University of Southern California Dissertations and Theses

Towards interference-aware protocol design in low-power wireless networks

(USC Thesis Other)

Towards interference-aware protocol design in low-power wireless networks

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content TOWARDS INTERFERENCE-AWARE PROTOCOL DESIGN
IN LOW-POWER WIRELESS NETWORKS
by
Dongjin Son
A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Ful¯llment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(ELECTRICAL ENGINEERING)
December 2007
Copyright 2007 Dongjin Son
Dedication
To my Parents, my wife Namhee, and my son Ian.
ii
Acknowledgments
IthasbeenmygreathonorandpleasuretoworkwithProf. BhaskarKrishnamachari
and Prof. John Heidemann at the University of Southern California (USC) and Infor-
mation Sciences Institute (ISI). They have worked together with me and guided me
withwiseandintelligentideasfortheentireperiodofthisthesiswork. Icouldlearnnot
only great research skills, but also their passion, joy for the work and deep love for their
family and people. I do not know how to thank them enough for their great support
and encouragement throughout my study at USC. They have been and will be my role
models for my research with their incredible enthusiasm and sincereness.
IwouldliketogivemyspecialthanktoProf. CauligiS.Raghavendra, Prof. Ahmed
Helmy, Prof. Gaurav S. Sukhatme, and Prof. Cyrus Shahabi for serving on my qualify-
ingexamanddissertationcommitteesandgivingmeinvaluablesuggestionsandsupport
for my dissertation. I would like to also thank my colleagues in Autonomous Networks
Research Group (ANRG) and ISI Laboratory for Embedded Networked Sensor Exper-
imentation (ILENSE). I have been very fortunate to meet these talented and sincere
people with great enthusiasm for their work. They have been very supportive with my
work and always inspire me to greater e®orts. I would like to give my special thanks
iii
to Prof. Wei Ye and earlier graduates Shyam and Marco for their marvelous help and
friendship. Outside of research group members, I want to appreciate my beloved mem-
bers in Good Shepherds Korean Christian Group at USC for their encouragement and
love. Finally,Iwanttothankmyfamily,especiallymymotherandwifewhohavealways
supported and believed in me, and my dear Lord.
iv
Table of Contents
Dedication ii
Acknowledgments iii
List Of Tables ix
List Of Figures xi
Abstract xv
Chapter 1 Introduction 1
1.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.1 Reliable Communication . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.2 Concurrent Communications Under Interference . . . . . . . . . 5
1.3 Thesis Statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.4 Summary of the Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.5 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.6 Organization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Chapter 2 Background on Wireless Communications 12
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2 Signal Propagation and Link Quality Models . . . . . . . . . . . . . . . 13
2.3 Interference Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4 Channel Capture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
Chapter 3 Related Work 19
3.1 Low-Power Wireless Channel Characteristics . . . . . . . . . . . . . . . 19
3.1.1 Understanding . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.1.2 Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.1.3 Evaluating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.2 Transmission Power Control . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.2.1 Topology Control . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
v
3.2.2 Channel Capacity . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.3 Concurrent Communication . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.3.1 Capture E®ects . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.3.2 MAC for Concurrent Communication . . . . . . . . . . . . . . . 28
Chapter 4 Transmission Power Control and Blacklisting based Low-
Power Wireless Link Quality Control 30
4.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.2 Transmission Power Control on a Single Wireless Link . . . . . . . . . . 32
4.2.1 Experiment Methodology . . . . . . . . . . . . . . . . . . . . . . 32
4.2.2 The E®ects of Transmission Power Control on Link Quality . . . 33
4.2.3 E®ects of Di®erent Transmitters on Link Quality . . . . . . . . . 36
4.2.4 E®ects of Di®erent Receivers on Link Quality . . . . . . . . . . . 37
4.2.5 E®ects of Wireless Link Distance (Path Loss) on Link Quality . 38
4.2.6 E®ects of Node Location (Multi-Path) on Link Quality . . . . . 39
4.2.7 The E®ects of Time (Environment) on Link Quality . . . . . . . 43
4.2.8 Selecting a Transmission Power Level . . . . . . . . . . . . . . . 44
4.3 PCBL: Transmission Power Control with Blacklisting. . . . . . . . . . . 45
4.3.1 Key Characteristics and Bene¯ts of Our Proposed Scheme . . . . 46
4.3.2 Basic PCBL Algorithm: Optimization Prior to Routing . . . . . 47
4.3.3 On-demandTransmissionPowerOptimizationforEachLong-lived
Communication . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.3.4 Experiment Results for a Single Data Flow . . . . . . . . . . . . 50
4.3.5 Experiment Results for Multiple Data Flows . . . . . . . . . . . 55
4.3.5.1 PCBL vs M-BL with a Collision Avoidance Scheme . . 56
4.3.5.2 PCBL vs M-BL without a Collision Avoidance Scheme 59
4.3.5.3 Multi-hop, Multi-°ow Experiments . . . . . . . . . . . . 62
4.3.5.4 Lessons from the PCBL and M-BL Comparison . . . . 63
4.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
Chapter 5 Experimental Study of Concurrent Transmission in Wireless
Sensor Networks 66
5.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
5.2 Experimental Methodology . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.2.1 Hardware and Software . . . . . . . . . . . . . . . . . . . . . . . 69
5.2.2 Measurement Design . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.2.3 A Regression Model Mapping SINR to PRR. . . . . . . . . . . . 73
5.3 Experimental Study of Single Interferers . . . . . . . . . . . . . . . . . . 74
5.3.1 Interference and Black-Gray-White Regions . . . . . . . . . . . . 74
5.3.2 SINR Threshold and Transmitter Hardware . . . . . . . . . . . . 79
5.3.3 E®ects of Location on PRR and SINR . . . . . . . . . . . . . . . 82
5.3.4 E®ect of Sender Signal Strength on the SINR Threshold . . . . . 83
5.3.5 Testbed Experiments . . . . . . . . . . . . . . . . . . . . . . . . . 87
vi
5.3.6 Modeling the SINR Threshold . . . . . . . . . . . . . . . . . . . 89
5.4 Experimental Study of Multiple Interferers . . . . . . . . . . . . . . . . 92
5.4.1 Joint Interference Estimator . . . . . . . . . . . . . . . . . . . . . 92
5.4.2 Additive Signal Strength Assumption . . . . . . . . . . . . . . . 93
5.4.2.1 Two interferer experiments . . . . . . . . . . . . . . . . 94
5.4.2.2 Additivity and RIS levels . . . . . . . . . . . . . . . . . 95
5.4.2.3 Additivity with Additional Interferers . . . . . . . . . . 96
5.4.3 Variation in JRIS Measurements . . . . . . . . . . . . . . . . . . 97
5.4.4 E®ects of Joint Interference . . . . . . . . . . . . . . . . . . . . . 99
5.5 Preliminary evaluation of 802.15.4 Radio . . . . . . . . . . . . . . . . . . 100
5.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
Chapter 6 Evaluating the Importance of Concurrent Packet Communi-
cation 105
6.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
6.2 Motivating Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
6.3 Mathematical Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
6.3.1 Power Setting for CCability . . . . . . . . . . . . . . . . . . . . . 113
6.3.2 Topology Condition for CC . . . . . . . . . . . . . . . . . . . . . 115
6.3.3 CCability with Limited Power Range . . . . . . . . . . . . . . . . 116
6.3.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
6.4 Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
6.4.1 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
6.4.2 De¯ning Regions of Placement and the CCable Ratio . . . . . . 120
6.4.3 Fixed Transmission Power Cases . . . . . . . . . . . . . . . . . . 122
6.4.4 User Controllable Parameters . . . . . . . . . . . . . . . . . . . . 123
6.4.4.1 Location Change . . . . . . . . . . . . . . . . . . . . . . 125
6.4.4.2 SINR Threshold . . . . . . . . . . . . . . . . . . . . . . 127
6.4.4.3 Comparing Fixed and Dynamic Power Control . . . . . 128
6.4.4.4 Power Control Granularity . . . . . . . . . . . . . . . . 130
6.4.5 Uncontrollable and Environmental Parameters . . . . . . . . . . 130
6.4.5.1 Path Loss Exponent . . . . . . . . . . . . . . . . . . . . 131
6.4.5.2 Path Loss Variance . . . . . . . . . . . . . . . . . . . . 132
6.4.6 Capturable Region . . . . . . . . . . . . . . . . . . . . . . . . . . 133
6.5 Testbed Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
6.5.1 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
6.5.2 Results from the Outside Scenario . . . . . . . . . . . . . . . . . 138
6.5.3 Results from the Inside Scenario . . . . . . . . . . . . . . . . . . 140
6.6 2D Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
6.6.1 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
6.6.2 CCability with Optimal Power Setting . . . . . . . . . . . . . . . 144
6.6.3 CCability at Di®erent Power Settings . . . . . . . . . . . . . . . 145
vii
6.7 Making CCable Decisions in Practice . . . . . . . . . . . . . . . . . . . . 146
6.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
Chapter 7 Towards Concurrent Communication in Wireless Networks 148
7.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
7.2 MAC Protocol Designs . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
7.2.1 Today's Practice: CS-RTS/CTS with Simple Power Control . . . 151
7.2.2 A Upper Bound on Performance with an Oracle . . . . . . . . . 151
7.2.3 Exploiting Power Control and Channel Capture . . . . . . . . . . 153
7.2.4 GAPC: Gain-Adaptive Power Control and Capture . . . . . . . . 154
7.2.5 Comparing MAC Protocols . . . . . . . . . . . . . . . . . . . . . 156
7.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
Chapter 8 Future Work and Conclusions 160
8.1 Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
8.2 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
References 164
viii
List Of Tables
4.1 Key ¯ndings from transmission power control study . . . . . . . . . . . 31
4.2 Packet reception rate (%) for the links between node 11 and node 31 at
increased transmission power levels (dBm) . . . . . . . . . . . . . . . . . 34
4.3 Standard deviations for the links with di®erent levels of PRR . . . . . . 44
4.4 Brief PCBL algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.5 PCBL algorithm for a long-lived communication . . . . . . . . . . . . . 50
4.6 The energy consumption di®erence in packet transmission compared to
the PCBL scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.7 Four experiment scenario comparison . . . . . . . . . . . . . . . . . . . . 54
4.8 PCBL and M-BL comparison with a collision avoidance . . . . . . . . . 57
4.9 PCBL and M-BL comparison without a collision avoidance . . . . . . . 60
4.10 PCBL and M-BL comparison with a collision avoidance . . . . . . . . . 62
5.1 Key ¯ndings from concurrent transmission study . . . . . . . . . . . . . 68
5.2 SINR-to-PRR mapping with region distinction . . . . . . . . . . . . . . 78
5.3 Parameter ¯
1
and 95% con¯dence intervals for two di®erent locations. . 81
5.4 ¯
1
, SINR threshold (SINR
µ
), and R
2
(goodness-of-¯t) value for sender SRC2
for SRC1-SRC2 pair experiments when we use a ¯xed ¯
0
. . . . . . . . . . . 84
ix
5.5 Comparison of JRIS(e) and JRIS(m) metric for JRIS estimation at two
di®erent individual RIS levels . . . . . . . . . . . . . . . . . . . . . . . . 96
6.1 Key ¯ndings from concurrent communication study . . . . . . . . . . . . 106
x
List Of Figures
1.1 E®ects of wireless communication models . . . . . . . . . . . . . . . . . 2
1.2 PC104 testbed at USC/ISI and a snapshot of link quality: weak, asym-
metric, and good links . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Stargate/Stayton testbed at USC/ISI and a snapshot of interference
caused by one sender's packet transmission . . . . . . . . . . . . . . . . 5
4.1 E®ects of transmitter hardware change . . . . . . . . . . . . . . . . . . . 35
4.2 E®ects of receiver hardware change . . . . . . . . . . . . . . . . . . . . . 37
4.3 E®ects of link distance change . . . . . . . . . . . . . . . . . . . . . . . . 39
4.4 RSS change at di®erent receiver positions . . . . . . . . . . . . . . . . . 40
4.5 E®ects of node location change . . . . . . . . . . . . . . . . . . . . . . . 40
4.6 PRR, RSS and noise level change over time at di®erent transmission
power levels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
4.7 Standard deviation change for di®erent PRR value . . . . . . . . . . . . 44
4.8 Packetdeliveryrate(PDR)fromtheexperimentswith¯vedi®erentpower
control schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.9 Topology changes with di®erent power control schemes . . . . . . . . . . 53
4.10 Stargate node locations for the multiple data °ow experiments. . . . . . 55
4.11 Packet delivery rate (PDR) from the experiments with three data °ows 57
xi
5.1 Overview of the testbed with experimental methodology used for time
synchronization, signal strength and PRR measurement . . . . . . . . . 70
5.2 E®ects of varying only one sender's transmission power level on the PRR
and RSSI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.3 Packet reception rate at di®erent RSS combination from two concurrent
senders. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
5.4 E®ectofdi®erentpacketsenderandinterfererhardwareonSINR-to-PRR
relationship . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.5 E®ect of di®erent packet sender and interferer location on SINR-to-PRR
relationship . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.6 Experiments with wide range of sender and interferer signal strength.
Sender: SRC2, Interferer: SRC1. . . . . . . . . . . . . . . . . . . . . . . 83
5.7 SINR-to-PRRrelationshipcategorizedfordi®erentreceivedsignalstrength
levels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
5.8 SINR threshold for 0.9 PRR change at di®erent received signal strength
level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.9 Testbed experiments with 12 neighbor nodes . . . . . . . . . . . . . . . 87
5.10 E®ects of introducing new capture-aware simulation model . . . . . . . 88
5.11 The number of CCable link comparison between the two simulation models 90
5.12 Two interferer experiments varying the strength of interference from one
interferer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
5.13 Experiment results with two interferers (IFR1 and IFR2) causing equiv-
alent RIS at the receiver . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.14 Frequency distribution of JRIS measurement values for two interferer
experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.15 SINRthresholdchangeswithdi®erentnumberofinterfererswhichchanges
the received interference strength . . . . . . . . . . . . . . . . . . . . . . 98
xii
5.16 SINR to PRR relationship: preliminary results with CC2420 radio . . . 102
6.1 Two concurrent packet communications at three di®erent locations . . . 107
6.2 Example scenario with two concurrent packet sender-receiver pairs vary-
ing R1-S2 distance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
6.3 CCability for di®erent schemes. SINR values are measured at -10 dBm
¯xed Tx power . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
6.4 The CCable transmission power relationship between two senders . . . . 112
6.5 Simulation topology: two sender-receiver pairs . . . . . . . . . . . . . . 118
6.6 Simulation result with area index at ¯xed transmission power . . . . . . 119
6.7 CCable regions with di®erent ¯xed transmission power levels . . . . . . 122
6.8 CCableregionsandoptimalTxPowerfortwosenders(S1andS2)varying
distance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
6.9 Comparison of RSS from two concurrent senders (S1 and S2) . . . . . . 125
6.10 CCable regions with di®erent SINR
µ
. . . . . . . . . . . . . . . . . . . 127
6.11 CCable region comparison with and without power control. . . . . . . . 128
6.12 CCable regions with 8 levels (Mica Z) and 25 levels Tx power control. . 129
6.13 CCable regions with di®erent path loss exponent . . . . . . . . . . . . . 130
6.14 CCable regions with di®erent path loss variance . . . . . . . . . . . . . . 133
6.15 Capturable Regions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
6.16 MicaZ experiment topology with two sender-receiver pairs . . . . . . . . 135
6.17 CCability in the outside testbed experiment as S2 is moved (presented
together with the expectation from simulation with our proposed formula)136
6.18 Experimental results at di®erent S2 locations with variable transmission
powers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
xiii
6.19 CCability from the inside scenario . . . . . . . . . . . . . . . . . . . . . 141
6.20 2D simulation results with optimal transmission power settings . . . . . 142
7.1 MAC power control comparison of CCability . . . . . . . . . . . . . . . 152
7.2 Comparison of CCable area with limited power levels for ¯ve MACs . . 156
7.3 CC plus capture rate comparison with limited power levels . . . . . . . 157
xiv
Abstract
Wireless sensor networks deployed densely for ¯ne-grained monitoring often experi-
ence high channel contention from concurrent packet transmission. The main cause of
concurrencyinthesenetworksistheburstynatureofevent-basedtra±c. Co-channelin-
terference from simultaneous transmission is inevitable in wireless communication, and
abetterunderstandingofitisessentialforreliableande±cientcommunicationprotocol
design.
The central thesis of this dissertation is that interference-aware communica-
tion protocol design can signi¯cantly improve the performance of low-power
wireless networks. We substantiate this thesis through four studies. The ¯rst is a
systematic experimental evaluation of low-power wireless links involving variable trans-
mission power. The second is an implementation of transmission power control with
blacklisting which ensures link reliability and low-interference. The third is an analysis
of the e®ects of concurrent packet transmission under single and multiple interferers.
And the ¯nal study is an evaluation of how power control can be used to improve
concurrent communication.
xv
These experimental studies and analyses provide fundamental insights and new
guidelines for interference-aware communication protocol design. They show that sig-
ni¯cant improvement in utilizing constrained wireless channel resources is possible by
embracing concurrency through power control and channel capture.
xvi
Chapter 1
Introduction
1.1 Overview
Theinstabilityandunpredictabilityoflow-powerwirelesschannelsduetofading,multi-
path, hardware non-ideality, and interference makes it extremely challenging to develop
e±cientandreliablecommunicationprotocols. Bythetermlow-power,wemeantheuse
oflowcostRFtransceiverswhichrequireverylowpowerconsumptionandarenormally
optimized for short-range communication with small number of components [10, 11].
Due to their limited cost and size, their overall performance can be di®erent from more
delicate and expensive high-power devices. However, early studies in the context of
mobileadhocnetworksandwirelesssensornetworkshaveoftenbeenbasedonidealized
and simpli¯ed simulation approximations. While such approximations can be valuable
at establishing bounds on performance and exploring algorithms at a high level, they
can provide misleading results if not used carefully [35, 44, 53].
The most common approximations incorporated in prior wireless communication
protocol design regarding link quality and interference are: (a) distance-based binary
link quality estimation, which assumes perfect reception within a ¯xed communication
range, and (b) packet collision from any concurrent transmission, which considers any
1
Figure 1.1: E®ects of wireless communication models
simultaneous transmission within a communication range from the receiver to be a
packet collision.
Recentempiricalstudieshaveshownthelimitationsofsuchsimpli¯edandconserva-
tive approximations and identi¯ed several important characteristics of low-power wire-
lesschannels[6, 25,47,93]. Duetodynamiclink qualityof low-powerwireless networks
understandingandusingrealisticmodelsofwirelesschannelisessentialforreliablecom-
munication protocol design. Understanding the e®ect of co-channel interference from
simultaneoustransmissionisimportantfordesigningmoree±cientcommunicationpro-
tocol in densely deployed wireless networks.
We empirically study and understand low-power wireless channels by varying many
related conditions and tunable parameters, especially transmission power, with and
without co-channel interference. Then we provide useful models and guidelines for
interference-aware protocol design, and also propose some protocols that improve the
performance of wireless communications through transmission power control and con-
current packet communication.
2
Figure1.2: PC104testbedatUSC/ISIandasnapshotoflinkquality: weak,asymmetric,
and good links
1.2 Motivation
We present the relationship between the communication models adopted for wireless
networks and communication protocol design in Figure 1.1. The model of the network
topology depends on the adopted models. This a®ects the design of communication
protocols, applications, and their theoretical performance analysis. Low-power wireless
networks have been studied and designed using conventional models and assumptions
that are rather simpli¯ed, idealized, and conservative. This is mainly due to the com-
plexities inherent in more accurate models and the lack of understanding of wireless
communication under di®erent realistic conditions.
The di®erences between the reality and commonly used models and assumptions
deterioratetheperformanceofcommunicationprotocolsinrealsystems. Inthissection,
we introduce two motivations and related problems that initiated our research and that
become the main objectives pursued throughout our study.
1.2.1 Reliable Communication
Ifwegiveanywirelessnetworkacloserlook, wecaneasily¯nddynamiccommunication
links of varying quality. Figure 1.2 shows a snapshot information of the link quality in
3
packet reception rate (PRR) for every link in our PC104 [23] testbed; We de¯ne and
indicate two types of unreliable links with PRR metric in this ¯gure: weak links (with
less than 10% PRR) and asymmetric links (with a good link only in one direction).
In our testbed, every pair of PC104 nodes are connected through a reliable commu-
nication route, in which every communication link is classi¯ed as a good link. However,
itisoftenthecasethatunreliablelinksareutilizedinsteadofgoodcommunicationlinks
and the throughput of the network is badly a®ected in our testbed experiments. We
have tested with two variants of the directed di®usion [36, 38] routing protocols; one
phase pull and two phase pull. While two phase pull checks the quality of the links in
communication route between the sender and sink in both directions, one phase pull
checks the link quality of the route only in one direction from the sink to the source.
The end-to-end packet delivery rates with single data °ow (at the same experiment
settingasSection4.3.4)rangebetween43to58%withonephasepulldi®usionand72to
83% with two phase pull di®usion experiments without any link quality control scheme.
The performance drop for two phase pull routing mainly originates from the weak links
and dynamic link quality and one phase pull loses many packets from asymmetric link
qualities in selected route.
As the experiment results show, having unreliable links could be worse than having
none of these links at all when bi-directional communication is required and there is a
good communication route to use between two nodes. Therefore, unreliable links need
to be either converted to good links or prevented from use.
If we use transmission power control, we can elevate the quality of wireless links.
However, it comes with some side e®ects. First, increased transmission power may
generate new weak links with extra signal power that is not yet enough to build new
reliable communication links. Secondly, increased transmission power uses up more
network capacity with stronger interfering signal strength. There is a trade-o® between
4
90
71
70
188
91
193
87
88
72
186
190
85
83
194
75
185
183
192
184
81
191
73
89
76
187
Figure 1.3: Stargate/Stayton testbed at USC/ISI and a snapshot of interference caused
by one sender's packet transmission
thelinkqualityandinterferencelevel,buttheactuale®ectoftransmissionpowerandits
optimalsettingisnotobviousfordi®erentsituations. Ourearlytestbedexperiencewith
some conventional wireless communication protocols motivates the study of low-power
wireless link with transmission power control and its e®ects; mainly to take the most
advantage from limited channel resource while improving communication reliability by
carefully adjusting transmission power.
1.2.2 Concurrent Communications Under Interference
When we think of a wireless communication, we often imagine it over clear channel
withoutanyinterference. However,itisnottrueinmanycasesregardlessofthestrength
of the interference, and it is too ideal and wasteful condition to ensure.
Figure1.3showsasnapshotofinterferencecausedbyonenode'spackettransmission
toitsneighbornodes. Interferencelevelisindicatedintwolevelsbasedonthemeasured
received signal strength (RSS) from this packet sender; Solid line means the receiver
can always receive a packet from this interferer meeting the minimum RSS requirement
for successful packet reception, and dotted lines means the lower than required RSS for
packet reception which provides intermediate link quality (i.e., packet reception rate
5
between 10% and 90%). Each packet transmission in wireless domain causes co-channel
interference to the network and the level of interference is important by deciding the
probability of packet collision under the concurrent transmission or the probability of
controlpacketreceptionusedbysomechannelacquisitionprotocolspriortodatapacket
transmission.
While interference from packet transmission is endemic in wireless communication,
there is no empirical understanding of the e®ects of interference on packet communi-
cation (i.e., packet delivery) in low-power wireless networks. It is often the case that
wireless communication protocols are designed to avoid any concurrent packet commu-
nication within the same channel to prevent any possible packet collision caused by
interference from other simultaneous transmissions. This conservative approach low-
ers the utilization of the channel while help improving reliable packet communication.
However,itisnotobvioushowmuchbene¯twecanobtainbyhavingbetterunderstand-
ing of co-channel interference and increasing the number of concurrent communications
under interference; particularly with how much additional complexity. In our study, we
want to build up the knowledge about the low-power wireless channel under co-channel
interference from concurrent packet transmission and provide a new design paradigm
towards concurrent communication for communication protocols.
1.3 Thesis Statement
The central thesis of the proposed dissertation is that interference-aware commu-
nication protocol design can signi¯cantly improve the performance of low-
power wireless networks.
We substantiate our thesis statement through the four studies summarized in the
followingsection. Thegeneralstrategywefollowisthatwe¯rstexertourselvestoobtain
a better empirical understanding of low-power wireless links under many conditions.
6
Thorough analysis and consequent modeling provides the basis of practical design and
implementation of interference-aware protocols.
1.4 Summary of the Work
First,weperformasystematicexperimentalstudytounderstandlow-powerwirelesslink
characteristics under di®erent settings that are closely related to wireless link quality
(in Section 4.2). Not only we do study the causes of dynamic link quality at static
power level, but we also study the e®ects of variable transmission power on the link
quality control and evaluate several link quality metrics. Even though there have been
some prior experimental studies [25, 47, 84, 93], there was no systematic study with
variabletransmissionpoweranditsroleasawirelesslinkqualitycontrollerundervarious
scenarios. This study provides empirical understanding of low-power wireless channel
under transmission power control, which is the key component of interference-aware
protocol design.
Second, we propose a transmission power control with blacklisting (called PCBL)
scheme based on the insights from the low-power wireless link study under variable
transmission power (in Section 4.3). PCBL addresses the problems caused by unreli-
able communication links prevalent in the low-power wireless networks by either adding
or subtracting transmission power to control signal and interference strength, and by
blacklisting unresolved unreliable links with power control. PCBL is implemented and
evaluatedonarealtestbedbothwithsingleandmultipledata°ows. PCBLprovidesen-
ergye±cient, low-interference(i.e., betterspatialreuse), andreliablemulti-hopwireless
communication under traditional collision avoidance scheme which do not allow plural
communications within the same channel.
Third, we study the e®ects of Concurrent packet transmission (CTX) |by two or
moresenderswithinmaximumcommunicationrangeofeachother'sreceiveratthesame
7
time over the same channel| under the case with both single and multiple interferers.
Nearly all wireless MAC protocols are designed today with the very conservative as-
sumption that concurrent transmissions should be prevented, because sender-receiver
pairswithinradiorangesendingonthesamechannelwillcorrupteachother'scommuni-
cation. While recent work has suggested that channel capture e®ects can be signi¯cant
in reality, we thoroughly study the e®ect of concurrent packet transmission with real
testbed experiments. We introduce a new metric and quantify the signi¯cance of cap-
turee®ectsontheabilitytohaveconcurrentcommunicationsamongtwosender-receiver
pairs that are within range of each other.
Wehavecon¯rmedthecapturee®ectandtheexistenceoftheSINRthresholdwhich
ensures the successful delivery of the strongest packet under the concurrent packet
communication situations with single and multiple interferers. We have also observed
some discrepancy between theory and e®ects of measured interference in low power
RF transceiver hardware. This study provides a better understanding of the e®ects
of concurrent transmissions, especially on packet delivery, and suggests richer interfer-
ence models and useful guidelines for interference-aware protocol design and improves
performance analysis of higher layer protocols.
Finally, we evaluate the importance of communication protocol design towards con-
current communication (CC)|allowing successful transmissions by two senders within
maximum communication range of each other's receiver at the same time over the same
channel|inlowpowerwirelessnetworks. Thelargescaleanddistributednatureoflow-
power wireless networks often requires multi-hop wireless communication. Traditional
collision avoidance schemes, such as RTS/CTS exchange, are proposed to solve hidden
node terminal problem for single hop, direct communication. However, this can be an
inappropriate solution for multi-hop communication. In a multi-hop wireless commu-
nication, resolving the exposed node problem, which de¯ne as the case with prohibited
8
feasible concurrent communication, is as signi¯cant issue as providing a collision-free
packet communication, because it can signi¯cantly lower the network throughput.
Instead of following conservative approach of traditional collision avoidance, we ¯rst
mathematically model the conditions for successful concurrent packet communication
(or packet collision) considering node topology and available power control capability.
BasedonthismodelandnewmetricforCC,weestimatethebene¯tsfromallowingcon-
current packet communications with thorough simulations and complementing testbed
experiments. We can observe a large range of topological settings where concurrent
communicationispossible(CCableregion)usinganoptimalCCabletransmissionpower
control scheme called oracle, which is based on our mathematical modeling of CCable
conditions. The Oracle scheme provides on upper bound performance of CCability, but
it requires accurate topology and tra±c information. As a practical alternative for the
oraclescheme,weintroduceasketchofnewMACprotocol,calledGain-AdaptivePower
Control (GAPC), which provides signi¯cant bene¯t with only local information.
1.5 Contributions
The contribution of our study is two-fold. First, we perform systematic measurement
studies in real testbeds to understand and characterize low-power wireless channel bet-
ter. Several related environmental parameters and tunable controls are investigated,
and some useful guidelines of interference-aware protocol design are identi¯ed (summa-
rized in Tables 4.1, 5.1, and 6.1). Extensive experimental studies provide fundamental
knowledge about wireless channels and contemporary low-power RF transceiver hard-
ware. Our empirical studies are especially focused on the e®ect of transmission power
control and the e®ects of concurrent packet transmission because they are key elements
of our interference-aware protocol design.
9
Second, we introduce practical models and protocols based on the observations and
analysis obtained from testbed experiments. We propose a new model for interference
simulation (in Section 5.3.6) and identify topological conditions for concurrent commu-
nication (in Section 6.3.2) and its corresponding optimal transmission power setting (in
Section 6.3.1). We design two interference aware protocols; First one is called PCBL,
which combines transmission power control with link level blacklisting (introduced in
Section4.3). Thisminimizesnetworkinterferencewhileimprovingreliabilityofcommu-
nication links in dense networks. The second one is called gain adaptive power control
(GAPC), which improves the probability of successful concurrent packet communica-
tion only with minimum local information (introduced in Section 7.2.4). We believe
this work establishes an essential direction for future wireless communication proto-
col, especially MAC design, away from the use of carrier sense and RTS/CTS to avoid
concurrent communication, instead embracing concurrency through power control and
channel capture.
1.6 Organization
The remainder of this dissertation is organized as follows. We ¯rst provide some back-
ground on wireless communication related to our study and discuss related work re-
spectively in Chapter 2 and 3. In Chapter 4, we study low power wireless links under
variable transmission power in real testbed and propose a link quality control scheme
which combines power control with link blacklisting. In Chapter 5, we present em-
pirical study of concurrent transmission in low-power wireless networks to understand
the e®ects of co-channel interference from simultaneous packet transmissions. Next, we
study the feasibility of concurrent communication on the same channel, and propose a
sketch of new medium access control protocol which allows concurrent communication
10
and improves channel capacity in Chapter 6 and 7. Finally, we identify some remaining
challenges and discuss directions for future work and then conclude.
11
Chapter 2
Background on Wireless Communications
2.1 Introduction
Wireless communication is a general trend in networking personal computers as well as
prevalent ad-hoc, embedded systems including sensor nodes. The main advantage and
attraction of wireless communication is its convenience and the freedom and mobility
obtained from untethered communication. Deployments can be easier and faster, and
management of the system can be more convenient especially in harsh and dynamic
environment.
However, as an unstable and insecure communication medium, the wireless domain
bringshugecomplexityincommunicationprotocoldesign. Butthiscomplexityhasbeen
mostly hidden under simpli¯ed approximations.
Recently, there are increasing number of research e®orts that revisit popular, tradi-
tional assumptions and unveil the complexities of wireless communication in order to
reduce the wastage of network capability due to simpli¯ed and inaccurate approxima-
tions, and to take advantage of special characteristics that exist in the wireless domain.
12
In this chapter, we discuss some foundational details mostly developed by other
researchers to provide background knowledge of wireless communication; especially fo-
cusingonthetopicrelatedtoourmainwork. Wewillalsoprovidesomebasicde¯nitions
we will use in the rest of the thesis.
2.2 Signal Propagation and Link Quality Models
Signal propagation in wireless communication is complicated due to signi¯cant change
inreceivedsignalmainlyfromlarge-scalepathlossandsmall-scalemultipathfading[65]
The simplest model is the free space propagation model. This predicts the received
signal strength solely based on the distance between the sender and receiver when there
exists clear line-of-sight path. However, the high variance in the e®ect of distance
makethismodelunsuitableforunderstandingandanalyzingintricatedetailsofwireless
communication.
The most commonly used model for many analytical studies on wireless network is
the exponential path loss model with log-normal fading (presented below). This model
re°ects the fading around the same distance from environmental diversity.
PL(d)
dB
= PL(d
0
)
dBm
+10nlog(d=d
0
)+X
¾
dB
(2.1)
P
r
(d)
dBm
= P
t
dBm
¡PL(d)
dB
Here P
t
and P
r
are the transmission and reception power in dBm. The sender-
receiver distance is d, and d
0
is the reference distance for path loss (PL). X
¾
is the
variance in path loss due to multipath fading, modeled as Gaussian random variable
with zero mean and standard deviation ¾
dB
. This model de¯nes the path loss and the
received signal strength (RSS) at the receiver at a given transmission power level. It
13
is hard to use this model to estimate link quality in real systems because radio and
environment speci¯c parameter values are necessary. However, this model is very useful
to create a realistic simulation of wireless channel and to analyze experimental results
to understand wireless link in more detail.
Link quality estimation is closely related topic to the signal propagation model.
Even though it is intuitive to use the signal propagation model to estimate the link
quality based on the estimated received signal strength, one of the most popular way of
link quality estimation has been only based on the maximum communication range of
the transmitter. This assumes 100% packet reception within this ¯xed communication
range and 0% packet reception outside of this range.
There are more realistic link quality metrics introduced to overcome problem of bi-
nary type link quality approximation, often simply based on link distance [98]. In the
testbed, successful packet reception rate (PRR) can be actually measured with multi-
ple packet transmission e®orts. Measured link quality is a good, realistic link quality
estimator, but not ideal metric to use due to its signi¯cant overhead, especially for
the dynamic environment. Some radios provide a link quality information; for example
CC2440 radio provide link quality indicator (LQI) values as well as RSS measurements.
Close correlation between RSS and PRR is experimentally presented in [16].
We will discuss further about some novel link quality estimators in related work (in
Section 3.1.3).
2.3 Interference Models
Developingarealisticinterferencemodelhasasigni¯cantimportanceinlow-powerwire-
less sensor networks. This is because of the high probability of concurrent packet trans-
missionsthatbecomeamajorsourceofinterference. Finegrainedmonitoringinwireless
sensornetworksnecessitatesdensedeploymentofinexpensivenodes,anditincreasesthe
14
chance of concurrent packet transmission. Moreover, sensor-actuated or user-initiated
(or sink-initiated) bursty nature of packet transmissions, which normally travel mul-
tiple hops, also increase channel contention and the probability of concurrent packet
transmissions within a shared wireless medium.
Interference models are introduced and used mainly for two purposes. First, to
de¯ne the conditions for a successful packet communication (or a packet collision), and
set the boundary of communication and interference. Second, they are used to estimate
and represent the level of interference on the network or on each node based on the
network topology either with or without considering tra±c condition.
There are again two kinds of conditions and requirements for a successful packet
communications. First are the conditions which decide the success of packet reception
at the receiver, and the second are the conditions implied by the upper layer proto-
col to meet their design requirements. For example, some MAC protocols only require
the clear channel at the receiver (called Receiver Con°ict Avoidance), while some re-
quires the clear channel at the sender as well as receiver for a feedback reception (called
Transmitter-Receiver Con°ict Avoidance). While we are interested in understanding
wireless channels and improving protocol design rather than studying the e®ect of in-
terference under a certain protocol, we mainly focus on the ¯rst condition in our work.
Therearetwowidelyutilizedmodelsforsuccessfulwirelesscommunication: thepro-
tocol model, and the physical model [32]. In the protocol model, which is implemented
by many state-of-the-art wireless network simulators, concurrent transmissions from
any node within a given range (referred to as the interference range) of the intended
