Persistence scheme, this mechanism appears to enact a much more course
grained level of control in this scenario, evidenced through the variability in the
progress of the data traces in Figure 6.4(c). Figures 6.4(g) and (h) indicate that
the preferred 0 value lies in the region of 0.40, with j3 = 0.40 the point where
throughput is equal on each connection. Using this value, Figures 6.4(e) and (f)
again illustrate the variation of Jain's index and the per connection throughput
throughout the experiment respectively. The fairness index is increased from
0.52 with 0 = 0 to 0.72 with fi = 0.40.
Enhanced CTS Suppression
Using the ECTS scheme, Figures 6.5(c) and (d) clearly illustrate how this mech
anism provides very fair channel access for both nodes employing a value of
ft = 0.75. Figures 6.5(e) and (f) illustrate the variation of Jain's index and the
per connection throughput throughout the experiment respectively, both ex-
Prevention Of Signal Strength Dependent Unfairness 146
(a) Data Trace - Evolution p = 0 (b) Data Trace - Access Times p = 0
2000
_ 1800
& 1600
m 1400
« 1200
g 1000
S 800
| 600
o 400
200
0
1000
Connection A Connection B
2 4 6 8 10 12 14 16 18 20
Time (sec)
(c) Data Trace - Evolution p = 0.25
3 CO Q TJ (1)
>
8 at
tr
900 •
800
700
600
500
400
300
200
100 r
Connection A Connection B
2 4 6 8 10 12 14 16 18 20
Time (sec)
(g) Jain's Fairness Index vs p
Connection A Connection B «
0 2 4 6 8 10 12 14 16 18 20
Time (sec)
(d) Data Trace - Access Times p = 0.25
1 •
X o c CO CO 0)
c
s LL
0
1.1 1
0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2
0.1 0
-
•
2 4 6 8 10 12 14 16
Time (sec)
(e) Jain's Fairness Index
'W flfLrlij.t! W* 't^W $&• ' V
'^-~i
Jain's, Index: Average 0.83
18
•
»\
•
20
•
•
Q.
I-•a
<o tn
E o Z
0
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
a E o z
0.6 0.55 0.5 0.45 0.4
0.35 0.3 0.25 0.2 0.15 0.1 0.05
0
Connection A Connection B *
' •
2 4 6 8 10 12 14 16 18 20
Time (sec)
(f) Normalised Throughput p = 0.25
0 2 4 6 8 10 12 14 16 18 20
Time (sec)
(h) Normalised Throughput vs p
Connection A Connection B
0.1 0.2 0.3 0.4 0.5 0.6
Figure 6.3 Three node topology, static scenario for p-Persistence on Backoff Count
down algorithm using U D P - /? = 0.25
Prevention Of Signal Strength Dependent Unfairness 147
CO
s m 3 ra Q •a
s CD
DC
2000
1800
1600
1400
1200
1000
800
600
400
200
(a) Data Trace - Evolution p = 0 (b) Data Trace - Access Times p = 0
1200
<g 1000
m
CO CO Q T3
>
"8 V
800
600 •
400 •
200
CD
E co U-
Connection A Connection B
CD TJ
c CO
"c 'ra ~3 0) O)
s
I
1.1
1
0.9
0.8
0.7
0.6 r
0.5
4 6 8 10 12 14 16 18 20
Time (sec)
(c) Data Trace • Evolution p = 0.40
Connection A Connection B
t/y./
-
/
y /'" /
y
i
1 1 -)
,/ ~~~
6 8 10 12 14 16 18 20
Time (sec)
(e) Jain's Fairness Index
2 4 6 8 10 12 14 16 18 20
Time (sec)
(g) Jain's Fairness Index vs p
100 Frame Window
•
* • + / \
/ \ / \ * \ •'
..+ ••+
'- ....-i- \
0.1 0.2 0.3 0.4 0.5 0.6
0) n
•Z 2
73 O
z
0.6
0.5
0.4
0.3
0.2
0.1
Connection A Connection B
0 2 4 6 8 10 12 14 16 18 20
Time (sec)
(d) Data Trace - Access Times p = 0.40
Connection A Connection B
0 2 4 6 8 10 12 14 16 18 20
Time (sec)
(f) Normalised Throughput p = 0.40
Connection A Connection B
4 6 8 10 12 14 16 18
Time (sec)
(h) Normalised Throughput vs p
0.6
3
n r: C7> 3 O sz r-
CI) O)
ni F o Z
05
0.4
0.3
0.2
0.1
0
Connection A Connection B
0.1 0.6
Figure 6.4 Three node topology, static scenario for Probabilistic Discard algorithm
using U D P - 0 = 0.40
Prevention Of Signal Strength Dependent Unfairness 148
hibiting significantly less variability when compared with the previous schemes.
Figure 6.5(g) indicates a clear peak in the fairness index for 0 between 0.7 and
0.9. However, the aggregate channel throughput in Figure 6.5(h) is significantly
lower at the preferred 0 = 0.75, being reduced from 0.58 with 0 = 0 to an ag
gregate of 0.24. This represents a significant drawback for this technique. The
fairness index however is significantly improved over the previous two schemes,
from 0.52 with 0 = 0 to 0.93 with 0 = 0.75.
6.5.4 Static Scenario Discussion
In all three cases, the fairness control schemes are able to provide improved
throughput for the weaker connections by preventing the stronger connection
from gaining unfair channel access. Examining the performance as a function
of 0 indicates there is a well defined peak in each case, corresponding to the
maximum in Jain's index, as well as the point where throughput is equal on each
connection. In each case, the optimal point with respect to the fairness index
corresponds to a point where aggregate throughput performance is quite low.
This is common across each scheme, with the Enhanced C T S scheme resulting in
the highest aggregate throughput reduction of 56%. The Probabilistic Discard
scheme suffers an aggregate throughput reduction of 6%, while the p-Persistence
of Backoff Countdown scheme results in an aggregate throughput reduction of
only 2%.
The reduced throughput obtained with the Enhanced CTS scheme represents a
significant disadvantage for this approach in comparison with the p-Persistence
and Probabilistic Discard schemes. Aggregate throughput is expected to be
reduced in this manner as a result of increased contention introduced by the
fairness control scheme. In multiple access systems, a single user is able to
achieve greater throughput than the sum of a number of users, due to the
increased contention for the channel as the number of users increases. In the
trials without fairness control, the stronger user is effectively a single user as
the contention information they infer from the network indicates there are no
Prevention Of Signal Strength Dependent Unfairness 149
(a) Data Trace - Evolution p = 0 (b) Data Trace - Access Times p = 0
CO CD
m
8 CO
Q •a
s
CO CD
8 to Q •a
> CD CJ CD
DC
4000
3500
3000
2500
2000
1500
1000
500
0
1000
900
800
700
600
500
400
300
200
100
0
Connection A Connection B
5 10 15 20 25 30
Time (sec)
(c) Data Trace - Evolution p = 0.75
Connection A Connection B
,y
.y
.*£ /-,*>*•'
10 15 20 25 30
Time (sec)
(e) Jain's Fairness Index
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
T3 O
z
T3 O
z
X CD TJ C CO
CD
c S U-
s •o
c CO
c CO —i CD CD
3
1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 (
1.1
1
0.9
0.8
0.7
0.6
0.5
w i t$ n mwv .n/fw W A A /"!<. rr> m
•
Jajn's Index: Average 0.93 i *
) 5 10 15 20 25
Time (sec)
(g) Jain's Fairness Index vs p
100 Frame Window
••-•-*
1 — - — t i • i • i i i i i
\T\i
-
3
* .
•
3
a a o 1-T3 CD W
"co E o z
0
3
a. CD 3 O J=
H T3 CD CO E o z
0.4
0.3
0.2
0.1
0.6
0.5
0.4
0.3
0.2
0.1
Connection A Connection B
5 10 15 20 25
Time (sec)
(d) Data Trace - Access Times p = 0.75
Connection A Connection B
Connection A Connection B «
10 15 20
Time (sec)
(h) Normalised Throughput vs p
Connection A Connection B
30
10 15 20 25 30
Time (sec)
(f) Normalised Throughput p = 0.75
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
P
Figure 6.5 Three node topology, static scenario for enhanced CTS Suppression algorithm using U D P - 0 = 0.75
Prevention Of Signal Strength Dependent Unfairness 150
other users contending for the channel. The weaker node in this case suffers
significant contention. In the case where control is employed, the stronger user
is forced to contend for the channel which has the effect of reducing overall
system throughput. A n obvious tradeoff exists between fairness and system
throughput and must be considered when formulating fairness goals for the
network.
