BANDWIDTH SCAVENGING FOR
DEVICE COEXISTENCE IN WIRELESS NETWORKS
By
Anthony Tyrone Plummer Jr.

A DISSERTATION
Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of
DOCTOR OF PHILOSOPHY
Electrical Engineering
2011

ABSTRACT
BANDWIDTH SCAVENGING FOR
DEVICE COEXISTENCE IN WIRELESS NETWORKS
By
Anthony Tyrone Plummer Jr.

The objective of this thesis is to develop a wireless channel access framework that allows
secondary user (SU) devices to coexist with primary user (PU) devices via utilizing unused
spectrum or whitespace found between the PUs’ transmissions. Those whitespaces typically last
for short durations (i.e. on the order of milliseconds) in between the active data transmissions by
the PUs. The key design objectives for an SU access strategy are to “scavenge” the maximum
amount of spatio-temporally fragmented whitespace while limiting the amount of disruptions
caused to the primary users (PU). These conflicting goals become particularly challenging
without deterministic prior knowledge about the future occurrences and durations of the
whitespaces. Our approach to address this problem is to develop stochastic whitespace access
mechanisms based on previously observed statistical model of the channel whitespace. The key
contributions of this thesis are as follows. First, it provides an extensive statistical analysis of the
whitespace characteristics using simulations and experiments on a prototype testbed, in the
presence of various primary user traffic scenarios. The simulated network allows explorations of
the effects of various traffic and topology conditions, and the experimental testbed provides
insights into real world whitespace measurements in the presence of various hardware and
operating system related limitations. The second contribution is an opportunistic access strategy

for the secondary users that is developed based on the measurement and modeling of whitespace
resulted from ad hoc mode 802.11 PU traffic. This opportunistic channel access, or scavenging,
during ultra-short and non-deterministic 802.11 whitespace is then evaluated for functionality
and performance through analytical modeling, network simulation, and testbed experiments. It is
demonstrated that the proposed strategy is able to consistently scavenge above 90% of the
available whitespace capacity, while keeping the primary users disruption less than 5%. The
third major contribution is to generalize the above bandwidth scavenging approach by extending
the proposed technique for arbitrary whitespace distributions. This generalization has resulted in
a new access strategy that can be applied to non-802.11 primary traffic and is able to handle
whitespace durations with multi-modal density functions. Effectiveness under arbitrary
whitespace profiles is achieved by introducing a new concept of transmission opportunities
within a given whitespace, and then developing a knapsack optimization based transmission
probability shuffling mechanism across the transmission opportunities with the whitespace.
Combining the above three components, the thesis offers a framework for SU support within a
PU network with arbitrary topology and traffic profiles.

Copyright by
Anthony Tyrone Plummer Jr.
2011

To My Wife
For All Her
Love And Support

v

ACKNOWLEDGEMENTS

I would like to thank my advisor Dr. Subir Biswas for supporting me through my work.
His guidance allowed me to flourish during my time graduate school. I would also like to thank
my committee Dr. Matt Mutka, Dr. Tongtong Li, and Dr. Percy Pierre for their time and support.
I would also like to especially thank Dr. Percy Pierre and Dr. Barbara O’kelly for their
guidance and support in helping me attend and be successful in graduate school.
Thanks must be also given to my lab mates Mahmoud Taghizadeh, Jayanthi Rao, Matthias
Schmidt, Muhannad Quwaider, Tao Wu and Fan Yu for all of the brainstorming and
implementation discussions. Last but not least, I would like give thanks to my wife Raenita
Fenner, my parents Anthony Plummer Sr. and Avis Plummer along with the rest of my friends
and family for supporting me through this process.

vi

TABLE OF CONTENTS

LIST OF TABLES ......................................................................................................................... xi
LIST OF FIGURES ...................................................................................................................... xii
Chapter 1:

Introduction ............................................................................................................. 1