receiver is considered to cause a packet collision that results in the loss of a packet from
its corresponding sender. This model is solely based on the distance between the nodes,
but many recent measurement studies [6, 12, 25, 47, 77, 93] provide conclusive proofs
that communication distance cannot be a good estimator of wireless link quality.
15
However, in the physical model, the decision of packet reception is based on the
signal-to-interference-plus-noise-ratio (SINR) instead of the physical distance between
the nodes. The physical model is capable of providing more complicate and realistic
modeling and analysis of interference e®ects because it can distinguish the e®ect of
di®erent number of interferers and its interfering signal strength which protocol model
cannot.
Design of wireless communication protocol grounds on a certain selected interfer-
ence model. While there are increasing number of interference-aware protocols over all
networkstacksbesideslinklayer, arealisticinterferencemodelis essentialfactor forthe
success of wireless communication protocols.
2.4 Channel Capture
In wireless communication, the phenomenon of strongest packet reception under con-
current packet transmissions within a same channel is called capture e®ect. This is also
known as physical layer capture [80] or FM capture [48].
There is a minimum ratio value between the signal and interference plus noise
strength which ensures the reception of the strongest packet. We de¯ne this as a signal-
to-interference-plus-noise ratio threshold (SINR threshold), and the following formula
presents this relation.
SINR
dB
=
P
t
(S)G
S;R
P
i2IFR
P
t
(i)G
i;iR
+N
R
¸SINR
µR
dB
(2.2)
Here S and R indicate an intended sender and its receiver. IFR means the set of
interferers,andiandiRindicateasingleinterfereranditscorrespondingreceiver. P
t
(s)
denotes the transmission power of the sender or interferer s, and G
s;r
denotes the gain
16
of the link between the sender s and receiver r. N
R
means the ambient noise level at
the intended receiver R. SINR
µR
means the SINR threshold required for the intended
receiver R.
SINR threshold varies with implementation details of modulation scheme. Modu-
lation techniques that increase the signal's resistance to interference, such as spread
spectrum technique, can greatly reduce the SINR threshold value, and improves the
successful packet communication under co-channel interference.
With a given SINR threshold value and interference plus noise level, a node can
control its channel accessability with transmission power control. Extra transmission
power can improve the SINR value and power control capability is very important in
exploiting capture e®ect by controlling the signal and interference levels.
One caveat of capture e®ect is that timing of packet transmission could be an im-
portant factor to the success of channel capture. This is related to the implementation
of physical layer which detects and delivers a packet to the upper layer of the proto-
col stack. Normally once a node detects the preamble of a packet it stops preamble
searching until the end of the detected packet reception time. Therefore if the strongest
packet arrives at the receiver later than weaker packet, there is no chance of detect-
ing strongest packet at the receiver [83]; Similarly if the strongest packet arrives ¯rst,
collision of other weaker packets cannot be detected at the receiver. Therefore, imple-
mentation of a packet detection scheme (such as [43, 83]) is required in the context of
simultaneous packet transmission, especially with partial overlapping of transmission
time, to utilize capture e®ect. To simplify and ignore this caveat from our study, we
always use or assume synchronized packet transmission from concurrent senders.
Fromseveralrecentempiricalstudies[39,43,79,83]withlow-powerRFtransceivers,
capturee®ectshavebeenexperimentallytestedandtheimplicationofthisphenomenon
on protocol design have been presented. In our work, we suggest a interference-aware
17
protocol design which takes advantages of capture e®ects. We use transmission power
control to improves the gains from exploiting capture e®ects.
18
Chapter 3
Related Work
There are three strands of research in the literature that are related to our work: (a)
recent studies of low-power wireless links based on testbed experiments (b) the large
literatureontransmissionpowercontrolinwirelessadhocandsensornetworks(c)study
of capture e®ect and wireless MAC protocols for concurrent communication.
3.1 Low-Power Wireless Channel Characteristics
3.1.1 Understanding
Therehavebeenmanyrecentstudiesempiricallypresentandaddressthecharacteristics
of low-power wireless links. Ganesan et al. present a large scale (about 150 nodes)
empirical study on a mote-based sensor network; identifying the presence of weak links,
link asymmetry and studying their impact on the performance of simple °ooding [25].
Zhao and Govindan perform a detailed study of wireless links with motes under
di®erent environments, distances, modulation schemes etc. and identify the existence
of a large gray region in distance between connected and disconnected regions where
links are highly variant and unreliable [93]. The transitional (i.e., gray) region is also
observed by Woo et al. who focus on the problem of neighborhood table management
19
and propose mechanisms to blacklist unreliable neighbors in order to provide reliable
delivery [84].
Cerpa et al. introduce a system that can be used to study the characteristics of
lowpowerwirelesschannel. Byusingtheproposedsystem, calledSCALE,theytestthe
spatialandtemporalcorrelationwithlinkquality. Theycouldnot¯ndaclearcorrelation
between the link distance and any of packet delivery rate, temporal variation of link
quality, and asymmetric links [6]. Later, they further studied temporal properties of
low power wireless links using SCALE system, and proposed two new routing protocols
based on their observations from the large data set [8].
Kotz et al. revisits some of the most common assumptions considered in wire-
less communications with empirical measurements taken from outdoor routing exper-
iments. They provide some recommendations to consider as well as the weakness of
these assumptions [44, 45]. We also performed systematic empirical studies especially
to understand better of low power wireless links under transmission power control (in
Section 4.2.2) and co-channel interference (in Chapter 5, Section 6.5) .
3.1.2 Modeling
Based on growing experiences with low-power radio channels, new wireless communi-
cation models have been also proposed. Cerpa et al. identi¯ed the properties of group
links as well as individual links and proposed a series of individual wireless link models
andwirelessnetworkgeneratorsthatcanbeusedfordesignandsimulationoflow-power
wireless network protocols [7].
Zuniga and Krishnamachari [98, 99] focus their study on the causes of link unre-
liability and asymmetry observed in the transitional (e.g., gray) region of low-power
20
wireless link quality. They quanti¯ed the impact of transitional regions and also pro-
posed a link quality simulation model mainly based on the link distance considering
related environmental and radio parameter settings.
Zhou et al. study the irregularity of propagated RF signals on di®erent direction.
Thisstudyprovidesausefulradioirregularitymodel(RIM)re°ectingrealisticlow-power
wireless channel behavior based on empirical results and analyze its impact on upper
layer protocols together with multiple solutions to address radio irregularity [94, 95].
In our study, we introduced several models for low power wireless links including
a regression model mapping SINR to PRR (in Section 5.2.3), SINR threshold simula-
tion model (in Section 5.3.6), and topology and power condition model for concurrent
communication for two senders (in Section 6.3). These models are mainly based on
empirical results from either Mica2 and MicaZ motes, but they can still provide some
useful guidelines and checklists for protocol design with di®erent types of hardware
platforms.
3.1.3 Evaluating
Therearemultiplewaysofevaluatingandquantifyingsinglewirelesslinkandmulti-hop
communication route with several proposed metrics.
Lal et al. introduce a link quality metric called link ine±ciency, which is the mean
number of transmission required for the link to deliver a packet. They propose a rule
to determine this link ine±ciency (or link cost) metric based on the measured signal-to-
noise-ratio (SNR) value. Using this rule together with their prior experimental results,
it requires only a few measurements of the channel to estimate link cost metric [47].
Aguayo et al. perform link measurements to study the causes of packet loss in a
802.11 mesh network (Roofnet). They experimentally study several packet loss related
factors such as SINR (which they refer to as S/N ratio), transmit bit-rate, interference,
21
andmulti-pathfading. Theirexperimentalresultsshowawide(greaterthan3dB)gray
regionofSINRwithintermediatevaluesofpacketdeliveryprobabilityevenforthesame
receiver. They argue that, for this reason, SINR cannot be used as a reliable predictor
of delivery rate in 802.11 networks [3].
DeCoutoetal. performedexperimentalstudyon802.11networkandshowthatwhy
multi-hop routing based on minimum hop count is not always an optimal answer [13].
They introduce a new link and path metric, called ETX (expected number of transmis-
sions), to improve the delivery performance of multi-hop packet routing [12].
Draves et al. compared three di®erent link quality metrics for multi-hop packet
routing, including ETX, per-hop round trip time (RTT), and per-hop packet pair delay
(PktPair), with minimum hop count metric on 802.11a network. There experimen-
tal results show that ETX performs best for static networks and minimum hop count
performs best with mobile nodes [19].
Seadaetal. proposeandmathematicallyanalyzedanewmetricwhichistheproduct
ofpacketreceptionrate(PRR)anddistance(PRR£DIST).Theirproposedmetrictakes
into account both link distance and quality of the link. With automatic repeat request
(ARG), PRR£DIST metric was optimal choice for a forward metric in their analysis.
Their claim is supported with simulations and testbed experiments [73].
SrinivasanandLevisevaluateanda±rmthevalueofRSSIasalinkqualityestimator
with CC2420 radio. They compared RSSI with link quality indicator (LQI) provided
by 802.15.4 radio [81]. RSSI and LQI are physical layer information for link quality
estimation, and recently Fonseca et al. proposed a new wireless link estimation which
use protocol independent feedbacks from multiple layers combining physical, link, and
network layer information. Proposed estimation technique can signi¯cantly improve
multi-hop packet delivery ratio and its cost [70].
22
Inourstudy,wemainlyuseeitherPRRorSINRmetrictoevaluatelinkquality. Our
experimental results show that PRR provides good link quality estimation with good
number of probe packets. This is because it is actual link quality measurement which
includes environmental and hardware factors in it. However, it requires costly process
in terms of packet exchange overhead and time, so we relate SNR/SINR value, which
is a physical layer metric, to PRR, which is a link layer metric (in Section 4.2.4, 5.3.1),
andcombinesblacklisting(inSection4.3)toonlyprovidereliablelinkstonetworklayer.
3.2 Transmission Power Control
Transmission power control plays a key role in interference-aware protocol design by
controllingtheintensityofthesignalandinterferencestrength. Theliteratureontrans-
mission power control, though quite vast, has hitherto focused on slightly di®erent
concerns and objectives. Two main research interests of the related work on power
control are energy e±cient topology control and channel utilization.
3.2.1 Topology Control
One of the main purpose of prior studies with transmission power control was topology
control. Theprimarygoalsoftopologycontrolareensuringdesirablenetworkconnectiv-
ity in energy-e±cient way and extending network lifetime. Kubisch et al. proposed two
distributed algorithms which ensures the network connectivity and increases the life-
time of the network [46]. A topology control scheme based on directional information is
discussedin[82], wheretransmissionpowerisincreaseduntilatleastoneneighbornode
is found in each direction. Kawadia and Kumar proposed several clustering and routing
protocols with power control mainly to improve channel capacity while reducing the
23
energy consumption for multi-hop packet communication [41]. Power controlled topol-
ogy control mechanisms based on geographical location information are also presented
in [52, 64, 69, 72] to ensure network connectivity with minimal energy consumption.
3.2.2 Channel Capacity
Theothermainpurposeoftransmissionpowercontrolwastoimprovechannelutilization
with a better spatial reuse.
In cellular network, transmission power control has been a key mechanism for im-
proving network capacity from early 1980's for channelized [24, 31, 59, 91] and CDMA
system [26, 61]. Zander proposed a distributed iterative power control algorithm based
on signal-to-interference ratio which signi¯cantly improves network capacity [90] and
synchronous and asynchronous convergence of iterative power control algorithms has
been also presented [24, 88]. Rulnick and Bambos proposed a power control scheme
to provide required level of quality of service for mobile terminals under time-varying
interference [71]. Later, Bambos and Kandukuri advanced this power control scheme
by both considering interference level from its transmission at increased power and the
backlog size from withholding its transmission under strong interference [4].
Inmulti-hoppacketcommunicationnetwork,thereisatrade-o®betweenthenumber
of hops in packet delivery and the level of interference when we adjust packet transmis-
sionpower. ElBattetal. proposedapowermanagementprotocoltoimproveend-to-end
throughput in wireless network by trading increased number of hops with reduced col-
lision and interference. They introduce a concept of cluster and adjust transmission
power for de¯ned clusters to reduce interference and improve network throughput while
providing appropriate network connectivity [21]. Similarly, transmission power control
24
schemestoincreasethenetworkthroughputbycontrollingthenumberofhopsinmulti-
hop packet delivery are also discussed in [28, 49, 55]. Park and Sivakumar proposed
transmission power control scheme which considers load condition in the network [62].
There are more than a few MAC protocols which utilize transmission power control,
mainly for energy saving, but which also help to improve spatial reuse of the network.
Basic power control approach is to use minimum required power for data packet com-
munication while preventing collision by transmitting RTS/CTS packets at maximum
power level [2, 29].
Jung and Vaidya proposed a power control MAC (PCM) which use di®erent trans-
mission power for control and data packet. PCM is mainly proposed to improve energy
e±ciencywhile addressingthe hiddennodeproblem from asymmetric powerallocations
fordi®erentdatapacketcommunications. PCMusesperiodicbusytonetoavoidpacket
collision from hidden terminals [40]. There are also some other protocols utilize busy
tonesforcollisionavoidanceanduseRTS/CTStypecontrolpacketstoestimateoptimal
transmission power levels [54, 86].
Transmission power control can cause unfairness problem as well as hidden node
problem. Sheth and Han proposed a reactive power controlled MAC protocol (SHUSH)
toresolvebothoftheseproblems. Thisprotocolgiveshigherpriorityfortheinterrupted
node which use extra power for the RTS and ¯rst frame of the data packet to silence
interferer[75]. ShihandChenalsoaddresshiddennodeproblemcausedbytransmission
power control mechanism. They propose a MAC protocol, called Collision Avoidance
Power Control (CAPC), which assigns extra transmission power to resist from possible
interference[76]. Thiscanpossiblyimprovethepossibilityofconcurrentcommunication,
but the main purpose of their additional power allocation is to mitigate hidden node
problem from asymmetric transmission power distribution.
25
Even though there have been extensive research e®orts with transmission power
control in wireless communication, there are few empirical studies that consider trans-
missionpowercontrol. Westudy(inChapter4)thee®ectsoftransmissionpowercontrol
on wireless link quality on real sensor network testbed with Mica2 motes. Based on the
insight obtained from testbed experiments, we propose a power control scheme with
link blacklisting to improve link reliability and energy e±ciency [77]. Later Lin et al.
proposed an adaptive transmission power control (ATPC) protocol based on the em-
pirical measurements from the MicaZ motes with 802.15.4 radios [11] which reacts to
the temporal change of the link quality with explicit on-demand feedback packets [50].
However, both of these works do not explicitly study the bene¯ts from transmission
power control for concurrent packet communication.
3.3 Concurrent Communication
As described in previous sections, a great deal of prior work has empirically studied
low-power wireless links channel. These works have improved our understanding of the
wireless communication and also provides better communication models and metrics.
However,mostofthesestudiesdonotconsiderthesituationwithco-channelinterference
from simultaneous transmission by multiple senders. In fact, a design goal of most
currentmedia-accessprotocolshavebeentoavoidconcurrenttransmissions,oftenwithin
a two-hop neighborhood of the sender. In this section, some of the works, which are
closely related to our study with concurrency in transmission, will be discussed mainly
focusingoncapturee®ectandMACprotocoldesigntowardsconcurrentcommunication.
26
3.3.1 Capture E®ects
Inwirelesscommunicationcommunity,capturee®ect,wherebyapacketwiththestronger
signal strength can be received in spite of a collision, has been a well known phe-
nomenon [18, 30, 48, 87] and various capture models have been proposed and evaluated
mostly for ALOHA networks [33, 60, 67, 68, 74, 92, 97] and recently for some 802.11
networks[34, 42]. The most common model use a constant threshold (called capture ra-
tio)foreachmodulationandcodingschemewiththeratioofthesignalstrengthandthe
summation of interference strength. However, these are primarily theoretical studies.
In densely deployed wireless sensor networks, concurrent packet transmission is en-
demic, and recently there have been more than a few empirical studies that explore the
implications of concurrent transmission. One recent paper by Whitehouse et al. [83]
does address wireless link quality in the presence of concurrent transmissions. They
propose a technique to detect and recover packets from collisions taking advantage of
capture e®ects. Their scheme works by allowing the detection of preambles even during
packet reception. They study the performance of the proposed scheme through exper-
iments with a single interferer and show that the simplistic protocol model (in which
the communication range is chosen to be the interference range) signi¯cantly overesti-
mates interference and can result in ine±cient MAC design. Our study (presented in
Chapter 5) complements their work by quantifying the SINR conditions under which
the capture e®ect can be observed (that are the conditions under which their proposed
scheme shows performance gains).
Kochutet al. empiricallystudycapturee®ectin802.11bandshowthatthestronger
signal can still capture a channel even when it does not arrive ¯rst at the receiver if
it is still earlier than the end of the ¯rst start frame delimiter of weaker signals. They
introducesome¯xesforwirelessnetworksimulatorsconsideringtheirnewcapturemodel
tomakethemmorerealistic[43]. However, unliketheultimatefocusinourwork, which
27
is having more concurrent communications, they both study the case where multiple
transmitters send to a common receiver mainly to test capture e®ects.
Zhou et al. proposed a radio interference detection protocol (RID) which can be
usedtomeasureandshareinterferenceinformationamongtheneighboringnodes. High
poweredcontrolpacketisintroducedforinterferencedetectionandcollectedinformation
is stored in the table and can be used for interference-aware protocol design [96].
Reis et al. introduce two physical layer models that provide e®ective prediction of
the probability of packet delivery under interference from concurrent transmission [66].
These models are based on the RF measurements from real 802.11 testbed.
Recently, Moscibroda et al. analytically and empirically study the inaccuracy and
ine±ciencyofprotocoldesignbasedongraph-basedmodel[58],andanalyzethecapacity
of wireless network with a physical model allowing concurrent communications [56, 57].
3.3.2 MAC for Concurrent Communication
Medium access control protocol manages channel access from multiple contenders, and
several MAC protocols are suggested to improve the number of concurrent communica-
tion taking advantages of capture e®ects in the wireless networks.
Acharya et al. suggested modi¯cation of 802.11 DCF to improve the spatial reuse
by allowing more concurrent communication. There are two main modi¯cations. First,
they add extra time space between the control and data packet. This allow time for
multiple RTS/CTS control packet exchange. Second, RTS/CTS packets include exact
time information for DATA and ACK packet. This extra information enables synchro-
nized transmission for multiple senders [1]. However, they do not utilize transmission
power control, and therefore topologies for concurrent communication is very limited.
Later, Muqattash and Krunz proposed a power control MAC protocol, called POW-
MAC. This protocol utilizes both transmission power control and extra time for a series
28
of control packet exchange before a data packet transmission, which they call access
window (AW).
Maniezzo et al. proposed an interference-aware (IA) MAC protocol which consider
measured SINR value for medium access decision. By measuring and including addi-
tional information about signal strength within their RTS/CTS type control packets,
they expect up to 30% performance enhancement compared to the basic 802.11 by im-
proving spatial reuse in the network. Their proposed scheme and analysis does not use
transmission power control and does not consider asymmetric link quality prevalent in
low-power radio [9, 51].
ElBatt and Ephremides introduce an algorithm which can identify a set of feasible
concurrent communication and appropriate transmission power settings to make these
communication successful. It uses two alternating phases of scheduling and power con-
trol. Scheduling phase selects a probable set of concurrent communications, and power
control phase ¯nds proper power for each communication. This protocol requires cen-
tral controller for scheduling phase and separate feedback channel to notify SINR value
measured at receiver to its corresponding transmitter [20].
Jamieson et al. [39] consider concurrent transmissions when they investigate MAC
protocolperformancebyturningonando®thecarriersensefunctionalityatdi®erentbit
rates in an 802.11 testbed. They argue that a capture-aware carrier sense mechanism
that considers the bit rates and SINR will improve network e±ciency. Our research
shares the same belief. We study SINR threshold for successful packet reception (in
Chapter 5), which provides useful background knowledge for the development of simi-
lar techniques for low-power wireless networks. Subsequently, we verify the feasibility
concurrentcommunication(inChapter6)andintroduceasketchofnewmediumaccess
control protocol (in Chapter 7).
29
Chapter 4
Transmission Power Control and Blacklisting based
Low-Power Wireless Link Quality Control
4.1 Overview
Protocoldesignanditsevaluationinlow-powerwirelessnetworkshaveconsideredsome-
what simpli¯ed and idealized wireless channel approximations. However, recent empir-
ical studies show that communication protocols in real system operation do not match
the results obtained from idealized simulations or analysis. Unstable link quality causes
dynamic network topology, ant this ends up with signi¯cant loss in performance and
di±culty in protocol design.
The central thesis of the work presented in this chapter (which appears in [16], [77])
is that e±cient control of the link quality is possible by combining transmission power
management with link blacklisting strategies. There has been extensive research on
transmission power control in wireless networks. However, to our knowledge, most
of these studies are based on theoretical analysis or simulations with idealized radio
models. In this chapter we instead take an experimental approach, thus capturing the
full complexities of radio propagation in our testbed. In addition, the primary foci of
prior studies have been the energy consumption and the network capacity gains from
transmissionpowercontrol; weprimarilyconsiderthereliabilityoftheresultingsystem.
30
Findings from single link experiments Section 4.2
Having unreliable links can be worse than having no links 4.2.2
Transmission (Tx) power is not an accurate link quality estimator 4.2.3
RSS is not a good link quality estimator for di®erent receivers 4.2.4
Link distance is not always a good link quality estimator 4.2.5
Node location signi¯cantly a®ects link quality due to multi-path 4.2.6
Link quality variation over time can be reduced with Tx power control 4.2.7
Tx power needs to be set high enough to reduce link quality variation 4.2.8
Findings regarding transmission power control with blacklisting Section 4.3
Blacklisting can satisfy protocol's idealized link quality assumptions 4.3.4
Power control can signi¯cantly improve the throughput of the network 4.3.5
Table 4.1: Key ¯ndings from transmission power control study
Our contribution in the work presented in this chapter is twofold. First, we provide
a thorough experimental study of how low-power wireless communication links behave
with respect to variable transmission power under di®erent settings. This gives us fun-
damentalbackgroundknowledgeofthee®ectsofpowercontrolwhichbecomesthemain
tunable parameter we use throughout the work in this dissertation. Second, we propose
a transmission power control scheme with blacklisting and evaluate its e®ectiveness in
link quality control under multi-hop packet delivery scenarios. We list key ¯ndings of
this chapter in Table 4.1.
Ourexperimentsinvestigatethepossiblereasonsoflinkqualityvariationandidentify
transmission power ranges where link quality shows high variation. Our observations
showthattheimpactoftransmissionpoweronqualityofagivenlinkisquitesensitiveto
manyfactorssuchasnodepositions,surroundingenvironment,andindividualhardware
di®erences. Wealso¯ndthatthequalityofeachlinkwithrespecttotransmissionpower
can change over time, and the dynamics of the variable power link quality are di®erent
for distinct links. We conclude that it is useful to develop a per-link quality control
mechanism that chooses a su±ciently high power to reduce link quality variation, while
using blacklisting to remove any links that cannot be made high-quality even with
power-control from the topology.
31
Based on our observations, we propose and evaluate a new transmission power con-
trol scheme called power control with blacklisting (PCBL). The distinguishing charac-
teristic of this scheme is its consideration of empirically determined link quality when
adjusting transmission power. It incorporates the following key elements: 1) packet-
based power control (considering both packet type and destination) 2) metric-based
link quality estimation 3) unreliable link removal (per link or per packet-based black-
listing).
The e®ectiveness of the transmission power control scheme is evaluated via further
testbedexperimentsthatconsiderbothsingleandmultiple°owscenarios(single-hopas
well as multi-hop). We also consider the performance both with and without collision
avoidance using RTS/CTS messages. In these experiments, we compare the PCBL
scheme with constant power schemes without blacklisting as well as maximum power
with blacklisting. We ¯nd that PCBL shows improved reliability and energy-e±ciency
under most settings.
4.2 TransmissionPowerControlonaSingleWirelessLink
In this section, we identify the aspects of low power RF wireless links that make many
previously proposed power control schemes di±cult to implement in practice. We per-
form systematic experiments on single wireless links varying several key parameters
under di®erent transmission power levels.
4.2.1 Experiment Methodology
The link quality measurements in our testbed show inconsistency for some links within
the transmission range. To identify the cause of this discrepancy and the e®ects of
transmission power change on the wireless link, we perform systematic experiments
varying some key parameters presumably related to the wireless link quality: hardware
32
di®erence, distance between the transmitter and receiver, locations of the nodes, and
time (i.e., surrounding environment change).
Our link quality experiments are performed on a Stargate [37] testbed with Mica2
motes which use CC1000 [10] radio operating at 433 MHz as a RF transceiver. The
Emstar [27] software platform is used for our experiments and data collection.
Experiment results present both packet reception rate (PRR) and received signal
strength (RSS) at the receiver given transmission power level at the sender. These
statistics are based on 50 packet experiments.
Thirteen di®erent transmission power levels ranging from -13 to 10 dBm are tested
in the indoor environment. We also vary node positions and the link distance between
the transmitter and receiver for some experiments. The link distance between the
transmitter and receiver is varied between 6 m and 20 m, and we present some selected
distances which show interesting results.
4.2.2 The E®ects of Transmission Power Control on Link Quality
The wireless link quality is closely related to the received signal strength and the trans-
mission power control can be used to adjust the quality of the communication links to
avoid asymmetric or weak links. We de¯ne the communication link between two nodes
as a weak link when the qualities of the links in both directions are below the required
reliability. We introduce a good link threshold (TH
g
), which states this reliability re-
quirement, based on PRR. We also de¯ne the link with reliable connection in only one
direction as a asymmetric link.
Table 4.2 shows the e®ect of transmission power increase on the quality of links
between node 11 and 31. The transmission power values in the table represent the
output powers of packet transmitters in dBm. Supported output power range for the
transceiver[10]ofmica2motesrangesbetween-20and10dBm. ThePRRvaluesinboth
33
Pwr 0 1 2 4 6 8 10 (dBm)
Link
11!31 54.3 86.3 92.4 100 100 100 100 (%)
31!11 0 27.2 83 85.7 96.8 100 100 (%)
Table 4.2: Packet reception rate (%) for the links between node 11 and node 31 at
increased transmission power levels (dBm)
directions are lower than 90% (which is our TH
g
for reliable communication) at default
transmissionpower0dBm. ThePRRofthelinkfromND11toND31(whichisdenoted
by LINK
11¡>31
) crosses over TH
g
at the transmission power 2 dBm and the PRR of
LINK
31¡>11
exceeds TH
g
at the transmission power 6 dBm. The symmetric and weak
links can become good quality with transmission power control as the example shown
in Table 4.2. Not only unreliable links can be converted to a reliable links, but new
communicationlinkscanbealsodiscoveredandusedforpacketdeliveryattheincreased
transmissionpowerlevel. Disconnectednodesinthesparsenodeareaofthenetworkand
in the harsh communication environment might be able to build their connection back
to the network at increased transmission power. The extra energy consumption needed
to convert an unreliable link to reliable link is often very small, especially when the
link quality is near the TH
g
value at default transmission power. The bene¯ts from the
converted reliable links often surpass the increased energy consumption. Our testbed
experiments (in Section 4.3) present the bene¯ts of converted, reliable wireless links
with proposed transmission power control scheme and discuss the relationship between
the link reliability and energy consumption.
When transmission power control is involved in the link quality management, def-
initions of asymmetric links and weak links should be updated. The reason is that
the classi¯cations made at default power may not be valid at other transmission power
levels. We could convert all of the ¯ve unreliable links identi¯ed in Figure 1.2 to good
links by increasing transmission power levels in our testbed experiments. Therefore, we
34
−12 −10 −8 −6 −4 0 2 4 6 8 10
0
10
20
30
40
50
60
70
80
90
100
Transmission power (dBm)
PRR (%)
From71
From72
From73
From88
From90
(a) Tx Power to PRR
−95 −90 −85 −80 −75 −70 −65
0
10
20
30
40
50
60
70
80
90
100
RSS (dBm)
PRR (%)
From71
From72
From73
From88
From90
(b) RSS to PRR
0 5 10 15 20 25 30
0
10
20
30
40
50
60
70
80
90
100
SNR (dB)
PRR (%)
From71
From72
From73
From88
From90
(c) SNR to PRR
Figure 4.1: E®ects of transmitter hardware change. Five di®erent transmitters to the
same receiver.
classify links into three di®erent types (i.e., good, weak, asymmetric) considering the
supported power control capability from the used hardware.
Even though ampli¯ed transmission power elevates the quality of wireless links, it
comes with some side e®ects. First, increased transmission power may generate new
weak links with increased signal strength that is not yet enough to build new reliable
links. We merge a link blacklisting together with our proposed transmission power
control scheme to address this problem. Secondly, increased transmission power uses
up more network capacity. There is a trade-o® between the improved link quality and
reduced network capacity.
Ourproposedtransmissionpowercontrolschemedoesnotalwaysincreasethetrans-
mission power. We only assign a minimum transmission power for the required commu-
nication reliability. Therefore, we lower the transmission power for the links which has
higher than necessary signal power at the receiver with a default transmission power.
Our proposed transmission power control scheme considers these two identi¯ed side
e®ects as well as the bene¯ts of transmission power control.
35
4.2.3 E®ects of Di®erent Transmitters on Link Quality
To see the e®ect of the transmitter hardware di®erences on wireless link quality, we
measure link qualities from ¯ve di®erent transmitters to a same receiver. From this
experiment, we want to study the signi¯cance of transmitter hardware non-ideality
from low-power, low-cost hardware, and its e®ect on link quality. Every transmitter
uses the same software settings and sends packets from the exact same location to the
staticreceiver. Bothtransmitterandreceiverarelocatedinthehallwayofthebuilding.
The link distance between the transmitter and receiver is 20 m and we measure the link
qualities in both PRR and RSS metric at the receiver for each transmitter varying the
transmission power.
FromtheexperimentresultsshowninFigure4.1(a), wecanseethatthelink quality
at the receiver becomes quite di®erent for di®erent transmitters even at the same node
location at the same output power level. When we plot the relationship between the
measured RSS and PRR in Figure 4.1(b), we can see that RSS to PRR relationship is
similar for the ¯ve di®erent transmitters. From this experiment, we can see that the
the di®erent link qualities observed with the PRR metric in Figure 4.1(a) results from
thedi®erentoutputsignalstrengthfromdi®erenttransmittersatthesametransmission
power setting. Hardware non-ideality causes this inconsistency in actual output power
among di®erent nodes. Figure 4.1(c) shows the signal-to-noise-ratio (SNR) to PRR
relationship. Thenoiselevelatthe same receiveris about the same andthe graph looks
very similar to RSS to PRR relationship, other than the changed x-axis unit.
Somepriorstudiesidenti¯edatransitional(orgray)regionwherePRRisdi®erentat
the same link distance [6, 25, 93]. Hardware non-ideality is factor causing transitional
region because it can distinguish link qualities at low transmission power level as the
Figure4.1shows. Fromtheseexperiments, wecanseethatthetransmissionpowerlevel
cannot be a good estimator of link quality due to hardware variance, and the level of
36
−12 −10 −8 −6 −4 −2 0 2 4 6 8 10
0
10
20
30
40
50
60
70
80
90
100
Transmission power (dBm)
PRR (%)
To71
To72
To73
To88
To90
(a) Tx Power to PRR
-95 -90 -85 -80 -75 -70 -65
0
10
20
30
40
50
60
70
80
90
100
RSS (dBm)
PRR (%)
To71
To72
To73
To88
To90
(b) RSS to PRR
0 5 10 15 20 25 30
0
10
20
30
40
50
60
70
80
90
100
SNR (dB)
PRR (%)
To71
To72
To73
To88
To90
(c) SNR to PRR
Figure 4.2: E®ects of receiver hardware change. Same transmitter to ¯ve di®erent
receiver.
link quality variance in real world is closely related to the default transmission power
selection.
4.2.4 E®ects of Di®erent Receivers on Link Quality
We also investigate the e®ects on the link quality when di®erent nodes are placed as
packet receivers. Similar to the experiments with di®erent transmitter hardware (pre-
sented in 4.2.3), we want to study the signi¯cance of receiver hardware non-ideality
from low-power, low-cost hardware, and its e®ect on link quality. We use the exact
same transmitter and receiver positions with 20 m link distance as our previous exper-
iments with transmitter change. Five di®erent receiver nodes are tested with a same
transmitter.
Figure 4.2(a) shows that link quality changes for di®erent receiver nodes even the
same transmitter transmits at the same output power level. RSS to PRR relationship
shown in Figure 4.2(b) still shows big di®erence for di®erent receiver nodes. This is
because the di®erence is not coming from the transmitter side.
37
When we compare the SNR to PRR relationship in Figure 4.2(c), there is much
smaller di®erence among di®erent receiver nodes. Therefore, we can see that the ob-
served di®erences in link qualities at di®erent receiver nodes can be attributed to the
di®erent level of ambient noise at the receiver.
InFigure4.1(a)and4.2(a),theareainthetransmissionpowerrangebetween-6and
10 dBm shows high variation of link qualities. In this transmission power range, the
quality of each link is di®erent at the same transmission power level and the di®erent
transmission power is required for each link to reach the same PRR level. We call the
rangeoftransmissionpowerthatgeneratesthiskindofvariationunreliable transmission
power range. Outside (either higher or lower side) of the unreliable transmission power
range, the link quality is the same regardless of the selected transmission power level.
The link quality di®erence observed in unreliable transmission power range can be
avoided by transmission power control in two ways. First, we can assign the same
transmission power outside of the unreliable transmission power range. Second, we can
assign a distinct transmission power for each link to provide a desired link quality level.
From the experiment results, we realize that the transmission power level or the
measured received signal strength (RSS) level may not be an accurate link quality
estimator for di®erent hardware. We can explicitly measure PRR with multiple packets
or use the SNR and PRR pair of information together for a better and e±cient link
quality estimation.
4.2.5 E®ects of Wireless Link Distance (Path Loss) on Link Quality
We empirically study the e®ects of the wireless link distance and transmission power
change on the link quality in this section. From this study, we want to examine the
correlation between link distance and link quality, and between transmission power and
38
−12 −10 −8 −6 −4 −2 0 2 4 6 8 10
0
10
20
30
40
50
60
70
80
90
100
Transmission power (dBm)
PRR (%)
6m
8m
10m
12m
14m
16m
(a) Tx Power to PRR
−95 −90 −85 −80 −75 −70 −65
0
10
20
30
40
50
60
70
80
90
100
RSS (dBm)
PRR (%)
6m
8m
10m
12m
14m
16m
(b) RSS to PRR
0 5 10 15 20 25 30
0
10
20
30
40
50
60
70
80
90
100
SNR (dB)
PRR (%)
loc1
loc2
loc3
loc4
loc5
loc6
(c) SNR to PRR
Figure 4.3: E®ects of link distance change
link quality. Experiments are performed in the hallway of the building where a clear
line of sight is available between the transmitter and receiver.
As the experiment result presented in Figure 4.3(a) shows, PRR changes as the link
distance and transmission power level change. The order of link distance that shows
better PRR at the same transmission power level is 6 m, 8 m, 12 m, 10 m, 16 m, 14 m
while this order changes among 10 m, 12 m, 16 m distances at di®erent transmission
power levels.
Thee®ectofpath-losscanbeobservedatarelativelycoarsegranularityeventhough
theorderisnotlineartothelinkdistance: closerdistance(6m,8m)showclearlybetter
link quality than longer distance (14 m, 16 m). The non-linear link quality order in our
experiment results can be attributed to the severe indoor multi-path e®ect. When we
plot the RSS to PRR and SNR to PRR relationship in Figure 4.3(b) and 4.3(c), we can
con¯rm that RSS as well as SNR is a good link quality estimator for the same hardware
pair. Experiment results show that link distance is not an accurate link quality metric.
4.2.6 E®ects of Node Location (Multi-Path) on Link Quality
To better understand the e®ects of node location and to see if how severe the multi-
path e®ect is, we performed a series of link quality measurements in the hallway of the
building.
39
2 4 6 8 10 12 14 16 18
−85
−80
−75
−70
−65
−60
−55
Receiver location index
RSS (dBm)
11m−P0
11m−P5
14m−P0
17m−P0
Figure 4.4: RSS change at 19 di®erent receiver locations at around 11,14,17m distance
betweenthetransmitterandreceiver. Thetransmitteruses0(and5onlyfor11m)dBm
transmission power (P0)
−12 −10 −8 −6 −4 −2 0 2 4 6 8 10
0
10
20
30
40
50
60
70
80
90
100
Transmission power (dBm)
PRR (%)
loc1
loc2
loc3
loc4
loc5
loc6
(a) Tx Power to PRR
−95 −90 −85 −80 −75 −70 −65