For TCP results in this scenario, the reader is referred to Appendix B. These
results exhibit lower throughput than the U D P results. This is due to the ineffi
ciency of the combined 802.11 M A C employing RTS/CTS with T C P Reno. The
static case investigated here may be representative of certain W L A N scenarios,
though is not a realistic scenario for a more general W L A N or M A N E T . Nodes
are able to move, and the propagation environment may change in a manner
not accounted for in our propagation model. Therefore, in the following section
we continue this investigation in a scenario where nodes move throughout the
experiment. This examines the dynamic behaviour of the algorithms, as well as
their ability to operate is scenarios where both hidden and in-range nodes are
present.
6.5.5 General Dynamic Case - Hidden and In-Range Nodes
In the scenario employed for this series of experiments the nodes move in to
wards the common node 15 seconds into the experiment. Through until 20
seconds, both nodes are no longer hidden from each other. At 20 seconds, both
nodes move back to become hidden again, though this time Connection B is
the stronger of the two. This scenario combines hidden nodes, in-range nodes,
as well as introducing variability into the received average signal strength as a
result of node mobility. In this case, nodes are physically moved at 10 m/sec
rather than having their signal power adjusted.
An example signal trace is shown in Figure 6.6. This illustrates the received
S N R of each connection throughout the experiment. Connection B commences
Prevention Of Signal Strength Dependent Unfairness 151
Figure 6.6 Example received SNR trace during dynamic experiment
as the weaker connection, evidenced by the weaker signal of approx 22dB op
posed to 27dB for connection A between 5 and 15 seconds on the S N R trace.
15 seconds into the experiment, both nodes move towards a point equidistant
from the common node. 20 seconds into the experiment, the nodes move back
outwards to the original (hidden) positions. The distance of each node from
the common node is swapped. Connection B then maintains a stronger average
signal strength of approximately 27dB as opposed to the 22dB average obtained
by Connection A.
In this scenario, using the equal throughput criteria to identify a preferred 0 is
not possible, as the dominating host changes throughout the experiment. This
results in relatively equal throughput for each connection throughout the exper
iment, as both hosts spend an equal time dominating the channel. Accordingly,
per connection throughput as a function of 0 does not vary as widely as in the
static scenario.
p-Persistence on Backoff Countdown
Figures 6.7(a) and (b) illustrate the traces without fairness control. In this
Prevention Of Signal Strength Dependent Unfairness 152
(a) Data Trace - Evolution p = 0 (b) Data Trace - Access Times p = 0
, ^ CD
m
ra CO
Q
>
8 cu DC
2000
1800
1600
1400
1200
1000
800
600 400
200
2000
_ 1800
8 1600
m 1400
« 1200
§ 1000
E 80° I 600 8 400 DC
200 0
X CD T3
1.1
1 (• 0.9 0.8 r 0.7 0.6
8 0.5
CO IX.
0.4 0.3 0.2 0.1 0
1.1
cu
- 0.9
CD
if 0.7
I 0.6 0.5
Connection A • Connection B *
5 10 15 20
Time (sec)
(c) Data Trace - Evolution p = 0.25
Connection A Connection B
10 15 20
Time (sec)
(e) Jain's Fairness Index
i / R ^ . f fc IVf"*"""^! .fMU/liW
5 10 15 20
Time (sec)
(g) Jain's Fairness Index vs p
.
•
•
\
/
/
100 Frame Window •
•
\ \
0.1 0.2 0.3 0.4 0.5 0.6
TJ O
z
0.5
o. 0.4 £: O)
1 0.3
| 0.2 co E o 0.1
0.6
3 n £ rn 3
o sz. •n CD CO a f-o
0 5
U.4
0.3
0.2
0.1
Connection A Connection B
5 10 15 20 25
Time (sec)
(d) Data Trace - Access Times p = 0.25
Connection A Connection B
10 15 20 25
Time (sec)
(f) Normalised Throughput p = 0.25
i 10 15 20
Time (sec)
(h) Normalised Throughput vs p
Connection A Connection B
0.1 0.2 0.3
P 0.4
30
30
0.5 0.6
Figure 6.7 Three node topology, dynamic scenario for p-Persistence on Backoff
Countdown algorithm using U D P 0 = 0.25
Prevention Of Signal Strength Dependent Unfairness 153
scenario, periods where either connection is able to dominate the radio resource
correspond to the period where that connection was the stronger of the two.
Between approximately 15 and 20 seconds, the resource is effectively shared as
both hosts move in towards the common node. Once the nodes are in range of
each other, reliable carrier sense information prevents either host from domi
nating the radio resource.
Employing a value of 0 = 0.25, Figures 6.7(c) and (d) illustrate the ability of
this mechanism to increase the access opportunities of the weaker connection,
and react to the changes in relative signal strength. The slope of both traces
in Figure 6.7(c) remains constant throughout the experiment, indicating the
ability of this mechanism to control unfair behaviour the network environment
changes. As each connection spends an equal time dominating the channel in
this scenario, determination of a preferred 0 from Figures 6.7(g) and (h) is
not as clear as in the static case. Figure 6.7(g) indicates that a peak in the
fairness index was present at in the region of 0 = 0.25, where the fairness index
has increased by 31%, and the aggregate average throughput suffers a 1 3 %
reduction.
Probabilistic Discard
In this case, it is again evident that the mechanism is able to control the stronger
connection. In this dynamic scenario the performance has improved over the
static scenario. The period during which the nodes are within range is obvious
in Figures 6.8(a), (c) and (e) between 15 and 20 seconds. Through this period,
the network is inherently fair as each node has reliable carrier sense information,
and the mechanism does not appear to adversely effect either trace throughout
this period. Figure 6.8(g) indicates the a preferred 0 lies between 0.3 and 0.4.
Using a value of 0 = 0.35, the fairness index is increased to 0.89, an increase of
37%. Figures 6.8(c), (e), and (f) illustrate the coarseness evident in the static
trial is not present in this dynamic scenario. However, throughput is reduced
by 3 9 % with 0 = 0.35
Prevention Of Signal Strength Dependent Unfairness 154
(a) Data Trace - Evolution p = 0 2000
_ 1800
|! 1600
g 1400
« 1200 g 1000 •a
s
(b) Data Trace - Access Times p = 0
CD
DC
800
600
400 •
200
0
1600
9 1400 CD
•S. 1200 co * 1000 8 CO Q T3
I 8 400 CD c 200
800
600
CD T3
CO CO CD
E co LL
1.1 1
0.9 0.8 0.7 0.6 0.5 0.4 0.3 r 0.2 0.1 0
1.1
x 1 CD
^ 0.9
CD
2 0.7
< 0.6
0.5
Connection A Connection B
25 5 10 15 20
Time (sec)
(c) Data Trace - Evolution p = 0.35
30
Connection A Connection B
10 15 20 25 30
Time (sec)
(e) Jain's Fairness Index
M M ppi w>, .n,\rvT^n~*'H n ufi,
Jain's Index: Average 0.89 i *
10 15 20 25
Time (sec)
(g) Jain's Fairness Index vs p
30
•
i r
/ /
h /
1_
100 Frame Window
.
•
\ + •' \
* v
i 1 1
3 •
1 •
O z
z>
s SI
H TJ CD CO
ra E o
0.6
0.5
0.4
0.3
0.2
0.1
0
0.1 0.2 0.3 0.4 0.5 0.6
0.6
0.5
0.4 f
0.3
0.2
0.1
0
Connection A Connection B
Connection A Connection B
5 10 15 20 25
Time (sec)
(f) Normalised Throughput p = 0.35
Connection A Connection B »
i 10 15 20 25
Time (sec)
(h) Normalised Throughput vs p
Connection A Connection B
5 10 15 20 25 30
Time (sec)
(d) Data Trace - Access Times p = 0.35
30
30
0.1 0.2 0.3 0.4 0.5 0.6
Figure 6.8 Three node topology, dynamic scenario for Probabilistic Discard algo
rithm using U D P 0 = 0.30
Prevention Of Signal Strength Dependent Unfairness 155
Enhanced CTS Suppression
The enhanced CTS results in Figure 6.9 illustrate the effectiveness of this mech
anism in controlling signal strength dependent unfairness. A 0 value of 0.55
provides an average fairness index of 0.91. However, the aggregate throughput
has been reduced while again reducing the aggregate throughput by 53%. The
average fairness index is increased by 40%, similar to the 3 7 % achieved with the
Probabilistic Discard technique. Figure 6.9(g) illustrates that a good fairness
outcome will be achieved in the range of 0 between 0.5 and 0.65.
6.5.6 Dynamic Scenario Discussion
The dynamic experiments have illustrated that each of the three mechanisms
are able to control hidden nodes responsible for significant unfairness suffered
by weaker hidden nodes. During periods where the stronger host changes, the
algorithm defined in Equation (6.2) is able to adapt quickly, allowing each
mechanism to then retain the fair behaviour during changes in the localised
topology. The reader is referred to Appendix B for T C P results in this scenario.