1.1
PRIORITIZED DEVICE COEXISTENCE IN WIRELESS NETWORKS .................................... 1
1.2
GENERAL DEFINITION OF PRIMARY AND SECONDARY USERS ...................................... 2
1.3
ROLE OF MAC LAYER TO SUPPORT DEVICE COEXISTENCE ......................................... 3
1.4
BANDWIDTH SCAVENGING FOR DEVICE COEXISTENCE ................................................ 4
1.5
STATE OF THE ART ....................................................................................................... 6
1.5.1 Traffic Models of Primary Users ............................................................................ 7
1.5.2 Random Access Based Primary User Networks ..................................................... 7
1.5.3 TDMA based Primary User Networks .................................................................... 9
1.5.4 Other Secondary User Access Strategies .............................................................. 10
1.5.5 Access Strategies Considering Multiple Secondary Users ................................... 11
1.6
CASE STUDY: COEXISTENCE IN 802.11 NETWORKS ................................................... 12
1.6.1 Motivation and Applications................................................................................. 12
1.6.2 Problem Definition................................................................................................ 13
1.7
CASE STUDY: TRAFFIC PROTECTION IN CSMA BASED SENSOR NETWORKS .............. 13
1.7.1 Motivation and Applications................................................................................. 13
1.7.2 Problem Definition................................................................................................ 14
Chapter 2:

Dissertation Scope and Contributions................................................................... 16

2.1
INTRODUCTION .......................................................................................................... 16
2.2
BANDWIDTH SCAVENGING COMPONENTS .................................................................. 17
2.3
SCOPE ........................................................................................................................ 18
2.4
CONTRIBUTIONS ........................................................................................................ 19
2.4.1 Coexistence in 802.11 Networks .......................................................................... 19
2.4.2 Traffic Protection in CSMA based Sensor Networks ........................................... 19
2.4.3 Generalization of Bandwidth Scavenging Concept for Wireless Networks ......... 19
Chapter 3:

802.11 Whitespace Analysis ................................................................................. 20

3.1
INTRODUCTION .......................................................................................................... 20
3.1.1 Motivation ............................................................................................................. 20
3.1.2 Problem Definition................................................................................................ 20
3.1.3 Related work ......................................................................................................... 21
vii

SIMULATION SETUP ................................................................................................... 22
3.2
3.2.1 NS2 Implementation ............................................................................................. 22
3.2.2 Sample Experiments with PUs and SUs topology ................................................ 25
3.3
AD-HOC PRIMARY USER NETWORK ........................................................................... 31
3.3.1 PU and SU Topologies.......................................................................................... 31
3.3.2 Traffic Profiles ...................................................................................................... 32
3.3.3 Channel Characteristics ........................................................................................ 33
3.4
WHITESPACE CHARACTERISTICS ............................................................................... 37
3.4.1 Whitespace Model ................................................................................................ 37
3.4.2 Whitespace Statistics ............................................................................................ 38
3.5
SUMMARY AND CONCLUSIONS................................................................................... 51
Chapter 4:

Contiguous SU Transmission Strategy (CSTS) .................................................... 52

4.1
INTRODUCTION .......................................................................................................... 52
4.1.1 Motivation ............................................................................................................. 52
4.1.2 Problem Definition................................................................................................ 52
4.1.3 Assumptions.......................................................................................................... 53
4.1.4 Related Work ........................................................................................................ 53
4.2
CSTS ACCESS ALGORITHM OVERVIEW ..................................................................... 55
4.3
FUNCTIONAL OVERVIEW............................................................................................ 58
4.4
ALGORITHM ANALYSIS .............................................................................................. 59
4.4.1 Primary Traffic Disruption ................................................................................... 59
4.4.2 Effective Secondary Throughput .......................................................................... 60
4.4.3 Dimensioning Wait-threshold µ ............................................................................ 62
4.4.4 Jmax Computation .................................................................................................. 63
4.4.5 Impacts of wait threshold µ .................................................................................. 63
4.5
SUMMARY AND CONCLUSION .................................................................................... 65
Chapter 5:

Performance Evaluation for Coexistence in 802.11 Network ............................. 66

5.1
INTRODUCTION .......................................................................................................... 66
5.1.1 Comparison Protocols ........................................................................................... 66
5.1.2 Simulation Implementation ................................................................................... 68
5.2
CSTS EVALUATION ................................................................................................... 69
5.2.1 Primary User Load ................................................................................................ 69
5.2.2 Secondary User Packet Size.................................................................................. 73
5.2.3 Dynamic PU Traffic Patterns ................................................................................ 76
5.2.4 Channel Sensing Rate ........................................................................................... 79
5.2.5 Disruption Bound .................................................................................................. 80
5.3
SUMMARY AND CONCLUSIONS................................................................................... 82
Chapter 6:
6.1

Traffic Protection in a Prototype CSMA based Sensor Networks ........................ 83
INTRODUCTION .......................................................................................................... 83
viii

6.1.1 Motivation ............................................................................................................. 83
6.1.2 Problem Definition................................................................................................ 84
6.1.3 Related Work ........................................................................................................ 85
6.2
TESTBED SETUP ......................................................................................................... 86
6.2.1 Crossbow TelosB Motes ....................................................................................... 88
6.2.2 Hardware Components and Network Topology ................................................... 88
6.3
PRIMARY USER TRAFFIC PROFILES ............................................................................ 89
6.4
PERFORMANCE EVALUATION OF CSTS ON PROTOTYPE TESTBED ............................. 92
6.4.1 Compared Protocols .............................................................................................. 92
6.4.2 Experiment Implementation.................................................................................. 92
6.5
PERFORMANCE WITH UNIFORM PU TRAFFIC PROFILE ................................................ 93
6.6
PERFORMANCE WITH PU TRAFFIC WITH LOW-RATE VIDEO ....................................... 97
6.7
PERFORMANCE WITH MULTI-MODAL PU TRAFFIC ..................................................... 99
6.8
SUMMARY AND CONCLUSIONS................................................................................. 101
Chapter 7:

Divided SU Transmission Strategy (DSTS) ....................................................... 103

7.1
INTRODUCTION ........................................................................................................ 103
7.1.1 Motivation ........................................................................................................... 103
7.1.2 Problem Definition.............................................................................................. 103
7.1.3 Related Work ...................................................................................................... 103
7.2
ACCESS BASED ON TRANSMISSION OPPORTUNITIES ................................................ 104
7.3
TRANSMISSION BITMAP VECTOR ............................................................................. 105
7.4
FUNCTIONAL OVERVIEW.......................................................................................... 107
7.5
PU DISRUPTIONS AND SU THROUGHPUT ................................................................. 107
7.5.1 Probability of TO Existence................................................................................ 107
7.5.2 Probability of Disruption .................................................................................... 108
7.6
COMPUTING TBV VIA SOLVING KNAPSACK PROBLEM ............................................ 109
7.7
SUB-OPTIMAL TBV COMPUTATION ......................................................................... 111
7.8
TBV SOLUTION EXAMPLE ....................................................................................... 112
7.9
PERFORMANCE EVALUATION OF THE DSTS PROTOCOL .......................................... 112
7.9.1 Comparison Protocols ......................................................................................... 113
7.9.2 Ad hoc 802.11 Whitespace ................................................................................. 115
7.9.3 CSMA Profile Using Test Bed Measured Whitespace ....................................... 123
7.10 SUMMARY AND CONCLUSION .................................................................................. 128
Chapter 8:

DSTS With Support for Multiple Secondary Users ............................................ 130

8.1
INTRODUCTION ........................................................................................................ 130
8.1.1 Motivation ........................................................................................................... 130
8.1.2 Problem Definition.............................................................................................. 130
8.2
TBVAP STRUCTURE ................................................................................................ 131
8.3
DELAY ANALYSIS .................................................................................................... 133
8.4
WHITESPACE ANALYSIS FOR MULTIPLE PUS ........................................................... 135
8.4.1 Network Setup .................................................................................................... 135
ix