0
10
20
30
40
50
60
70
80
90
100
RSS (dBm)
PRR (%)
loc1
loc2
loc3
loc4
loc5
loc6
(b) RSS to PRR
0 5 10 15 20 25 30
0
10
20
30
40
50
60
70
80
90
100
SNR (dB)
PRR (%)
loc1
loc2
loc3
loc4
loc5
loc6
(c) SNR to PRR
Figure 4.5: E®ects of node location change
40
We placed a transmitter in the middle of the hallway and measured a link quality
in both PRR and received signal strength (RSS) at the receiver located at around 11,
14, 17 m distance from the transmitter. At each of these three chosen distance, we
draw a perpendicular line (about 147 cm long) connecting two walls of the hallway
and performed experiments placing the same receiver at 19 di®erent locations from the
leftmost position 1 to the rightmost position 19 (at every 7.5 cm interval) on this line.
Figure 4.4 shows that even for the links at around the same distance (i.e., on the
same line), the RSS changes depending on the speci¯c node positioning. The measured
range of RSS on the same line (i.e., RSS
max
¡RSS
min
) was between 7.9 and 12.7 dBm
in the four experiments results presented in this ¯gure. Multi-path e®ect causes this
severe link quality variation. We can even observe that there is a better link quality
at 17 m distance than 11 m distance at the same transmission power level (0 dBm in
this example) due to multi-path e®ect at some combination of receiver node placements
(e.g., when two receivers are placed at the position 5 of 11 m and 17 m distance).
When we compare the RSS change at 0 and 5 dBm transmission power at 11 m
distance, we can clearly see the improvement of the link quality with increased output
power. However,themeasurementsshowtheRSSimprovementvariesbetween3.51and
7.49 at 19 di®erent positions at the same amount of transmission power change.
Figure 4.5 shows a similar experiment results at 10 m link distance. A new pair
senderandreceivernodesisusedandthereceiverisplacedatsixdi®erentnodelocation
on the same line at two inch intervals. We can see wide variation of link qualities,
between -7 and 8 dBm transmission power levels.
Fromtheseresultswehaveshownthat(1)themulti-pathe®ectsaresevereinindoor
inbothcaseswithalineofsightlinkbetweenthetransmitterandreceiver,(2)severelink
quality variation can be expected with small movement of sensor nodes with low-power
wireless links, (3) we can expect signi¯cant link quality improvement in terms of PRR
41
1400 1600 1800 2000 2200 2400 0200 0400 0600
0
20
40
60
80
100
Time (hhmm)
PRR (%)
P−4
P0
P4
P8
P10
(a) Link quality (in PRR) change over time
1400 1600 1800 2000 2200 2400 0200 0400 0600
−100
−95
−90
−85
−80
−75
−70
Time (hhmm)
RSS (dBm)
P−4
P0
P4
P8
P10
NOISE
(b) Received signal strength (RSS) and noise level change over time
Figure 4.6: PRR, RSS and noise level change over time at di®erent transmission power
levels
42
with small increase in transmission power for the links in the unreliable transmission
power range, (4) the e®ect of transmission power control (i.e., the change of RSS at the
receiver)varieswithdi®erenthardware(i.e.,transmitterandreceiverpair)anddi®erent
node location.
4.2.7 The E®ects of Time (Environment) on Link Quality
Wecontinuouslymeasuredthelinkqualityofthetestbedfor18hoursbetween13:00pm
and7:30amtoseethechangeoflinkqualityovertime. Fromthisexperiments,wewant
to see the link quality variation from surrounding environment change and correlation
betweenthelinkqualityvarianceanddefaulttransmissionpowerlevel. Packetreception
rate (PRR), received signal strength (RSS) and ambient noise are measured every eight
minutes with 50 packets between -4 and 10 dBm transmission power levels.
Figure 4.6 presents link quality snapshots of communication link (LINK
72¡>91
) in
the Stargate testbed (shown in Figure 4.10). We can clearly see high variations in the
link quality at the same transmission power level. Especially, there is much higher
variation in link quality (both in PRR and RSS) during the daytime (between 14:00
and18:00)thanatnighttime(between2:00and6:00). Webelievethedynamicchanges
in surrounding environment during the daytime (mainly from the movement of objects
near the communication) causes much severe link quality variation.
When we compare the link quality variation between the cases with di®erent trans-
mission power levels in Figure 4.6(a), we can see that the level of link quality variation
is higher at di®erent transmission powers. When we use the transmission power level
outside unreliable transmission power range (e.g., transmission power of 8 dBm), this
link can be converted to a reliable link regardless of the time change. Even at the high
transmission power level, the change in RSS is similar to the low power cases as shown
in Figure 4.6(a). However, the PRR is less sensitive to the RSS variation at this level of
43
50 60 70 80 90 98
0
5
10
15
20
25
30
35
40
45
PRR
Stardard dev.
Figure 4.7: Standard deviation change for di®erent PRR value
PRR (%) 60-70 70-80 80-90 90-100 90-95 95-98 98-99 99-100
STDEV (%) 40.5 23 18.8 3.4 19.8 10.8 2.2 0.89
Table 4.3: Standard deviations for the links with di®erent levels of PRR
signal strength and can tolerate dynamic environmental change with the extra received
signal strength than required for reliable communication.
4.2.8 Selecting a Transmission Power Level
Figure 4.7 and Table 4.3 show the link quality variation for the links with di®erent
PRR levels in our seven day measurements in the testbed. We can see that the links
with lower than 90% PRR shows very high variation in link quality. When we take a
closer look at the links with link quality higher than 90%, the links with lower than
98% PRR still show relatively high quality variation. It is because the value is still
within the unreliable transmission power range where a small signal strength change
can signi¯cantly a®ect the link quality. Experiment results imply that it is better to
use only links with higher PRR close to 100% for a consistent link quality.
However, a blacklisting-only approach without transmission power control scheme
often chooses 80% or 90% PRR as a blacklisting threshold (BL
µ
) value. If we choose a
44
higherblacklistingthresholdtoincreasethelinkreliability,wewillwastemanylinkswith
stillreasonablygoodquality. Therefore, ablacklistingschemeunderstatictransmission
power su®ers from the frequent link quality changes. Fluctuation in link quality around
theblacklistthreshold(BL
µ
)mayresult in frequenttopology changethat can harmthe
performance of upper layer protocols.
With transmission power control scheme, it is feasible to use a higher blacklisting
threshold (e.g., 98% or even 100% PRR) because converting moderate quality links to
good, reliable links is often achievable with even small transmission power increase.
Wehaveusedapacketreceptionrate(PRR)asametrictoestimatealinkqualityand
collected a PRR at every transmission power level, and proper transmission power level
isselected basedonthemeasuredPRRat di®erenttransmissionpowerlevels. However,
it is hard to evaluate link quality with the PRR metric due to the high overhead of
PRR measurement from the repeated transmissions of multiple probe packets. If the
received signal strength (RSS) information is available at the receiver, we can use the
RSS information to maintain proper link quality easily and quickly in response to the
environmental change. The use of RSS for a link quality control is possible because the
PRRisproportionaltothemeasuredRSSwithinsomevariationrange(aswecanseein
Figure 4.1), and we can relate RSS value to the PRR for the same receiver. Therefore,
we can use the measured RSS as a good indicator of the link quality once we ¯gure
out the relationship between RSS and PRR for the given receiver (as we discussed in
Section 4.2.4).
4.3 PCBL: Transmission Power Control with Blacklisting
Basedontheexperimentalobservations, weintroduceanewtransmissionpowercontrol
mechanism.
45
4.3.1 Key Characteristics and Bene¯ts of Our Proposed Scheme
We propose a transmission power control scheme with the following key characteristics.
(1) Transmission power control for link quality management:
The primary purpose of transmission power control is to provide reliable commu-
nication links to the link users. Every communication including broadcast as well as
unicast is always using reliable links which meets user's requirement and expectation.
PCBLcanensurereliablebi-directionalcommunicationlinksandthecollisionavoidance
scheme implemented based on RTS/CTS handshakes could perform better by eliminat-
ing asymmetric and weak links.
(2) Packet-based transmission power control:
Apropertransmissionpowercanbeassignedtoeachpacketbasedonthedestination
andtypeofthepacketconsideringlinkqualityrequirements(i.e.,thelinkqualitycontrol
threshold: LQ
µ
). Wecanexpectreducedenergyconsumptionforpackettransmissionby
using minimum transmission power which meets LQ
µ
. The reduced interference from
minimizing the transmission power for each communication can improve the spatial
reuse of the network as well. We can also provide customized reliability to the packets
with di®erent importance.
(3) Metric-based link quality estimation:
Link quality is empirically measured based on the packet reception rate (PRR)
metric rather than distance-based link quality approximation. We observe that the
link distance is not an accurate metric of link quality in our experimental study (in
Section4.2.5). PCBLutilizesPRRmetricempiricallymeasuredatdi®erenttransmission
powerfordi®erentlinkstore°ectthediverselinkqualitiesevenatthesamelinkdistance
in real world.
46
If received signal strength (RSS) information is available, (RSS,PRR) pair informa-
tion can be used together with transmission power level for more e±cient and faster
link quality estimation and corresponding control.
(4) Blacklisting at adjusted transmission power level:
Not every link can be converted to a good link with transmission power control
even at the maximum transmission power level. Even new weak or asymmetric links
can be generated at adjusted transmission power level. We combine link blacklisting
approachtogetherwithtransmissionpowercontrolschemetoavoidtheuseofremaining
unreliable links at new transmission power control even for the broadcast packet. Both
link-based and packet-based blacklisting schemes will be discussed.
4.3.2 Basic PCBL Algorithm: Optimization Prior to Routing
We¯rstexplainthebasicstepsofimplementingourproposedtransmissionpowercontrol
with blacklisting scheme (PCBL) in this section. A brief version of the algorithm is
presented in Table 4.4.
First of all, each node measures the quality of links to its neighbor nodes in PRR
metric for all discrete (or pre-selected) transmission power levels P = fP
min
· P
i
·
P
max
g, where i is a transmission power in dBm. Let PRR(r)
P
i
denote the PRR at the
output power of i dBm at the receiver node r. At the receiver, it records the RSS value
for each transmission power level from each sender.
To select a unicast transmission power level, a link quality control threshold value
(LQ
µ
) in PRR metric needs to be selected according to the required level of link relia-
bility (e.g., LQ
µ
Ã 0:95). Simply, a common LQ
µ
value can be used for every link, or
each node can use di®erent LQ
µ
values for di®erent links or even for di®erent type of
packets based on the importance of each packet communication.
47
Step 1: Collect link statistics in PRR metric at every selected
transmission power level
Step 2: Select a unicast transmission power for each link (i.e., for
each one hop neighbor) which minimize energy consumption while
providing required link reliability
Step 3: Blacklist remaining or new unreliable links after transmission
power control
Step 4: Select a broadcast transmission power for each node with the
maximum unicast transmission power level
Table 4.4: Brief PCBL algorithm
We de¯ne Utx
r
as the minimum transmission power which satis¯es the LQ
µ
is
assigned for each link (i.e., for each receiver r) or for each packet type as a unicast
transmission power: Utx
r
= min P
i
, such that PRR(r)
P
i
> LQ
µ
. Otherwise, Utx
r
=
P
max
.
Basedontherequiredlinkreliabilityandtheintendedconnectivitylevel,PRR-based
blacklist threshold (BL
µ
) is selected. PCBL can use either a link-based blacklisting or
packet-based blacklisting scheme.
Link-based blacklisting blacklists every link with lower PRR than BL
µ
, thus deter-
mining the topology of the network. Each link can select a di®erent BL
µ
for further
optimization if necessary. For example, sparse part of the network might be better to
use a lower blacklisting threshold. Packet-based blacklisting uses adaptive BL
µ
value
for each packet (e.g., based on the type of the packet or the type of application) rather
then for each link in the network. Packet-based blacklisting is necessary when the re-
quirements for the link qualities are di®erent for each application or for each type of
packet. Finer control of transmission power in packet-based blacklisting can provide a
better utilization of the network.
LQ
µ
is used to control the quality of the link and BL
µ
is employed to ensure the
minimum reliability of the link. LQ
µ
should be greater than equal to BL
µ
and the gap
48
betweenLQ
µ
andBL
µ
reducesthevariationofthelinkavailabilitywhichmaylowerthe
performanceofupper-layerprotocolsfromthefrequentchangesofthenetworktopology
during the operation.
In the last step, each node (i) selects a broadcast transmission power (Btx
i
) with
the maximum unicast transmission power for all non-blacklisted links assigned in steps
2 and 3: B
i
= max Utx
r
;for8 neighbor r, where PRR(r)
Utxr
> BL
µ
. This ensures
eachsendertransmitsbroadcastpacketswithenoughtransmissionpowertoreachevery
neighbor node.
Our goal in transmission power control is to assign a minimum transmission power
that provides required link quality for each packet transmission and also to remove the
negative e®ects caused by unreliable links. We choose a transmission power, which
is close to the optimal, for each link, for unicast transmission, and for each node, for
broadcasttransmission,withourproposedalgorithm(inTable4.4). Transmissionpower
selection refers to the PRR values, which are realistic link quality metrics, collected at
di®erent transmission power levels. Selected transmission powers satisfy the required
link quality and also minimize the interference to the network. Adjusted transmission
power from the default can expose new links that are not previously visible. The
unreliable links that cannot be converted to good links even with transmission power
control and newly generated weak and asymmetric links at changed transmission power
levels are blocked from the use with blacklisting scheme.
4.3.3 On-demand Transmission Power Optimization for Each Long-
lived Communication
Collection of link statistics before the start of each data communication is unacceptable
forsomeapplicationwherethepromptdeliveryofcollectedinformationiscritical. Keep
49
Step 1: Collect link statistics only at the maximum transmission
power level (P
max
)
Step 2: Blacklist unreliable links before ¯nding a routing path
Step 3: Find a delivery path between the source and sink with
a chosen routing protocol
Step 4: Identify unicast transmission powers to use only for
the links in the delivery path
Table 4.5: PCBL algorithm for a long-lived communication
maintaining up-to-date link quality information for every link in the network ahead of
time is unnecessary and ine±cient when communication is infrequent.
Adi®erenttransmissionpowercontrolapproachcanbetakentoreducetransmission
delay and overhead of the link statistics collection and the summary of this algorithm is
presented in Table 4.5. We can collect link statistics only for the links participating the
packetcommunicationbetweenthegivensourceandsink. Withthismodi¯edapproach,
eachnodeconvergestoaclosetooptimaltransmissionpowerlevelafterareliablerouting
path is set up by a routing protocol with small prior link statistics collection e®orts.
When we ¯nd an optimal unicast power level for each link, we can collect the link
quality during the idle data communication period or we can lower the transmission
power from the P
max
to the lower level and decide proper transmission power level
based on the number of retransmission experienced at each transmission power level.
4.3.4 Experiment Results for a Single Data Flow
We evaluate the performance of proposed transmission power control with blacklisting
schemeinthePC104testbedshowninFigure1.2. Eventhoughweusedrelativelysmall,
manageablenumberofnodesforourexperiments,thistestbedsatis¯esourexperimental
conditionasitstillpossessesmanyunreliablelinks. Largertestbedexperimentsincreases
the number of hops for packet delivery, and additional hops increase the probability of
50
0
10
20
30
40
50
60
70
80
90
100
Power control schemes
Packet delivery rate (%)
OPP−P0 TPP−P0 OPP−P5 TPP−P5 OPP−P10 TPP−P10 M−BL PCBL
Figure 4.8: Packet delivery rate (PDR) from the experiments with ¯ve di®erent power
control schemes
including unreliable links in the delivery path. Therefore, we can expect even further
performance drop in multi-hop packet communication without any link quality control
scheme from the larger testbeds.
Inourexperiments,twonodeslocatedfarthestinthetestbedareselectedasapacket
sender (node 21) and a receiver (node 26). Directed di®usion [38] is used as a routing
protocol and fully active mode S-MAC [89] is used as a medium access control protocol.
We compare the following eight scenarios, categorized based on ¯ve di®erent trans-
mission power control schemes, in this experiments: (1) OPP-P0 and TPP-P0: One
Phase Pull (OPP) and Two Phase Pull (TPP) di®usion routing at default transmis-
sion power of 0 dBm (2) OPP-P5 and TPP-P5: OPP and TPP routing at increased
transmission power of 5 dBm (3) OPP-P10 and TPP-P10: OPP and TPP at the maxi-
mum transmission power of 10 dBm (4) M-BL: TPP with Blacklisting at the maximum
transmission power level (i.e., 10 dBm) and ¯nally, (5) PCBL: TPP with our proposed
scheme for transmission power control with blacklisting. We set LQ
µ
to 98% PRR and
BL
µ
to 90% PRR in this experiment. We perform experiment with only TPP for M-BL
51
and PCBL schemes because both schemes remove asymmetric links and OPP and TPP
are expected to perform equally in terms of PRR when there is no asymmetric links.
The experiment results show average end-to-end packet delivery rate (PDR) over ¯ve
1200 second-long experiments.
Figure 4.8 presents PDRs measured from the eight di®erent testbed experiments.
Errorbarsshowstandarddeviations. First,wewanttoseeifhowmuchimprovementwe
can expect in multi-hop packet communication by increasing the default transmission
power of each node instead of using distinct transmission power for each link or for
each packet. When we compare the PDRs for OPP di®usion at di®erent transmission
power levels, we can see that PDR gets higher at the higher transmission power level.
We can observe PDR improvement by increasing transmission power from 0 to 5 dBm.
Improved delivery rate mainly comes from the improved link quality of the weak links,
which are part of the packet delivery route. However, the improvement is negligible
when the default transmission power was increased from 5 to 10 dBm, the maximum
available output power.
When we compare the performance of OPP and TPP at 0 dBm transmission power,
TPP shows higher PDR because it avoids using asymmetric links that OPP di®usion
selected as a part of the routing path. At the higher transmission power of 5 dBm,
PDR even gets worse because this power level generates new unreliable links, that are
utilizedbytheroutingprotocol. Atthemaximumtransmissionpowerlevel, TPPshows
about the same PDR as with 0 dBm transmission power, and still loses about 21% of
the total packets.
The PDR in directed di®usion is highly dependent on the reliability of the delivery
route that is selected for use. In the wireless network which only uses a single hop
data communication, simply increasing the default transmission power is an e®ective
way to improve PDR. In a multi-hop wireless data communication, whether unreliable
52
(a) without Tx power control
(b) with M-BL scheme (c) with PCBL scheme
Figure 4.9: Topology changes with di®erent power control schemes: a solid arrow rep-
resents a realiable link with over 90% PRR and a dotted arrow represents a link with
0 < PRR(%) < 90. Each link is also marked with corresponding (transmission power,
observed PRR) pair information
wireless links exist or not is an important factor which decides the reliability of data
communication. Increasingdefaulttransmissionpowercanconvertsomeoftheweakand
asymmetric links to reliable links, and help discover new links which was not available
at lower transmission power level, but it is not an e®ective way to improve PDR in
most cases because (1) it may not make every unreliable link reliable, and (2) it may
also generate new unreliable links at new transmission power level, and (3) it uses up
more network capacity because an increased transmission power level has larger spatial
footprint.
TPP with M-BL and TPP with PCBL result in close to 100% packet delivery rate:
99.2% and 98.7% respectively. Both schemes provide comparable link qualities at ad-
justed transmission power levels. Therefore, the same links are blacklisted and most
likely both schemes generate the same network topology. Figure 4.9 shows a simpli¯ed
four node example from the testbed that visually compares the links under three dif-
ferent schemes. The main di®erences between PCBL and M-BL are found in (1) the
amount of energy consumption and (2) the level of interference in packet transmission.
53
Di®erence Unicast Broadcast Total Per Packet
M-BL +75.4% +53.2% +67% +66.2%
TPP-P0 +3.5% -40.3% -13% +10.8%
Table 4.6: The energy consumption di®erence in packet transmission compared to the
PCBL scheme
Expr Flows Description
No. A B
1 89Ã 75 72Ã 73 low interference
2 91Ã 70 72Ã 73 medium interference
3 85Ã 81 72! 87 one °ow gets stronger interference
4 73Ã 87 88! 72 strong interference
Table 4.7: Four experiment scenario comparison
We compare the energy consumption in packet transmission between the PCBL
scheme with the case without any power control scheme (i.e., TPP-P0) and also with
the case with M-BL scheme in table 4.6. We add up the energy consumption from
every broadcast (control packets sent by the routing protocol) and unicast packet (data
packet) transmissions for TPP-P0 and M-BL schemes, and compare those against our
PCBL scheme. We exclude the energy consumption from MAC control packets (i.e.,
RTS, CTS, and ACK) from the calculation.
In our testbed experiments with single data °ow, M-BL scheme shows 67% more
energy consumption than PCBL and both unicast and broadcast packets used up much
more energy than PCBL because it transmits every packet at maximum transmission
power. Original TPP scheme (TPP-P0), which transmits packets at default power of
0 dBm, consumes 13% less energy than PCBL scheme. When we compare energy con-
sumption for each successful packet delivery, however, PCBL saves 10.8% transmission
power compared to TPP-P0. This energy savings from PCBL shows an example of
the compensation gained from the increased network reliability that exceeds the extra
energy consumption in packet transmission.
54
Figure 4.10: Stargate node locations for the multiple data °ow experiments.
4.3.5 Experiment Results for Multiple Data Flows
In a single °ow packet communication experiment, both PCBL and M-BL perform well
in terms of packet delivery rate. To see the e®ect of over-ampli¯ed transmission signal
from the M-BL scheme, we performed experiments with multiple data °ows in this
section.
First, we run four node experiments with two concurrent data communication °ows
(called °ow A and B) in our Stargate testbed (shown in Figure 4.10). Two packet
senders are synchronized to start the packet transmission and continuously transmit
packets one after another. We test PCBL and M-BL schemes with both turning on and
o® the collision avoidance functionality. In this experiment, we use 100% PRR for LQ
µ
and 90% PRR for BL
µ
in this experiment for PCBL scheme. We repeat experiments
with four di®erent sender and receiver node pairs (i.e., four di®erent pairs of data °ow)
asshowninFigure 4.10anddescribedinTable4.7. Di®erentnodelocationschangethe
signalstrengthatthereceiverandthelevelofinterferencebetweenthe°ows. Werepeat
experiments¯vetimesforeachsettingandeachresultshowsthemeanvalue. Duetothe
non-linearityinthemeasuredsignalstrengthvalueatthehigherthan-55dBm, weused
estimated RSS value based on the output power level for SNR and SINR calculation for
the values measured in this region.
55
4.3.5.1 PCBL vs M-BL with a Collision Avoidance Scheme
We enabled a collision avoidance scheme in S-MAC [89] which is a carrier sensing fol-
lowed by a RTX/CTS/DATA/ACK sequence in this experiment.
Table 4.8 compares the experiment results between the PCBL and M-BL. The colli-
sion avoidance scheme prevents packet collisoin by disallowing concurrent packet trans-
mission within the same channel. Therefore, the PDR is always 100% for both PCBL
and M-BL. However, additional packet sender and receiver in the same channel com-
pete for the limited bandwidth and reduce the throughput of both data °ows. Each
additional active sender or receiver occupies the channel by deferring any other packet
transmission with RTS or CTS packet transmission when we use a collision avoidance
scheme.
Whenthereisnocompetingpacketsender, asendercouldtransmit6.7datapackets
(excluding control packets) per second in our experiments with 230B data packets.
Experiment1resultsshowcloseto6.7packetspersecondthroughputforboth°owswith
PCBLscheme. Inotherwords,thereisalmostnointerferencecomingfromtheother°ow
and concurrent packet communication was possible with PCBL in this case. However,
the throughput for M-BL in Experiment 1 was 3.9 packets per second. Reduced packet
transmission rate with M-BL means the data and control packet communications from
the other °ow caused interference (i.e., share the channel together) and lowered the
throughput of communication links.
As we change the node location from Experiment 1 to Experiment 2, the level
of interference between the two °ows is increased. The strength of interference (i.e.,
the signal strength of the data and control packet transmitted from the other °ow) is
importantevenwhenthecollisionavoidanceschemeisturnedonbecauseitchangesboth
the probability of the control packet reception and the probability of a packet collision
when the control packet is lost and concurrent packet transmission is not prevented.
56
Experiment PCBL MBL
w/CS °ow A °ow B °ow A °ow B
1 data rate (pkts/sec) 6.7 6.5 3.9 3.9
SNR (dB) 33.47 40.15 50.97 49.79
PRR (%) 100 100 100 100
2 data rate (pkts/sec) 5.57 5.29 3.71 4.51
SNR (dB) 27.57 38.4 39.13 49.83
PRR (%) 100 100 100 100
3 data rate (pkts/sec) 4.42 4.33 3.54 4.08
SNR (dB) 27.11 27.5 29.35 42.38
PRR (%) 100 100 100 100
4 data rate (pkts/sec) 2.94 3.33 3.78 3.77
SNR (dB) 29.35 31.08 42.59 43.91
PRR (%) 100 100 100 100
Table 4.8: PCBL and M-BL comparison with a collision avoidance
0
10
20
30
40
50
60
70
80
90
100
Flow
Packet delivery rate (%)
M−BL
PCBL
Flow 1 & 2 Flow 3
Figure 4.11: Packet delivery rate (PDR) from the experiments with three data °ows
57
Depending on the number of competing sender and receiver nodes and the quality
of signals from these nodes in the same channel, the throughput of each link changes.
As the Table 4.8 shows, the throughput of the PCBL drops as the interference from
the other °ow increases. However, the throughput of the M-BL does not change much
with experiments at four di®erent locations. The extra transmission power used in M-
BL help the delivery of the control packet to the other °ow and prevent most of the
concurrent packet transmissions.
Withacollisionavoidancescheme,interferenceonlyincreasesthedelayofthepacket
communication instead of causing a packet lost from the collision. However, when mul-
tiple senders compete for the same channel for a long period of time, the delayed trans-
mission can lead to the packet drops from the queue over°ow. We performed another
multi-hop packet communication experiments with multiple continuous data °ows in
the testbed (shown in Figure 1.2). Three data °ows are involved in this experiment.
Node 17and38sendpacketstonode33(°ow1&2)andnode20sendspacketstonode
12 (°ow 3) at 1 packet-per-second send rate for 700 seconds.
Figure4.11presentsexperimentresultsbasedonthe¯verepeatedexperiments. This
¯gure shows the PDR for °ow 1 & 2 and °ow 3 under di®erent power control schemes.
The standard deviation from repeated experiments is marked with an error bar. PDR
in °ow 3 are similar for both schemes: 97.9% for TPP with M-BL and 97.6% for TPP
with PCBL. The PDR for °ow 1 & 2, however, shows 21% di®erence in favor of TPP
with PCBL: 95.5% for PCBL and 74.5% for M-BL. Packets from node 17 and 38 are
all delivered through node 11 and the wireless channels around node 11 involves four
times more tra±c than the tra±c between node 20 and 12. The interference from over-
ampli¯ed transmission power in M-BL saturates the wireless channel around node 11
andcausemorepacketdropswithM-BLin°ow1&2while°ow3couldstillgetenough
channel access with both schemes.
58
The collision avoidance scheme can prevent packet collisions, but the stronger inter-
ference from M-BL use up more channel capacity and builds up the queue size and ends
up with packet drop from queue over°ow. Therefore, even with the collision avoidance
scheme, interference can still a®ect the delivery ratio of the packet from the ine±cient
spatial reuse of the network.
4.3.5.2 PCBL vs M-BL without a Collision Avoidance Scheme
In this section, we compare the performance of PCBL and M-BL scheme in the situa-
tions where the interference from eachcommunication can in°uence the communication
of others. We perform this experiment to identify performance of our proposed power
controlschemeunderdi®erentMACdesign. Wecompletelydisabledcollisionavoidance
(including carrier sensing, random back-o®s, and RTS/CTS) in the experiments pre-
sented in this section. This allows for a sender to transmit packets even when it hears a
on-going packet communication in the same channel. In other words, concurrent packet
transmission is always allowed regardless of the current channel condition.
There is no di®erence in packet send rate (packets per second) for di®erent schemes
anddi®erent°owsinthisexperimentbecauseeachsendercantransmitapacketanytime
regardless of the channel condition without collision avoidance scheme. Therefore, the
send rate is about two times faster compared to the case with a collision avoidance
scheme in our experiment.
Table4.9showsthePRRtogetherwithsignal-to-interference-plus-noise-ratio(SINR)
rather than SNR because there exists an interference from the other °ow. We can see
that SINR value becomes lower as the distance between the two °ows gets closer. Over-
all, higher number experiments experience greater interference from the other °ow and
thereforepresentlowerSINRvalues. Whenthelevelofinterferenceisrelativelylow,like
59
Experiment PCBL MBL
w/o CS °ow A °ow B °ow A °ow B
1 data rate (pkts/sec) 13.7 13.7 13.7 13.7
SINR (dB) 33.47 40.15 36 32.63
PRR (%) 100 100 100 100
2 data rate (pkts/sec) 13.7 13.7 13.7 13.7
SINR (dB) 27.57 38.4 13.83 31.63
PRR (%) 100 100 100 100
3 data rate (pkts/sec) 13.7 13.7 13.7 13.7
SINR (dB) 9.56 12.16 -2.68 25.47
PRR (%) 92.7 90 0 100
4 data rate (pkts/sec) 13.7 13.7 13.7 13.7
SINR (dB) -1.62 2.36 3.29 -0.70
PRR (%) 0 0 100 0
Table 4.9: PCBL and M-BL comparison without a collision avoidance
theExperiment1and2, theintendedsender'ssignalstrengthismuchstrongerthanthe
interference from the sender in the other °ow. As experimental studies prove [78, 85],
stronger signal could be delivered under the concurrent packet transmission situation
and both Experiment 1 and 2 show 100% packet reception rate for both PCBL and
M-BL for both °ows. This phenomenon is called capture e®ect [85].
In Experiment 3, °ow A and B are placed in a roughly parallel position while the
receiver of the °ow A is somewhat closer to the sender of the °ow B. With PCBL, both
receivers get much stronger signals from each sender and provide reliable concurrent
communication for both °ows. When the M-BL scheme changes the sender's transmis-
sion power to the maximum level, the interference strength from the sender in °ow B
(node 72) becomes even stronger than then signal strength from the intended sender in
°ow A (node 81). The SINR at the receiver (node 85) becomes negative and no packets
can be delivered successfully in °ow A. However, the packets in °ow B can be reliably
delivered because the sender in °ow A is located much further than the intended sender
of °ow B.
60
When both °ows are located very close, as in the Experiment 4, there is high proba-
bility that the interference strength from the other °ow is strong enough to disturb the
intendedpacketcommunicationevenwithPCBLscheme(whichcauseslessinterference
thanM-BL).WithPCBL,neither°owcouldsuccessfullydeliverpacketstothereceiver.
With M-BL, the interference strength is even higher than the case with PCBL on each
data °ow. The SINR at the receiver of the °ow B becomes negative and the PRR of
the °ow B becomes zero. However, the signal strength from the sender in °ow A is
even stronger than the high interference from °ow B and it can capture the channel at
the maximum transmission power level of both senders in this case. Experiment 4 is a
specialexampleofthebene¯tsofextrasignalstrengthfromover-ampli¯edtransmission
power level. However, better performance of M-BL is coming from the fact that the
di®erence in the RSS between two senders are fortunately big enough to allow capture
e®ect at the maximum transmission power level. Extra signal strength from the M-BL
increases the interference to the network as well as the signal strength at the intended
receiver.
In general, PCBL performs better if the interference from the concurrent transmis-
sionbecomesnegligibleatcontrolledpowerlevelattheotherreceiver. PCBLcanreduce
energy consumption and interference in addition to equal or better performance in net-
work throughput to M-BL. M-BL is better only when the interference at controlled
transmission power from the PCBL still causes strong interference to each other and
results in packet collisions, while M-BL provides a channel capture for either communi-
cation.
61
PCBL MBL
PDR w/ CA w/o CA w/CA w/o CA
Upper °ow (%) 53 67 54 34
Lower °ow (%) 45 53 37 46
AVG (%) 49 60 45 40
Table 4.10: PCBL and M-BL comparison with a collision avoidance
4.3.5.3 Multi-hop, Multi-°ow Experiments
Weperformedmulti-hop, multi-°owpacketcommunicationexperimentsinthe14nodes
testbed to study the e®ects of interference from PCBL and M-BL on network through-
put. Multi-hop reduces the throughput because intermediate nodes need to share the
channel for both packet transmission and reception and multi-°ow also reduces the
throughput from the increased amount of data packets and interference. The testbed is
divided into upper and lower sections and there are two senders and a receiver located
in each section. Each section is separated with a dotted line in Figure 4.10 and two
senders and a receiver are marked with S1, S2, and R respectively. We intentionally
blacklist incoming packets from the other section to maintain two separate data °ows
one at each section. This is because we only want to evaluate the e®ect of interference
in the multi-hop packet communication using either PCBL or M-BL.
Each sender continuously transmits a 230 byte packet at one second intervals and
the packet reception rates are measured at the receiver nodes. We use two-phase-pull
directed di®usion routing protocols for a route-setup, and the actual data packet com-
munication is enforced to be started after a successful communication route discovery.
By doing this, the performance of the routing protocol is excluded from the comparison
between the PCBL and M-BL.
Table 4.10 shows the packet delivery rate (PDR) in our testbed experiment. There
are two °ows, one in the upper section and the other one in the lower section of the
62
testbed. Each°owhastwosendersandareceiverasdescribedearlier. Whenwecompare
thePDRbetweenthecaseswithandwithoutacollisionavoidancescheme,PCBLshows
better performance without a collision avoidance scheme. On the contrary, M-BL can
deliver more packets when it performs with a collision avoidance scheme.
Reduced interference from PCBL increases the probability of concurrent packet
transmission in the wireless network, and we can reduce the number of missed con-
current communication opportunities by disabling the conservative collision avoidance
schemeimplementedinS-MAC.ThisisthereasonwhyPCBLwithoutacollisionavoid-
ance scheme can improve the packet delivery rate when the network is saturated with
continuous multihop packet transmissions from multi-data sources like our experimen-
tal scenario. The extra transmission power used for M-BL causes stronger interference
to more neighbor nodes and it drops the PDR and throughput of the network due to
higher probability of packet collision without a collision avoidance scheme.
The comparison between PCBL and M-BL in our multi-hop experiment shows that
bothperformsimilarunderthecollisionavoidancescheme(whilethePCBLshowsslight
advantage) due to the limited bandwidth of the network from multi-hop and extra con-
trol packets. When we disable the collision avoidance scheme, PCBL delivers more
packets, bene¯ting more from the removed control packets by gaining more bandwidth
for each node at lower interference level than M-BL.
4.3.5.4 Lessons from the PCBL and M-BL Comparison
We now summarize the lessons learned from Section 4.3.5.1 to 4.3.5.3.
² Bene¯ts of PCBL: Reduction in transmission power consumption and interfer-
ence. PCBL improves the spatial reuse of the wireless channel and enables more
simultaneous communications with less energy consumption.
63
² Weakness of PCBL: Vulnerability to the dynamic environment change. Envi-
ronmental change can lead to the link quality change even though it is not severe
compared to the case without transmission power control scheme. By setting the
linkqualitycontrol thresholdoutside of therange of link qualityvariation, wecan
avoid this problem and provide comparable performance with M-BL in providing
consistent link quality.
² Bene¯ts of M-BL: Endurance of the minor interference and environmental
change. Extra transmission power (i.e., stronger than RSS
µ
signal strength)
ensures reliable communication under some environmental change or minor in-
terference that is not prevented from the collision avoidance scheme.
² Weaknessof M-BL: Ine±ciency in energy consumption and spatial reuse of the
wireless channel. Extra signal strength cause more interference to the network
and the throughput of the network becomes lower as the interference strength
gets stronger.
ThemainadvantageofM-BLismoreconsistentlinkqualitywhenithashigherthan
RSS
µ
signal strength at the receiver. For PCBL, we can add some safeguard against
environmental change and every link can provide consistent link quality by setting the
RSS
µ
even higher than the RSS
µ
outside of the link quality variance range. PCBL can
reduce(orevenremove)thevarianceinthelinkqualityundertheenvironmentalchange
with a little extra energy consumption.
4.4 Summary
In this chapter, we have presented an experimental study of the e®ects of transmission
power control on low-power wireless links. Our study identi¯es the causes of high
varianceinlinkqualityunderdi®erentenvironmentalconditionsandhardwaresettings.
64
We repeat experiments for various experimental settings varying transmission power
levels.
Based on our better empirical understanding of wireless link behavior under power
control, we propose a packet-based link quality control scheme called PCBL. It con-
verts unreliable asymmetric and weak links to reliable wireless links which provide
a consistent link quality. We incorporate a blacklisting approach together with our
power control scheme to address the problem of remaining unreliable links at adjusted
transmission power setting. Blacklisting improves reliability by preventing the use of
unreliable links and ensuring minimum link reliability for the network. The proposed
transmission power control with blacklisting scheme (PCBL) provides energy-e±cient
link quality control with minimal channel interference, and provides a more stable and
reliable network topology. Throughout the study presented in this dissertation, we use
transmission power control as a main tunable parameter for interference-aware protocol
design improving reliability and channel capacity.
65
Chapter 5
Experimental Study of Concurrent Transmission in
Wireless Sensor Networks
5.1 Overview
There is growing awareness that realistic understanding of wireless links are essential
for developing e±cient protocols for wireless networks and evaluating them meaning-
fully [44]. In the previous chapter, we have presented some e®orts coherent to this
argument by evaluating a single wireless link without interference. However, a good
understanding of interference is essential, not only to improve the evaluation of existing
protocols under medium-to-high tra±c loads, but also to aid in the future design of
novel interference-aware protocols for wireless networks.
Most research considering network interference normally assumes one of two inter-
ference models: the protocol model or the physical model [32]. In the protocol model,
which is implemented by many state-of-the-art wireless network simulators, concurrent
transmissionsfromanynodewithinagivenrange(referredtoastheinterferencerange)
of a receiver will cause a collision that results in the loss of a packet from its corre-
spondingsender. ArecentstudybyWhitehouse et al.[83]hasarguedthatthisprotocol
model signi¯cantly overestimates packet loss during concurrent transmissions and can
therefore result in the design of ine±cient medium access protocols. In the physical
66