Table 6.1 summarises the percentage change in aggregate throughput and fair
ness index achieved with each scheme. This summary illustrates that while the
Enhanced C T S Suppression provides the greatest improvement in the fairness
index, there is a corresponding significant reduction in aggregate throughput.
This is a significant disadvantage for the Enhanced C T S Suppression mech
anism. The results in Table 6.1 illustrate that the p-Persistence on Backoff
Countdown mechanism offers a preferred combination of fairness improvement
and corresponding throughput reduction.
In each case the unfairness control schemes are again shown to be capable of
providing improved throughput for the weaker connections by preventing the
stronger connection from gaining unfair channel access. The well defined peak
in the fairness index as a function of 0 is present in each case, as with the static
scenarios.
Prevention Of Signal Strength Dependent Unfairness 156
2000 _ 1800 S 1600 m 1400 « 1200 § 1000 S 800 §3 600 | 400 200
(a) Data Trace - Evolution p = 0 (b) Data Trace - Access Times p = 0
1200
<g 1000
£ 800 CO CO Q •o CD
> CD CJ CD
DC
600
400
200
x CD TJ c
1.1 1
0.9 0.8 0.7
» 0.6 SS 0.5 I 0.4 "- 0.3
0.2 r 0.1 0
Connection A Connection B
25 5 10 15 20 Time (sec)
(c) Data Trace - Evolution p = 0.55
30
Connection A Connection B
10 15 20 25 30 Time (sec)
(e) Jain's Fairness Index
!Wf\, m i n./Mi* If j
in H \i
Jain's Index: Average 0.91
10 15 20 Time (sec)
(g) Jain's Fairness Index vs p
25 30
•a
o Z
co E o Z
0.5
0.4
0.3
0.2
0.1
Connection A Connection B
Connection A Connection B
10 15 20 25 Time (sec)
(f) Normalised Throughput p = 0.55
Connection A Connection B
i 10 15 20 25 Time (sec)
(h) Normalised Throughput vs p
Connection A Connection B
5 10 15 20 25 30 Time (sec)
(d) Data Trace - Access Times p = 0.55
30
30
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Figure 6.9 Three node topology, dynamic scenario for Enhanced CTS Suppression
algorithm using UDP 0 = 0.55
Prevention Of Signal Strength Dependent Unfairness 157
Table 6.1 Comparison of improvement in fairness index and reduced aggregate normalised throughput for static and dynamic UDP scenarios with each fairness control technique
Mechanism
p-Persistence
on Backoff
Countdown
Probabilistic
Discard
Enhanced
CTS Suppression
Static Scenario Fairness
Index
Increase
60%
39%
79%
Aggregate
Throughput
Reduction
2%
6%
56%
Dynamic Scenario
Fairness
Index
Increase
31%
37%
40%
Aggregate Throughput
Reduction
13%
39%
53%
6.6 Conclusions and Recommendations
In this chapter we have considered three schemes to correct for the relative
signal strength based unfairness identified, analysed, and modelled in previous
chapters. The investigation presented here was designed to establish the per
formance of each mechanism across a range of realistic network topologies and
signal strength scenarios.
The mechanisms proposed comprise two components. The first is a technique
to determine a probability variable for each hidden node in proportion to the
relative signal strength amongst all identified hidden nodes. This component is
common in all three mechanisms, and may be tuned through the capture param
eter, 7 to suit any specific P H Y displaying relative signal power based unfair
behaviour. The probability variable is then employed by each mechanism to
provide weaker connections with additional transmission opportunities. Three
mechanisms have been proposed, namely: p-Persistence on Backoff Countdown,
Selective Discard, and Enhanced CTS Suppression.
Prevention Of Signal Strength Dependent Unfairness 158
Through both the static and dynamic experiments outlined in previous sections,
the performance of the three fairness control mechanisms has been established.
The results of this investigation illustrate that all three mechanisms provide
a significant increase in the fairness properties and from this aspect would
be suitable for implementation in a physical system. However, the Enhanced
C T S Suppression technique results in an unacceptable reduction in aggregate
throughput. The Probabilistic Discard technique provides a smaller increase in
fairness index than the p-Persistence on Backoff Countdown technique, and a
larger reduction in aggregate throughput in dynamic scenarios. Combined with
the relative simplicity of the Probabilistic Discard technique, this implies that
the choice becomes one of trading implementation complexity and flexibility
against fairness and throughput performance.
In terms of performance, implementation complexity, and flexibility, results for
the p-Persistence on Backoff Countdown, Probabilistic Discard, and Enhanced
C T S mechanisms can be summarised in the following manner:
• The Enhanced CTS Suppression mechanism results in the greatest im
provement in the fairness index, though has the greatest impact on aggre
gate throughput. The range of 0 values required for optimal performance
is greater with this technique than the other mechanisms, ranging from
0.55 in the dynamic scenario up to 0.75 in the static scenario. This scheme
requires careful selection of 0 to achieve optimal performance. The im
plementation of this technique will make use of a planned enhancement
to the IEEE 802.11 standard allowing a node to forcibly suppress another
specified node, and therefore represents a flexible technique requiring no
additional changes to the standard outside those already planned. This
technique offers increased flexibility over the other mechanisms through
an implementation employing an enhancement to the IEEE 802.11 stan
dard. The reduction in throughput observed with this scheme represents a
significant disadvantage compared to the p-Persistence on Backoff Count
down and Probabilistic Discard techniques.
Prevention Of Signal Strength Dependent Unfairness 159
• The p-Persistence on Backoff Countdown mechanism results in a smaller
increase in the fairness index than the Enhanced C T S mechanism, being
approximately equal to the Probabilistic Discard technique in the dynamic
scenario. This technique has the smallest impact on aggregate throughput
of the three. This technique did not require any adjustment in the control
parameter as dynamic node behaviour was introduced, maintaining a con
stant value of 0 = 0.25 to achieve the optimal fairness outcome. However,
implementation of this scheme represents a significant problem in cases
where inter-operability and standards compliance are significant issues.
The mechanism proposed is to include the variable in a field within an
A C K frame, as it is directed, and therefore only interpreted by the receiv
ing node. However, this raises compatibility issues for stations employing
this technique with legacy IEEE 802.11 stations, who will be unable to
interpret an extended A C K frame format. In these cases, another option
is distribution of the probability variable within a new M M P D U control
frame format.
• The Probabilistic Discard mechanism whilst representing the most course
grained technique the three approaches, still provides good improvement
in the fairness properties while having and a moderate impact on aggregate
throughput. A control parameter of between 0 = 0.35 and 0 = 0.40 was
found to be suitable for the dynamic and static scenarios respectively.
The simplicity with which this scheme may be implemented represents
the biggest single advantage over other schemes. N o explicit information
exchange is required, as each node simply drops D A T A and R T S frames
from offending nodes. This mechanism trades simplicity against improved
performance, control, and flexibility.
The most important aspect of each scheme is the selection of the 0 value required
to optimise the fairness performance. The experiments have identified a defined
range of 0 values for each mechanism, suitable for the scenarios investigated
here. Investigation of other factors affecting the selection of an appropriate 0
Prevention Of Signal Strength Dependent Unfairness 160
is an issue requiring future study, including the development of an adaptive
technique to control 0. Further, the scalability of each mechanism with respect
to the number of nodes requires investigation.
The major recommendations for each technique arising from this investigation
are:
• In scenarios where aggregate throughput is significant, the p-Persistence
on Backoff Countdown mechanism appears to be the preferred alternative.
This is provided that compatibility and implementation issues can be
addressed.
• In scenarios where simplicity is the overriding concern, the Probabilis
tic Discard scheme has been shown to perform well in terms of fairness
outcomes.
• In scenarios where per connection fairness is the overriding factor, the
Enhanced C T S Suppression mechanism represents a more suitable option.
This comes at the expense of a significant reduction in aggregate through
put. The ability to tune the Enhanced C T S Suppression mechanism to
improve the aggregate throughput though adaption of the suppression
period, or the control parameter 0, requires further investigation.
Chapter 7
Conclusions
7.1 Overview
The popularity of high speed local area wireless networks has been driven in
recent years by the introduction of the IEEE 802.11 M A C / P H Y protocol. Ac
cordingly, there has been continued research effort examining many aspects of
the performance of wireless M A C protocols with respect to the quality of service
they are able to provide. While many early problems have been covered by ex
isting literature, this thesis has presented an investigation of capture effects and
fairness behaviour arising from an investigation of physical system performance.
This thesis has followed a path of identification, empirical investigation, anal
ysis, modelling, leading to the identification and evaluation of options for the
prevention of capture effects present in a range of physical IEEE 802.11 network
scenarios. This chapter presents a summary of the major results presented in
this thesis.