8.4.2 Primary User Traffic Profile ............................................................................... 136
8.4.3 Whitespace Analysis ........................................................................................... 136
8.4.4 Whitespace Capacity........................................................................................... 138
8.5
PERFORMANCE EVALUATION OF TBVAP ................................................................ 140
8.5.1 Impacts of Primary User Load ............................................................................ 140
8.5.2 Impacts of Secondary User Load ........................................................................ 142
8.5.3 Delay Performance.............................................................................................. 145
8.6
SUMMARY AND CONCLUSION .................................................................................. 146
Chapter 9:
9.1
9.2

Summary and Conclusions ................................................................................. 147
CONTRIBUTIONS ...................................................................................................... 147
FUTURE WORK ........................................................................................................ 150

Appendix A:

TelosB Based TestBed Implementation.......................................................... 154

A.1 INTRODUCTION ........................................................................................................ 154
A.2 PRIMARY USER NETWORK (PUN) ........................................................................... 154
A.2.1
Mote Side Implementation .............................................................................. 155
A.2.2
PC Side Implementation ................................................................................. 158
A.2.3
Traffic Generation Evaluation ........................................................................ 166
A.3 SECONDARY USER NETWORK (SUN)....................................................................... 170
A.3.1
RSSI Measurement ......................................................................................... 170
A.3.2
Evaluation of the RSSI Sampling ................................................................... 173
A.3.3
Mote Side Implementation .............................................................................. 177
A.3.4
PC Side Implementation ................................................................................. 185
Appendix B:

LIST OF PUBLICATIONS ............................................................................ 188

Bibliography ............................................................................................................................... 190

x

LIST OF TABLES

Table 5-1: Simulator implementation steps .................................................................................. 68
Table A-1 RSSI sampling specifications .................................................................................... 171
Table A-2: Access strategies and states ...................................................................................... 180

xi

LIST OF FIGURES

Figure 1-1: Different whitespace profiles and their CDFs.............................................................. 5
Figure 2-1: Architectural components of Bandwidth Scavenging ................................................ 17
Figure 2-2: Application of the Bandwidth Scavenging Concept .................................................. 18
Figure 3-1: Primary and Secondary Users grid topology ............................................................. 23
Figure 3-2: Primary and Secondary Users interference ranges .................................................... 23
Figure 3-3: Primary and Secondary Users interference ranges for three 2-hop PU horizontal
connections ................................................................................................................................... 25
Figure 3-4: Secondary Users whitespace measurements for three 2-hop PU horizontal
connections ................................................................................................................................... 27
Figure 3-5: Secondary Users whitespace measurements over entire network for three 2-hop PU
horizontal connections .................................................................................................................. 27
Figure 3-6: Primary and Secondary Users interference ranges for five 4-hop PU horizontal
connections ................................................................................................................................... 28
Figure 3-7: Secondary Users whitespace measurements over entire network for five 4-hop PU
horizontal connections .................................................................................................................. 28
Figure 3-8: Primary and Secondary Users interference ranges for seven 6-hop PU horizontal
connections ................................................................................................................................... 29
Figure 3-9: Secondary Users whitespace measurements over entire network for seven 6-hop PU
horizontal connections .................................................................................................................. 29
Figure 3-10: Primary and Secondary Users interference ranges for two 5-hop PU horizontal
connections ................................................................................................................................... 30
Figure 3-11: Secondary Users whitespace measurements over entire network for two 5-hop PU
horizontal connections .................................................................................................................. 30
Figure 3-12: PUs and SUs topology and traffic flow for linear chain. (TOP1) ............................ 31
xii