model, a packet from the sender is lost at the receiver only if the signal-to-interference-
plus-noise-ratio (SINR) falls below a given threshold. To our knowledge, the physical
model, which is widely used in communication theory, has not been previously tested
rigorously through real experiments in the context of low-power wireless networks.
Severalrecentempiricalstudiesinthecontextofwirelesssensornetworkshavegiven
us an understanding of the complex non-ideal behavior of low-power wireless links [7,
25, 47, 77, 93]. However, most of these empirical studies have focused on single links,
without concurrent transmissions from interfering nodes.
In the work presented in this chapter (which appears in [79]), we systematically
study the e®ects of concurrent transmissions through experimental measurements with
low-power Mica2 motes equipped with CC1000 radios. Our experiments involve the
measurement of received signal and interference strengths as well as packet reception
rates under carefully designed single-interferer and multiple-interferer scenarios. We
¯nd the simplistic interference range-based protocol model to be inadequate from this
empirical study, which agrees the results from Whitehouse et al. [83]. Our experimental
results con¯rm some key aspects of the SINR-based physical model, while suggesting
signi¯cant ways in which it can be enhanced for applicability in real deployments.
There are several concrete ¯ndings from our experimental study that o®er useful
insights; these are summarized in Table 5.1. Our measurements, conducted with Mica2
motes,con¯rmthatguaranteeingsuccessfulpacketreceptionwithhighprobabilityinthe
presence of concurrent transmissions requires that the SINR exceed a critical threshold.
However, groups of radios show a wide gray region of about 6 dB. We ¯nd that this
large gray region occurs because the SINR threshold can vary signi¯cantly depending
on the measured signal power and radio hardware (but not depending signi¯cantly on
the location). By contrast, we ¯nd that the gray region is quite narrow for a speci¯c
hardware combination at a ¯xed signal strength level. We ¯nd that it is harder to
67
Finding Section
Single interferer e®ects 5.3
Capture e®ect is signi¯cant 5.3.1
SINR threshold varies due to hardware 5.3.2
SINR threshold does not vary with location 5.3.3
SINR threshold varies with measured RSS 5.3.4
Groups of radios show»6 dB gray region 5.3.5
New SINR threshold model 5.3.6
Multiple interferer e®ects 5.4
Measured interference is not additive 5.4.2
Measured interference shows high variance 5.4.3
SINR threshold increases with more interferers 5.4.4
Table 5.1: Key ¯ndings from concurrent transmission study
estimate the level of interference in the presence of multiple (two or more) interferers
for two reasons: (a) the joint interference measurements show a much higher variation
when there are multiple interferers, and (b) the measured joint interference strength
is not always the sum of the individual interference strengths. We also ¯nd that the
measured SINR threshold generally increases with the number of interferers.
Therestofthechapterisorganizedasfollows: Wepresentourexperimentalmethod-
ologyinSection5.2. Wediscusstheresultsfromexperimentsinvolvingasingleinterferer
inSection5.3,andthoseinvolvingmultipleinterferersinSection5.4. Finally,wepresent
our conclusions and discuss future work in Section 5.6.
5.2 Experimental Methodology
In this section, we discuss some key aspects of our experimental methodology. In Sec-
tion 5.2.1, we discuss the hardware and software used. We describe our experimental
design for carrying out synchronized measurements in Section 5.2.2. We conclude this
sectionbydiscussingtheregressionmodelweuseformappingSINRtopacketreception
rates in Section 5.2.3.
68
5.2.1 Hardware and Software
Our study is based on systematic experiments on a PC104 [23] testbed running Linux.
The experiments are conducted in a controlled indoor o±ce environment where sur-
rounding objects are static, with minimal time-varying changes in the wireless channel
due to multi-path fading e®ects. Any code that can be used commonly by all PC104
nodes is accessed on a central computer through an NFS-mounted directory. We use
Mica2 motes, with the Chipcon CC1000 [10] radio operating at 433 MHz, as an RF
transceiver on the PC104 node. This device provides 38.4 Kbps data rate with Manch-
esterencodingandusesnon-coherentFSKmodulationscheme. WeusetheLinux-based
Emstar software framework to take advantage of its interactive interface with sensor
nodes in the testbed [27].
We use the S-MAC protocol [89], con¯gured in fully-active mode without sleep
cycles. To study collisions in a controlled manner we intentionally disable carrier
sense and random backo® in the MAC. Disabling the collision avoidance scheme im-
plemented in S-MAC allows us to freely transmit concurrent packets even when there is
on-goingpackettransmissioninthesamewirelesschannel. WealsoomittheMAC-level
RTS/CTS/DATA/ACK sequence by sending packets as broadcasts, avoiding the com-
plications of ARQ. We thus disable much of the MAC functionality in order to focus on
the fundamental behavior of wireless links in the presence of interference.
There are several other important wireless platforms, including IEEE 802.11 and
IEEE 802.15.4. As an experimental study, we can only a±rm that our results apply
to the CC1000 radio. However, hardware variation and large gray regions have been
previously observed for 802.11 radios [3] and it is likely that low power 802.15.4 radios
will show similar results. We have some preliminary results for 802.15.4 in Section 5.5.
69
Figure 5.1: Overview of the testbed with experimental methodology used for time syn-
chronization, signal strength and PRR measurement
5.2.2 Measurement Design
Our study requires a careful con¯guration to synchronize both packet transmissions as
well as measurements of signal strength and packet loss. Figure 5.1 shows our experi-
mentalcon¯guration. Eachexperimentinvolvesfourtypesofnodes: asender,areceiver,
one or more interferers, and a special synchronizer node. The synchronizer broadcasts
a sync packet just before each single or concurrent packet transmission. This serves
to synchronize the clock of every node in the testbed. The sync packet is a kind of
reference broadcast [22]. Each transmitting node (sender or interferer) sets its packet
transmission time and the receiver sets the received signal strength measurement time
based on this reference time.
In our controlled experiments the hardware identity and locations of the sender,
interferer, and receiver is ¯xed, but we vary the transmit power of the sender and
interferersoversomerange. Weplacenodesono±cetablesataboutonemeterinheight.
Every transmitter, including a sender and one or more interferers, is placed about the
samedistancefromareceivernodeforminganisoscelestriangleatbetween¯vetoseven
70
meters in our experiments. For each speci¯c combination of transmit power settings,
there is a series of packet transmission epochs. In each epoch, there is the following
sequence of transmissions, each interleaved with a sync packet (see Figure 5.1): (i) the
sender transmits alone; (ii) each interferer in turn transmits alone; (iii) all interferers
transmit concurrently; (iv) the sender transmits concurrently with all interferers. The
receivermeasuressignalstrengthinthemiddleofeachsingleorconcurrenttransmission,
exceptthe¯nalone,whichisusedtorecordwhetherthepacketwasreceivedsuccessfully
or not. We also measure a signal strength right after each individual packet reception
when there is no signal on the channel. This approach measures ambient noise levels
during experiments.
If a total of n packet transmission epochs are used for a particular transmit power
combination, the packet reception rate (PRR) for that combination is calculated as
the total number of packets received successfully divided by n. We typically use 75
epochs to estimate PRR with a precision of about 1.3%. In addition, ambient noise
measurements at the receiver are taken at the end of reception of each of the single
packet transmissions.
Due to jitter in the testbed system, transmission start times vary with a mean of
3ms. Further,obtainingreliablesignalstrengthmeasurementscantakeupto7ms(this
is not a controllable parameter in the CC1000 radios [10]). Hence the signal strength
measurement times need to be carefully chosen at the receiver to ensure any intended
collision occurs. We take measurements in the middle of long packet transmission pe-
riods. With 230 byte packets, packet transmission time is about 97 ms and so we can
tolerate substantial jitter.
Assecondpotentialtimingproblemcanoccurdependingonwhen packetstransmis-
sions begin. When the sender and all interferers are transmit concurrently, variation in
the transmission starting times can cause the sender packet to arrive 8 ms or later than
71
the ¯rst interferring packet. In such cases we observe that the packet is never recog-
nizedatthereceiver,evenifitssignalisstrongenoughtooverwhelmtheinterferer. This
problem occurs because our implementation of the radio's physical layer requires that
packet data immediately follow the start symbol of the packet. It will refuse to shift to
a later, stronger packet once it has read the start symbol of the earlier packet. The 8
msperiodcorrespondstothetransmissiontimerequiredforthe18bytepreambleand2
byte start words. This problem was identi¯ed by Whitehouse et al. [83]; they solved it
bymodifyingtheMACsoftwaretoretrainwhenit encounterssubsequentstartsymbols
of higher power. We became aware of this approach mid-way through our work. To
keep a consistent methodology, rather than modify our MAC to retrain, we detect and
¯lteroutcaseswhenthestrongestpacketarriveslaterthan8ms. Todothisweaddtwo
timestamps to each packet, recording transmission start and completion times. Fortu-
nately, because timing error is normally distributed with a mean of 3 ms, few packets
arrive late. From timestamps in logs, about 3% of epochs must be discarded due to
late arrival of the strongest packet. By removing these packets, we should get loss rates
comparable to a MAC that can retrain on later packets as proposed by Whitehouse et
al.
Signalstrengthmeasurementsareusedtoestimatethereceivedsignalstrength(RSS)
and received interference strength (RIS) for the concurrent packet transmissions at the
end of the epoch. Measured signal strengths include the strength of the transmission
and any ambient noise. Received signal strength measurements are taken in ADC
counts and converted to dBm following the manufacturer's documentation [10, 14].
This documentation also indicates that signal strength measurements are inaccurate
when they exceed -55 dBm. We con¯rmed this claim with tests and therefore drop
measurements beyond this threshold.
72
Given the RSS, we de¯ne JRIS as the measured joint received interference strength
when all interferers transmit concurrently. If N is the average ambient noise level
measured at the receiver, we can then calculate the signal-to-interference-plus-noise-
ratio (SINR) as:
SINR
dB
=10log
10
10
RSS
dBm
=10
¡10
N
dBm
=10
10
JRIS
dBm
=10
(5.1)
NotethatwebaseourSINRvaluesfrommeasurementstakendirectlyatthereceiver.
This approach is central to the experimental nature of our work. Alternatives such as
measuring transmit power at the sender would require the use of theoretical models
of path loss and ambient noise, neither of which we know for our environment. While
our approach avoids inaccurate signal strength estimation due to mismatches between
model and environment, we do not claim that the measured signal strength values
represent\true"signalstrengths, sincethatwouldrequireacalibratedcomparisonwith
a highly accurate RF measurement device. Instead, we claim that they represent signal
strengths as measured by actual radios. Our results may not directly apply to future
radios with more accurate measurements of signal strength, however we believe our
¯ndings have great utility with regard to practical protocols which must depend on
similar measurements in real deployments.
5.2.3 A Regression Model Mapping SINR to PRR
Whileallofour¯ndingsarebasedonrawmeasurements,weaddregressionlinesinsome
of the graphs to clarify the SINR-to-PRR relationship. The link layer model presented
by Zuniga and Krishnamachari [98], especially SNR to PRR conversion formula, is the
basis for our regression model.
PRR = (1¡
1
2
exp
¡¯
0
SINR+¯
1
)
8(2f¡l)
(5.2)
73
Thisregressionmodelisintendedfornon-coherentFSKmodulationandManchester
encoding that is used in Mica2 motes. We introduce the parameters ¯
0
and ¯
1
to ¯t
the experimental dataset to the regression model. The ¯
0
value controls the shape of
the regression curve and ¯
1
induces horizontal shifts of the curve. Based on repeated
experiments, we determined that a constant ¯
0
value provides excellent ¯ts (e.g., see
Table 5.4); ¯nd the optimal ¯
0
for each experiment improved our R
2
values by at most
0.01. We therefore hold ¯
0
constant at 2.6 in all our single-interferer ¯gures. The
parameter f is the frame size (230 bytes for our experiments) of the packet and l is the
preamble size in bytes (20 bytes).
5.3 Experimental Study of Single Interferers
In this section, we describe our systematic experiments to understand how concurrent
packet transmissions a®ect packet reception when there is a single sender and a single
interferer. We begin by studying how di®erent transmit powers cause di®erent regions
ofreception, fromgoodtonoisytobad(orwhitetograytoblack)(inSection5.3.1). We
then de¯ne the signal-to-interference-plus-noise-ratio (SINR) threshold for good recep-
tion and show that it varies with hardware combinations (in Section 5.3.2) and signal
strength(inSection5.3.4),anddoesnotvarystronglyduetolocation(inSection5.3.3).
Next,wecomplementourdetailedstudiesbasedonsmallnumbersofnodeswithalarger
12-node experiment (in Section 5.3.5). Finally, from these results we propose a realistic
simulation model (in Section 5.3.6).
5.3.1 Interference and Black-Gray-White Regions
It is well known that stronger packets can be received even in the face of weaker,
concurrent transmissions, and this result has recently been con¯rmed and exploited
experimentally [83]. We begin our study with experiments to carefully quantify this
74
−16 −14 −12 −10 −8 −6 −4 −2 0 2
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Tx power of SRC1 (dBm)
PRR
SRC1
SRC2
(a) Transmission power level to PRR
−16 −14 −12 −10 −8 −6 −4 −2 0 2
−100
−95
−90
−85
−80
−75
−70
−65
−60
−55
Tx power of SRC1 (dBm)
RSSI (dBm)
SRC1
SRC2
NOI
(b) Transmission power level to RSS
Figure 5.2: E®ects of varying SRC1's transmission power level on the PRR and RSSI.
Ambient noise level (NOI) at the receiver is shown together. Error bars show 95%
con¯dence intervals
75
capture e®ect as a function of the measured signal strengths from concurrent packet
transmitters over a wide range of transmission powers.
In these experiments we consider two transmitting nodes, SRC1 and SRC2. By
de¯nition, we call the stronger signal source the sender and the weaker signal source
the interferer. From this de¯nition these roles change with the varying transmission
powers. To study how these roles change, we vary transmission powers as both sources
send 230-byte packets and calculate packet reception rate (PRR), here over 60 epochs.
Figure 5.2(a) presents the packet reception ratio (PRR) of SRC1 and SRC2 as the
transmit power of SRC1 varies. Here we ¯x the transmission power level of SRC2 at
-4 dBm and vary the output power of SRC1 from -17 dBm to 2 dBm. Without inter-
ference, either source has reliable communications with the destination. However, the
experiment shows that three distinct regions occur as SRC1's transmit power varies.
Beginning at the left of the graph, when SRC1 is less than -10 dBm, SRC2's trans-
missions are always received. In the middle of the graph, when SRC1 transmits at
powers between -7 and -5 dBm, packets from neither of the senders are recognized at
the receiver. At the right of the graph, with SRC1 at -1 dBm or more, SRC1 is always
successful. This experiment shows two clear regions of packet capture, for SRC2 at the
left, and SRC1 at the right. We call these regions the white regions, where one source
is assured reception even in the face of a concurrent transmission. These regions can be
compared to the black region in the middle where neither transmission is received. Fi-
nally, weobservetwogray regionsatintermediatepowerlevels(from-10to-7dBmand
-4to-1dBm), wherepacketsreceptionisintermittent. Wede¯nethegrayregionasany
combination of sender and interferer transmit power levels that result in PRRs between
10% and 90%. Our de¯nition was inspired by the notion of the gray area described by
Zhao and Govindan [93]. As with their de¯nition, our gray region corresponds to high
76
variation in packet reception. However, the gray area de¯ned in their work refers to a
spatial distance range, and is not related to power levels.
To measure the level of interference in the channel we directly measure the received
signal strength (RSS) in Figure 5.2(b). Recall that we measure RSS values at the re-
ceiver, ¯rst taking separate measurements for each transmitter and then during the
concurrent transmission. Measured ambient noise during our experiment shows consis-
tent values, with a standard deviation of less than 1 dBm. The measured noise °oor is
also much lower than the interference level in all our experiments and contributes little
to the SINR.
We align the x-axes of Figures 5.2(a) and 5.2(b) to relate RSS to PRR. We observe
that when the RSS of both sources become similar (within 0.6 dBm, when SRC1 is
around -6 dBm), packet reception for both transmitters is zero as the transmissions
corrupt each other. Further from this point, more packet receptions are observed as the
received signal strength di®erence between two transmitters increases.
Table5.2reproducesthePRR,RSSI,andtransmitpowervaluesfromFigure5.2and
adds calculated signal-to-interference-plus-noise-ratio (SINR) values. SINR represents
the di®erence between the sender (by de¯nition, the strongest transmitter) and the
interferer. We categorize each SINR value based on the corresponding PRR as being in
a black, gray, or white region for the dominant source.
For simplicity, Figure 5.2 varied only one source's transmission power while holding
the other constant. By contrast, Figure 5.3 shows measured results when the transmit
powers of both sources are varied. This extensive set of experiments con¯rms that the
results of Figure 5.2 hold regardless of which transmitter is varied or what power levels
are considered. A horizontal or vertical slice through this ¯gure would show white
regions for either SRC1 or SRC2, a black region in the middle, and gray regions on the
77
Tx Pwr RSS of
of SRC1 SINR PRR Region
SRC1 (dBm) (dB)
-17 -76.55 9.51 1
-14 -74.07 7.08 1 white (SRC1)
-12 -72.59 5.87 1
-10 -71.09 4.21 0.98
-8 -69.76 3.00 0.72 gray (SRC1)
-7 -68.22 1.56 0
-6 -66.33 0.58 0 black
-5 -65.78 1.73 0 (neither)
-4 -63.99 2.98 0.03
-3 -63.01 3.98 0.22 gray (SRC2)
-2 -61.96 5.02 0.82
-1 -60.36 6.54 0.98
0 -59.64 7.08 1 white (SRC2)
1 -58.13 8.75 1
2 -36.85 9.93 1
Table5.2: SINR-to-PRRmappingwithregiondistinction. RSSofSRC2isstaticaround
-66.8 dBm and ambient noise is around -94.6 dBm
−75 −70 −65 −60 −55
−75
−70
−65
−60
−55
SRC1 RSS (dBm)
SRC2 RSS (dBm)
0.1 PRR
0.9 PRR
Figure 5.3: Packet reception rate at di®erent RSS combination from SRC1 and SRC2.
Black-gray-white regions are marked with cross, triangle, and circle respectively
78
border. We also observe that the edge of the gray region is not strictly linear as the
transmit power varies. We will study this issue in more detail in Section 5.3.4.
Figures 5.2 and 5.3 show that concurrently transmitted packets are all corrupted
when they have nearly equivalent signal strength at the receiver. However, there is
a signi¯cant range of transmission powers in which the capture e®ect occurs and the
stronger packet is received successfully. These results lend further evidence to show
thatthesimplisticprotocolinterferencemodelcanbehighlymisleading. Capture-aware
MAC schemes are indeed likely to provide signi¯cant improvements in e±ciency.
These observations motivate us to analyze various factors that impact relationship
between SINR and PRR. We de¯ne the SINR threshold as the minimum SINR which
guarantees a reliable packet communication with PRR¸ 0.9. In the following sections,
we examine the impact of hardware combinations, node locations, and signal strength
variations on the measured SINR threshold. In particular, we seek to know whether
there is a constant SINR threshold for all scenarios.
5.3.2 SINR Threshold and Transmitter Hardware
Section 5.3.1 demonstrated the packet capture e®ect and de¯ned the SINR threshold.
WenextstudySINRthresholdtoseeifitisa®ectedbyvarianceintransmitterhardware.
In this experiment, we use di®erent Mica2 motes with the same type of CC1000 radio.
We consider two pairs of nodes, SRC1-SRC2 and SRC1-SRC3. As in Section 5.3.1,
we hold one transmitter's received signal strength constant at -66 dBm and vary the
others from -66 to -77 dBm. We then measure the SINR threshold.
Figure 5.4 presents these experimental results. On the left side of the graphs, SRC1
isthesenderandSRC2orSRC3istheweakerinterferer. Ontherightside,theopposite
holds, with SRC1 being weaker. The x-axis shows the SINR (the negative signs on the
lefthandsideshouldbeignoredasanartifactofthepresentation). Inaddition,thesolid
79
−10 −8 −6 −4 −2 0 2 4 6 8 10
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
SINR (dB)
PRR
SRC2
(with SRC1)
SRC3
(with SRC1)
SRC1
(with SRC2)
SRC1
(with SRC3)
Figure 5.4: E®ect of di®erent packet sender and interferer hardware on SINR-to-PRR
relationship
and dotted lines ¯t our regression model (de¯ned in Section 5.2.3) to the experimental
data.
First, we compare the experiment results from SRC1-SRC2 pair, shown as the solid
line model and asterisk points. The SINR threshold values are di®erent for each trans-
mitter; SRC1 has an SINR threshold of 3.4 dB and SRC2 has an SINR threshold of 5.3
dB. There is a nearly 2 dB di®erence between these thresholds. When we compare the
experiment results with di®erent pairs of hardware (i.e., between the solid and dotted
regression lines), we can see that SRC1 requires a stronger signal strength to reach the
same level of PRR at the same receiver when the interferer is changed from SRC2 to
SRC3. SRC1's regression line (shown in the left side of the ¯gure) moves about 1 dB
to the left with interferer SRC3 and SRC3 requires about 1.7 dB lower SINR threshold
compared to SRC2 when the same node SRC1 is the interferer. These results indicate
stronglythatthespeci¯chardwarecombinationofsenderandinterfererchangethemea-
sured SINR threshold. (We rule out location di®erences as an alternative explanation
in Section 5.3.3.)
80
−10 −8 −6 −4 −2 0 2 4 6 8 10
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
SINR (dB)
PRR
Original Location
Swapped Location
Figure 5.5: E®ect of di®erent packet sender and interferer location on SINR-to-PRR
relationship
Location Source ¯
1
(95% con¯dence)
Original SRC1 -0.914 (§0:108)
SRC2 3.802 (§0:127)
Swapped SRC1 -0.587 (§0:157)
SRC2 3.774 (§0:147)
Table 5.3: Parameter ¯
1
and 95% con¯dence intervals for two di®erent locations
Note that since our SINR calculations in all cases are based on measurements at the
same receiver, we can rule out di®erences that have to do with transmit-side calibration
settings, receiver sensitivities, or di®erences in the magnitude of the path loss from
di®erent transmitter locations. We speculate that the hardware-combination-speci¯c
variations in the SINR-threshold result from distorted signals due to non-linear e®ects
in the radio transmitters. Even at the same measured signal strength at the receiver,
thesignalsfromdi®erentsourcesmayhavedi®erentlevelsofdistortion,inturna®ecting
the packet reception di®erently.
81
5.3.3 E®ects of Location on PRR and SINR
One possible explanation for the variations in hardware shown in Section 5.3.2 could
be that the nodes were in di®erent locations, resulting in di®erent multi-path e®ects
on the channel. We therefore next study the e®ect of location on the SINR-to-PRR
relationship.
To study the possible e®ect of packet sender and interferer location on the SINR-
to-PRR relationship, we swap the location of SRC1 and SRC2 and performed the same
experiments as in Section 5.3.2. Swapping the sender locations changes the channels
observed between the two transmitters and the receiver. Figure 5.5 compares the ex-
periment results from new, swapped location with previous experiment results at the
original node location. There is no noticeable di®erence in SINR-to-PRR relationship
between these two set of experiment results. When we compare the parameter ¯
1
value
used for each regression model (presented in table 5.3), ¯
1
values are very close for the
same sender, not for the same location. But, SRC1 ¯
1
value is still located a little bit
outsideof95%con¯denceintervalof¯
1
valueusedforswitchedlocation. Thisdi®erence
is from the e®ect of location change but it is minor compared to the hardware e®ect, as
can be observed from the corresponding curves in ¯gure 5.5.
From this comparison, we can verify that the main di®erence in SINR threshold
betweentwonodesisfromthetransmitterhardware(orsignaldistortionlevel)di®erence
rather than their location di®erence. We have run similar experiments with a two
additionalpairsofnodes,aswellaswithdi®erentlocationsforthenodesusedabove. We
consistently ¯nd that location change does not make distinguishable di®erence in SINR
threshold. However, all our experiments have been carried out in an o±ce environment.
An area of future work is to study if these results apply in other environments, both
indoors and outdoors.
82
0 1 2 3 4 5 6 7 8 9 10
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
SINR (dB)
PRR
Figure 5.6: Experiments with wide range of sender and interferer signal strength.
Sender: SRC2, Interferer: SRC1
5.3.4 E®ect of Sender Signal Strength on the SINR Threshold
Our studies with two senders showed that the edge of the white region does not exhibit
a linear relationship with unit slope (see Figure 5.3), which would be expected if the
SINR threshold remained a constant regardless of the measured signal strength. In
Section 5.3.2, we showed that di®erent transmitter hardware results in di®erent SINR
thresholds. We next study more carefully how the measured sender signal strength
a®ects the SINR threshold.
Here we vary the transmission power level of both packet sender and interferer over
a wide range so that the received signal strength range varies from -91 to -52 dBm at
the intended receiver. Figure 5.6 shows these experimental results, where SRC1 is an
interferer and SRC2 is a packet sender.
This ¯gure shows a gray region that is about 4.2 dB wide from SINR values of just
above 2 to above 6 dB. This wide range applies even though locations and hardware are
both constant|the only di®erence we have made for this experiment was to vary the
transmit signal strength of the sender.
83
0 1 2 3 4 5 6 7 8 9 10
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
SINR (dB)
PRR
−56.0
−57.5
−59.0
−60.5
−62.0
−63.5
−65.0
−66.5
−68.0
−69.5
−71.0
1
11
1
2
3
4
5
6
7
8
9
10
11
RSS (dBm)
2 3 4
10 9 8 7 6 5
Figure 5.7: SINR-to-PRR relationship categorized for di®erent received signal strength
levels. Experiment results in each category are represented with a regression line.
Sender: SRC2, Interferer: SRC1
Index RSS range ¯
1
SINR
µ
R
2
(Fig 5.7) (dBm) (dB)
1 -55.2 { -56.7 0.425 3.99 0.998
2 -56.7 { -58.2 3.827 5.30 0.982
3 -58.2 { -59.7 6.894 6.48 0.993
4 -59.7 { -61.2 7.183 6.59 0.992
5 -61.2 { -62.7 6.873 6.47 0.987
6 -62.7 { -64.2 6.373 6.28 0.995
7 -64.2 { -65.7 3.856 5.31 0.963
8 -65.7 { -67.2 3.802 5.29 0.979
9 -67.2 { -68.7 2.589 4.82 0.997
10 -68.7 { -70.2 1.232 4.30 0.997
11 -70.2 { -71.7 0.223 3.91 0.992
Table 5.4: ¯
1
, SINR threshold (SINR
µ
), and R
2
(goodness-of-¯t) value for sender SRC2 for
SRC1-SRC2 pair experiments when ¯
0
is set to 2.6
84
To better understand the data in Figure 5.6, we collected the RSS values into 1.5
dB intervals (10 raw ADC counts) and then ¯t our regression model to each set of
experimentaldata. Table5.4showstheRSSrangesandcorrespondingmodelparameters
(¯
1
) and SINR thresholds, along with goodness-of-¯t (R
2
) data. (We use a constant 2.6
of¯
0
basedontheanalysisfromtheexperimentaldatasetasdescribedinSection5.2.3.)
This table shows that our model provides an excellent ¯t to the data, even with a
constant value for ¯
0
, since the worst case R
2
¯t value is 0.963. We therefore conclude
that our regression model can accurately summarize the experimental data. We also
observethatthemodelparameter¯
1
variesnon-linearlyoverthesemeasuredRSSvalues.
This variation in ¯
1
shows that the SINR threshold also varies with measured RSS in
some non-linear manner, even when hardware and location are unchanged.
To investigate how the SINR value relates to transmission power we plot the regres-
sion models in Figure 5.7. These show that the SINR threshold is highest at medium
measured RSS values and lowest when the measured RSS value is strong or weak. For
example, in Figure 5.7 the ¯tted model shifts to the right (higher SINRs) as the RSS
shrinks from -56.0 to -60.5 dBm (see arrows 1, 2, 3, and 4), then shifts back to the left
as RSS reduces further to the lowest observed values of -71.0 (arrows 5 through 11).
Tocon¯rmthatthisexperimentalresultwasnotpeculiartoourhardwareorlocation,
werepeatedsimilarexperimentswithseveralotherpairsofnodes. Wedonotreproduce
the raw SINR-PRR graphs, but instead ¯t a model to each experiment and compute
the SINR threshold. Figure 5.8 shows how the SINR threshold value (for 0.9 PRR)
changes over di®erent levels of sender signal strength for three di®erent pairs of node
experiments: SRC1 with each of SRC2, SRC3, and SRC4. For each pair of nodes, the
¯gureshowstwolines,onelineeachforwhenoneofthetransmittersbehavesasapacket
sender while the other behaves as an interferer.
85
−74 −72 −70 −68 −66 −64 −62 −60 −58 −56
0
1
2
3
4
5
6
7
Received signal strength (dBm)
SINR threshold
SRC1(SRC1−SRC2)
SRC2(SRC1−SRC2)
SRC1(SRC1−SRC3)
SRC3(SRC1−SRC3)
SRC1(SRC1−SRC4)
SRC4(SRC1−SRC4)
Figure 5.8: SINR threshold for 0.9 PRR change at di®erent received signal strength
level
All six SINR thresholds in Figure 5.8, show maximum values when the sender's
signal strength (measured at the receiver) is around -61 dBm. In this region, the SINR
threshold,the¯
1
parametervalue,andthewidthoftheblackregionareallhighest. This
resultsuggestthatMACprotocolsdesignedtoexploitthecapturee®ectandsimulations
designed to realistically model wireless collisions both must consider the magnitude of
the signal strengths in addition to the ratio of signal and interference powers. We
believe that curves such as those plotted in Figure 5.8 can be used as the basis for
realistic simulations.
An important open question is understanding what physical phenomena causes this
variation in SINR threshold. One possibility is that the radio transfer function exhibits
nonlinear e®ects that a®ect signals with high and low signal strengths; another is that
the RSSI measurement process itself is skewed at these extremes. A more detailed
understanding of the causes of this RSS-SINR-PRR relationship is an area of future
work.
86
0 1 2 3 4 5 6 7 8 9 10
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
SINR (dB)
PRR
Signal strength
Hardware
Figure 5.9: Testbed experiments with 12 neighbor nodes
5.3.5 Testbed Experiments
To con¯rm that our hardware and signal strength e®ects on SINR apply generally, we
performed testbed experiments that consider a wider range of hardware combinations
and RSS levels. We randomly deployed 12 PC104 nodes in two large rooms where the
distance between the intended receiver and the farthest node in the testbed was around
18 meters. We selected an intended receiver node and a time synchronizer (using the
procedure described in Section 5.2.2) and performed pairwise experiments with the
remaining 10 nodes in the testbed. For each pair, one node is the sender (with stronger
RSS) and the other node behaves as an interferer.
We set the interferer's transmission power constant at -8 dBm so that it has a
constantreceivedinterferencestrength(RIS)atthereceiver. MeasuredRISvaluesfrom
di®erent interferers range from -81 to -63 dBm, but we observe a change of up to 1
dBm RIS from the same interferer at di®erent times, presumably due to time-varying
changes in the environment. We then vary the transmission power of the sender from
strength equal to the interferer's RIS value until a power level where the RSS is strong
enough to provide reliable (close to 100%) packet reception.
87
90
71
70
188
91
193
87
88
72
186
190
85
83
194
75
185
183
192
184
81
191
73
89
76
187
(a) Interference from the sender
90
71
70
188
91
193
87
88
72
186
190
85
83
194
75
185
183
192
184
81
191
73
89
76
187
(b) CCable links under old model
90
71
70
188
91
193
87
88
72
186
190
85
83
194
75
185
183
192
184
81
191
73
89
76
187
(c) CCable links with new model
Figure 5.10: E®ects of introducing new capture-aware simulation model
We calculate SINR values based on the measured RSS and RIS pair information as
well as the measured ambient noise and plot the SINR-to-observed-PRR relationship in
Figure5.9. ExperimentalresultsshowalargevariationintheSINR-to-PRRrelationship
(or in SINR threshold values). We speculate that this is because di®erent interferers
in the testbed generate signals with di®erent distortion levels and di®erent RISs at the
intended receiver. Also, di®erent senders have di®erent SINR thresholds for the same
interferer.
The change in RIS level causes a similar e®ect as the RSS level change (presented in
Section 5.3.4). This change is because di®erent interference levels require di®erent RSS
levelstoprovidethesameleveloflinkreliability. Foronepairofsenderandinterferer,we
intentionallychangethedefaulttransmissionpoweroftheinterferer(whichresultsinthe
RISbetween-74.2and-60.5dBm)toseethee®ectofRISchangeontheSINRthreshold
apart from the hardware e®ect. Figure 5.9 marks these results with triangles. This RIS
88
level change causes a change in SINR threshold similar to our previous observations,
with a 1.9 dB gray region.
In the ¯gure, the circles represent experiment results corresponding to having dif-
ferent sender hardware for a given ¯xed interferer. This sender hardware change results
in about 3.1 dB gray region. The width of gray region varies between 1.6 and 3.6 dB
for di®erent individual interferers with 9 di®erent senders. Overall, we observe a 6.1 dB
wide gray region in the testbed experiments.
Thus, the testbed experiments con¯rm the two identi¯ed causes of SINR threshold
variation (hardware combination and measured signal strength). These causes can ex-
plain the high variation in SINR-to-PRR mapping observed in previous experimental
studies [3], and strongly suggest that constant SINR-to-PRR mappings will not model
all realistic situations. Upper layer protocols designed based on the constant SINR
threshold assumption may therefore be ine±cient or work incorrectly.
5.3.6 Modeling the SINR Threshold
Now that we have identi¯ed that hardware and signal strength each a®ect the SINR
threshold, we propose a simple simulation model for single-interferer scenarios that
considers these e®ects. We also show that this model can allow very di®erent commu-
nications patterns than simpler models of intereference.
Based on the collected data in the testbed (shown in Figure 5.8), we model the RSS
and SINR threshold relationship with a quadratic function. We then allow hardware
choicetoshiftthismodelwithanormaldistributionaroundourobservedmean,selected
onceeachsimulationforeachpairofnodes. Wehaveveri¯edthataquadratic¯tssignal
strength reasonably well, but con¯rming the normal distribution of hardware is an area
of future work. (We do not have enough hardware combinations to con¯rm normality
89
20 40 60 80 100 120 140 160
0
20
40
60
80
100
120
140
160
Link number
Number of CCable links
Capture−aware
Capture−unaware
Figure5.11: ThenumberofCCablelinkcomparisonbetweenthetwosimulationmodels
at this time.) The model for SINR threshold (SINR
µ
) for sender S at a given RSS is
therefore:
SINR
µ
(S;RSS)=®
2
RSS
2
+®
1
RSS+®
0
+³
S
(5.3)
where ³
S
»N(0;¾
2
)
Where we set ®
2
= ¡0:0305, ®
1
= ¡3:855, ®
0
= ¡116:91. The hardware e®ect is
modelledasazero-meanGaussianrandomvariable³
S
withavarianceof¾
2
=1:33,that
moves the curve up and down. This single-interferer model represents one application
of our experiments to modeling the reception of real radios in simulation.
Figure 5.10 shows the e®ects of using our newly proposed capture-aware simulation
model compared to the traditional packet collision model which assumes a collision if
there is a concurrent packet transmission within range. This ¯gure visually compares
90
the concurrently communicatable links (in short, CCable links) together with the com-
munication from the node 183 to 81 (comm
183¡>81
, two nodes in the middle of the
testbed joined by thicker line) under the two di®erent simulation models.
Figure5.10(a)showsthecommunicationlinksfromthesendernode183toitsneigh-
bor nodes. Each link represents the interference from the sender as well. Solid lines
show over 90% PRR links and dotted lines show the links with between 10% and 90%
PRR. We use link qualities empirically measured in the 25 node testbed shown in the
¯gure. We measure both PRR and RSS at 0 dBm transmission power with 50 packets.
Figure 5.10(b) shows the CCable links together with the comm
183¡>81
with the
traditional simulation model. Solid lines show the CCable links that are available with
over 90% PRR and the dotted lines represent the links that might be CCable when the
packet from the node 183 is not received due to unreliable connection.
InFigure5.10(c),wepresentthelinksthatcanbeCCabletogetherwiththecomm
183¡>81
based on our capture-aware SINR threshold model. To be on the safe side from the ob-
served hardware variation, we added an extra 4 dB to the calculated SINR threshold
value from our model, which ensures the SINR threshold to be at least the maximum
SINR threshold value we observed in our testbed experiment. This guarantees that the
CCable links can be used regardless of hardware variation.
As we can easily compare between Figures 5.10(b) and 5.10(c), our new capture-
aware SINR threshold model shows signi¯cantly more CCable links. We performed
the same comparison for all links with PRR over 90% (174 total links) in the testbed
assuming every reliable link can transmit a packet using the link. Figure 5.11 com-
pares the number of CCable links between the new capture-aware model (top solid line)
and the traditional capture-unaware simulation model (bottom dahsed line). The new
capture-awaremodeltypicallyprovidesabout 3timesmoreCCablethantheold model.
This example concretely illustrates the utilityof our experimentalstudy in enabling the
91
development and evaluation of novel capture-aware MAC protocols. It also suggests
that that current RTS/CTS based medium access protocols are overly conservative, a
potential area of future work.
5.4 Experimental Study of Multiple Interferers
In this section, we consider concurrent packet transmissions involving more than two
transmitting nodes (i.e., involving two or more interferers). In Section 5.4.1, we de¯ne
how we empirically measure the joint interference as well as a conventional estimator
assuming additive interference strengths. We then show that the measured joint in-
terference generally does not match the additive assumption (Section 5.4.2). We then
show in Section 5.4.3 that it is di±cult to estimate the joint interference in the presence
of more than one interferer, because measurements show high variance. Finally, we
investigate the impact of multiple interferers on the SINR threshold in Section 5.4.4.
5.4.1 Joint Interference Estimator
When there is a single interferer (IFR) (i.e., a concurrent packet transmitter), we can
estimate the interference strength from this interferer based on the individually mea-
sured received interference strength (RIS). We now consider how joint interference may
be estimated in the presence of multiple interferers.
The following two metrics are estimators of joint interference, with n interferers and
k measurements from a given setup:
JRIS(e)=
n
X
i=1
RIS
IFRi
JRIS(m)=
P
k
i=1
JRIS
i
k
(5.4)
92
−16 −14 −12 −10 −8 −6 −4
−90
−85
−80
−75
−70
−65
Tx power of IFR2 (dBm)
RIS (dBm)
Min
Max
JRIS(e)
JRIS(m)
IFR1
IFR2
Figure 5.12: Two node experiments IFR2 at -75 dBm and IFR1 between -82{ -70 dBm
RIS at the receiver
JRIS(e)
1
istheestimationbasedonthesummationofindividualRISmeasurements
from each interferer where RIS measurement for each interferer is taken without any
interference in the same channel. JRIS(e) is a conventional way to calculate the inter-
ference from multi-sources in theoretical studies.
JRIS(m) uses the mean of multiple JRIS measurements as the estimator of joint
interference. JRIS(m) is a more practical method to estimate the joint interference
frommultipleinterferersusingthecollected, actualJRISmeasurementsinrealsystems.
5.4.2 Additive Signal Strength Assumption
We ¯rst investigate the following question: \is the additive signal strength assumption
valid in the measurements with low-power RF radios?". Here, our aim is to examine
1
Note that we must compute in linear units of power, so we convert values to milliwatts for addition,
then back to dBm.
93
−75 −73 −71 −69 −67 −65 −63 −61 −59 −57
−90
−85
−80
−75
−70
−65
−60
−55
−50
Tx power of IFR2 (dBm)
RIS (dBm)
Max
Min
JRIS(e)
JRIS(m)
IFR1
IFR2
Figure5.13: Experimentresultswithtwointerferers(IFR1andIFR2)causingequivalent
RIS at the receiver
the validity of using the measurement-based JRIS(m) as an interference estimator in
practice.
5.4.2.1 Two interferer experiments
We carefully design experiments (as described in Section 5.2.2) to measure the JRIS at