7.2 Significant Results
The literature review in Chapter 2 identified a need for an experimental inves
tigation of the fairness properties of the C S M A / C A M A C protocol employing
the R T S / C T S / D A T A / A C K handshake in a real propagation environment, with
161
Conclusions 162
particular reference to the capture behaviour of the system in terms of both
m o d e m and protocol capture. This is addressed in Chapter 3 through a de
tailed series of experiments in hidden terminal topologies with the IEEE 802.11
M A C / P H Y protocol. This investigation has uncovered a reliable and repeat-
able relationship between the relative signal strength of hidden connections and
the ability of a host to capture the radio channel. In all experiments involving
a measured difference in signal power of greater than 5dB, the stronger of two
hidden connections is able to gain preferential access to the channel, despite the
use of the R T S / C T S handshake to reserve transmission opportunities equally
for both nodes. The effect is observed with all current 802.11 DSSS PHY's,
and with the two radio front end circuits employed in all current 802.11 radio
interfaces.
Experiments performed with TCP illustrate that the adverse interaction be
tween the T C P and M A C retransmission timers is not present in cases where
the signal strength is equal on each connection. Experiments performed using
greedy U D P sources confirm this behaviour, with the stronger connection gain
ing significantly greater throughput than the weaker connection. Again, the
measured relative signal power threshold of 5dB was observed. In cases with an
equal received signal power on each connection, effective sharing is observed.
These results suggest a complex m o d e m capture behaviour biased in favour of
packets arriving at the receiver with a higher signal power.
Following the empirical identification of the relative signal strength dependence
in Chapter 3, Chapter 4 presents an analysis of the mechanisms behind this
behaviour. The scenario under investigation is one where a radio frame is as
sumed to be under reception when an interfering frame arrives at the receiver.
Expressions relating the B E R experienced by the original signal at the output
of the correlation receiver to the relative signal power between the interfering
and original signals are derived through a modification of the techniques used
to determine the magnitude of multiple access interference in C D M A systems.
This investigation illustrates that an interfering signal arriving with a relative
Conclusions 163
power of greater than 2dB will result in the effective corruption of the original
signal. Conversely, the interfering frame will suffer little impact from the origi
nal frame provided the receiver is able to resolve the two signals. This threshold
matches the empirical measurements presented in Chapter 3.
The results of Chapters 3 and 4 raise significant issues for the validity of sim
ulation models for physical IEEE 802.11 networks. In particular, the ability of
current receiver models to accurately model the effects identified and analysed
in previous chapters must be established. Chapter 5 presents an investigation
of the behaviour of packet capture models in terms of their ability to match
empirical data when employed in a simulation environment. Further, network
fairness at the transport layer is introduced as a metric to compare simulation
with packet capture models against empirical data. In response to an apparent
inability of current packet capture models to accurately reflect the empirical
data, a new model is proposed based on the physical operation of an IEEE
802.11 radio interface. The new model, termed Message Retraining, is shown
to be significantly more accurate in matching the fairness behaviour of empirical
data in terms of both magnitude of the fairness index, and the fairness horizon
or timescale over which the network is considered fair. This model is shown
to provide an accurate basis for the development of mechanisms to prevent
unfairness of this type arising in hidden terminal topologies.
Employing the model developed and validated in Chapter 5, Chapter 6 intro
duces a number of techniques to prevent unfair behaviour arising in hidden
terminal topologies as a result of relative signal power differences. Chapter 6
presents a utility function based mechanism, employing the average relative
signal power for identified hidden neighbours to determine a probability vari
able for each hidden connection. This is coupled with one of three mechanisms
designed to provide additional transmission opportunities for weaker connec
tions. Performance analysis indicates that all three techniques are effective in
providing improved fairness outcomes for network scenarios involving hidden
terminals, though a significant reduction in aggregate throughput was observed
Conclusions 164
for the Enhanced C T S Suppression technique. This reduced throughput, com
bined with implementation issues dictate that either the Probabilistic Discard
or p-Persistence on Backoff Countdown technique be employed in a physical
network. In terms of implementation, the p-Persistence on Backoff Countdown
technique requires the distribution of a probability variable to identified hidden
nodes. In comparison to the Probabilistic Discard technique which requires no
additional messaging. The three mechanisms aim to return the network to the
state where each connection has reliable carrier sense information. The mecha
nisms presented in Chapter 6 do require further investigation, though the results
obtained are very promising, considering the minimal cost required to provide
a significantly improved fairness outcome.
7.3 Further Work
This thesis has addressed a number of significant issues associated with capture
effects and related fairness properties in physical IEEE 802.11 networks. How
ever, there are still a number of issues that require further investigation. These
are described below.
• Extension of multiple access analysis to consider an OFDM physical layer.
Almost all new wireless P H Y proposals are based on an O F D M signalling
technique. The extension of the multiple access interference analysis pre
sented in Chapter 4 will be of significant benefit in identifying similar
behaviour in future O F D M systems. Further, due to a lack of O F D M
hardware available on the market at the time, the empirical performance
of an O F D M system in the hidden terminal topologies requires investiga
tion when suitable IEEE 802.11a O F D M systems are available.
• An opportunity exists to extend the development of the Message Retrain
ing capture model to determine the capture probability as a function of
the number of interfering hidden nodes. This may then be used with an
Conclusions 165
appropriate analytical model of the IEEE 802.11 M A C protocol to deter
mine the aggregate channel throughput achieved.
• An assumption made during the performance investigation of the fairness
control schemes is that all hidden nodes are identified to the common node.
Further investigation of suitable mechanisms to identify hidden nodes in
a general topology network is an issue requiring investigation. Potential
interactions between a hidden terminal identification mechanism and the
fairness control schemes proposed require examination.
• The scalability of each fairness control mechanism proposed in Chapter 6
with respect to the number of supported nodes requires investigation.
• A thorough investigation of the stability of each fairness control mecha
nism with respect to the control parameter, 0 will enable each mechanism
to enact more precise control.
• An adaptive mechanism to control 0 will require development.
• A detailed investigation of the interaction between MAC level QoS schemes
proposed for IEEE 802.11 and the fairness control mechanisms proposed
in Chapter 6 will be required before the fairness control techniques can
be employed in scenarios requiring service differentiation.
• A limitation in the performance analysis of the fairness control schemes
in Chapter 6 is the inability to test each scheme when employed in an
IEEE 802.11 radio interface. Investigation of the performance of the
Enhanced C T S Suppression, Probabilistic Discard, and p-Persistence on
Backoff Countdown mechanisms when implemented in an 802.11 radio
interface will provide additional insight into the range of 0 values over
which each scheme will provide reliable, stable operation.
• Merging fairness control mechanisms with schemes designed to control
general protocol or topology dependent unfairness is also a significant issue
requiring investigation. This may include the ability to incorporate traffic
Conclusions 166
observations as a decision making variable. This has the potential to allow
a combined scheme to provide a generalised fairness control technique.
• The investigation of power control mechanisms to prevent the unfair be
haviour observed in this thesis is an area for future investigation. The
results presented in Chapters 3 and 4 indicate that very fine power control
system capable of maintaining the range of received powers at a common
node within a 3dB range may provide a means of controlling the packet
capture problem. This may be possible in a traditional wireless L A N base
station - client topology, particularly given current power control efforts
within the IEEE 802.11 W G . Designing a power control mechanism to
achieve this aim in a general topology M A N E T style network is a non-
trivial problem.
• Detailed investigation of the Point Co-ordinate Function and the Hybrid
Co-ordinate Function M A C protocols within 802.11 in similar scenarios to
those investigated here. As these M A C protocols are centrally controlled,
hidden stations will not be forced to contend for channel access. This has
the potential to remove the relative signal strength basis, though other
issues with polling beacon reception may arise. This area requires detailed
investigation.
• Refinement and extension of the interpretation of fairness in both a per
user and global sense is required.
Bibliography
Abramson, N. (1970). The ALOHA System - Another Alternative for Com
puter Communications. In 1970 Fall Joint Computer Conference, AFIPS
Conference Proceedings, volume 37, pages 281-285.
Abramson, N. (1977). The Throughput of Packet Broadcasting Channels. IEEE
Transactions on Communications, 25(1):117—128.
Anastasi, G., Lenzini, L., and Mingozzi, E. (2000). HIPERLAN/1 M A C Proto
col: Stability and Performance Analysis. IEEE Journal on Selected Areas
in Communications, 18(9):1787-1798.
Arnbak, J. C. (1987). Capacity of Slotted ALOHA in Rayleigh fading channels.
IEEE Journal on Selected Areas in Communications, 5:261-269.
Bensaou, B., Wang, Y., and Ko, C. C. (2000). Fair medium access in 802.11
based wireless ad-hoc networks. In Mobile and Ad Hoc Networking and
Computing, 2000. Mobihoc. 2000, pages 99-106, Boston, MA.