the intended receiver. First, we run some experiments with two concurrent interferers
(IFR1 and IFR2) to see the e®ect of combined interference on the JRIS values. IFR2
uses constant transmission power and the RIS from IFR2 is around -75 dBm at the
receiver. IFR1 varies its transmission power between -17 to -4 dBm and this power
adjustment results in the RIS between -82 to -70 dBm at the receiver.
Figure 5.12 presents the following information:(1) IFR1 and IFR2: mean RIS at the
receiver from each interferer (IFR1 and IFR2) measured individually without any inter-
ference (2) JRIS(e): joint interference estimation based on the additive signal strength
assumption(3)JRIS(m): meanmeasuredJRISfrombothinterferers(4)Min-Max: min-
imum to maximum value range of JRIS measurements in two dotted lines. Each data
94
point represents a mean measurement value over 100 experiments with 230B packets.
Error bars show 95% con¯dence intervals for JRIS(m) values.
While it is intuitive to see the dominance of stronger interference signal over the
weakerinterferenceduetothelogarithmicnatureofdBmunit,westillexpecttomeasure
ahigherJRIS(m)valuefromtheintensi¯edjointinterferencethansingleRISwhenboth
interferers have equivalent RIS at the receiver, as with the JRIS(e) estimates. However,
the JRIS(m) value follows the single stronger RIS value within the 95% con¯dence
interval even at the point where both interferers have about the same individual RISs
at the receiver (e.g. when transmission power of IFR1 is -10 dBm in Figure 5.12).
Even though JRIS(e) value is normally considered as an estimator of joint inter-
ference, our experiments show that the measured JRIS(m) values are generally always
lower than the estimated JRIS(e) values.
5.4.2.2 Additivity and RIS levels
To investigate if the observation from -75 dBm individual RIS level holds at di®erent
interference strength levels, we perform further experiments with two interferers at
multiple RIS levels between -76 and -59 dBm. Figure 5.13 shows the experiment results
for the cases when both interferers generate equivalent RIS at the receiver at di®erent
interferencestrengthlevels. WhiletheJRIS(m)valuenormallyfollowsthestrongerRIS
value when the RIS values are not equal as well as at extreme values of RIS when they
are equal for all interferers, in this experiment we ¯nd some intermediate RIS levels
(around -68 dBm) where the JRIS(m) value is larger than the stronger value. However,
it is still the case that the JRIS(m) is smaller than the JRIS(e) value.
95
# of Individual JRIS(e) JRIS(m)
IFRs RISs (dBm) (dBm) (dBm)
1 -72.9 | | | -72.9 -72.9
2 -72.9-73.4 | | -70.1 -72.7
3 -73.0-73.5-73.3 | -68.5 -70.4
4 -72.9-73.5-73.5-73.0 -67.2 -68.9
(a) RIS from each interferer around -73 dBm
# of Individual JRIS(e) JRIS(m)
IFRs RISs (dBm) (dBm) (dBm)
1 -68.8 | | | -68.8 -68.8
2 -69.0 -68.7 | | -65.8 -67.1
3 -69.1 -68.6-68.7 | -64.0 -64.2
4 -68.9 -69.0-68.8-68.2 -62.7 -63.7
(b) RIS from each interferer around -68.8 dBm
Table 5.5: Comparison of JRIS(e) and JRIS(m) metric for JRIS estimation at two
di®erent individual RIS levels
5.4.2.3 Additivity with Additional Interferers
To see the e®ect of additional interferers on JRIS(m) and JRIS(e), we have performed
experiments with one to four interferers each with equivalent individual RIS levels. We
incorporate the change in JRIS(m) value at di®erent RIS levels (identi¯ed in previous
section) by repeating the same experiments at the following two RIS levels: -73.0 dBm
(where JRIS(m) is close to the single strongest RIS) and -68.8 dBm (where JRIS(m) is
higher than the single strongest RIS). We have individually measured RIS values from
each interferer and the JRIS value from di®erent number of concurrent interferers over
the 75 packet experiments for each setup.
WhenwecomparetheJRIS(e)andJRIS(m)attwodi®erentRISlevelsinTable5.5,
there are smaller di®erences between the two interference estimators at -68.8 dBm in-
dividual RIS. This is in agreement with our previous results that shows higher signal
strength additivity at -68.8 dBm than at -73 dBm (presented in Figure 5.13). These
96
−78 −76 −74 −72 −70 −68 −66 −64 −62 −60
0
10
20
30
40
50
60
70
RIS (dBm)
Frequency
JRIS
RIS1
RIS2
Figure 5.14: Frequency distribution of JRIS measurement values for two interferer ex-
periments
results with multiple interferers also con¯rm our previous observation that the JRIS(e)
estimates stronger interference than measured by JRIS(m).
5.4.3 Variation in JRIS Measurements
When we look at the each JRIS measurement value rather than the mean value (i.e.,
JRIS(m)), thereissigni¯cantvariationintheJRISmeasurementsespeciallywhenIFR1
and IFR2 have close interference strength at the receiver. The wide minimum to maxi-
mumJRISvaluerange(inFigure5.12and5.13)clearlyrepresentsasigni¯cantvariation
in JRIS measurements. The standard deviation of the JRIS measurements is around 3
dBm (2.75 to 3.65 dBm) over the experiments with di®erent levels of two equivalent in-
terference strength (shown in Figure 5.13). And the minimum-to-maximum JRIS range
is consistently very wide throughout the experimented signal strength levels.
Figure 5.14 shows one example of the frequency histogram from the 300 JRIS and
RISmeasurementsfromtwointerfererexperiments. WhileRISmeasurementsfromeach
interferer (RIS1 and RIS2) are clustered together near the mean value (-68.2 and -68.5
97
−74 −72 −70 −68 −66 −64 −62 −60
3
4
5
6
7
8
9
10
11
12
Received interference strength (dBm)
SINR threshold
1
2
3
4
2
3
4
1
2
3
2 3
1
2
2
1 IFR cases
JRIS(m) JRIS(e)
JRIS(m)
JRIS(e)
JRIS(m)
JRIS(e)
−73 dBm
−68.8 dBm −64.1 dBm
Figure5.15: SINRthresholdchangeswithdi®erentnumberofinterfererswhichchanges
the received interference strength
dBm respectively), the JRIS values are widely distributed around its mean value (-66.2
dBm). This histogram clearly shows the wide variation from the multiple interferers in
the JRIS measurements (where the standard deviation is 3.02 dBm) compared to the
singleinterferencecases(wherethestandarddeviationis0.30and0.37respectively)and
some additive behavior (about 2 dBm increase in JRIS(m)) from multiple interference
atthegivenindividualRISlevel. TheJRISvaluesarestillnormallydistributed. Similar
frequency distributions are observed from the experiments with two to four interferers.
In wireless communication protocols, collecting the received signal strength indi-
cation (RSSI) is a natural way to estimate the current interference plus noise level.
However, single RSSI measurement (which we call RIS for interference measurement)
cannot be an appropriate estimator of current interference if there is any possibility of
having multiple interferers, due to the signi¯cant variance in the measurement values.
98
5.4.4 E®ects of Joint Interference
BycomparingJRIS(m)andJRIS(e),wehaveevaluatedhowmeasuredjointinterference
levels from multiple interferers compare to estimated joint interference. We next relate
this back to the SINR threshold for reliable packet reception.
To evaluate the SINR threshold with multiple interferers, we vary both the number
ofinterferersandtheindividualRISlevels. Weconsiderfrom1to4interferers, andRIS
levels of -73, -68.8, and -64.1 dBm, matching the experiments in Table 5.5 and adding
the -64.1 dBm level.
Figure5.15showstheexperimentresults,comparingtheSINRthresholdagainstthe
receivedinterferencestrength(RIS).Wemarkeachdatapointwiththenumberofinter-
ferers in each experiment and also indicate the method of joint interference estimation
(either JRIS(e) or JRIS(m)) for each branch. The experiments in the same branch use
thesameindividualRISlevel. AsindicatedinSection5.4.1,JRIS(e)valuesarepredicted
from individually measured RIS values, while JRIS(m) are joint measurements.
Wedrawthreeconclusionsfromthisexperiment. First,weconsiderhowSINRvaries
as we add interferers at a given RIS level. We have three examples in the strings of
experiments starting at -73, -68.8, and -64.1 dBm. Regardless of the estimator used
(JRIS(m) or JRIS(e)), we observe that additional interferers raises the SINR level re-
quiredtosuccessfullyreceiveapacket. Thistrendisclearestforthe-73dBmcasewhere
1 to 4 interferers are considered, but it holds for all three cases.
Second, we can compare SINR threshold for two di®erent estimators JRIS(e) and
JRIS(m) (i.e., the dotted and solid lines in the ¯gure). We ¯nd that JRIS(e) has
consistentlylowerSINRthresholdthanJRIS(m). RecallfromSection5.4.2thatJRIS(e)
has a consistently higher estimation of interference level. A lower SINR threshold with
higher interference estimation sounds counterintuitive, but this is a consequence of the
way in which the SINR threshold is calculated. We have the measured received signal
99
strength and its corresponding packet reception rate from the experiments. The only
di®erence is in the interference level obtained by the two di®erent estimators. We
calculatetheSINRthresholdwiththispre-identi¯edRSSandtheestimatedinterference
with both methods, taking into account the ambient noise level. Hence the JRIS(e)
estimator, which o®ers a higher level of interference, results in a lower SINR threshold.
This illustrates the point that a careful selection of interference estimator is important
because that can signi¯cantly a®ect the calculated SINR threshold value.
Finally, we can compare SINR threshold values as the JRIS increases. JRIS will
rise either due to increase in the individual RIS in our three sets of experiments, and
also due to increase in the numbers of interferers. In Section 5.3.4 and 5.3.5 we show
that SINR threshold values changes at di®erent signal strength levels. We highlight
the variation in SINR threshold with a single interferer at di®erent RIS levels with
an arched, dashed line at the bottom. We may perhaps expect multiple interferers
to generally follow a similar trend. Unfortunately we do not have enough data to
conclusively support or refute this trend for multiple interferers. The trend in two
interferers shows a monotonically decreasing trend but this could be due to missing
points at lower power levels. Investigating this further is an area of future work.
5.5 Preliminary evaluation of 802.15.4 Radio
We performed brief experiments with MicaZ motes equipped with CC2420 [11] radios
toverifythatourresultsapplytootherlow-powerradiossuchas802.15.4. TheCC2420
uses O-QPSK modulation with direct sequence spread spectrum (DSSS), unlike the
CC1000's FSK, and it operates at 2.4GHz at 250Kbps instead of 465MHz at 56Kbps.
We performed experiments using the same methodology from Section 6.4.1 with the
802.15.4 radios. We use one synchronizer and one receiver and two concurrent packet
senders, one as a sender and the other as a interferer. Each concurrent packet sender
100
varies its transmission power between -25 and 0 dBm at eight di®erent power levels.
For one set of experiments, we run experiments at 64 di®erent transmission power
combinations of two concurrent senders (SRC1 and SRC2). We measure a PRR with
50 data packets at each transmission power level.
We performed 25 sets of repeated experiments at two di®erent concurrent sender
locations. The distance from the senders to a receiver was 3 and 4 meters respectively
forSRC1andSRC2forthe¯rstnodelocation,andonlySRC1isrepositionedto5meter
distance from the receiver in the second node location. We use a 128 byte packet size
(the maximum packet size for 802.15.4 radio) for our experiments. The experiment was
performed in a closed room with no movement.
Figure 16 shows the SINR-to-PRR relationship as both SRC1 and SRC2 power
varies. As in Figure 5.4, we can see that SRC1 captures the channel on the left of the
¯gure, whileSRC2transmitssuccessfullyontherightwhenitspowerisstronger. These
results con¯rm that the capture e®ect we observe in the CC1000 also occurs with an
802.15.4 radio in the CC2420, in spite of a higher bit rate and di®erent modulation
scheme. We also observe that some hardware variation still exists in this new radio (as
weobservedpreviouslyinSection5.3.2). ThiscanbeseenaroundSINR0inFigure5.16,
when on the right, SRC1 is able to capture the channel at SINR values between 0 and
1, while on the left, SRC2 is unable to receive until SINR is greater than 1dB (to the
left of -1 dB on the graph). The minimum SINR value which always provides 90% PRR
was 3.87 dB for SRC2 and 2.69 dB for SRC1.
Finally, two di®erences between the radios. While we observe around 4 dB gray
region for the CC1000 (Figure 5.6) with received signal strength change, the CC2420
shows 2{3 dB gray region, also providing lower SINR threshold. Most of the time
higher than 2 dB of SINR value consistently provide reliable packet communication
in our experimental results. A likely explanation for this di®erence is that the DSSS
101
−10 −8 −6 −4 −2 0 2 4 6 8 10
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
SINR (dB)
PRR
Figure 5.16: SINR to PRR relationship: preliminary results with CC2420 radio
modulation is better at rejecting noise than the simpler approach in the CC1000. Also,
we did not observe a strong relationship between SINR threshold and received signal
strength (Section 5.3.4).
Although these preliminary results suggest that several of our ¯ndings hold on this
newer radio, additional experiments are needed to draw more careful conclusions.
5.6 Summary
In this chapter we have presented experimental analysis of the e®ects of concurrent
packet transmissions in low-power wireless link communications. We have con¯rmed
the capture e®ect and the existence of the SINR threshold which ensures the successful
delivery of the strongest packet under the concurrent packet communication situations.
Our main contributions and ¯ndings are as follows:
² We have performed the ¯rst systematic experimental study which veri¯es a dif-
ference between the conventional approximation of the interference e®ect and the
102
real-world behavior of concurrent packet transmissions. Our experimental study
provides new guidelines for more realistic simulation models.
² Our study shows that the SINR threshold is not a constant value, but that it
depends on the transmitter hardware and the signal strength level. While the
combinations of di®erent hardware and signal strength in the testbed generate
large (about 6 dB)grayregion with mixedreception rate at thesame SINR value,
the gray region is small for a ¯xed hardware combination at the same signal
strength level.
² Upper layer protocols that assume a constant SINR threshold can fail or be in-
e±cient due to the signi¯cant variation in SINR threshold. Protocols designed
considering capture e®ects and variability in SINR threshold will be more de-
pendable and e±cient.
² Single RSSI value measurement is not always a good estimator of current inter-
ference level because there is a large variation in measured signal strength in a
multiple interference situation.
² The measured interference from multiple transmitters is generally less than the-
oretically predicted by the assumption that interference is additive. For a given
measured signal strength, therefore, the measured joint interference results in
higher calculated SINR threshold values than predicted by theory.
² The SINR threshold generally increases with the number of interferers.
Understanding the e®ects of concurrent packet transmissions in low-power wireless
networks provides a fundamental knowledge for more e±cient interference-aware pro-
tocol design. In this study, we quanti¯ed SINR threshold for some selected hardware
andobservedsigni¯cantcapturee®ectfromconcurrenttransmissions. Thisimplieshigh
103
possibility of concurrent packet communication. With careful selection of transmission
power for simultaneous transmitters, we can provide successful packet reception (i.e.,
channel capture) at multiple receivers at the same time.
104
Chapter 6
Evaluating the Importance of Concurrent Packet
Communication
6.1 Overview
Concurrent packet transmission | transmission by two or more senders within max-
imum communication range of each other's receiver at the same time over the same
channel | has been considered harmful and avoided in wireless communication. Pro-
tocols such as 802.11 explicitly prevent concurrent transmission with carrier sensing
and by exchanging RTS/CTS messages (as proposed earlier [5]). Approaches such as
RTS/CTS handshake reduce multi-hop wireless throughput by blocking all other trans-
mission around both the sender and receiver.
The possibility and promise of concurrent packet transmission has recently been
demonstrated in several studies [39, 43, 66, 79, 83] including our thorough experimental
studypresentedinthepreviouschapter. Thebene¯tsofconcurrentcommunicationseem
obvious, when it is possible, because it will improve spatial reuse, resulting in higher
throughput especially when tra±c is heavy and network density is high. However, it is
notobviousthathowoftenconcurrenttransmissionispossible. Forexample, ifonepair
of nodes always transmits at the lowest power level possible, then any other concurrent
transmissions will increase the e®ective noise and force the original pair to raise its
105
Contributions Section
CCable link and optimal Tx power decision rule 6.3
New metric for CCability (concurrent communication) 6.4.2
Quanti¯cation of CCability: User Controllable parameters
Higher Tx power increases CCability for ¯xed power 6.4.3
Shorter link distance increases CCability 6.4.4.1
Lower SINR threshold increases CCability 6.4.4.2
Tx Power control increases CCability 6.4.4.3
Quanti¯cation of CCability: Uncontrollable parameters
E®ect of Path loss exponent on CCability is dependent 6.4.5.1
on location and power °exibility
CCable links are prevalent in real-world 6.4.5.2
Concurrent transmission is at least capturable in general 6.4.6
Experiments with MicaZ motes validate simulation results 6.5
Table 6.1: Key ¯ndings from concurrent communication study
power. This coupling may mean that there is never a signi¯cant bene¯t to concurrent
transmission.
To illustrate this question, Figure 6.1 shows three con¯gurations where two pairs of
nodes send concurrently, S1 to R1 and S2 to R2. We assume a nominal radio range of 8
m at -10 dBm transmission power. In case (a), the pairs are 10 m apart and can easily
send concurrently, even with an RTS/CTS-based MAC protocol, since nodes are out of
range of each other. On the other hand, in case (c), R1 and S2 are only 2 m apart, and
no attempt at concurrent transmission will be successful. Although S1's transmission
can be received at R1, if S2 attempts to send concurrently, S1 must raise its power due
to interference from S2, and this higher power forces S2 to raise its power, ad in¯nitum.
However, there are intermediate cases where concurrent transmission is possible (as
suggested in [79]). For example, when the R1-S2 distance is 7 m as in case (b), then
802.11's RTS or CTS will block concurrent transmission. However, with proper MAC
support both S1 and S2 can simultaneously capture the channel at its corresponding
receiver. We look at this example and others in more detail in this chapter.
106
(a)
-5 m
S1
-0 m
R1
10 m
S2
15 m
R2
(b) S1 R1
7 m
S2
12 m
R2
(c) S1 R1
2 m
S2
7 m
R2
Figure 6.1: Two concurrent packet communications at three di®erent locations
Table6.1summarizesthecontributionsoftheworkpresentedinthischapter(which
appears in [15]). First, we develop a simple decision rule to decide when concurrent
communication is possible while minimizing transmission power. With this decision
rule, we can determine if concurrent transmission is ever possible for a given topology,
and then compute the optimal (minimal) transmission power settings for successful
concurrent communication, if possible.
Through simulations, we then systematically quantify opportunities for concurrent
communication with and without transmission power control as many radio and envi-
ronmental parameters vary, including node position, mean and variance of path loss,
signal-to-interference-plus-noise-ratio (SINR) threshold, and granularity and range of
transmission power control.
We also introduce and use a new metric to estimate the feasibility and bene¯ts from
allowing concurrent transmission for di®erent environmental and hardware conditions.
Our simulations show that often, 40{75% of the time, depending primarily on distance
and location, two pairs of nodes can communicate concurrently. We can observe large
region where concurrent communication is possible even with ¯xed transmission power,
but dynamic power control signi¯cantly improves concurrent communications. We vali-
datekeyresultsthroughexperimentsoverMicaZmoteswith802.15.4radios,con¯rming
concurrent transmission is possible and validating our simulation results.
107
S1 R1
S11
(Dist: 5 m)
I12
S2 R2
S22
(Dist: 5 m)
I21
(dist: vary)
Figure 6.2: Example scenario with two concurrent packet sender-receiver pairs varying
R1-S2 distance. PL
0
= 45, n = 4, SINR
µ
= 4, X
¾
= 0, N
1
= N
2
=¡95 dBm, Fixed
Tx power=¡10 dBm
6.2 Motivating Example
We de¯ne two (or more) transmissions as concurrently communicatable or CCable if
they can both successfully be received at the same time. Traditional MAC protocols
suchas802.11ensurecollision-freepacketcommunicationbycarriersensingfollowedby
RTS/CTS handshakes that bar communication from nodes within one hop of either the
sender or receiver, while other protocols (for example, TRAMA [63]) adopt a TDMA-
based schedule with two-hop neighbors in mind. In this chapter, we explore the use of
channelcaptureandtransmissionpowerselectiontoallowCCablecommunicationwhen
nodes are within range of each other.
In this section we explore how transmission power and node location interacts to al-
lowCCablecommunicationinsomecases. TointroduceCCablecommunicationwe¯rst
use simulation to explore how transmission power and source and destination location
a®ects the ability to communicate.
Figure 6.1 showed several scenarios where both senders (S1 and S2) concurrently
transmit packets to their corresponding receivers (R1 and R2). We generalize this
example with variable distance between R1 and S2 in Figure 6.2. We denote the signal
strength from sender i to receiver j as S
ij
, and the interference it generates at the other
receiver k as I
ik
. We de¯ne ambient noise at each receiver as N
j
. Throughout this
simulation,weusethesamelinkdistanceof5mbetweeneachsenderandreceiver. Under
108
1 2 3 4 5 6 7 8 9 10
−30
−25
−20
−15
−10
−5
0
5
10
15
Distance R1−S2 (m)
SINR (dB)
SINR(R1)
SINR(R2)
SINR threshold = 4
Capturable
CCable
(w/o PC)
CSMA
RTS/
CTS
CCable
(w/ PC)
Figure6.3: CCabilityfordi®erentschemes. SINRvaluesaremeasuredat-10dBm¯xed
Tx power
the constant radio and environmental parameter settings (presented in the caption of
Figure 6.1, we only vary the distance between the R1 and S2 and calculate signal
and interference value based on an exponential path loss model (details of this model
are explained in Section 6.3). We consider the SINR value greater than equal to 4
as a threshold for successful communication (i.e., SINR
µ
= 4) in this simulation. We
considerthesignalsarrivedfromunintendedsendersasinterferenceinSINRcalculation
at each receiver.
Figure 6.3 shows simulations as the S2-R2 pair of nodes moves right and left. We
compare the distance from R1 to S2 against the SINR values of the intended transmis-
sions when both senders transmit at a power of -10 dBm. At this transmission power,
either communication would succeed if it occurred separately, but here we can identify
four di®erent regions of communication when both senders transmit at the same time.
Starting at the right of the ¯gure, when R1-S2 is 9 m or further away, we call
this the CSMA RTS/CTS CCable region. Intuitively, at these distances, nodes just
cannotheareachotherandsospatialreuseallowsthemtooperateindependently. With
109
MAC protocols such as 802.11, carrier sensing and RTS/CTS handshake are used to
prevent concurrent transmission. The carrier sense threshold is always set to less than
equal value to the RSS which ensures successful packet reception. We use the received
signal strength (RSS) of -91 dBm as a packet reception threshold in our simulation
(based on the empirical results with MicaZ mote). We also choose the same RSS value
as packet reception threshold as the carrier sense threshold, which is the maximum
plausible value in this example. With lower carrier sense threshold (which is more
general setup in conservative 802.11 MAC), the 802.11 CC region becomes smaller. In
other words, longer than 9 m link distance between R1-S2 is required for 802.11 CC in
real implementations.
We call the next region from the right, when R1-S2 is just less than 7 m to 9 m,
CCable without transmission power control. Concurrent communication is possible
in this region at constant transmission power because both receivers can capture the
intended packet, since both have a higher SINR value than SINR
µ
at the given default
transmission power.
Next,wemarkCCableregionwhenwetakeadvantageoftransmissionpowercontrol.
There is an even more expanded CCable region, starting from about 3 m, compared to
the case with a static power. We will show how to select the power appropriately to
enable CC in this region, in the next section. The combined region from 3 m to 9 m
thus represents the zone where a sophisticated MAC could allow communication that
would be prevented by MACs that use CSMA and RTS/CTS, similar to 802.11.
The next region, with R1-S2 distance starting at about 1.5 m, is a case where
one sender (in this case S2) can capture the channel, but the other sender cannot
communicate successfully (only one receiver's SINR exceeds SINR
µ
). We de¯ne this
region as capturable region.
110
Finally, in the leftmost region, where R1-R2 is less than 1.5 m, neither pair can
communicate. The receivers are located too close together and neither can capture the
channel. Both receivers have SINR values lower than SINR
µ
and their transmissions
will always collide and corrupt each other. In this region, both senders need to increase
their transmission powers to meet the SINR
µ
condition for successful communication.
But, if one sender attempts to increase its transmission power to capture the channel,
thisfurtherincreasestheinterferenceattheotherreceiver. Theothersenderrequiresto
increaseitstransmissionpowertoovercomethisextrainterferenceanditonlyneutralizes
the bene¯ts from the transmission power control. Therefore, CC is never possible even
with transmission power control in this region.
These di®erent regions suggest the complex interaction between concurrent senders.
We next de¯ne and model concurrent communication more formally.
6.3 Mathematical Modeling
We begin by modeling mathematically when concurrent transmissions can occur for the
case of two senders and two receivers. For our modeling, we use the exponential path
loss model with log-normal fading [65, 98]:
PL(d)
dB
= PL(d
0
)
dBm
+10nlog(d=d
0
)+X
¾
dB
(6.1)
P
r
(d)
dBm
= P
t
dBm
¡PL(d)
dB
Here P
t
and P
r
are the transmission and reception power in dBm. The sender-
receiver distance is d, and d
0
is the reference distance for path loss (PL). X
¾
is the
variance in path loss due to multipath fading, modeled as Gaussian random variable
111
−25 −20 −15 −10 −5 0 5 10
−25
−20
−15
−10
−5
0
5
10
Transmission power of S1 (dBm)
Transmission power of S2 (dBm)
SINR at R1 = SINRth1
SINR at R2 = SINRth2
S2 Capture
power settings
CCable power settings
S1 Capture
power settings
(a) d
R1¡S2
=10 m, lf
1
=4, lf
2
=2
−25 −20 −15 −10 −5 0 5 10
−25
−20
−15
−10
−5
0
5
10
Transmission power of S1 (dBm)
Transmission power of S2 (dBm)
SINR at R1 = SINRth1
SINR at R2 = SINRth2
S2 Capture power settings
CCable
S1 Capture
power settings
(b) dR1¡S2 =4 m, lf
1
=1:8, lf
2
=0:8
−25 −20 −15 −10 −5 0 5 10
−25
−20
−15
−10
−5
0
5
10
Transmission power of S1 (dBm)
Transmission power of S2 (dBm)
SINR at R1 = SINRth1
SINR at R2 = SINRth2
S2 Capture power settings
No CCable
power settings
S1 Capture
power settings
(c) dR1¡S2 =2 m, lf
1
=1:4, lf
2
=0:4
Figure6.4: The CCabletransmissionpowerrelationshipbetweenS1and S2. PL
0
=45,
n=4, SINR
µ
=4, X
¾
=0
with zero mean and standard deviation ¾
dB
. This model de¯nes the path loss and the
received signal strength (RSS) at the receiver for a given transmission power level.
112
6.3.1 Power Setting for CCability
Forconcurrenttransmissiontobepossible,thereceivedSINRmustbeabovethethresh-
old for each receiver (SINR
µr
for receiver r):
S
11
dBm
¡10log(10
I
21
dBm
=10
+10
N
1
dBm
=10
) ¸ SINR
µ1
dB
(6.2)
S
22
dBm
¡10log(10
I
12
dBm
=10
+10
N
2
dBm
=10
) ¸ SINR
µ2
dB
For a given distance-based path loss model, such as the one we described in Equa-
tion 6.1, we get the following non-linear inequalities relating the transmission powers of
both senders, given a sender x to receiver y distance of d
xy
and transmission power of
P
t
(s) for sender s:
P
t
(S1) ¸ PL(d
11
)+SINR
µ1
+ (6.3)
10log(10
(Pt(S2)¡PL(d
21
))=10
+10
N
1
=10
)
P
t
(S2) ¸ PL(d
22
)+SINR
µ2
+
10log(10
(P
t
(S1)¡PL(d
12
))=10
+10
N
2
=10
)
Wecanvisualizethesenon-linearinequalitiesasregionsinaplotwheretheaxesrep-
resent the transmission powers P
t
(S1) and P
t
(S2). The intersection of regions would
then indicate when both conditions are satis¯ed simultaneously, i.e. when concurrent
transmissions are possible. From the above equation, we see that the shape of these
regions would be primarily determined by the path loss model and the inter-node dis-
tances. Figure 6.4 shows these regions for three di®erent node topologies.
113
Figure 6.4(a) shows regions corresponding to the non-linear inequalities for the sce-
nario shown in Figure 6.2 at 10 m of R1-S2 distance. Each line indicates the sender's
optimaltransmissionpowerwhichmeetstheSINRthresholdrequirementatitsintended
receiver with equality. The line with circles shows calculated S1's optimal transmission
powers if the S2's transmission power varies between -25 and 10 dBm as shown in
the Y-axis. The region to the bottom-right of this curve represents all combinations
of transmission powers that allow receiver R1 to capture the message. The line with
crosses similarly shows S2's calculated optimal powers for di®erent S1's transmission
power selections. The region to the top-left of this curve shows all combinations of
transmit powers that allow receiver R2 to capture the message. The overlapping re-
gion, therefore, shows the combination of transmission powers that allow for concurrent
transmission (i.e., these are the CCable power settings).
As shown in the plots in Figure 6.4, the extent and the existence of the overlapping
CCable region depends upon the inter-node distances. In particular, compared to (a),
(b) shows a smaller CCable region requiring higher transmit powers as the R1-S2 dis-
tance becomes smaller; when the R1-S2 distance is reduced even further in (c), we ¯nd
that the two regions no longer overlap.
The crossing point of the two lines in Figure 6.4 provides the optimal S1 and S2's
transmission power combination, which is the minimum transmission power setting for
CC. We can actually solve analytically for this crossing point (when it is possible) by
treating the inequalities from Equation 6.3 as simultaneous non-linear equations. This
114
yields the following expressions for the optimal transmission power settings for S1 and
S2:
P
t
(S1)=PL(d
11
)+SINR
µ1
+ (6.4)
10log(10
(P
t
(S2)¡PL(d
21
))=10
+10
N
1
=10
)
P
t
(S2)=10log(10
(PL(d
11
)¡PL(d
12
)+SINR
µ1
+N
1
)=10
+10
N
2
=10
)
¡10log(10
¡(SINR
µ2
+PL(d
22
))=10
¡10
(PL(d
11
)¡PL(d
12
)¡PL(d
21
)+SINR
µ1
)=10
)
Equation 6.4 provides the optimal transmission power to use for each sender S1
and S2 without exhaustive trial and error. Optimal power setting consumes minimum
energy for concurrent communication causing minimal interference to the network.
6.3.2 Topology Condition for CC
We can get some analytical insight into the impact of topology by deriving a neces-
sary and su±cient condition for CCability. In order to ensure that the simultaneous
non-linear equations have a bounded solution, it can be shown that the following topol-
ogy condition is necessary and su±cient (this condition ensures that the logarithm in
equation 6.4 has a positive argument):
SINR
µ1
+SINR
µ2
< (6.5)
PL(d
12
)¡PL(d
11
)+PL(d
21
)¡PL(d
22
)
115
Adopting the exponential path loss model from Equation 6.1, this can be written
as:
SINR
µ1
+SINR
µ2
< 10n(log(
d
12
d
11
)+log(
d
21
d
22
)) (6.6)
Let us de¯ne the location °exibility lf
i
for each sender i as the ratio of the distance
between a sender and its intended receiver to the distance between a sender and its
unintended receiver (i.e., interfered node). Thus lf
1
=
d
12
d
11
and lf
2
=
d
21
d
22
. The lf value
indicates the endurance level to the additional interference and noise under concurrent
transmission. Depending on the lf value of each sender (the higher the better), the
possibility of CC and the area of the CCable second sender location changes. This can
be seen in Figure 6.4.
6.3.3 CCability with Limited Power Range
Wehavenowshownhowtodetermineoptimaltransmissionpowerforconcurrenttrans-
mission: evaluate the topology condition to determine if CCability is possible (Equa-
tion 6.5, and if so, compute the optimal transmission powers with Equation 6.4). Real
hardware, however, has limited control over transmission power in terms of supported
range and granularity. If the optimal power computed above is supported, we are done.
If not, we next consider how to adapt to constrained choice of power settings:
If either optimal transmission power level is greater than that supported by the
hardware, CC is not possible.
Ifeitheroneoftheoptimaltransmissionpowersislowerthansupportedpowerrange,
we set the transmission power of this node to the minimum supported transmission
power for that node and calculate the transmission power of the other sender with
116
Equation 6.4. CC is possible only if the calculated transmission power is within the
supported power range.
Ifbothselectedtransmissionpowersarebelowthesupportedminimumpowerlevels,
we select the one with higher di®erence between the optimal transmission power and
its minimum supported power (let's call this ¯rst sender). It attempts to send at its
minimumsupportedpowerlevel,andwecomputetheotherrequiredtransmissionpower
accordingly. Ifthisexceedsitsrange,CCisnotpossible. Otherwiseweusethesuggested
power for the second sender, or bring it to the minimal supported range if it was lower
thanwhatissupported. Thisisbecausetheincreaseofthesecondsender'stransmission
power level to its minimum supported power range is still less than the increase of the
¯rst node's transmission power. Therefore, the ¯rst sender can tolerate the increase of
the second sender's transmission power level.
Thebasicruleisthattheincreaseofthesameamountoftransmissionpowerforboth
CCablesendersfromtheCCablepowerlevelalwaysallowsCCattheirnewtransmission
power level, if new power levels are supported. This is because the e®ect from the noise
decrease at higher transmission power or higher received power level. Therefore, the
same amount of signal and interference increase always ends up with higher SINR at
the receiver.
6.3.4 Summary
To summarize, the following are the two controllable factors that play an important
role in CCability. First, CCability depends on the location °exibility. Higher location
°exibility increases the possibility of CC, represented by a greater gap between the two
linesinS1andS2'soptimalpowerplot. Second, CCabilitydependsonthetransmission
power °exibility, which means the range of controllable transmission power (i.e., the
117
0 m -50 m 50 m
S1 R1
S11
S2 R2
Figure 6.5: Simulation topology: two sender-receiver pairs
minimum and maximum transmission power level). Higher transmission power °exi-
bility improves the CCability by increasing the chance of meeting the required CCable
transmission power for each sender. Therefore, we can expect higher CCability due to
higher °exibility from location and transmission power range.
6.4 Simulation
In this section we analyze the feasibility of concurrent packet transmission through sys-
tematic simulations to understand the e®ects of user controllable and radio parameters
such as node position, SINR threshold (SINR
µ
), range and granularity of transmis-
sion power control), and uncontrollable and environmental parameters like path loss
exponent (n), and the variance in path loss due to multipath fading (X
¾
) on CCability.
6.4.1 Methodology
When we simulate a speci¯c topology for a CCability test, the main variable parameter
of interest is the relative positions of the senders and receivers, rather than exact node
locations. To make exploration of the topology space manageable, we consider only
four nodes (two senders and two receivers) and place them on a line over 100 m as
shown in Figure 6.5. Even with this simple line topology we can test large number of
distance combinations; we will consider more general 2D topologies in our future work.
To characterize the topology we name the two sender-receiver pairs S1-R1 and S2-R2.
We de¯ne the origin of the line as the location of the receiver R1. In each simulation
118
−50 −40 −30 −20 −10 0 10 20 30 40 50
−50
−40
−30
−20
−10
0
10
20
30
40
50
R2 location (m)
S2 location (m)
II*
RI
RRT
RRA
SSA
SST
SI
IR IS
SR
RS
CCable
Region
CCable
Region
(a) Area index
−50 −40 −30 −20 −10 0 10 20 30 40 50
−50
−40
−30
−20
−10
0
10
20
30
40
50
R2 location (m)
S2 location (m)
Collision
Region
(b) Collision Region
0 m -50 m 50 m
S1 R1
IIT
IIA
SSA
SST
RRT
RRA
IS
SI
RI
IR
RS
SR
(c) Example for each area index
Figure 6.6: Simulation result example with area index: ¯xed transmission power of 10
dBm for (a) and (b). S1;R1=(¡10;0), n=4, SINR
µ
=4, X
¾
=0
119
we position S1 and R1 and then vary the locations of S2 and R2, testing for CCability.
We typically ¯x other parameters (S1-R1 distance, transmission power, and noise) then
plot CCability as a function of locations of S2 and R2, showing the minimum required
transmission power of either S1 or S2 for concurrent communication.
Our tested CC related parameter values change for di®erent hardware (especially
radio, antenna) and environmental conditions. Because there are many parameters to
explore,wegenerallyholdall¯xedbutoneforeachsection. Wealwaysuseambientnoise
Nof¡95dBmandpathlossatreferencedistancePL(d
0
)=¡35,andthefollowingisthe
most common setup for other parameters: path loss exponent n = 4, SINR
µ
= 4 dB,
X
¾
=0dB.OurradiosaremodeledontheChipconCC2420RFtransceiver,an802.15.4
radio widely deployed in the MicaZ and Telos-B motes. When we consider controllable
transmission power, we normally limit them between¡25 dBm and 0 dBm as with this
radio.
6.4.2 De¯ning Regions of Placement and the CCable Ratio
Tosimplifydiscussion,webeginbypresentinganexampleandshowingpotentialrelative
placement of the two pairs of nodes. Figure 6.6(a) shows one set of simulation results.
In this chapter, we list the locations of sender and receiver in order in parentheses (in
meters on the line) followed by sender and receiver id. This ¯gure shows a sample
simulation result when both senders use a ¯xed transmission power level of 0 dBm and
S1;R1 = (¡4;0). X-axis shows the R2 location in meters and Y-axis shows the S2
location. S1 and R1 locations are ¯xed for each simulation set.
To compare CCability with an RTS/CTS-based protocol we bound the nominal
communication range (without collision) with horizontal and vertical lines. Vertical
lines indicate the one-hop area around R1 that would be blocked by its CTS, and
horizontallinesshowthesameregionaroundR2. Wede¯necollisionregion astheregion
120
whereconcurrentpackettransmissionisprohibitedbyRTS/CTS-basedprotocoltoavoid
packet collision. Each sender only transmits a packet when its intended receiver is
locatedwithinitscommunicationrange. Therefore,theactualcollisionregionissmaller
than the whole areas within two vertical and horizontal lines. Figure 6.6(b) shows an
example of traditional collision region when S1;R1 = (¡4;0). For each simulation S1
and R1 locations are constant and the collision area is set based on these static node
locations.
Figure 6.6(a) shows two dark CCable regions. These regions let us quantify the
bene¯ts of CCability. We de¯ne CCable ratio as CCable region within collision region.
CCableratio = CCable region / Collision region
This ratio re°ects the fraction of area where a MAC protocol that supports concur-
rent transmission can send when a traditional MAC protocol would prohibit concur-
rent communication. A larger CCable ratio potentially allows greater overall network
throughput and more spatial reuse.
Next, to explain why these regions are CCable, Figure 6.6(c) shows twelve di®erent
con¯gurations of S2 and R2 relative to S1 and R1, and labels each with a three letter
code. The ¯rst two letters of each code indicate the location of S2 and R2 relative
to S1-R1: I means inside S1-R1, S means outside S1-R1 on S1 side, R means outside
S1-R1 on R1 side. We use the third character to indicates the direction of the S2-R2
communication, if necessary: A is away from S1, T is towards S1, or * is either.
Returning to Figure 6.6(a), we see that the regions which are CCable are typically
RR* or and SS*, where S2-R2 are on either side of S1-R1. They must be far enough
away not to interfere: CC is possible when S2-R2 are at 7 m and 12 m, and fails
when they are at 2 m and 7 m. These are similar cases as Figure 6.1(b) and (c).
121
However, here we can see all possible combinations of distance for successful concurrent
packetcommunication. Webroadenthisdiscussionaswegotoconsiderotherparameter
settings and node locations.
−50 −40 −30 −20 −10 0 10 20 30 40 50
−50
−40
−30
−20
−10
0
10
20
30
40
50
R2 location (m)
S2 location (m)
−20
−20
−10
−10
−10
−10
0
0
0
0
Figure 6.7: CCable regions with di®erent ¯xed transmission power levels. S1;R1 =
(¡4;0), n=4, SINR
µ
=4, X
¾
=0
6.4.3 Fixed Transmission Power Cases
Many protocols are designed assuming that nodes always transmit at the same power,
because some radios do not support and protocols do not exploit power control. We
consider ¯xed transmission power here, and generalize to controllable power in the next
section.
Figure 6.7 shows CCability as S2 and R2 move for three ¯xed transmission power
levels of ¡20, ¡10, and 0 dBm. We hold all other parameters constant as the caption
shows. This¯gureshowsthattheCCableregionislargerathighertransmissionpowers.
As transmission power grows from ¡20 to 0 dBm, we see the CCable ratio grow from
0.26 to 0.44.
122
The CCable ratio grows for larger powers because the higher interference from
stronger transmission is more than o®set by the increased signal strength. In addi-
tion, the larger transmission power increases the size of collision region that would
be reserved with an RTS/CTS protocol. Considering node placement (de¯ned in Fig-
ure 6.6(c)), with ¯xed transmission power we only see CCability in the RR* and SS*
regions (we will later show that power control allows CCability elsewhere). This limi-
tation is because the °exibility is limited when transmission power is ¯xed | there is
no power °exibility | so communication distance is the only factor that contributes
the signal and interference strength. Therefore, the sender always needs to be located
closer than the interferer for every receiver to have a positive SINR value which meets
the intended receiver's SINR threshold.
With ¯xed power, communication is only CCable when these conditions hold:
SINR
µ1
· 10nlog
µ
d
12
d
11
¶
=10nlog(lf
1
) (6.7)
SINR
µ2
· 10nlog
µ
d
21
d
22
¶
=10nlog(lf
2
)
As this condition implies, the leftover SINR value (i.e., signal strength) for each
communication cannot be used to increase the CCability in ¯xed power case; while we
can adjust SINR values between two receivers with distinct transmission power settings
for each sender (i.e., with power control) to improve CCability.
6.4.4 User Controllable Parameters
Userscancontrolsomeaspectsoftheirnetwork,includingtheirlocationandtheirchoice
of radio. We next consider CCability when we vary these parameters.
123
−50 −40 −30 −20 −10 0 10 20 30 40 50
−50
−40
−30
−20
−10
0
10
20
30
40
50
R2 location (m)
S2 location (m)
S1,R1=(−2,0)
(a) Optimal Tx Power of S1
−50 −40 −30 −20 −10 0 10 20 30 40 50
−50
−40
−30
−20
−10
0
10
20
30
40
50
R2 location (m)
S2 location (m)
S1,R1=(−2,0)
(b) Optimal Tx Power of S2
−50 −40 −30 −20 −10 0 10 20 30 40 50
−60
−50
−40
−30
−20
−10
0
10
20
X Coordinate of S2 (m)
Transmission power (dBm)
Tx Pwr of S1
Tx Pwr of S2
min power
max power
(c) Optimal Power Settings
−25 −20 −15 −10 −5 0
−25
−20
−15
−10
−5
0
Transmission power of S1 (dBm)
Transmission power of S2 (dBm)
SINR at R1 = SINRth1
SINR at R2 = SINRth2
S2 Capture
power settings
S1 Capture power settings
CCable
power settings
(d) CCable Power Settings
−50 −40 −30 −20 −10 0 10 20 30 40 50
−50
−40
−30
−20
−10
0
10
20
30
40
50
R2 location (m)
S2 location (m)
S1,R1=(−12,0)
(e) Optimal Tx Power of S1
−50 −40 −30 −20 −10 0 10 20 30 40 50
−50
−40
−30
−20
−10
0
10
20
30
40
50
R2 location (m)
S2 location (m)
S1,R1=(−12,0)
(f) Optimal Tx Power of S2
−50 −40 −30 −20 −10 0 10 20 30 40 50
−60
−50
−40
−30
−20
−10
0
10
20
X Coordinate of S2 (m)
Transmission power (dBm)
Tx Pwr of S1
Tx Pwr of S2
min power
max power
(g) Optimal Power Settings
−25 −20 −15 −10 −5 0
−25
−20
−15
−10
−5
0
Transmission power of S1 (dBm)
Transmission power of S2 (dBm)
SINR at R1 = SINRth1
SINR at R2 = SINRth2
No CCable power settings
S2 Capture power settings
S1
Capture
(h) CCable Power Settings
Figure 6.8: CCable regions and optimal Tx Power for S1 and S2 varying distance:
S1;R1 = (¡2;0) for (a)¡(d), S1;R1 = (¡12;0) for (e)¡(h), n = 4, SINR
µ
= 4,
X
¾
=0, S2;R2=(x;¡24) for (c) and (f), S2;R2=(¡5;10) for (d) and (h)
124
−5 −2 0 10
−90
−80
−70
−60
−50
−40
−30
−20
−10
0
Location: X Coordinate (m)
RSS (dBm)
RSS from S1
RSS from S2
Figure 6.9: Comparison of RSS from S1 and S2. S1;R1 = (¡2;0), S2;R2 = (¡5;10),
n=4, SINR
µ
=4, X
¾
=0
6.4.4.1 Location Change
One of the most important parameters is node location. An important design consid-
eration of any sensornet deployment is how many nodes will be deployed and where.
Withstationarynodes,particularlocationswillalloworprecludeCCability;whennodes
moving or are randomly positioned, we can at least characterize the probability of con-
current communication. To systematically explore the e®ect of node location we ¯x R1
at 0 m and move the other nodes. For a given experiment we typically ¯x S1 (and so
the S1-R1 distance), then test all combinations of S2-R2 placement.
Figure 6.8 presents CCable regions and optimal S1 and S2 transmission powers
settings for two di®erent S1-R1 distances of 2 m (top row) and 12 m (bottom row).
Transmission power is shown as grayscale on the right and center graphs (darker values
indicate greater transmission power, red is the darkest indicating the maximum power),
andcalledoutspeci¯callyforoneS2,R2caseintherightmostgraphs(6.8(d)and6.8(h)).
In general, S1's optimal transmission power is stable (Figures 6.8(a) and 6.8(e))
becauseneitherS1norR1moveinthesesimulations. However,therearesomelocations
where interference from S2 forces S1 to increase its transmission power: darker region
125
in Figure 6.8(a), for example, S2;R2 = (4;27). S2's transmission power spans a much
larger range (Figures 6.8(b) and 6.8(f)), mainly due to changes in the S2-R2 distance.
These ¯gures show considerable amount of CCable region inside the collision region:
CCable ratio values are 0.77 for an S1-R1 distance of 2 m and 0.43 for 12 m. Greater
CCabilityispossiblewhenS1andR1arecloserbecausetheycancommunicateatlower
power. In other words, longer distances imply higher interference and lower location
°exibility for CC.
Figure 6.8(c) and 6.8(g) show the S1 and S2's optimal transmission power setting
for concurrent transmission both when we relax the power control limitation (shown
in ¯lled symbols) and when we have ¡25 dBm and 0 dBm constrained power range
(shown in empty symbols). These ¯gures presents the optimal power change due to
hardware limitations. We can also see that at higher S1-R1 link distance of 12 m,
optimal transmission power (noticeably for S1) increases. Worse location °exibility
for S1 increases required transmission power for concurrent communication. Lower
°exibility means reduced chance of concurrent transmission under the same condition.
We can see that CCability is greatly reduced at longer distance (in Figure 6.8(h)).
Finally, when we compare CCable region in Figure 6.8 to ¯xed power (Figure 6.7),
we see that concurrent communication is sometimes possible in the SR or RS regions
with power control. SR and RS communications mean that S2 and R2 are on opposite
sides of S1-R1|sometimes S2-R2 can transmit over the heads of S1-R1! This unin-
tuitive communication becomes possible with a proper transmission power setting for
each sender; If S2 selects an appropriate transmission power which only increases the
interference, but does not corrupt the packet from S1, while S2 can still provide strong
enough signal strength for a packet reception at R2. Figure 6.9 compares the RSS from
each sender at di®erent node location under the same SR scenario presented in Fig-
ure6.8(d), and using theoptimaltransmissionpowershownin this ¯gure: ¡18:67dBm
126
for S1 and¡6:77 dBm for S2. In this case, both R1 (at 0 m) and R2 (at 10 m) receive
stronger signal strength from its intended sender, also meeting SINR threshold of 2 dB.
This ¯gure visually explains how concurrent communication is possible for this unlikely
situation with transmission power control.
−50 −40 −30 −20 −10 0 10 20 30 40 50
−50
−40
−30
−20
−10
0
10
20
30
40
50
R2 location (m)
S2 location (m)
(a) SINR
µ
=2
−50 −40 −30 −20 −10 0 10 20 30 40 50
−50
−40
−30
−20
−10
0
10
20
30
40
50
R2 location (m)
S2 location (m)
(b) SINR
µ
=5
−25 −20 −15 −10 −5 0
−25
−20
−15
−10
−5
0
Transmission power of S1 (dBm)
Transmission power of S2 (dBm)
RSS at R1 = SINRth1
RSS at R2 = SINRth2
SINR threshold = 5
SINR threshold = 2
CTXable
power settings
(c) CCable Power Settings
Figure 6.10: CCable regions with di®erent SINR
µ
: S1;R1 = (¡4;0), n = 4, X
¾
= 0,
S2;R2=(15;6) for (c)
6.4.4.2 SINR Threshold
Arecentworkhasshownthatdi®erenthardwarerequiresdi®erentsignal-to-interference-
plus-noise ratios (SINR
µ
) to reliably send data [79]. In this section, we want study the
e®ects of SINR
µ
on CCability.
Figure 6.10 shows the CCable region and CCable transmission power range for two
di®erent SINR
µ
values of 2 and 5 (¯xing S1 and R1 and -4 m and 0 m). We see that
a larger SINR
µ
value reduces the CCable region. A larger SINR
µ
reduces the power
°exibility of both senders, since each must have greater \headroom" to successfully
communicate (due to the summation of SINR thresholds in Equation 6.6 as shown in
Figure 6.10(c)). The CCable ratio of these cases are 0.75 and 0.61 for SINR
µ
of 2 and
5 respectively, so this loss of °exibility translates into 14% less opportunity for CCable
communication for this speci¯c case.
127
The e®ect of SINR threshold value changes under di®erent communication environ-
ments. When we reduce the path loss exponent from 4 to 3, the CCable ratio di®erence
increases to 0.25 (0.77 and 0.52 for SINR
µ
of 2 and 5 respectively). Lower n value
decreases the signal strength di®erence at the same link distance and the link distance
between the sender and interferer need to be greater for the same SINR
µ
. In other
words, the same SINR gap (or location °exibility) covers larger distance (or region).
Therefore, SINR threshold plays more signi¯cant role under lower n value situation.
This simulation shows that the impact of SINR threshold change (due to hardware
di®erences) cannot be ignorable, and it varies under di®erent environmental conditions.
−50 −40 −30 −20 −10 0 10 20 30 40 50
−50
−40
−30
−20
−10
0
10
20
30
40
50
R2 location (m)
S2 location (m)
Figure 6.11: CCable region comparison with and without power control. S1;R1 =
(¡15;0), n=4, SINR
µ
=4, X
¾
=0
6.4.4.3 Comparing Fixed and Dynamic Power Control
Now that we understand the e®ect of node location, we next quantify the advantage
of dynamic power control over ¯xed transmission power. To do so we compare the
relative sizes of CCable region with and without power control at three di®erent S1-R1
distances: 5 m, 10 m, and 15 m.
128
Figure 6.11 compares the CCable region with ¯xed power control (the dark gray
regions) with the additional area added with dynamic power control (the light gray
regions) when the S1-R1 distance is 15 m. For the ¯xed-power case S1 sends at the
maximum transmission power of 0 dBm, since this provides the largest CCable region
(Section6.4.3). InthiscasetheCCableratiois0.22withoutpowercontroland0.36with
power control; when the S1-R1 distances are 5 m and 10 m we see similar tendencies
(¯xed vs. dynamic CCable regions of 0.42 vs. 0.63 at 5 m and 0.33 vs. 0.48 for 5 m and
10 m, respective).
From this comparison and the simulation results presented in previous sections, we
conclude that power control provides signi¯cantly greater CCability than ¯xed power
controlfora giventopology. TheCCableregiondi®erencewithandwithoutpowercon-
trolbecomeshigherwhentheS1-R1distanceisgreater. Thelocation°exibilitybecomes
much worse at longer link distance, and the ¯xed power scheme cannot overcome lower
location °exibility because it does not have any power °exibility like dynamic power
control case.
−50 −40 −30 −20 −10 0 10 20 30 40 50
−50
−40
−30
−20
−10
0
10
20
30
40
50
R2 location (m)
S2 location (m)
C1:(−4,0)
(a) -4m, 8 levels
−50 −40 −30 −20 −10 0 10 20 30 40 50
−50
−40
−30
−20
−10
0
10
20
30
40
50
R2 location (m)
S2 location (m)
C1:(−4,0)
(b) -4m, 25 levels
Figure6.12: CCableregionswith8levels(MicaZ)and25levels(¯ne)Txpowercontrol
at 4 m link distance between S1 and R1: PL
0
=35, n=3.5, SINR
µ
=4, X
¾
=0
129
6.4.4.4 Power Control Granularity
Wehavealsosimulatedthee®ectsofthegranularityoftransmissionpowercontrol. The
Chipcon CC1000 supports transmission power levels between ¡20 dBm and 10 dBm,
selectable at 1 dBm increments across most of this range. The newer Chipcon CC2420
provides a similar range (from ¡25 dBm to 0 dBm), but it support only 8 distinct
settings over this range (¡25,¡15,¡10,¡7,¡5,¡3,¡1, and 0 dBm).
Figure 6.12 compares the simulation results between the case with 8 levels and 25
levels of ¯ner transmission power control at the same¡25 dBm to 0 dBm power range.
Finer level of transmission power control slightly increases CCable region (about 3% in
CCable ratio) with more possible transmission power combinations in general, and it
alsolowersthetransmissionpowerforsomeCCablelocations. Therefore, wecanexpect
some minor bene¯ts to the CCability and energy consumption with ¯ner transmission
power control.
−50 −40 −30 −20 −10 0 10 20 30 40 50
−50
−40
−30
−20
−10
0
10
20
30
40
50
R2 location (m)
S2 location (m)
(a) n=3.5
−50 −40 −30 −20 −10 0 10 20 30 40 50
−50
−40
−30
−20
−10
0
10
20
30
40
50
R2 location (m)
S2 location (m)
(b) n=4.5
−25 −20 −15 −10 −5 0
−25
−20
−15
−10
−5
0
Transmission power of S1 (dBm)
Transmission power of S2 (dBm)
SINR at R1 = SINRth1
SINR at R2 = SINRth2
n=4.5
n=3.5
CCable
power settings
(c) CCable Power Settings
Figure 6.13: CCable regions with di®erent path loss exponent. S1;R1 = (¡6;0),
SINR
µ
=4, S2;R2=(15;6) for (c)
6.4.5 Uncontrollable and Environmental Parameters
While some parameters are under user control, wireless propagation itself is known to
be highly variable and unpredictable. We next consider CCability as we vary path loss
130
exponent and path loss variance from the exponential path loss model with log-normal
fading (presented in Equation 6.1). These parameters correspond to greater variability
in propagations.
It is well known that this model only approximates wireless propagation, so we
supplement these results with testbed experiment in Section 6.5.
6.4.5.1 Path Loss Exponent (n)
Propagationenvironmentdistinguishesreceivedsignalstrengthandqualityforthesame
hardwareatdi®erentnodelocationsinwirelesscommunication. Thepathlossexponent
n is a primary parameter that determines signal strength in di®erent communication
environments. According to the prior study by Rappaport [65], we can model di®erent
environments with path-loss exponents between 1.6 to 6. A larger n increases the path
loss, decreasing the viable reception distance and, for a given distance, decreasing the
received signal strength and interference.
Figures 6.13(a) and 6.13(b) compare CCability with path loss exponents (n) of 3.5
and 4.5. Note that the lower n corresponds to a larger e®ective transmission range,
as shown by the dashed line box indicating the collision region. When we compare
Figure 6.13(a) with Figure 6.13(b), we can clearly see the di®erence in CCable location
betweenthesetwocases,butnotintheCCableratiovalue. WeobservethesameCCable
ratio of 0.56 from both cases with di®erent path loss exponent of 3.5 and 4.5.
When we compare the possible transmission power combination in 6.13(c), we can
see the higher n value increases the minimum transmission power for CC (transmission
power for S1 and S2 respectively changes from -26.8 dBm and -21.3 dBm to -19.9 dBm
and -12.4 dBm) because it reduces the communication range at the same transmission
power level, but higher n value also increases the possibility for CC by increasing the
e®ect of location °exibility in right side of Equation 6.6.
131
Eventhoughhighern valuecanincreaseCCableregion(accordingtoEquation6.6),
this advantage comes together with higher transmission power requirement for CC at
the same topology. Therefore, higher path loss exponent can increases CCable region
onlywhenitsupportsincreasedtransmissionpowerrequirement. However,transmission
power range is very limited especially with low-power wireless networks in general.
On the contrary, lower n value decrease the minimum transmission power required
for CC, but if the path loss exponent value becomes too low, it greatly reduces the
location°exibility(i.e., distancee®ect), andthiscanmakeCCabletwocommunications
(at higher n) non-CCable. Therefore, the relationship between the path loss exponent
and CCability varies depends on the given location and power °exibility.
6.4.5.2 Path Loss Variance: X
¾
Thereissigni¯cantevidencethatwirelesschannelsvarygreatlyovertimeandlocation[3,
77, 84, 93]. A common way to model this variance analytically is using a zero-mean
Gaussian random variable with standard deviation ¾, or X
¾
. Although this model does
notcaptureallreal-worldbehavior,weuseitheretosimulatecontrollablelevelsofpath-
loss variance. We vary the ¾ value and compare the results in Figure 6.14. This ¯gure
shows the simulation results when we introduce non-zero X
¾
value and use randomized
antenna gain for each di®erent node location. This ¯gure implies that there are wide
area where the CC is unexpected or inconsistently available (e®ectively gray regions of
CCability).
In addition, we observe that cross-pair communication, where S2-R2 surround S1-
R1, are more prevalent at higher path loss variance. We can observe the SR type
communicationatX
¾
=2(inFigure6.14(a)),andbothSRandRS typecommunication
at X
¾
= 4 (in Figure 6.14(b)). But, there is no cross-pair type communication for the
132
−50 −40 −30 −20 −10 0 10 20 30 40 50
−50
−40
−30
−20
−10
0
10
20
30
40
50
R2 location (m)
S2 location (m)
(a) X
¾
=2
−50 −40 −30 −20 −10 0 10 20 30 40 50
−50
−40
−30
−20
−10
0
10
20
30
40
50
R2 location (m)
S2 location (m)
(b) X
¾
=4
Figure 6.14: CCable regions with di®erent path loss variance X
¾
: S1;R1 = (¡4;0),
n=4, SINR
µ
=5
same con¯guration (shown in Figure 6.10(b)) without path loss variance. We conclude
that, in practice, CCability will depend strongly on current environmental conditions.
6.4.6 Capturable Region
We mainly discussed about CCable region in our simulation results. However, we ob-
served large portion of power settings that allows only one successful communication
under concurrent packet transmission (for example, we can see large S1 or S2 capture
powersettingsinFigure6.4). Understandingcapturable situationismeaningfulforboth
unicast and especially for broadcast communication.
Figure 6.15 presents CCability obtained from the simulation with two concurrent
packetsendersS1andS2,withcorresponding(onlyforunicastcommunication)receivers
R1 and R2. Both senders use a ¯xed transmission power of 0 dBm, which simulates
the maximum power available for MicaZ motes. We indicate di®erent types of capture
regions in di®erent colors. First, black shows the unicast capture region where at least
one of unicast communications (S1-R1 or S2-R2) is successful; this combines both S1
and S2 capturable regions. Unicast capture regions cover 90% of the collision region.
133
−50 −40 −30 −20 −10 0 10 20 30 40 50
−50
−40
−30
−20
−10
0
10
20
30
40
50
R2 location (m)
S2 location (m)
Figure 6.15: Capturable Regions. S1;R1 = (¡4;0), n = 4, SINR
µ
= 5, X
¾
= 0, Tx
power =0 dBm
Grayindicatesthebroadcast capture region wheretransmittedpacketcanbereceivedby
anyneighbornode(i.e.,eitherR1orR2). Grayregionismeaningfultobroadcasttypeof
communication, and corresponds to 97% of the collision region (black region is a subset
ofgrayregion). Lightgrayshowsthecollisionregionwheresuccessfulcommunicationis
not possible. There are only 3% of the collision region in which actual packet collision
happens.
These simulation results imply that we can expect signi¯cant number of successful
packet delivery under the traditional packet collision situation if MAC supports appro-
priate functionalities for concurrent packet communication (such as [83]).
6.5 Testbed Experiments
While the simulations in Section 6.4 are invaluable at systematically exploring the pa-
rameter space, multiple empirical studies suggest that analytic models do not capture
the complexity in wireless propagation [3, 77, 84, 93].
134
-3.6 m
loc1
-3.0 m
loc2
-2.4 m
loc3
-1.8 m
loc4
-1.2 m
loc5
-0.6 m 0 m 0.6 m
loc6
1.2 m
loc7
1.8 m
loc8
2.4 m 3.0 m
loc9
3.6 m
loc10
S1 R1 S2 R2
(a) Outside (S2 moves)
-2.0 m -1.0 m 0 m 1.0 m
loc1
2.0 m
loc2
3.0 m
loc3
S1 R1 S2 R2
(b) Inside (R2 moves)
Figure 6.16: MicaZ experiment topology with two sender-receiver pairs. Experimented
S1 locations: loc1{loc10 for scenario 1 and loc1{loc3 for scenario 2
Wethereforenextstudykeyparametersinatestbedwithrealsensornodestoverify
the ¯ndings of our simulations. We use low-power MicaZ motes equipped with CC2420
radios [11] to measure the received signal and interference strength and to test the
CCability under concurrent packet transmission situation with di®erent node topolo-
gies. The main objective of our experimental study is to demonstrate the feasibility of
concurrent transmission in the real systems.
6.5.1 Methodology
Our testbed experiments follow the methodology of recent studies of concurrent trans-
mission [79]. Like our simulation study, we use two sender-receiver pairs of nodes,
S1-R1 and S2-R2. To coordinate the senders, our experiments add a ¯fth node, the
synchronizer, that transmits a a packet to synchronize the concurrent packet senders.
We disable carrier sensing and random backo® functionality from the MAC layer to
allow concurrent packet transmission from multiple senders.
We consider two scenarios, outside and inside, as shown in Figure 6.16. In the
outside scenario S2 is always outside the S1-R1 pair, and S2 moves. We vary the S2-R2
distance, considering ten di®erent positions of S2, roughly every 60 cm. This scenario
corresponds to cases SR and RRA in Figure 6.6(c). In the second experiment, inside,
135
−25 −15 −10 −7 −5 −3 −1 0
−25
−15
−10
−7
−5
−3
−1
0
Transmission power of S1 (dBm)
Transmission power of S2 (dBm)
SINR at R1 = SINRth1
SINR at R2 = SINRth2
(a) S2 at -3.0m
−25 −15 −10 −7 −3 −1 0
−25
−15
−10
−7
−3
−1
0
Transmission power of S1 (dBm)
Transmission power of S2 (dBm)
SINR at R1 = SINRth1
SINR at R2 = SINRth2
(b) S2 at -2.4m
−25 −15 −10 −7 −5 −3 −1 0
−25
−15
−10
−7
−5
−3
−1
0
Transmission power of S1 (dBm)
Transmission power of S2 (dBm)
SINR at R1 = SINRth1
SINR at R2 = SINRth2
(c) S2 at -1.8m
−25 −15 −10 −7 −5 −3 −1 0
−25
−15
−10
−7
−5
−3
−1
0
Transmission power of S1 (dBm)
Transmission power of S2 (dBm)
SINR at R1 = SINRth1
SINR at R2 = SINRth2
(d) S2 at -1.2m
−25 −15 −10 −7 −5 −3 −1 0
−25
−15
−10
−7
−5
−3
−1
0
Transmission power of S1 (dBm)
Transmission power of S2 (dBm)
SINR at R1 = SINRth1
SINR at R2 = SINRth2
(e) S2 at 0.6m
−25 −15 −10 −7 −5 −3 −1 0
−25
−15
−10
−7
−5
−3
−1
0
Transmission power of S1 (dBm)
Transmission power of S2 (dBm)
SINR at R1 = SINRth1
SINR at R2 = SINRth2
(f) S2 at 1.2m
−25 −15 −10 −7 −5 −3 −1 0
−25
−15
−10
−7
−5
−3
−1
0
Transmission power of S1 (dBm)
Transmission power of S2 (dBm)
SINR at R1 = SINRth1
SINR at R2 = SINRth2
(g) S2 at 1.8m
−25 −15 −10 −7 −5 −3 −1 0
−25
−15
−10
−7
−5
−3
−1
0
Transmission power of S1 (dBm)
Transmission power of S2 (dBm)
SINR at R1 = SINRth1
SINR at R2 = SINRth2
(h) S2 at 3.0m
Figure 6.17: CCability in the outside testbed experiment as S2 is moved (presented
together with the expectation from simulation with our proposed formula). Circles
are CCable, triangles and squares are S1 or S2 capturable, and Xs indicate a collision.
Simulation results at the same topology are presented together with two dotted lines.
136
we place S2 between S1 and R1 so that the S2-R2 communication crosses S1-R1. We
then move R2 to three positions, from 1 to 3 m beyond R1; this corresponds to the case
IR.
The MicaZ supports 8 di®erent transmission power levels from ¡25 to 0 dBm. For
each position experiment, we ¯rst measure the signal and interference strength with 10
packets and then test the CCability with 25 concurrent packet transmissions for every
64 di®erent combinations of two senders' transmission power settings. We repeat the
same experiment twice for each topology to verify that the results are consistent; the
results were similar and we show only one experiment here.
−3.6 −3 −2.4 −1.8 −1.2 −0.6 0 0.6 1.2 1.8 2.4 3 3.6
0
10
20
30
40
50
60
S2 location (m)
Number of power combinations (our of 64)
CC
S1 Capture
S2 Capture
Collision
(a) CCability
−3.6 −3 −2.4 −1.8 −1.2 −0.6 0 0.6 1.2 1.8 2.4 3 3.6
−95
−90
−85
−80
−75
−70
−65
−60
−55
−50
−45
S2 location (m)
RSS (dBm)
S11
I21
S22
I12
(b) RSS at TxPwr =¡5 dBm
−3.6 −3 −2.4 −1.8 −1.2 −0.6 0 0.6 1.2 1.8 2.4 3 3.6
−10
0
10
20
30
40
50
S2 location (m)
SINR (dB)
SINR at R1
SINR at R2
topology condition
(c) SINR at TxPwr =¡5dBm
−3.6 −3 −2.4 −1.8 −1.2 −0.6 0 0.6 1.2 1.8 2.4 3 3.6
−25
−15
−10
−7
−5
−3
−1
0
S2 location (m)
Transmission power (dBm)
Expriment S1
Expriment S2
Simulation S1 (w/ pwr limit)
Simulation S2 (w/ pwr limit)
(d) Optimal transmission power
Figure 6.18: Experimental results at di®erent S2 locations with variable transmission
powers.
137
6.5.2 Results from the Outside Scenario
Figure6.17showstheabilityofnodesinourtestbedtoconcurrentlycommunicate(CC)
or capture the channel in our experiments at eight di®erent locations of S2 (out of 10
due to space). Each ¯gure shows 64 di®erent CCability tests with power of each of the
two senders on each axis. These are all supported power combinations from two MicaZ
senders. Results of each test are shown by di®erent symbols: ¯lled circles are CCable,
whileemptytrianglesorsquaresindicatecapturebyS1orS2, andXsindicatecollisions
where neither receiver can capture data.
To compare our experiments with simulation, we predict the CCable power settings
through simulation and plot these as two lines (as in Figure 6.4). The simulations
require parameters for the channel propagation model that we do not know, so we use
actuallymeasuredpathloss ateachlocationusingthedatapresentedinFigure6.18(b).
We also used observed values for SINR threshold (2 dB for MicaZ) and ambient noise
level for each node (¡96:3 dBm for R1 and ¡96 dBm for R2). We can see that our
simulation results provides very close match of experimentally observed CCability. We
will discuss the implication of this in Section 6.7.
Nine out of ten con¯gurations supported concurrent communications at some power
settings. OnlyR2placementat0.6m(veryclosetoR1)wasunabletoconcurrentlycom-
municate. Thisexperimentdemonstratesthelargeopportunityforconcurrenttransmis-
sion if MAC support for packet capture and appropriate power selection was available,
and RTS/CTS was revised. Nevertheless, current MAC protocols would prohibit many
of these opportunities to transmit in their carrier sense checks or through an RTS/CTS
handshake.
Figure 6.18(a) summarizes these experiments by comparing the number of power
con¯gurations that support CCability, capture, or collision out of the 64 possible power
138
combinations at each location. In some ways this the fraction of CCable power combi-
nations is not a useful metric, since an intelligent MAC would not select transmission
powerrandomly,butinsteadcouldselectwhateverpowerlevelwasbest(soideally,even
a single CCable con¯guration could be exploited). However, the percentage of CCable
con¯gurations does characterize the level of °exibility in selecting transmission powers,
theprobabilityofagivenoutcome(CC,capture,orcollision)witha¯xedpowerscheme,
and perhaps the degree of tolerance to environmental noise and interference for each
location.
We can also see from Figures 6.17 and 6.18(a) that evenif two transmissions are not
CCable, almost always one or the other can be delivered with the capture e®ect. The
SINR threshold of the MicaZ around 2 dB [79]. The low number of collisions in this
experiment shows that it is rare for RSSs from both senders to fall within this 2 dB
range. In our experiments, only 3% of power con¯gurations resulted in collisions. This
observation con¯rms our simulation results presented in Section 6.4.6. Older radios
sometimes have larger SINR thresholds (the 2 to 6 dBm of the Mica2 [79]) and so may
show larger collision regions.
To measure the e®ect of location on the signal and interference strength we plot the
measuredRSSandSINRvaluesat¯xedpowerlevel. Figure6.18(b)showsthemeasured
received signal strength (RSS) for each pair S11, S22, I12, I21 at ¯xed transmission
powerlevelof¡5dBm,andFigure6.18(c)showscalculatedSINRvalueateachreceiver
based on these measurements. First, we can see that measured RSS does not always
correspond to link distance as we expect. This variation is due to environmental factors
such as multi-path re°ections.
We show the topology condition (the di®erence between the right hand side and the
lefthandsideoftheinequalityinEquation6.5)asthesolidline. Apositivevaluemeans
the topology condition is satis¯ed and the communication may be CCable; as we can
139
see, this condition is only negative when S2 is 0.6 m, consistent with our ¯ndings in
Figure6.18(a). InFigure6.18(c)wecanobserveoneofthereceiverdoesnotsatisfy2dB
SINRthresholdatthefollowingthreeS2locations: ¡3m,¡1:2m,and0:6m. However,
concurrent communications become possible with power control for every experiment
other than when S2 is located at 0.6 m.
Figure 6.18(d) shows the optimal (i.e., minimum) transmission power selected for
each sender S1 and S2 for concurrent communication based on our experiment results,
together with optimal transmission power selected from our proposed formula (Equa-
tion6.4),butwithlimitedpowerrangebetween-25dBmand0dBmat1dBmintervals.
We can see that experiments with S2 locations at -3 m, -1.8 m, and -1.2 m, which have
much higher SINR at R1 (shown in Figure 6.18(c)), use higher optimal transmission
power for S2 to redistribute leftover SINR (i.e., signal strength) value to make CC
possible. We can also see that the simulated optimal power matches our experiments.
There is a di®erence in the plot, but this di®erence is coming from the limited power
levelsupportfromourtestednodes(MicaZ),andnodescanchoosethesamepowerlevel
as actual experiment based on simulation results.
6.5.3 Results from the Inside Scenario
InthesecondscenarioinFigure6.16(b),weplaceS2insidetheS1-R1pair. Experimental
results (presented in Figure 6.19) show that CC is possible even with this con¯guration
if nodes can control their transmission power. We observed that CC is possible when
R2 was at 2 m or 3 m in our experiments (presented in Figure 6.19(a)). However, we
do not see any concurrent communication when R2 was at 1 m. This experimental
result con¯rms what is predicted in simulation based on measured path loss; From
Figure 6.19(b) we can see that topology condition fails only for location 1 m. This ¯nal
140
1 2 3
10
20
30
40
50
60
R2 location (m)
Number of power combinations (out of 64)
CC
S1 Capture
S2 Capture
Collision
(a) CCability
1 2 3
−20
−15
−10
−5
0
5
10
15
20
R2 location (m)
SINR (dB)
SINR at R1
SINR at R2
topology condition
(b) SINR, TxPwr: 0 dBm
Figure 6.19: CCability from the inside scenario
case illustrates the case where CC is not possible because the location fails to satisfy
the topology condition.
Thesetwotestbedexperimentscon¯rmourkeysimulationresults: ¯rst,thatconcur-
rent communication is highly probable in many previously restricted cases with tradi-
tional802.11likemediumaccesscontrol. Second,completecollisionsandfullcorruption
of both packets is rare and often at least one sender can capture a packet. Finally, con-
trollable transmission power signi¯cantly improves CCability.
6.6 2D Simulations
Wepresentedoursimulationresultsfromlinetopologiesmainlytounderstandthee®ects
ofrelatedradioandenvironmentalparametersonconcurrentpacketcommunications. In
thissection,wehavesettheseparametervaluesconstantwhileonlyvaryingthelocations
ofsendersandreceiversontwodimensionalspace. Withmorecompletetwodimensional
simulations,wecanquantifyandvisuallyandnumericallycomparethebene¯tsinterms
of spatial reuse from concurrent packet communications under di®erent power control
schemes.
141
−50 −40 −30 −20 −10 0 10 20 30
−40
−30
−20
−10
0
10
20
30
40
X Coordinate of R2 (m)
Y Coordinate of R2 (m)
(a) CCable Region (S1)
−50 −40 −30 −20 −10 0 10 20 30
−40
−30
−20
−10
0
10
20
30
40
X Coordinate of R2 (m)
Y Coordinate of R2 (m)
(b) CCable Region (S2)
−30 −20 −10 0 10 20 30
−30
−20
−10
0
10
20
30
X Coordinate of S2 (m)
Y Coordinate of S2 (m)
PC CTxability
S1:(−6,0)
(c) CCability varying S2 location
−50 −40 −30 −20 −10 0 10 20 30
−40
−30
−20
−10
0
10
20
30
40
X Coordinate of R2 (m)
Y Coordinate of R2 (m)
(d) Max Power Collision Region
−50 −40 −30 −20 −10 0 10 20 30
−40
−30
−20
−10
0
10
20
30
40
X Coordinate of R2 (m)
Y Coordinate of R2 (m)
(e) CCable Region (Min S2)
−50 −40 −30 −20 −10 0 10 20 30
−40
−30
−20
−10
0
10
20
30
40
X Coordinate of R2 (m)
Y Coordinate of R2 (m)
(f) Min Power Collision Region
−50 −40 −30 −20 −10 0 10 20 30
−40
−30
−20
−10
0
10
20
30
40
X Coordinate of R2 (m)
Y Coordinate of R2 (m)
(g) CCable Region
−50 −40 −30 −20 −10 0 10 20 30
−40
−30
−20
−10
0
10
20
30
40
X Coordinate of R2 (m)
Y Coordinate of R2 (m)
(h) CCable Region MAX Power
Figure 6.20: 2D simulation results with optimal transmission power settings. S1=(-
6 m,0 m), R1=(0 m,0 m), S2=(-11 m,0 m) for all ¯gures except (c) where S2 location
varies within the presented coordinate space, R2 varies within the presented space for
all ¯gures
142
6.6.1 Methodology
Ourfocusoftwodimensionalsimulationisnotonthee®ectsofdetailparametersettings
anymore,andweusethefollowingstaticvaluesandonlyvarythelocationofnodesintwo
dimensional spaces: PL
0
= 35, n = 3:5, SINR
µ
= 2, X
¾
= 0, N
1
= N
2
=¡95 dBm.
We still use the static location S1 and R1 communication for each set of simulations
(the e®ect of changing this link distance has been presented in Section 6.4.4.1). Our
simulation use S1 location at x, y coordinates of -6 m, 0 m and R1 location at 0 m,
0 m on two dimensional space. We tested every S2 and R2 location combination within
the square region ranged between -50 m and 50 m range in both vertical and horizontal
space (i.e., x and y axis). Test location interval is 2 m for S2 and 1 m for R2.
We use three di®erent power control schemes: (1) OptiPC: our proposed optimal
powercontrolschemeforconcurrentcommunicationproposedinSection6.3(2)MinPC:
controltransmissionpowertotheminimum(supported)levelthatisgoodforthepacket
reception at its own intended receiver under no interference (3) MaxP: use the maxi-
mum transmission power supported by the device. We limit the supported power for
transmission between -25 dBm and 0 dBm with 1 dBm level power control capability
matching the power range of MicaZ motes.
OptiPC uses the minimum transmission power which allows concurrent communi-
cations considering the interference from simultaneous transmissions. MinPC use the
minimumrequiredtransmissionpowerforitsownintendedreceiverignoringthepossible
interference. MinPCminimizesitsinterferencetothenetwork,butitalsominimizesthe
communication reliability under co-channel interference. MaxP always use the maxi-
mum supported power for packet transmissions. We can consider MaxP as a good,
simpleapproachtotake, especiallyforsparsenetworkwhereconcurrentcommunication
is a rare event. MaxP provides reliable communication links while minimizing energy
e±ciency and spatial reuse.
143
6.6.2 CCability with Optimal Power Setting
In this section we present 2D simulation with optimal transmission power settings to
estimatethebene¯tsfromappropriatepowersettingsintermsofspatialreuse. The¯rst
two rows of Figure 6.20 present the 2D simulations results using our proposed optimal
transmission power setting scheme for two concurrent packet transmissions from S1 to
R1 and S2 to R2. These ¯gures showthe sample case of S2 at x, y coordinates of -11 m,
0 m. Color marked areas indicate the R2 locations which allow successful concurrent
communication and the color intensity of each point means the intensity of optimal
power required for S1 in Figure 6.20(a) and for S2 in Figure 6.20(b) (darker values
indicate greater transmission power, red is the darkest indicating the maximum power).
Figure 6.20(a) and Figure 6.20(b) mark CCable region with OptiPC scheme in S1
and S2's transmission power respectively. We can observe about 93% of CCable R2
locations within the collision region with optimal transmission power settings; We use
normalized collision region considering the maximum transmission power of the node
for the comparison with the case without transmission power control (Shown in Fig-
ure 6.20(d). Extra transmission power assigned to overcome interference (i.e., darker
color) can be geographically identi¯ed near the communication from S1 to R1 as well
as non-CCable regions.
We repeated simulations varying the S2 locations and calculated the CCability test-
ing every R2 locations like previous examples (shown in (a) and (b)). Figure 6.20(c)
plots the average CCability we observed for each S2 location. Each S2 point in this
¯gure presents the probability of successful concurrent communication when R2 loca-
tion is randomly selected within the collision region. On average, 89% R2 locations are
CCable for S2 location changes between -35 and 35 m of both x and y directions. This
resultprojectsustheexpected,highimprovementinspatialreusefromtwodimensional
space when we use appropriate transmission power control.
144
6.6.3 CCability at Di®erent Power Settings
In this section we present 2D simulations results with several di®erent power control
schemes to compare their performance for CCability. The last two rows of the Fig-
ure 6.20 show the simulation results when we use MinPC and MaxP power control
schemes; CCable regions colored based on S2's transmission power. MinPC adjusts
senders transmission power to have the SINR at the receiver equal to the SINR thresh-
old. Therefore, theoretically MinPC scheme cannot endure any interference because
that will drop the SINR below the SINR threshold. However, the discrete 1 dBm level
(or even coarser level) of power control and limited power control range used in our
assumption and found in general for low-power radio may still give some room for extra
interference. Out of the total collision region with MinPC (presented in Figure 6.20(f),
only very small fraction (about 1%) of R2 locations can support concurrent communi-
cation. Figure 6.20(g) shows every R2 location which supports successful concurrent
communications with MinPC scheme, including the non-Collision region, but the CCa-
ble region is still quite small compared to the OptiPC scheme.
Figure 6.20(h) presents the CCable R2 locations when we use MaxP scheme. In
this example, some extra transmission power is used to sustain communication under
interference, but still only 15% of collision region can be CCable while causing high
energyconsumptionfortransmissionandunnecessarilyhighinterferencetothenetwork.
Fromthiscomparison,wecanseethatmoresophisticatedtransmissionpowercontrol
can signi¯cantly improve CCability; Simple minimum power control scheme or maxi-
mum static power scheme does do very well in this regard. RTS/CTS based collision
avoidance often overestimates the e®ect from concurrent transmission and designing
a MAC protocol incorporating a more sophisticated power control could be useful to
improve spatial reuse of the network.
145
6.7 Making CCable Decisions in Practice
Section 6.3.3 described how we select optimal transmission powers to enable CCability
when possible in simulation. Making this decision in practice is considerably more
di±cultforseveralreasons: thedecisionmustbemadeatdistributednodeswithlittleor
nocommunication, nodelocationislikelyunavailableorinaccurate, andnoiselevelsare
constantly changing. Even if locations are known, distance is not an accurate estimator
of signal and interference strength as we observe in our testbed experiments and others
have observed in the past [3, 77, 84, 93].
However, if we can actually measure the path loss at a given location, we can avoid
these real-world complexities and use this measurement directly our proposed formulae
(Equation 6.4 and 6.5). This simpli¯cation is possible because these model parameters
are used only to estimate path loss. Actual path loss information can be collected with
a single RSS measurement at any transmission power level (path loss = Tx power ¡
RSS), suggesting that a CCable decision is feasible in real systems.
Ourinitialtestbedexperimentssuggestaclosematchbetweensimulationandtestbed
experiments (Figure 6.17). We design a practical MAC that supports CCability based
on our ¯ndings in this chapter. We presented our new MAC design in Chapter 7.
6.8 Summary
Inthischapter,wehavepresentedthe¯rste®orttoquantifytheopportunityforconcur-
rent communication in low-power wireless networks. We proposed a simple rule which
determines when successful concurrent communication is possible, and how to select an
optimal transmission power for concurrent communication given global knowledge or
146
with measured path loss. Through simulation we systematically explored the parame-
ter space, varying node position, mean and variance of path loss, signal-to-interference-
plus-noise-ratio (SINR) threshold, range and granularity of transmission power control.
We veri¯ed the key results of our simulations through testbed experiments with MicaZ
motes,demonstratingthatconcurrentcommunicationisoftenpossiblewithappropriate
power control and capture by at least one receiver is almost always possible.
147
Chapter 7
Towards Concurrent Communication in Wireless Networks
7.1 Overview
Avoidingcollisionsisoneofthekeyrolesofmedia-access(MAC)protocols. Researchin
MACAW[5]andstandardssuchas802.11employcarriersenseandexchangeofrequest-
to-send (RTS) and clear-to-send (CTS) packets to prevent concurrent communications
andhiddenterminalcasesthatmightcorruptcommunication. Nevertheless, concurrent
communication|allowingtransmissionbytwosenderswithinmaximumcommunication
rangeofeachother'sreceiveratthesametimeoverthesamechannel|canbebene¯cial,
provided both receivers can successfully receive what is sent. We present the e®ects
of concurrent packet transmission in Chapter 5 and study the feasibility of concurrent
communicationwithquanti¯cationsofthebene¯tsfromconcurrentcommunicationwith
appropriate power setting in Chapter 6. The bene¯ts of concurrent communication
come because carrier sense and RTS/CTS greatly reduce opportunities for spatial reuse
of the channel. In a multi-hop network, RTS/CTS-enforced-silence reduces end-to-end
throughput. And for networks designed for relatively small data payloads, such as