Bharghavan, V., Demers, A., Shenker, S., and Zhang, L. (1994). M A CAW:
A Media Access Protocol for Wireless LAN's. In SIGCOMM 94, pages
212-225. ACM.
Cali, F. Conti, M. and Gregori, E. (2000). IEEE 802.11 Protocol: Design and
Performance Evaluation of an Adaptive Backoff Algorithm. IEEE Journal
on Selected Areas in Communications, 18(9): 1774-1786.
Chandra, A., Gummalla, V., and Limb, J. O. (2000). Wireless Collision Detect
(WCD): Multiple Access with Receiver Initiated Feedback and Carrier De-
167
BIBLIOGRAPHY 168
tect Signal. In IEEE International Conference on Communications, ICC,
volume 1, pages 397-401.
Chandran, K., Raghunathan, S., Venkatesan, S., and Prakash, R. (1998). A
Feedback Based Scheme For Improving TCP Performance In AdHoc Wire
less Networks. In Papazoglou, M., editor, Distributed Computing Systems,
volume 1, pages 472 - 479, Dallas, Tx, USA. IEEE.
Chesson, G., Diepstraten, W., Hoeben, M., Singla, A., Teunissen, A., and
Wentink, M. (2001). IEEE 802.11-01/131rl Proposed VDCF Draft Text.
IEEE 802 LAN/MAN Standards Committee, P802.ll Working Group,
Task Group E: M A C QoS Enhancements.
Cheun, K. and Kim, S. (1998). Joint Delay-Power Capture in spread-spectrum
packet radio networks. IEEE Transactions on Communications, 46(4):450
-453.
Choi, S., Gray, S., Kasslin, M., Mangold, S., Soomro, A., Myles, D., Skellern,
D., and Ecclesine, P. (2001). IEEE 802.11-01/169r2 Transmitter Power
Control (TPC) / Dynamic Frequency Selection (DFS) Joint Proposal for
802.11h W L A N . IEEE 802 LAN/MAN Standards Committee, P802.ll
Working Group, Task Group H: Spectrum and Transmit Power Manage
ment in 5GHz BAND.
Colvin, A. (1983). CSMA with Collision Avoidance. Computer Communica
tions, 6(5):227-235.
Davis, D. H. and Gronemeyer, S. (1980). Performance of slotted ALOHA ran
dom access with delay capture and randomised time of arival. IEEE Trans
actions on Communications, 28:703 - 710.
Deng, J. and Haas, Z. J. (1998). Dual Busy Tone Multiple Access (DBTMA):
A New Medium Access Control for Packet Radio Networks. In IEEE In
ternational Conference on Universal Personal Communications, ICUPC,
volume 2, pages 973-977, Florence, Italy.
BIBLIOGRAPHY 169
Deng, J. and Haas, Z. J. (1999). Dual Busy Tone Multiple Access (DBTMA):
Performance Results. In IEEE Wireless Communications and Networking
Conference, WCNC, volume 3, pages 1328-1332, New Orleans, LA, USA.
European Telecommunications Standards Intsitute (1998). Broadband Radio
Access Networks (BRAN): High Performance Radio Local Area Networks
(HIPERLAN).
Fischer, M. (2001). IEEE 802.11-01/110rl Hybrid Co-ordination Function
(HCF) Proposed Updates to Normative Text. IEEE 802 LAN/MAN Stan
dards Committee, P802.ll Working Group, Task Group E: M A C QoS En
hancements.
Fullmer, C. and Garcia-Luna-Aceves, J. (1995). Floor Acquisition Multiple
Access (FAMA) for packet radio networks. In SIGCOMM 95, pages 262-
273, Cambridge MA. ACM.
Fullmer, C. L. and Garcia-Luna-Aceves, J. J. (1997a). Complete Single-Channel
Solutions to Hidden Terminal Problems in Wireless LANs. In IEEE Inter
national Conference on Communications, ICC, volume 2, pages 575-579,
Montreal, Canada. IEEE.
Fullmer, C. L. and Garcia-Luna-Aceves, J. J. (1997b). Solutions to Hidden
Terminal Problems in Wireless LANs. In SIGCOMM 97, volume 1, pages
39-49, Cannes, France. ACM.
Fullmer, C. L. and Garcia-Luna-Aceves, J. J. (1998). Performance of Floor
Acquisition Multiple Access in Ad Hoc Networks. In IEEE Symposium on
Computers and Communications, volume 1, pages 63-68, Athens, Greece.
IEEE.
Garcia-Luna-Aceves, J. J. and Tzamaloukas, A. (1999). Reversing the Collision
Avoidance Handshake in Wireless Networks. In ACM/IEEE International
Conference on Mobile Computing and Networking, Mobicom, pages 120-
131. ACM/IEEE.
BIBLIOGRAPHY 170
Geraniotis, E. and Pursley, M. B. (1982). Error Probability for Direct Sequence
Spread Spectrum Multiple Access Communications - Part II: Approxima
tions. IEEE Transactions on Communications, 30(5):985-995.
Geraniotis, E. and Soroushnejad, M. (1987). Collision Resolution Algorithms
for Networks with Spread Spectrum Capture Capability. Technical report,
University of Maryland, Institute for Systems Research.
Gerla, M., Bagrodia, R., Tang, K., and Wang, L. (1999a). TCP Over Wireless
Multi-hop Protocols: Simulation and Experiments. In IEEE International
Conference on Communications, volume 1, Vancouver, Canada. IEEE.
Gerla, M., Tang, K., and Bagrodia, R. (1999b). TCP Performance in Wireless
Multihop Networks. In 2nd IEEE Workshop on Mobile Computing Systems
and Applications, volume 1, pages 41 - 50. IEEE.
Goodman, D. J. and Saleh, A. A. M. (1987). The Near/Far Effect in Local
A L O H A Radio Communications. IEEE Transactions on Vehicular Tech
nology, 36(l):19-27.
Hansen, C, Fischer, M., Gubbi, R., and Kim, J. (2001). IEEE 802.11-01/217
TPC/DFS Proposal for 802.llh. IEEE 802 LAN/MAN Standards Com
mittee, P802.ll Working Group, Task Group H: Spectrum and Transmit
Power Management in 5GHz BAND.
Holland, G. and Vaidya, N. (1999). Analysis of TCP Performance over Mobile
Ad-Hoc Networks. In Fifth Annual International Conference on Mobile
Computing and Networking, Mobicom 99, Seatle.
Institution of Electrical and Electronic Engineers (1999a). Part 11: Wireless
LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifi
cations.
Institution of Electrical and Electronic Engineers (1999b). Part 11: Wireless
LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifi
cations, Higher Speed Physical Layer Extension in the 2.4 GHz Band.
BIBLIOGRAPHY 171
Institution of Electrical and Electronic Engineers (1999c). Part 11: Wireless
LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifi
cations, Higher Speed Physical Layer in the 5 GHz Band.
Jain, R., Chiu, D., and Hawe, W. (1984). A Quantitative Measure Of Fairness
And Discrimination For Resource Allocation In Shared Computer Systems.
Technical Report TR-301, Digital Equipment Corporation.
Karkkainen, K. H. (1992). Mean-Square Cross Correlation As A Performance
Measure for Spreading Code Families. In IEEE 2nd Int. Symp. on Spread
Spectrum Techniques and Applications, Yokohama, Japan.
Karn, P. (1990). MACA - A New Channel Access Method for Packet Radio.
In AARL/CRRL Amateur Radio 9th Computer Networking Conference,
pages 134-40.
Kim, D. I., Kim, I. K., and Scholtz, R. A. (1995). Counting Collision Free Trans
missions in Common-Code SSMA Communications. IEEE Transactions on
Communications, 43(2/3/4).
Kleinrock, L. and Tobagi, F. A. (1975). Packet Switching in Radio Channels:
Part I-Carrier Semse Multiple-Access Modes and Their Throughput-Delay
Characteristics. IEEE Transactions on Communications, 23(12):1400-
1416.
Koksal, C. E., Kassab, H., and Balakrishnan, H. (2000). An Analysis of Short
Term Fairness in Wireless Media Access Protocols: Extended Version of
short paper. In ACM Sigmetrics.
Lau, C. and Leung, C. (1992). Capture Models for Mobile Packet Radio Net
works. IEEE Transactions on Communications, 40(5):917-925.
Lawrence Berkeley Laboratories (1999). tcpdump. http://www.tcpdump.org.
Lu, S., Bharghavan, V., and Srikant, R. (1999). Fair Scheduling in Wireless
Packet Networks. IEEE/ACM Transactions on Networking, 7(4):473-489.
BIBLIOGRAPHY 172
Luo, H., Lu, S., and Bharghavan, V. (2000). A New Model for Packet Scheduling
in Multihop Wireless Networks. In ACM Mobicom.