802.15.4, the RTS/CTS exchange is avoided as control overhead.
Recent work has begun exploiting the richness of real-world wireless propagation
and richer MAC protocols. Experiments have shown that MAC protocols can exploit
148
channel capture, either by retraining mid-reception [43, 83, 96], or using more aggres-
sive carrier sense [39]. Other work has shown that power control can allow transmission
\over the heads" of intermediate nodes [58]. Experiments have evaluated power control
to maximize spatial reuse [77, 50], and to develop better models of wireless propaga-
tion [56, 66, 79]. Experimental work has also suggested the importance of SINR-based
channel models that represent the intermittent, power- and location-sensitive reception
inherent in concurrent wireless communication [58, 79]. This range of work provides
components for interference-aware protocol design and has shown the feasibility of con-
current communication with modern radios that provide per-packet power control and
MAC-levelchannelcapture. Whileverypromising,thisworkhasyettosuggestaspeci¯c
new MAC protocol or quantify trade-o®s.
This chapter (the work presented in this chapter appears in [17]) seeks to answer
the following question: how close can a practical MAC approach optimal?
Our main contribution is to relate these bounds on concurrent communication to
what can be accomplished in a real-world MAC protocol. Our optimal bounds require
perfect knowledge of all channel state: all concurrent communication, node locations,
and noise; information impossible to maintain in a realistic network, and complex and
expensive to approximate. On the other hand, a very simple MAC might send at the
lowest possible power to maximize channel reuse. We evaluate the bene¯ts of designs
thatemploydi®erentamountsofinformationrelativetoouroptimalperformancebound
(Section 7.2).
While our work presented in this chapter does not advocate a complete new MAC,
we show that two pairs of transmitters can communicate concurrently more than 80%
of the time with su±cient source separation, given perfect channel knowledge. We also
show that a practical gain adaptive power control-based MAC protocol can provide
much of this bene¯t (73% on average). These results suggest that future MAC designs
149
should embrace concurrent communication through power control and channel capture
and shift away from carrier sense and RTS/CTS.
7.2 MAC Protocol Designs
We have established that for the vast majority of topologies where senders have reason-
able separation, concurrent communication is possible given complete information (in
Chapter6). Yethowclosepracticecancometothisboundisnotclear, sinceapractical
MAC protocol must make control decisions based only on prior knowledge and local
information.
Wenextconsider¯vedi®erentpowercontrolalgorithmsforMACprotocols. Carrier
sense with RTS/CTS at maximum and minimum power represents the current state-
of-the-art. We present an oracle algorithm, to provide an upper bound on performance
given unachievably perfect information. We then introduce two simple MAC protocols
thatuseonlylocalandpriorinformation. MinPC sendsatminimumpowerwithchannel
capture; a very simple way to improve spatial reuse given prior knowledge of node loca-
tions. Finally, gain-adaptive power control (GAPC) adds a transmit-power-dependent
boost to MinPC to overcome some potential interference.
We evaluate these protocols through simulation using the SINR-based model that
we validated in Section 6.5 with testbed experiments. We use an exponential path-loss
model with the option of log-normal multipath fading in our simulations to obtain the
pair-wise link gains. In each simulation we consider two sender-receiver pairs. We ¯x
thelocationofonepairandthesecondsender,movethesecondreceiveroverallpossible
locationswithpossiblereception, andmeasurewhichreceiverscancaptureconcurrently
sent packets.
Figure7.1showsCCabilityforeachprotocol,inthis¯gureblackindicatestheCCable
region,grayshowswhereonecommunicationortheotheriscapturable,andwhiteshows
150
inability to communicate, either due to power limitations (outside the circle) or due to
collisions (inside). The above two rows of this ¯gure show simulation results without
consideringlinkqualityvarianceatthesamedistance,andbottomtworowsshowresults
with variance in link gain due to model multi-path fading. As can be seen, while fading
e®ectsdointroduceadegreeofnoise,theydonotfundamentallychangetheresults. For
ease of exposition, therefore, we ignore fading in subsequent discussion as we consider
each design alternative.
7.2.1 Today's Practice: CS-RTS/CTS with Simple Power Control
We begin by evaluating traditional control methods. Figures 7.1(a) and 7.1(b) show
the behavior of a traditional carrier-sense with RTS/CTS MAC. Simplest is to always
transmit RTS/CTS at maximum power to block any potential receivers. As shown in
Figure 7.1(a), this case always allows one sender, but never concurrent communication.
Slightly better is to send at minimum possible power needed to reach the indented
receiver (assuming unicast communication). Taking this step requires that each node
maintain a list of neighbors and estimates of the transmit power needed to exceed their
SNR threshold. We assume this information is collected and reasonably stable, a valid
assumptionforslow-fadingenvironmentswithlittlemobility[77]. Inthiscase,thesmall
black crescent in Figure 7.1(b) shows that even with CS-RTS/CTS we can get some
concurrent communication when R2 is located near S2|approximately 2% of the area.
While better than maximum power, we suggest that this gain is too modest relative to
the measurement overhead to motivate use of power control with CS-RTS/CTS.
7.2.2 A Upper Bound on Performance with an Oracle
Given perfect knowledge of the gain (or path loss) between the nodes in the network,
any concurrent communication, and noise, one can compute the optimal (minimal)
151
−60 −50 −40 −30 −20 −10 0 10 20 30
−40
−30
−20
−10
0
10
20
30
40
X Coordinate of R2 (m)
Y Coordinate of R2 (m)
(a) CS-RTS/CTS(Max)
−60 −50 −40 −30 −20 −10 0 10 20
−40
−30
−20
−10
0
10
20
30
40
X Coordinate of R2 (m)
Y Coordinate of R2 (m)
(b) CS-RTS/CTS(MinPC)
−60 −50 −40 −30 −20 −10 0 10 20 30
−40
−30
−20
−10
0
10
20
30
40
X Coordinate of R2 (m)
Y Coordinate of R2 (m)
(c) Oracle
−60 −50 −40 −30 −20 −10 0 10 20 30
−40
−30
−20
−10
0
10
20
30
40
X Coordinate of R2 (m)
Y Coordinate of R2 (m)
(d) MinPC
−60 −50 −40 −30 −20 −10 0 10 20 30
−40
−30
−20
−10
0
10
20
30
40
X Coordinate of R2 (m)
Y Coordinate of R2 (m)
(e) GAPC
−60 −50 −40 −30 −20 −10 0 10 20 30
−40
−30
−20
−10
0
10
20
30
40
X Coordinate of R2 (m)
Y Coordinate of R2 (m)
(f) CS-RTS/CTS(Max)
−60 −50 −40 −30 −20 −10 0 10 20 30
−40
−30
−20
−10
0
10
20
30
40
X Coordinate of R2 (m)
Y Coordinate of R2 (m)
(g) CS-RTS/CTS(MinPC)
−60 −50 −40 −30 −20 −10 0 10 20 30
−40
−30
−20
−10
0
10
20
30
40
X Coordinate of R2 (m)
Y Coordinate of R2 (m)
(h) Oracle
−60 −50 −40 −30 −20 −10 0 10 20 30
−40
−30
−20
−10
0
10
20
30
40
X Coordinate of R2 (m)
Y Coordinate of R2 (m)
(i) MinPC
−60 −50 −40 −30 −20 −10 0 10 20 30
−40
−30
−20
−10
0
10
20
30
40
X Coordinate of R2 (m)
Y Coordinate of R2 (m)
(j) GAPC
Figure 7.1: MAC power control comparison of CCability. The bottom two rows show
the case with fading variance. S1 = (¡6;0), R1 = (0;0), S2 = (¡21;0), and vary R2
between -35 m and 35 m both in X and Y directions, n=3:5, SINR
µ
=2, X
¾
=0 for
the top two rows, X
¾
=3 for the bottom two rows. Black = CCable, Gray = Capture,
White = No communication.
152
transmit power for concurrent communication. We describe this approach in previous
chapter (in Section 6.3.1). While gain can be observed in the network and its variance
estimated, one cannot have perfect knowledge of all concurrent communications. We
next describe an oracle algorithm that uses this perfect knowledge establish an upper
bound on the bene¯t we can expect from concurrent communication. While this oracle
usesperfectknowledge, itisstillsubjecttohardwarelimitationsofdiscretepowerlevels
and minimum and maximum transmit power.
Figure 7.1(c) shows sample results with the oracle. Compared to CS-RTS/CTS at
di®erent power levels (Figures 7.1(a) and 7.1(b)), we can see that there is considerable
room for concurrent communication. There is a small hole near the center, occupying
about 3% of the possible area, where only one communication can be allowed. In this
regionreceiverR2istooclosetosenderS1orreceiverR1forbothtocommunicategiven
a maximum transmit power. We call this region the region of impossible concurrency.
Of course, this scenario represents just one topology. However, we observe similar
results provided the two sources are separated by some minimum distance (outside the
region of impossible concurrency). We examine di®erent source and receiver placement
in more detail below (Figures 7.2 and 7.3).
For this con¯guration, the oracle allows concurrent communication over 97% of the
R2locations. Thisevaluationdemonstratesthepotentialofchannelcaptureandpower-
control, if we de¯ne a MAC algorithm that uses practical information.
7.2.3 Exploiting Power Control and Channel Capture
Ignoring interference, we maximize spatial reuse by always sending at the minimum
power that will reach the intended receiver. Our capture(MinPC) algorithm takes this
approach to power control, and employs MAC-level channel capture [83]. It is therefore
equivalenttoCS-RTS/CTS(MinPC),butreplacingCS-RTS/CTSwithchannelcapture.
153
As with CS-RTS/CTS(MinPC), we assume nodes maintain a list of neighbors and can
track the gain needed to reach them.
In theory, sending at the minimum transmit power cannot tolerate any level of
interference. However, in practice, real hardware can be set only at discrete power
levels, providing some level of protection to noise. (We use discrete levels at 1 dBm
increments in this simulation; the CC2420 radio provides slightly coarser levels.)
Figure 7.1(d) shows a moderate size region where concurrent communication is pos-
sible, 20% of the total area in this case. Comparing this to CS-RTS/CTS (MinPC)
demonstrates the advantage of channel capture over communications prohibition. In
addition, the large grey region shows that, even when concurrent communication is not
possible, at least one receiver or the other will get their data through. In this case, CC
or capture is 87% of the total area.
Thepenaltyofallowingconcurrentcommunicationisthesmallwhitecrescentregion
wheretransmitpowersareevenlymatchedatthereceivers,resultingincollisionswithout
capture. With RTS/CTS, one sender or the other would win the contention and send,
but with capture we depend on random backo® and retry when nodes are at this range.
7.2.4 GAPC: Gain-Adaptive Power Control and Capture
Whilediscretepowerlevelsprovidesomebu®eragainstnoisewiththecapture(MinPC)
algorithm, concurrent communication provides strong sources of interference that limit
the ability of min-power to approach oracle. This problem is particularly noticeable at
theedgesofS2'srange,whereinterferenceforS1preventsS2reception. InFigure7.1(d)
this case appears as the large grey doughnut surrounding the black CCable region.
This conditions can be overcome by systematically adding a boost of power in in-
verse proportion to the gain needed to reach the intended receiver. We call this algo-
rithm Gain-Adaptive Power Control. Since gain is roughly proportional to distance,
154
this means that short-distance transmissions get large boosts while longer transmission
gets relatively less gain. Our intuition for this scheme comes from observations in a
detailed simulation study (presented in Chapter 6) [15]: we found that short distance
communication is often overwhelmed by interference from longer transmissions.
If we de¯ne P
max
and P
S;R
as the maximum possible transmit power and the power
needed for source S to reach receiver R, then we can de¯ne the power boost ² as:
²=(P
max
¡P
S;R
)²
ratio
(7.1)
In this equation, ²
ratio
represents the fraction of remaining power to allocate to a
transmission. Large values of ²
ratio
will quickly assign all headroom to transmissions
and will increase the bonus given to shorter links. We varied ²
ratio
and found that
moderate values (0.3 to 0.7) provided the best levels of CCability (values that near 0
provide no boost, while values approaching 1 always operate at maximum power). We
adopt ²
ratio
=0:5 as a reasonable, robust choice.
Figure 7.1(e) evaluates gain-adaptive power control with ²
ratio
= 0:5. We see that
this approach comes very close to optimal: concurrent communication is possible with
the receiver in 76% of the area compared to the oracle algorithm, much closer than
capture(MinPC).
The cost of gain-adaptive control relative to the oracle can be seen in two locations.
The moderate-size grey area when R2 is placed near (0;0) is larger than optimal. This
areacorrespondstocaseswhereR1andR2arecompetingandthepowerboostprevents
concurrentcommunication. Inthisregionitisbestifonlyonesendertransmits. Second,
communicationin the narrowgrayringaround the edgeof the oracle cannot be reached
with gain-adaptive control because of slightly higher interference from the S1-R1 pair.
155
−30 −20 −10 0 10 20 30
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Coordinate X of S2 location (m)
CCable Area fraction
CS−RTS/CTS(Max)
CS−RTS/CTX(MinPC)
Oracle
MinPC
GAPC (0.5)
Figure 7.2: Comparison of CCable area with limited power levels (-25 dBm to 0 dBm)
for ¯ve MACs.
These results suggest that gain-adaptive power control is a practical scheme that
gets a signi¯cant fraction of optimal performance.
7.2.5 Comparing MAC Protocols
Figure7.1comparesthe¯veMACalgorithmsforaparticulartopology. Wecanquantify
the bene¯t of concurrent communication by observing the ratio of area of concurrent
communication(anythingblack)tothetotalreachableareawhenthereisnointerference
(indicated by thin black circles, also equal to the the gray area with CS-RTS/CTS at
maximum power in Figure 7.1(a)). We de¯ne this ratio as the CCable area.
Figure 7.2 compares this CCable area for each of the MAC schemes we consider.
This graph provides a single slice through the 2-D simulation with nodes placed at
S1 = (¡6;0), R1 = (0;0), S2 at coordinate (x, 0), with the x-coordinate indicated
on the horizontal axis of the graph, evaluated for all R2 locations over all potentially
receivable locations. Each point on the ¯gure represents the fraction of R2 locations
that allow concurrent communication for a given S2 x position.
156
−30 −20 −10 0 10 20 30
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Coordinate X of S2 location (m)
CC plus Capture area fraction
CS−RTS/CTS(Max)
CS−RTS/CTS(MinPC)
Oracle
MinPC
GAPC(0.5)
Figure 7.3: CC plus capture rate comparison with limited power levels (-25 dBm {
0 dBm)
Even with power control, Figure 7.2 shows that CS-RTS/CTS provides little spatial
reuseofthechannelthroughconcurrentcommunication;thisresultisconsistentwithits
design goal of preventing any possible interference. Shifting to a channel-capture-based
MAC (min-power) provides considerably greater opportunity for concurrent communi-
cation. Finally, GAPC comes relatively close to the best possible oracle result. We ¯nd
that its CCable area is within 73% of the optimal, averaged over all S2 locations, and
is as high as 95% in regions with su±cient source separation.
We can also observe where concurrent communication is possible. All algorithms,
including the oracle, fail when both senders (or receivers) are nearly in the same place.
In this region of impossible concurrency, no level of power is su±cient to capture the
channel. The algorithms di®er mainly in the width of this region|more sophisticated
algorithms are closer to the oracle's best-possible result.
We can use this same methodology to evaluate not just opportunities for concurrent
communication, but for CC or channel capture. Figure 7.3 shows that the oracle per-
formancealmoststrictlydominatesCS-RTS/CTSbythismetric|CS-RTS/CTSalways
getsexactlyonepacketthrough,whiletheoraclealwaysgetsoneortwo packetsthrough.
157
Figure 7.2 shows where two are possible, while in region of impossible concurrency it
gets one packet through. By comparison, the realizable algorithms capture(MinPC)
and GAPC can get one or two packets through 77{88% of the time. Even when con-
currency is impossible, most often one sender captures the channel. The 12{23% gap
represents lost capacity in conditions where packet collision allows neither sender to
communicate. Most of this loss occurs at near the edge of maximum communication
range where GAPC cannot boost power adequately to exceed interference because of
hardware limitations.
7.3 Summary
It is now widely understood that wireless propagation is much more than receive/no-
receive links. Prior work (in Chapter 6) has demonstrated that channel capture are
possible with an appropriate power control [15].
The work described in this chapter has established that the bene¯t of exploiting
these characteristics is signi¯cant. We provide a theoretical upper bound on perfor-
mance given realistic hardware and perfect knowledge. We then showed that a practi-
cally implementable power control algorithm, the GAPC scheme, can get near-optimal
performance, averaging 73% of area of concurrent communication obtained the oracle,
with successful capture in 77{88% of the cases.
While promising, our work is still a preliminary step. We focused on two pairs of
concurrent communications; we believe the results generalize to n-node communication
(through preliminary simulations not shown here), but through evaluation is future
work. More importantly, full implementations of the MAC algorithms that we propose
are necessary to provide full experimental validation of our conclusions.
158
We believe this work establishes an essential direction for future MAC research,
away from the use of carrier sense and RTS/CTS to avoid concurrent communication,
instead embracing concurrency through power control and channel capture.
159
Chapter 8
Future Work and Conclusions
We close by listing some open research questions and problems that can be addressed
in our future work, and by making some concluding statements.
8.1 Future Directions
First, an important challenge is to develop a good wireless channel (or link quality)
metric under dynamic environments. We ¯rst started our research by understanding
thebehavioroflow-powerwirelesslinksinstaticenvironments(inChapter4). Wetested
some of the link quality metrics and studied the in°uence of power control mostly in
this static condition. However, it is often the case to ¯nd an environment with dynamic
communication channels, and there is no systematic study of identifying appropriate
metric to use in this condition. We think our measurement study that was mostly
limitedtoastaticenvironmentcanbeextendedbymeasuringchannelbehaviorundera
dynamicenvironmentwitharealistictra±cmodel. Identifyingagoodprobabilisticlink
quality metric for dynamic environments and studying optimal settings of transmission
powercanleadtosigni¯cantchangeandimprovementincommunicationprotocoldesign.
A second future direction is modeling co-channel interference from more than two
concurrent transmissions. In our experimental study (presented in 5.4.2), we observe
160
thatadditivee®ectofmultipleinterferenceconsideredintheoryisnotalwaysvalidinreal
systems. However, it was di±cult to identify real causes of this phenomenon; though
we ¯nd that some of this behavior is attributed to non-ideal hardware used in low-
powerwirelessnetworks. Thatisthemaindi±cultyofproposinginterferencesimulation
models for more than two concurrent packet transmission. To further enhance the
accuracy and e±ciency of interference-aware protocol design, it is essential to have
more precise modeling of multiple interference on packet communication with some
state-of-the-art low-power wireless radios.
Third, it would be useful to extend our proposed topology condition and rules to
optimal transmission power setting for concurrent packet transmission for the case with
more than two interferers in the same channel. Once we obtain more accurate model of
multiple interference, one can use it to extend our process of making concurrent packet
communication decision to entire network.
Finally, it is important to develop new communication protocols which allow con-
current packet communication within the same channel. We have proposed a sketch of
MAC protocol GAPC, that take advantages of capture e®ect and transmission power
control, which improves the spatial reuse of the network only with local information.
It would be good to test some variations of GAPC taking into account some extra in-
formation other than channel gain; probably with some extra decision rules of power
setting for the second attempt of packet transmission at failure. Medium access con-
trol protocols and multi-hop packet communication (such as scheduling and routing)
protocols could be possible immediate candidates.
As a large research direction, one can think of extending this new design paradigm
towards concurrent communications for systems with advanced hardware that supports
multiple channels, smart antenna, advanced modulation techniques such as orthogonal
161
frequency division multiplex (OFDM), or for di®erent networks such as 802.11 (wireless
LAN), 802.16 (WiMAX).
8.2 Conclusions
Early empirical studies in low-power wireless networks revealed that the reality of wire-
lesscommunicationisquitedi®erentfromcommonapproximationsacceptedforaproto-
coldesignandevaluation(whicharenormallydonewithsimulationsduetothedi±culty
of real deployment).
In this dissertation, we presented experimental studies that lead to a better un-
derstand low-power wireless network, especially the e®ects of transmission power con-
trol and co-channel interference. Based on our empirical studies, we proposed some
interference-aware protocols that improve the performance of wireless communication
in terms of reliability and channel capacity of wireless networks.
First, we performed a systematic empirical study of low-power wireless links, espe-
cially with transmission power control to test its value as a link quality controller. Our
studyidenti¯es the causes of high variance in link qualityunder di®erent environmental
conditions and hardware settings, and identi¯ed the transmission power range where
the link quality dynamically changes (i.e., unreliable transmission power range).
Based on this empirical understanding of wireless links and e®ects of power control
on wireless channel, link quality control scheme with packet-based power control with
link blacklisting (PCBL) has been introduced. PCBL converts unreliable asymmetric
and weak links to reliable wireless links which provide a consistent link quality. We
incorporate a blacklisting approach together with our power control scheme to address
the remaining unreliable link problem at adjusted transmission power level. We imple-
mented and tested performance of PCBL in real testbed with actual routing protocol
162
toverifyitsperformanceandbene¯ts. PCBLperformsenergy-e±cientlinkqualitycon-
trol which provides more consistent and reliable network topology, and it also improves
spatial reuse of the network while minimizing channel interference.
Our ¯rst empirical study and protocol design is still limited by the prohibition of
concurrent packet transmission within the same channel with the simpli¯ed collision
avoidance scheme often implemented in medium access control layer. So next, we per-
formed and presented experimental analysis of the e®ects of concurrent packet trans-
missions in low-power wireless networks. We have con¯rmed the capture e®ect and the
existence of the SINR threshold which ensures the successful delivery of the strongest
packet under the concurrent packet communication situations with single and multiple
interferers. Our systematic experimental study veri¯es di®erences between the con-
ventional approximation of the interference e®ect and the actual impact of concurrent
transmission on packet delivery. Our experimental study provides new guidelines for
more realistic simulation models and capture-aware protocol design.
After con¯rming capture e®ects and seeming low probability actual packet collision
upon concurrent packet transmission, we have presented the ¯rst e®ort to quantify the
opportunity for concurrent communications (CC) in low-power wireless networks. We
proposed a simple rule to determine when communication is CCable and to select op-
timal transmission power given global knowledge or with measured path loss. Through
simulationwesystematicallyexploredtheparameterspace,varyingnodeposition,mean
andvarianceofpathloss,signal-to-interference-plus-noise-ratio(SINR)threshold,range
and granularity of transmission power control. We veri¯ed the key results of our simu-
lations through testbed experiments with MicaZ motes, demonstrating that concurrent
communication is often possible and capture by at least one receiver is almost always
possible.
163
Based on our mathematical modeling of CCable conditions and empirical experi-
ence, we proposed a simple sketch of MAC protocol, called GAPC, which controls its
transmission power based on the expected gain at intended receiver. GAPC uses only
localized information, but performs close to the optimal power setting that requires
global knowledge of the network.
From extensive experimental, simulation-based, analytical evaluation, we believe
that interference-aware communication protocol design can signi¯cantly improve the
performance of wireless communication protocol, mainly by embracing concurrency in-
stead of avoiding it, through capture e®ect and optimally tuned transmission power
settings.
164
References
[1] A. Acharya, A. Misra, and S. Bansal. MACA-P: A MAC for concurrent trans-
missions in multi-hop wireless networks. In IEEE PerCom, pages 505{508, Mar
2003.
[2] S.Agarwal,R.H.Katz,S.V.Krishnamurthy,andS.K.Dao. Distributedpowercon-
trol in ad-hoc wireless networks. In IEEE International Symposium on Personal,
Indoor and Mobile Radio Communications, pages 59{66, Oct 2001.
[3] D. Aguayo, J. Bicket, S. Biswas, G. Judd, and R. Morris. Link-level measurements
from an 802.11b mesh network. In ACM SIGCOMM, pages 121{132, Aug 2004.
[4] N.BambosandS.Kandukuri. Powercontrolledmultipleaccess(PCMA)inwireless
communication networks. In IEEE Infocom, pages 219{228, Mar 2000.
[5] Vaduvur Bharghavan, Alan Demers, Scott Shenker, and Lixia Zhang. MACAW: A
mediaaccessprotocolforwirelesslan's. In ACM sigcomm, pages212{225, London,
UK, Sep 1994.
[6] A. Cerpa, N. Busek, and D. Estrin. SCALE: a tool for simple connectivity assess-
ment in lossy environments. Technical Report CENS TR 03-0021, UCLA, 2003.
[7] A. Cerpa, J. L. Wong, L. Kuang, M. Potkonjak, and D. Estrin. Statistical model
of lossy links in wireless sensor networks. In IPSN, Apr 2005.
[8] A. Cerpa, J. L. Wong, M. Potkonjak, and D. Estrin. Temporal properties of low
powerwirelesslinks: Modelingandimplicationsonmulti-hoprouting. InMobiHoc,
pages 414{425, May 2005.
[9] M.Cesana,D.Maniezzo,P.Bergamo,andM.Gerla. Interferenceaware(IA)MAC:
anenhancementtoIEEE802.11bDCF. In IEEE Vehicular Technology Conference,
Oct 2003.
[10] Chipcon. CC1000singlechipverylowpowerrftransceiver. http://www.chipcon.
com/files/CC1000_Data_Sheet_2_3.pdf.
[11] Chipcon. CC2420 2.4 GHz ieee 802.15.4/ zigbee-ready RF transceiver. http:
//www.chipcon.com/files/CC2420_Data_Sheet_1_4.pdf.
165
[12] D. Couto, D. Aguayo, J. Bicket, and R. Morris. A high-throughput path metric
for multi-hop wireless routing. In IEEE Mobicom, pages 134{146, Sep 2003.
[13] D. Couto, D. Aguayo, B.A. Chambers, and R. Morris. Performance of multihop
wireless networks: Shortest path is not enough. In First Workshop on Hot Topics
in Networking (HotNets-1), Oct 2002.
[14] Crossbow. MPR/MIB user's manual 7430-0021-06. http://www.xbow.com/
Support/manuals.htm, Aug 2004.
[15] B. Krishnamachari D. Son and J. Heidemann. Evaluating the importance of con-
current packet communication in wireless networks. Technical Report ISI-TR-639,
USC/ISI, Apr 2007.
[16] B. Krishnamachari D. Son and J. Heidemann. Experimental study of transmission
powercontrolandblacklistingbasedlinkqualitycontrolinwirelesssensornetworks.
Technical Report ISI-TR-629, USC/ISI, Jan 2007.
[17] J. Heidemann D. Son and B. Krishnamachari. Towards concurrent communication
in wireless networks. Technical Report ISI-TR-646, USC/ISI, Jul 2007.
[18] D. Dardari, V. Tralli, and R. Verdone. On the capacity of slotted aloha with
Rayleigh fading: The role played by the number of interferers. IEEE Communica-
tion letters, 4, 5:155{157, May 2000.
[19] R. Draves, J. Padhye, and B. Zill. Comparison of routing metrics for static multi-
hop wireless networks. In ACM Sigcomm, pages 133{144, Aug 2004.
[20] T. ElBatt and A. Ephremides. Joint scheduling and power control for wireless
ad-hoc networks. In IEEE Infocom, pages 976{984, Jun 2002.
[21] T. ElBatt, S.V. Krishnamurthy, D. Connors, and S. Dao. Power management
for throughput enhancement in wireless ad-hoc networks. In IEEE International
Conference on Communications (ICC), pages 1506{1513, Jun 2000.
[22] J. Elson, L. Girod, and D. Estrin. Fine-grained network time synchronization
using reference broadcasts. In Fifth Symposium on Operating Systems Design and
Implementation (OSDI), pages 147{163, Dec 2002.
[23] ISI Laboratory for Embedded Networked Sensor Experimentation (ILENSE).
PC104 based nodes. http://www.isi.edu/ilense/testbed/pc104/.
[24] G. Foschini and Z. Miljanic. A simple distributed autonomous power control al-
gorithm and its convergence. IEEE Transactions on Vehicular Technology, 42,
4:641{646, Nov 1993.
166
[25] D. Ganesan, D. Estrin, A. Woo, D. Culler, B. Krishnamachari, and S. Wicker.
Complex behavior at scale: An experimental study of low-power wireless sensor
networks. Technical Report CS TR 02-0013, UCLA, 2002.
[26] K. Gilhousen, I. Jacobs, R. Padovani, A. Viterbi, L. Weaver, and C. Wheatley.
On the capacity of a cellular CDMA system. IEEE Transactions on Vehicular
Technology, 40,2:303{312, May 1991.
[27] L. Girod, J. Elson, A. Cerpa, T. Stathopoulos, N. Ramanathan, and D. Estrin.
EmStar: a software environment for developing and deploying wireless sensor net-
works. Technical Report CENS TR34, UCLA, Dec 2003.
[28] J.GomezandA.T.Campbell. Acaseforvariable-rangetransmissionpowercontrol
in wireless multihop networks. In IEEE Infocom, pages 1425{1436, Mar 2004.
[29] J. Gomez, A.T. Campbell, M. Naghshineh, and C. Bisdikian. Conserving trans-
mission power in wireless ad hoc networks. In IEEE International Conference on
Network Protocols (ICNP), pages 24{34, Nov 2001.
[30] D.J. Goodman and A.A.M. Saleh. The near/far e®ect in local ALOHA radio com-
munications. IEEE Transactions on vehicular technology, 36, 1:19{27, Feb 1987.
[31] S. Grandhi, R. Vijayan, D. Goodman, and J. Zander. Centralized power control in
cellularradiosystems. IEEE Transactions on Vehicular Technology, 42, 4:466{468,
Nov 1993.
[32] P. Gupta and P.R. Kumar. The capacity of wireless networks. IEEE Transactions
on information theory, 46, 2:388{404, Mar 2000.
[33] I.M.I. Habbab, M. Kavehrad, and C-E. W. Sundberg. ALOHA with capture over
slow and fast fading radio channels with coding and diversity. IEEE Journal on
selected areas in communications, 40, 3:79{88, Jan 1989.
[34] Z. Hadzi-Velkov and B. Spasenovski. Capture e®ect with diversity in ieee 802.11b
dcf. InIEEEInternationalSymposiumonComputersandCommunication(ISCC),
pages 699{704, Jun 2003.
[35] J. Heidemann, N. Bulusu, J. Elson, C. Intanagonwiwat, K.C. Lan, W. Ye Y. Xu,
D. Estrin, and R. Govindan. E®ects of detail in wireless network simulation. In
SCS Multiconference on Distributed Simulation, pages 3{11, Jan 2001.
[36] J. Heidemann, F. Silva, and D. Estrin. Matching data dissemination algorithms to
application requirements. In ACM Sensys, pages 218{229, Nov 2003.
[37] I-LENSE. Stargate based nodes. http://www.isi.edu/ilense/testbed/
stargate/.
167
[38] C. Intanagonwiwat, R. Govindan, and D. Estrin. Directed di®usion: A scalable
and robust communication paradigm for sensor networks. In Mobicom, pages 56{
67, Aug 2000.
[39] K. Jamieson, B. Hull, A. Miu, and H. Balakrishnan. Understanding the real-world
performance of carrier sense. In SIGCOMM'05 Workshop, pages 52{57, Aug 2005.
[40] E-S. Jung and N.H. Vaidya. A power control MAC protocol for ad hoc networks.
In ACM Mobicom, pages 36{47, Sep 2002.
[41] V. Kawadia and P.R. Kumar. Power control and clustering in ad hoc networks. In
IEEE Infocom, pages 459{469, Apr 2003.
[42] J.H.KimandJ.K.Lee. Capturee®ectsofwirelessCSMA/CAprotocolsinrayleigh
and shadow fading channels. IEEE Transactions on vehicular technology, 48,
4:1277{1286, Jul 1999.
[43] A. Kochut, A. Vasan, A.U. Shankar, and A. Agrawala. Sni±ng out the correct
physical layer capture model in 802.11b. In IEEE International Conference on
Network Protocols (ICNP), pages 252{261, Oct 2004.
[44] D. Kotz, C. Newport, and C. Elliott. The mistaken axioms of wireless-network
research. Technical Report TR2003-467, Dartmouth, Jul 2003.
[45] D. Kotz, C. Newport, R.S. Gray, J. Liu, Y. Yuan, and C. Elliott. Experimental
evaluation of wireless simulation assumptions. In MSWiM, pages 78{82, Oct 2004.
[46] M. Kubisch, H. Karl, A. Wolisz, L.C. Zhong, and J. Rabaey. Distributed algo-
rithms for transmission power control in wireless sensor networks. In Wireless
Communications and Networking (WCNC), pages 558{563, Dec 2003.
[47] D. Lal, A. Manjeshwar, F. Herrmann, E. Uysal-Biyikoglu, and A. Keshavarzian.
Measurement and characterization of link quality metrics in energy constrained
wireless sensor networks. In IEEE Globecom, pages 446{452, Dec 2003.
[48] K. Leentvaar and J. Flint. The capture e®ect in fm receivers. IEEE Transactions
on Communications, 24, 5:531{539, 1976.
[49] L. Li and P. Sinha. Throughput and energy e±ciency in topology-controlled
multi-hop wireless sensor networks. In Wireless Sensor Networks and Applications
(WSNA), pages 132{140, Sep 2003.
[50] S. Lin, J. Zhang, G. Zhou, L. Gu, T. He, and J.A. Stankovic. ATPC: Adaptive
transmissionpowercontrolforwirelesssensor. InSensys,pages223{236,Nov2006.
[51] D. Maniezzo, P. Bergamo, M. Cesana, and M. Gerla. How to outperform
IEEE802.11: Interference aware (IA) MAC. In MED-HOC NET, Jun 2003.
168
[52] K. Manousakis and J.S. Baras. Clustering for transmission range control and con-
nectivity assurance for self con¯gured ad hoc networks. In Military Communica-
tions Conference (MILCOM), pages 1042{1047, Oct 2003.
[53] R. Min and A. Chandrakasan. Top ¯ve myths about the energy consumption of
wireless communication. In ACM Sigmobile Mobile Computing and Communica-
tions Review, pages 65{67, Jan 2003.
[54] J.P. Monks, V. Bharghavan, and W.-M.W. Hwu. A power controlled multiple
accessprotocolforwirelesspacketnetworks. In IEEE Infocom, pages219{228, Apr
2001.
[55] J.P. Monks, J.P. Ebert, A. Wolisz, and W.W. Hwu. A study of the energy saving
and capacity improvement potential of power control in multi-hop wireless net-
works. In Local Computer Networks (LCN), pages 550{559, Nov 2001.
[56] T. Moscibroda. The worst-case capacity of wireless sensor networks. In IPSN,
pages 1{10, Apr 2007.
[57] T. Moscibroda, Y.A. Oswald, and R. Wattenhofer. How optimal are wireless
scheduling protocols. In IEEE Infocom, pages 1433{1441, May 2007.
[58] T. Moscibroda, R. Wattenhofer, and Y. Weber. Protocol design beyond graph-
based models. In HotNets-V, pages 25{30, Nov 2006.
[59] T. Nagatsu, T. Tsuruhara, and M. Sakamoto. Transmitter power control for cel-
lular land mobile radio. In Globecom, pages 1430{1434, Nov 1983.
[60] C.Namislo. AnalysisofmobileradioslottedALOHAnetworks. IEEETransactions
on vehicular technology, 33, 3:199{204, Aug 1984.
[61] R.W. Nettleton and H. Alavi. Power control for spread-spectrum cellular mobile
radio system. In IEEE VTC, pages 242{246, May 1983.
[62] S.J. Park and R. Sivakumar. Load-sensitive transmission power control in wireless
ad-hoc networks. In Global Telecommunications Conference, pages 42{46, Nov
2002.
[63] Venkatesh Rajendran, Katia Obraczka, and J.J. Garcia-Luna-Aceves. Energy-
e±cient, collision-free medium access control for wireless sensor networks. In ACM
Sensys, pages 181{192, Nov 2003.
[64] R. Ramanathan and R. Rosales-Hain. Topology control of multihop wireless net-
works using transmit power adjustment. In IEEE Infocom, pages 404{413, Mar
2000.
[65] T. Rappaport. Wireless Communications: Principles and Practice. Prentice Hall,
1996.
169
[66] C. Reis, R. Mahajan, M. Rodrig, D. Wetherall, and J. Zahorjan. Measurement-
based models of delivery and interference in static wireless networks. In ACM
Sigcomm, pages 51{62, Sep 2006.
[67] L.G. Roberts. ALOHA packet system with and without slots and capture. Com-
puter Communication Review, 24, 4:28{42, 1975.
[68] R.C.RobertsonandT.T.Ha. Amodelforlocal/mobileradiocommunicationswith
correctpacketcapture. IEEE Transactions on communications,40,4:847{854,Apr
1992.
[69] V. Rodoplu and T.H. Meng. Minimum energy mobile wireless networks. Selected
Areas in Communications, 17, 8:1333{1344, Aug 1999.
[70] R. Ronseca, O. Gnaawali, K. Jamieson, and P. Levis. Four-bit wireless link esti-
mation. In Hotnets, Nov 2007.
[71] J.M. Rulnick and N. Bambos. Mobile power management for wireless communica-
tion networks. Wireless Networks, 3, 1:3{14, Mar 2000.
[72] M. Sanchez, P. Manzoni, and Z.J. Haas. Determination of critical transmission
range in ad-hoc networks. In MMT, pages 293{304, Oct 1999.
[73] K. Seada, M. Zuniga, A. Helmy, and B. Krishnamachari. Energy-e±cient forward-
ing strategies for geographic routing in lossy wireless sensor networks. In ACM
Sensys, pages 108{121, Nov 2004.
[74] A.U.H. Sheiki, Y.D. Yao, and X. Wu. The ALOHA systems in shadowed mobile
radiochannelswithsloworfastfading. IEEETransactionsonvehiculartechnology,
39, 4:289{298, Nov 1990.
[75] A. Sheth and R. Han. SHUSH: Reactive transmit power control for wireless MAC
protocols. In IEEE Wireless Internet Conference, pages 18{25, Jul 2005.
[76] K-P. Shih and Y-D. Chen. CAPC: A collision avoidance power control MAC pro-
tocol for wireless ad hoc networks. IEEE Communications Letters, 9, 9:859{861,
Sep 2005.
[77] D. Son, B. Krishnamachari, and J. Heidemann. Experimental study of the e®ects
oftransmissionpowercontrolandblacklistinginwirelesssensornetworks. In IEEE
SECON, pages 289{298, Oct 2004.
[78] D. Son, B. Krishnamachari, and J. Heidemann. Experimental analysis of con-
current packet transmissions in low-power wireless networks. Technical Report
ISI-TR-609, USC/ISI, 2005.
170
[79] D. Son, B. Krishnamachari, and J. Heidemann. Experimental study of concurrent
transmission in wireless sensor networks. In ACM Sensys, pages 237{250, Nov
2006.
[80] M.SoroushnejadandE.Geraniotis. Probabilityofcaptureandrejectionofprimary
multiple access interference in spread spectrum networks. IEEE Transactions on
Communications, 39, 6:986{994, Jun 1991.
[81] K. Srinivasan and P. Levis. RSSI is under appreciated. In Emnets, May 2006.
[82] R. Wattenhofer, L. Li, P. Bahl, and Y.M. Wang. Distributed topology control for
power e±cient operation in multihip wireless ad hoc networks. In IEEE Infocom,
pages 1388{1397, Apr 2001.
[83] K. Whitehouse, A. Woo, F. Jiang, J. Polastre, and D. Culler. Exploiting the
capturee®ectforcollisiondetectionandrecovery. InIEEE Workshop on Embedded
Networked Sensors (EmNetS-II), pages 45{52, May 2005.
[84] A. Woo, T. Tong, and D. Culler. Taming the underlying challenges of reliable
multihop routing in sensor networks. In ACM Sensys, pages 14{27, Nov 2003.
[85] A. Woo, K. Whitehouse, F. Jiang, J. Polastre, and D. Culler. The shadowing
phenomenon: implicationsofreceivingduringacollision. InUCB Technical Report
UCB//CSD-04-1313, 2004.
[86] S.L. Wu, Y.C. Tseng, and J.P. Sheu. Intelligent medium access for mobile and ad
hoc networks with busy tones and power control. IEEE Journal on Selected Areas
in Communications, 18,9:1647{1657, Sep 2000.
[87] Y.D Yao and A.U.H. Sheikh. The cpature e®ect in frequency-hop spread-spectrum
multiple-access communications over fading channels with near/far problem. In
IEEE International Conference on Communications (ICC), pages 194{198, Jun
1988.
[88] R. Yates. A framework for uplink power control in cellular radio systems. IEEE
Journal on Selected areas in Communications, 13, 7:1341{1347, Sep 1995.
[89] W.Ye,J.Heidemann,andD.Estrin. Anenergy-e±cientMACprotocolforwireless
sensor networks. In IEEE Infocom, June 2002.
[90] J. Zander. Distributed cochannel interference control in cellular radio systems.
IEEE Transactions on Vehicular Technology, 41, 3:305{311, Aug 1992.
[91] J. Zander. Performance of optimum transmitter power control in cellular radio
systems. IEEE Transactions on Vehicular Technology, 41, 1:57{62, Feb 1992.
171
[92] K.ZhangandK.Pahlavan. Relationbetweentransmissionandthroughputofslot-
ted ALOHA local packet radio networks. IEEE Transactions on communications,
40, 3:577{583, Mar 1992.
[93] J. Zhao and R. Govindan. Understanding packet delivery performance in dense
wireless sensor networks. In ACM Sensys, pages 1{13, Nov 2003.
[94] G.Zhou,T.He,S.Krishnamurthy,andJ.A.Stankovic. Impactofradioirregularity
on wireless sensor networks. In ACM MobiSys, pages 125{138, Jun 2004.
[95] G. Zhou, T. He, S. Krishnamurthy, and J.A. Stankovic. Models and solutions
for radio irregularity in wireless sensor networks. ACM Transactions on Sensor
Networks, 46, 2:221{262, Mar 2006.
[96] G. Zhou, T. He, J.A. Stankovic, and T. Abdelzaher. RID: Radio interference
detection in wireless sensor networks. In IEEE Infocom, pages 891{901, Mar 2005.
[97] H. Zhou and R.H. Deng. Capture model for mobile radio slotted ALOHA systems.
IEEE Transactions on communications, 145, 2:91{97, Apr 1998.
[98] M. Zuniga and B. Krishnamachari. Analyzing the transitional region in low power
wireless links. In IEEE SECON, pages 517{526, Oct 2004.
[99] M. Zuniga and B. Krishnamachari. An analysis of unreliability and asymmetry in
low-power wireless links. ACM Transactions on Sensor Networks, 3, 2, Jun 2007.
172