Mud, R., Boer, J., Kamerman, A., Van Driest, H., Diepenstraten, W., Kop-
meiners, R., and Von Bokhorst, H. (1999). Wireless LAN With Enhanced
Capture Provision. US Patent No US5987033.
Muir, A. and Garcia-Luna-Aceves, J. J. (1997). Group allocation multiple access
with collision detection. In INFOCOM 97, Sixteenth Annual Joint Con
ference of the IEEE Computer and Communications Societies, volume 3,
pages 1182 -1190.
Nandagopal, T., Kim, T.-E., Gao, X., and Bharghavan, V. (2000). Achieving
M A C Layer Fairness in Wireless Packet Networks. In A CM Mobicom,
Boston, MA.
Nasipuri, A., Zhuang, J., and Das, S. R. (1999). A Multichannel CSMA M A C
Protocol for Multihop Wireless Networks. In IEEE Wireless Communica
tions and Networking Conference, WCNC, volume 3, pages 1402-1406.
Ozugur, T., Naghshineh, M., Kermani, P., and Copeland, J. A. (1999). Fair
Media Access Methods for Wireless LAN's. In Proc. of IEEE Globecom,
volume 2, pages 570-579.
Ozugur, T., Naghshineh, M., Kermani, P., Olsen, C. M., Rezvani, B., and
Copeland, J. A. (1998). Balanced Media Access Methods for Wireless
Networks. In Proc. of ACM/IEEE MobiCom, volume 1, pages 21-32.
Pagtzis, T., Kirstein, P., and Hailes, S. (2001). Operational and Fairness Issues
with Connection-less Traffic over IEEE 802.11b. In IEEE International
Conference on Communications, ICC 2001, volume 1.
Proakis, J. G. (1995). Digital Communications. McGraw Hill, third edition.
Pursley, M. B. (1977). Performance Evaluation for Phase Coded Spread Spec
trum Multiple Access Communication - Part 1: System Analysis. IEEE
Transactions on Communications, 25(8):795-799.
BIBLIOGRAPHY 173
Rappaport, T. S. (1996). Wireless Communications: Principles and Practice.
Prentice Hall.
Sherman, M. (2001). IEEE 802.11-01/130r5 Proposed Normative Text for ERTS
and ECTS. IEEE 802 LAN/MAN Standards Committee, 802.11 Working
Group, Task Group E: M A C QoS Enhancements.
Sobrinho, J. and Krishnakumar, A. S. (1999). Quality of Service in Ad Hoc Car
rier Sense Multiple Access Wireless Networks. IEEE Journal on Selected
Areas in Communications, 17(8):1353-1368.
Soroushnejad, M. and Geraniotis, E. (1991). Probability of Capture and Rejec
tion of Primary Multiple Access Interference in Spread Spectrum Networks.
IEEE Transactions on Communications, 39(6):986-994.
Swan, A. and Raman, S. (2000). A Study of Short Term Fairness in Wireless
M A C Protocols. In Technical Report.
Talucci, F. and Gerla, M. (1997). MACA-BI (MACA By Invitation) A Wireless
M A C Protocol for High Speed Ad-Hoc Networking. In 6th Annual Con
ference on Universal Personal Communications, volume 2, pages 913-917.
IEEE.
Talucci, F., Gerla, M., and Fratta, L. (1997). MACA-BI (MACA By Invitation)
A Receiver Oriented Access Protocol for Wireless Multihop Networks. In
Personal, Indoor & Mobile Radio Communications - Waves of the Year
2000, volume 2, pages 435-439. IEEE.
Tang, K., Correa, M., and Gerla, M. (2001). Effects of Ad Hoc MAC Layer
Medium Access Mechanisms Under TCP. To Appear: ACM/Baltzer Mobile
Networks and Applications, MONET.
Tang, K. and Gerla, M. (1999). Fair Sharing of MAC under TCP in Wire
less Adhoc Networks. In Multimedia, Mobility and Teletraffic for Wireless
Networks (MMT'99), Venice, Italy. IEEE.
BIBLIOGRAPHY 174
Tang, Z. and Garcia-Luna-Aceves, J. J. (2000). Collision Avoidance Transmis
sion Scheduling for Ad Hoc Networks. In IEEE International Conference
on Communications, volume 2, pages 183-190.
TGe (2001). DRAFT: Medium Access Control (MAC) Enhancements for Qual
ity of Service. IEEE 802 LAN/MAN Standards Committee, 802.11 Working
Group, Task Group E: M A C QoS Enhancements.
TGh (2001). DRAFT: Medium Access Control (MAC) and Physical Layer
(PHY) Spectrum and Transmit Power Management Extensions on the
5GHz Band in Europe. IEEE 802 LAN/MAN Standards Committee, 802.11
Working Group, Task Group H: Spectrum and Transmit Power Manage
ment in 5GHz BAND.
Tobagi, F. A. and Kleinrock, L. (1975). Packet Switching in Radio Channels:
Part II-The Hidden Terminal Problem in Carrier Sense Multiple-Access
and the Busy Tone Solution. IEEE Transactions on Communications,
23(12):1417-1433.
Tobagi, F. A. and Kleinrock, L. (1976). Packet Switching in Radio Channels:
Part III- Polling and (Dynamic) Split-Channel reservation Multiple Access.
IEEE Transactions on Communications, 24(8):832-844.
Tzamaloukas, A. and Garcia-Luna-Aceves, J. (1999). Poll-Before-Data Multiple
Access. In International Conference on Communications, ICC, volume 2,
pages 1207-1211. IEEE.
Tzamaloukas, A. and Garcia-Luna-Aceves, J. J. (2001). A Receiver-Initiated
Collision-Avoidance Protocol for Multi-Channel Networks. In INFOCOM,
pages 189-198, Anchorage, AK, USA. IEEE.
Tzamaloukas, A. E. (December 2000). Sender and Receiver Initiated Multiple
Access Protocols For Ad Hoc Networks. PhD thesis, University of Califor
nia, Santa Cruz.
BIBLIOGRAPHY 175
UCB/LBNL/VINT (1999). ns - Network Simuator. UCB/LBNL/VINT,
http://www-mash.cs.berkeley.edu/ns/.
Ware, C. and Dutkiewicz, E. (2001). IEEE 802.11-01/232 Packet Capture UDP
Experiments. IEEE 802 LAN/MAN Standards Committee, P802.ll Work
ing Group, Task Group E: M A C QoS Enhancements.
Ware, C. G., Judge, J., Chicharo, J. F., and Dutkiewicz, E. (2000). Unfair
ness and Capture Behaviour in 802.11 Adhoc Networks. In International
Conference on Communications, volume 1, New Orleans. IEEE Press.
Ware, C. G., Wysocki, T., and Chicharo, J. (2001a). Hidden Terminal Jam
ming Problems in IEEE 802.11 Mobile Ad Hoc Networks. In International
Conference on Communications, Helsinki. CRDOM.
Ware, C. G., Wysocki, T., and Chicharo, J. (2001b). On The Hidden Terminal
Jamming Problem in IEEE 802.11 Ad Hoc Networks, submitted to: IEEE
Transactions on Vehicular Technology.
Ware, C. G., Wysocki, T., and Chicharo, J. (2001c). Simulating Capture Be
haviour in IEEE 802.11 Radio Modems. Journal of Telecommunications
and Information Technology.
Ware, C. G., Wysocki, T., and Chicharo, J. (2001d). Simulation of Capture
Behaviour In IEEE 802.11 Radio Modems. In Vehicular Technology Con
ference, Fall 2001, New Jersey. IEEE.
Ware, C. G., Wysocki, T., and Chicharo, J. F. (2001e). Modelling Capture
Behaviour In IEEE 802.11 Radio Modems. In International Conference on
Telecommunications, Bucharest. IEE / IEEE. CDROM.
Webster, M. and Halford, S. (2000). IEEE 802.11-00/389rl Overview of OFDM
for High Rate Extension. IEEE 802 LAN/MAN Standards Committee,
P802.ll Working Group, Task Group G: High Rate Extensions in 2.4 GHz
Band.
BIBLIOGRAPHY 176
Wu, S., Tseng, Y., and Sheu, J. (2000). Intelligent Medium Access for Mobile
Ad Hoc Networks with Busy Tones and Power Control. IEEE Journal on
Selected Areas in Communications, 18(9).
Appendix A
Hidden Node Detection Mechanisms
As mentioned in Section 6.3.1, identification of hidden nodes is a key issue
for the performance of any technique designed to control unfair behaviour as a
result of hidden terminals. In this appendix, we outline proposed mechanisms to
identify potential hidden node pairs within the 802.11 framework. This can then
be employed to instigate the fairness control algorithms under development.
Within the context of a C S M A / C A M A C , a potential mechanism to identify
hidden nodes exists through observation of message exchange semantics within
either the R T S / C T S / D A T A / A C K or D A T A / A C K handshake. Figure A.l below
outlines the scenario for the identification of a hidden node. In this context,
nodes 1 and 3 are hidden nodes, with node 1 referred to as the hidden node for
node 3, via node 2.
An STA is able to identify a potential hidden STA if it receives a significant
number of C T S frames addressed to another STA, without observing the cor
responding R T S or D A T A frames. If such a case is observed, the intended
recipient of the C T S frame is the potential hidden STA, and the transmitter of
the C T S frame the node via which the hidden STA is present. Referring to Fig
ure A.l, STA 3 is able to identify STA 1 as hidden when C T S and A C K frames
addressed to STA 1 are observed without having observed the corresponding
177
Hidden Node Detection Mechanisms 178
^^-- Hidden Nodes ____
® © ®
i RTS TA=1
; CTS RA = I
! DATA TA=1
! ACK RA = 1
C T S RA = I ;
A C K RA=1 !
I I I
Figure A.l Hidden Terminal Message Exchange Semantics. Node 3 observes node 1 as hidden via CTS and A C K frames
RTS or DATA frames from STA 1.
In the case where no RTS/CTS handshake is employed, the messaging exchange
involves D A T A and A C K frames only. Each STA will be required to observe
A C K frames, then having observed no corresponding D A T A frames the intended
recipient of the A C K can be identified as a hidden node. Figure A.l provides
an overview of the message exchange observations required to identify a hidden
node.
Observation of messaging exchange sequences can be implemented using either
timing flags, or through a table of known neighbours. 802.11 places tight tim
ing constraints on the transmission of C T S and A C K messages (Institution of
Electrical and Electronic Engineers, 1999a). A n RTS must be followed by a
C T S after a single SIFS period. Likewise, an A C K frame must be returned a
single SIFS period after the completion of the D A T A frame. These constraints
may be exploited to identify a potential hidden node. If a node observes a C T S
frame without having observed the immediately previous RTS, followed by an
A C K without having observed the preceding D A T A frame, the Receiving Ad-
<u
Hidden Node Detection Mechanisms 179
dress (RA) of the C T S and A C K frame is a potential hidden node. This can be
achieved by recording the SA of the R T S or D A T A frame and the Duration/ID
field.
When a CTS or ACK frame is observed on the WM, the RA and Duration/ID
are compared against the address and duration recorded previously. Inconsis
tencies between the address or the duration fields indicate that the C T S may
have origininated from a hidden node. This is outlined in Figure A.2. Upon
receipt of a C T S frame, the Duration field should match the duration from the
corresponding R T S frame, less the time required to transmit the C T S frame and
a single SIFS period. If the observed Duration is inconsistent with the cached
value, the C T S frame is directed towards a potentially hidden host. A similar
process can be followed to determine if the Duration field within the A C K frame
is consistent with the value cached from a corresponding D A T A frame.
Alternatively, the cache of known neighbours introduced in Section 6.3, can be
utilised to identify potential hidden nodes. Upon reception of a C T S or A C K
frame the R A is checked against the cache of known neighbours. If the address
is unknown, then a potential hidden node has been identified. The algorithm
required to implement this technique is outlined graphically in Figure A.3.
Both the address based, and the timing method will require a threshold number
of observations to prevent 'false alarm' hidden node indications in a practical
implementation. Once a node has identified a hidden node, the address of the
hidden node must be communicated to the common node, via which the hidden
node is present. This can be achieved using a hidden station identification
management frame proposed in (TGh, 2001). Unfortunately, within the 802.11
standard an SA field is not included within a C T S or A C K frame. Without
the S A a node cannot identify the common node via which the hidden node is
observed. Referring to Figure A.l for example, node 3 is unable to identify node
2 as the common node via which node 1 is hidden. To overcome this limitation, a
node will be required to broadcast the hidden station identification notification
to all neighbours. A node receiving the hidden station management notification
Hidden Node Detection Mechanisms 180
(a) Nodes 1, 2, 3 Not Hidden
® RTS SA = l
CTS RA = !
DATAS A = 1
ACK R A = j
" R T S SA = 1 !
;CTSRA=I i
JDATASA=J
;ACKRA = 1 I
& co
,f to
n co
CTS/ACK immediately
preceeded by RTS/DATA
Consistent SA/Duration
(b) Nodes 1 and 3 hidden
® © ® ! RTS S A = 1 \
i i /
| C T S R A = 1 j
: DATAS A = 1 : ; 1 I ;
! ACK R A = 1 , ,
,6 CTSRA-J I
I C O " j
t 6
M 1
r & A C K R A _ j |
1
CTS or A C K not immediately
preceeded by RTS or D A T A
. Inconsistent SA/Duration
Figure A.2 STA 3 observes STA 1 as hidden through the timing constraints places
on CTS and A C K frames. In (a) STA 3 is not hidden from STA 1, hence Duration
and SA fields are consistent. In case (b) where STA 3 is hidden from STA 1, the
cached Duration and SA fields will be inconsistent with the values observed in the
CTS or A C K frames
Hidden Node Detection Mechanisms 181
® (a) Nodes 1, 2, 3 Not Hidden
Known STA List
~4
1
R T S SA = 1
CTS RA _ j
DATAS A = 1
ACK R A _ i
' R T S S A = 1 i
iCTS*A = > i JDATASA = 1
|ACKRA=1 i
Node 3 Action
STA 1 updated in Known STA List
_ STA 1 matched in known STA list
- STA 1 updated in known STA list
- STA 1 matched in known STA list
(b) Nodes 1 and 3 hidden
® ! R TS S A = I
| CTSRA=1
1 1
! \
! DATAS A = 1 J J
! ACKR A = 1 ji
l CO
co
I CO
' 55
tl
CTSRA=1 |
ACKR^S^ = j \
1
Node 3 Action
STA 1 not added to known STA list
STA 1 not matched in known STA list
STA 1 not added to known STA list STA 1 not matched in known STA list
STA 1 identified by STA 3 as hidden STA
Figure A.3 STA 3 observes STA 1 as hidden as no RTS or DATA frames have been
received to update known neighbour list in STA 3. In (a) STA 3 is not hidden from
STA 1, hence STA 3 is able to identify and maintain STA 1 in the known neighbours
list. In case (b) where STA 3 is hidden from STA 1, the R A within CTS and A C K
frames will not be found in the known neighbours list
Hidden Node Detection Mechanisms 182
which also contains both the SA and hidden node address within the notification
in the cache of known neighbours, is therefore the common node between the
hidden node pair. The common node is then able to apply a fairness control
mechanism as outlined in Sections 6.4.1, 6.4.2, and 6.4.3.
Appendix B
Additional Fairness Algorithm Results
B.l 3 Node TCP Results
The results presented here are the result of additional trials investigating the
performance of each fairness control mechanism with T C P traffic streams. Both
the static and dynamic scenarios of Sections 6.5.3 and 6.5.5 employed.
B.l.l Static topology Results
Additional TCP based results for the 3 node static hidden terminal topology of
Section 6.5.3.
Figure B.l p-Persistence on Backoff Countdown
Figure B.2 Probabilistic Discard
Figure B.3 Enhanced CTS Suppression
183
Additional Fairness Algorithm Results 184
(a) Data Trace - Evolution p = 0
1800
^ 1600
§• 1400
§ 1200
8 1000 co Q T3
>
'8 <D
OC
(b) Data Trace - Access Times (5 = 0
800
600
400
200
0
1400
•vr 1200 CD
m 1000
w 800 a ° 600
I 400 o CD
tt 200
1.1 1
0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0
Connection A Connection B
5 10 15 20 25 30
Time (sec)
(c) Data Trace - Evolution p = 0.25
Connection A Connection B
10 15 20 25 30
Time (sec)
(e) Jain's Fairness Index
r.vifl9^.ffiv^py^M^--^"!xv4y!^-ft-!