Abstract (if available)

Abstract Wireless sensor networks deployed densely for fine-grained monitoring often experience high channel contention from concurrent packet transmission. The main cause of concurrency in these networks is the bursty nature of event-based traffic. Co-channel interference from simultaneous transmission is inevitable in wireless communication, and a better understanding of it is essential for reliable and efficient communication protocol design.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Transport layer rate control protocols for wireless sensor networks: from theory to practice

PDF

Robust routing and energy management in wireless sensor networks

PDF

Algorithmic aspects of throughput-delay performance for fast data collection in wireless sensor networks

PDF

Realistic modeling of wireless communication graphs for the design of efficient sensor network routing protocols

PDF

Reliable and power efficient protocols for space communication and wireless ad-hoc networks

PDF

Time synchronization and scheduling in underwater wireless networks

PDF

Aging analysis in large-scale wireless sensor networks

PDF

Reconfiguration in sensor networks

PDF

Multichannel data collection for throughput maximization in wireless sensor networks

PDF

Congestion control in multi-hop wireless networks

PDF

A protocol framework for attacker traceback in wireless multi-hop networks

PDF

On location support and one-hop data collection in wireless sensor networks

PDF

Gradient-based active query routing in wireless sensor networks

PDF

Efficient and accurate in-network processing for monitoring applications in wireless sensor networks

PDF

Distributed wavelet compression algorithms for wireless sensor networks

PDF

Relative positioning, network formation, and routing in robotic wireless networks

PDF

Design of cost-efficient multi-sensor collaboration in wireless sensor networks

PDF

Understanding the characteristics of Internet traffic dynamics in wired and wireless networks

PDF

Cooperation in wireless networks with selfish users

PDF

Dynamic routing and rate control in stochastic network optimization: from theory to practice

Asset Metadata

Creator Son, Dongjin (author)

Core Title Towards interference-aware protocol design in low-power wireless networks

School Viterbi School of Engineering

Degree Doctor of Philosophy

Degree Program Electrical Engineering

Publication Date 11/30/2007

Defense Date 09/07/2007

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag concurrent transmission,interference,OAI-PMH Harvest,power control,wireless sensor networks

Language English

Advisor Krishnamachari, Bhaskar (committee chair), Heidemann, John (committee member), Raghavendra, Cauligi S. (committee member), Sukhatme, Gaurav S. (committee member)

Creator Email dongjins@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-m956

Unique identifier UC1320530

Identifier etd-Son-20071130 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-599225 (legacy record id),usctheses-m956 (legacy record id)

Legacy Identifier etd-Son-20071130.pdf

Dmrecord 599225

Document Type Dissertation

Rights Son, Dongjin

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Repository Name Libraries, University of Southern California

Repository Location Los Angeles, California

Repository Email cisadmin@lib.usc.edu