Jam's Index: Average 0.96
10 15 20
Time (sec)
(g) Jain's Fairness Index vs p
25 30
0.35
•o
o Z
3 n sz o> 3 p -C
•n
<» <a a E o Z
0.4
0.35
OH
0.2b
0.2
0.15
0.1
0.05
0
Q. J= D> 3
P CD CO
ra E o z
0.3
0.25
0.2
0.15
0.1
0.05
0
Connection A Connection B
5 10 15 20 25
Time (sec)
(d) Data Trace - Access Times p = 0.25
Connection A Connection B
10 15 20 25
Time (sec)
(f) Normalised Throughput p = 0.25
Connection A Connection B
10 15 20
Time (sec)
(h) Normalised Throughput vs p
Connection A Connection B *
30
30
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
Figure B.l 3 node topology, static scenario for p-Persistence on Backoff Countdown
algorithm using TCP - optimal 0 = 0.25
Additional Fairness Algorithm Results 185
(a) Data Trace - Evolution (3 = 0 1800
_ 1600
J[ 1400 § 1200
8 1000 CO
° 800 § 600 CD
8 400 = 200
(b) Data Trace - Access Times p = 0
1200
« 1000 S. m CO
c3 Q T3 CD
> CD U CD
CE
800
600
400
200
Connection A Connection B
5 10 15 20 25 30 Time (sec)
(c) Data Trace - Evolution p = 0.40
Connection A Connection B
A • r
< > • ' '
•
•
10 15 20 25 Time (sec)
(e) Jain's Fairness Index
30
CD n
E 2
z 1
13 O
z
X CO 13
CO CO CD
CO LL
1.1 1
0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 C
r-
-AA
•
) 5
1 1 1 1
\ /us j A^IM n w « u n i i Nrv**rnr\!$
\ U v \Z V j
Jam's Index: Average 0.96 t *
10 15 20 25
Time (sec)
(g) Jain's Fairness Index vs p
3
3 Q. SI O) 3 H 13
"co E o z
0
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
o sz O)
s sz CD CO CO
F o
0.5 0,45
0.4 0.3b 0.3
0.25
0.2 0.15
0.1 0.05
0
Connection A Connection B
Connection A Connection B
10 15 20 25 Time (sec)
(f) Normalised Throughput p = 0.40
5 10 15 20 25 30 Time (sec)
(d) Data Trace - Access Times p = 0.40
30
Connection A Connection B «
5 10 15 20 25 30 Time (sec)
(h) Normalised Throughput vs p
Connection A Connection B »
_3_ 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
P
Figure B.2 3 node topology, static scenario for Probabilistic Discard algorithm using TCP - optimal 0 = 0.40
Additional Fairness Algorithm Results 186
(a) Data Trace - Evolution p = 0
3000
« 2500 % * 2000 co ™ 1500 t> CD
>
8 CD
DC
1000
500
CD
%. m CO
ra D ID
§ CD
o CD
DC
1000
900
800
700
600
500
400
300
200
100
1.1 1
0.9 x 0.8 CD
S 0.7 • 0.6 8 0.5
co LL
0.4 • 0.3 0.2
0.1 r 0
Connection A Connection B
5 10 15 20 25 30
Time (sec)
(c) Data Trace - Evolution p = 0.55
Connection A Connection B
-•
./ y f
,,y jy
. *
/
L
'
/
J'" f
'
1
/ " •
y y-y •
s ^
-
-
,
10 15 20
Time (sec)
(e) Jain's Fairness Index
25 30
f. A f\ ,*\A », r\ n n^M A
Jam's Index: Average 0.89
10 15 20
Time (sec)
(g) Jain's Fairness Index vs p
25 30
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
1 •
3 r> sz O)
o s:
<i> CO CO b o Z
0.4
0.35
OH 0.25
0.2
0.15
0.1
0.05
0
0.5
0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
(b) Data Trace - Access Times p = 0
Connection A Connection B
5 10 15 20 25 30
Time (sec)
(d) Data Trace - Access Times p = 0.55
Connection A Connection B
K « *XM X
10 15 20 25
Time (sec)
(f) Normalised Throughput p = 0.55
30
Connection A Connection B
10 15 20
Time (sec)
(h) Normalised Throughput vs p
Connection A Connection B
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
P
Figure B.3 3 node topology, static scenario for Enhanced CTS Suppression algorithm
using T C P - optimal 0 = 0.55
Additional Fairness Algorithm Results 187
B.l.2 Dynamic Topology Results
Additional TCP based results for the dynamic 3 node topology of Section 6.5.5
Figure B.4 p-Persistence on Backoff Countdown
Figure B.5 Probabilistic Discard
Figure B.6 Enhanced CTS Suppression
Additional Fairness Algorithm Results 188
(a) Data Trace - Evolution p = 0
1600
~ 1400 CD
S. 1200 m * 1000 CO
g 800 "§ 600
'8 400 CD
*• 200
(b) Data Trace - Access Times p = 0
1600
~ 1400 £ •5. 1200 m * 1000 co ™ 800 •a
I 8 400 CD 01 200
600 •
CD T3
Connection A * Connection B »
5 10 15 20 25
Time (sec)
(c) Data Trace - Evolution p = 0.05
Connection Connection
(g) Jain's Fairness Index vs p
30
•a
o Z
XX .. .
_ ™ ™ ™
Connection A Connection B
13 O
z
0
1.1 1
0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2
0.1 0 C )
'
I
5
5 10 15 20 25 30
Time (sec)
(e) Jain's Fairness Index
, rr "Ti! \ t | \ j:; .* a ,=1 .'
Jain's Index: Average 0.96 , *
10 15 20 25 3
Time (sec)
3 Q. SZ CD 3
8 s: 1-
•a CD CO
n E o Z 0
0.3
0.25
0.2
0.15 r
0.1
0.05 r
0
CO 3
s SZ
^-T3 CD CO
ro E b z
0.3
0.25
0.2
0.15
0.1
0.05 I-
0
5 10 15 20 25
Time (sec)
(d) Data Trace - Access Times p = 0.05
Connection A Connection B
Connection A Connection B
10 15 20
Time (sec)
(h) Normalised Throughput vs p
Connection A Connection B
0.1 0.2 0.3
P 0.4 0.5
30
10 15 20 25 30
Time (sec)
(f) Normalised Throughput p = 0.05
0.6
Figure B.4 3 node topology, static scenario for p-Persistence on Backoff Countdown
algorithm using TCP - optimal 0 = 0.05
Additional Fairness Algorithm Results 189
(a) Data Trace - Evolution p = 0
CO CO Q 13
§ CD O CD
DC
1800 ^ 1600 CO
§, 1400 § 1200
1000 800
600
400
200
0
1400
IS 1200 CD
§• 1000 *: co 0! Q
CD O CD
DC
800
600
400
200
2.
1.1 i
0.9 0.8 0.7 I-0.6 0.5 0.4 0.3 0.2 0.1 0
1
x 0.9 S - 0.8 r
CD
S 0.6 CD
* 0.5
0.4
Connection A Connection B
25 5 10 15 20 Time (sec)
(c) Data Trace - Evolution p = 0.30
30
Connection A Connection B
5 10 15 20 Time (sec)
(e) Jain's Fairness Index
"JA rf "> ft.
Jajn's Index: Average 0.90
10 15 20 Time (sec)
(g) Jain's Fairness Index vs p
25 30
100 Frame Window
0.1 0.2 0.3 0.4 0.5 0.6
P
T3 O
z
0.3
3 C) SZ n> 3 n sz T) CD (0 ffl h Z
0 25
0.2
0.15
0.1
0.05
0
(b) Data Trace - Access Times p = 0
Connection A Connection B
5 10 15 20 25 Time (sec)
(d) Data Trace - Access Times p = 0.30
Connection A Connection B
10 15 20 25 Time (sec)
(f) Normalised Throughput p - 0.30
Connection A Connection B
30
30
5 10 15 20 25 30 Time (sec)
(h) Normalised Throughput vs p
Connection A Connection B »
0.1 0.2 0.3 0.4 0.5 0.6
P
Figure B.5 3 node topology, dynamic scenario for Probabilistic Discard algorithm
using T C P - optimal 0 = 0.30
Additional Fairness Algorithm Results 190
(a) Data Trace - Evolution p = 0
1800
^ 1600 • CO
£ 1400 • §_ 1200 • 8 1000 •
800 •
600
400 I-
200
0
•a
I '<u o CD
DC
Connection A Connection B
1400
•S? 1200 CD
§ 1000
«T 800 co ° 600 2 5 400 o CD
C 200
1.1 1
0.9 x 0.8 CU
f 0.7 • « 0.6 • 8 0.5
I 0.4 "- 0.3
0.2 0.1 0
25 5 10 15 20
Time (sec)
(c) Data Trace - Evolution p = 0.25
Connection A Connection B
10 15 20 25
Time (sec)
(e) Jain's Fairness Index
( (I ft/W M
Jain's Index: Average 0.96
10 15 20 25
Time (sec)
(g) Jain's Fairness Index vs p
30
30
O
z
3 O
T3 CD .CO
« E h_
o z
0.3
0.2
0.1
(b) Data Trace - Access Times p = 0
Connection A Connection B
Connection A Connection B
_j i_
Connection A Connection B
\ / \
5 10 15 20 25 30 Time (sec)
(d) Data Trace - Access Times p = 0.10
0 5 10 15 20 25 30
Time (sec)
(f) Normalised Throughput p = 0.25
5 10 15 20 25 30
Time (sec)
(h) Normalised Throughput vs p
Connection A Connection B «
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
P 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Figure B.6 3 node topology, dynamic scenario for Enhanced C T S Suppression algo
rithm using T C P - optimal 0 = 0.25