Figure 3-13: PUs and SUs topology and traffic flow for intersecting chain. (TOP2) .................. 32
Figure 3-14: PUs and SUs topology and traffic flow for parallel chain. (TOP3) ......................... 32
Figure 3-15: Channel measurements for 0.5 inter-packet time over 10 seconds .......................... 34
Figure 3-16: Channel measurements for 0.1 inter-packet time over 10 seconds .......................... 34
Figure 3-17: Channel measurements for 0.1 inter-packet time over 0.3 seconds ......................... 35
Figure 3-18: Channel measurements for 0.1 inter-packet time over 0.05 seconds ....................... 35
Figure 3-19: Channel measurements for 0.1 inter-packet time..................................................... 36
Figure 3-20: On-Off whitespace model ........................................................................................ 38
Figure 3-21: Sensing period .......................................................................................................... 38
Figure 3-22: Impacts of Uniform whitespace pdf distribution ..................................................... 40
Figure 3-23: Impacts of Poisson whitespace pdf distribution....................................................... 40
Figure 3-24: Uniform whitespace cdf distribution within the interval 0 ms to 1 ms .................... 41
Figure 3-25: Poisson whitespace cdf distribution within the interval 0 ms to 1 msTCP and Video
Distributions.................................................................................................................................. 41
Figure 3-26: Impacts of TCP whitespace pdf distribution ............................................................ 43
Figure 3-27: Impacts of video stream whitespace pdf distribution............................................... 43
Figure 3-28: TCP whitespace cdf distribution within the interval 0 ms to 1 ms .......................... 44
Figure 3-29: Video Stream whitespace cdf within the interval 0 ms to 1 ms ............................... 44
Figure 3-30: Impacts of data rate for Uniform pdf distribution .................................................... 45
Figure 3-31: Impacts of data rate for Poisson pdf distribution ..................................................... 45
Figure 3-32: Impacts of data rate for Uniform and Poisson cdfs within the interval 0 ms to 1 ms
....................................................................................................................................................... 46
Figure 3-33: Impacts of sensing period on Uniform whitespace pdf distribution ........................ 46
Figure 3-34: Impacts of sensing period on Poisson whitespace pdf distribution ......................... 47
xiii

Figure 3-35: Impacts of sensing rate for Uniform and Poisson cdfs within the interval 0 ms to 1
ms .................................................................................................................................................. 48
Figure 3-36: Available channel capacity in terms of fraction of whitespace for the Uniform
traffic profile with various inter-packet arrival times and topologiesAvailable Whitespace
Capacity ........................................................................................................................................ 48
Figure 3-37: Available channel capacity in terms of fraction of whitespace for the Poisson traffic
profile with various inter-packet arrival times and topologies ..................................................... 50
Figure 3-38: Available channel capacity in terms of fraction of whitespace for the TCP and
Video Stream traffic profile .......................................................................................................... 50
Figure 4-1: Different stages of CSTS access ................................................................................ 56
Figure 4-2: CSTS access algorithm .............................................................................................. 58
Figure 4-3: Functional diagram of the SU access strategy ........................................................... 59
Figure 4-4: Evaluation of the impact of µ in CSTS ...................................................................... 65
Figure 5-1: Primary user topologies and the traffic flows ............................................................ 67
Figure 5-2: Performance for uniformly distributed PU traffic...................................................... 71
Figure 5-3: Performance for Poisson distributed PU traffic ......................................................... 73
Figure 5-4: Impacts of the SU packet size on EST and PTD on TOP1 ........................................ 74
Figure 5-5: Impacts of the SU packet size on EST and PTD on TOP2 ........................................ 75
Figure 5-6: Impacts of the SU packet size on EST and PTD on TOP3 ........................................ 75
Figure 5-7: Impacts of the SU packet size on EST and PTD for Video Stream ........................... 76
Figure 5-8: Performance with time-varying w(n) for TOP2 and TOP3........................................ 78
Figure 5-9: Performance with time-varying w(n) for TOP1 and TOP3........................................ 78
Figure 5-10: Impacts of channel sensing interval Tp .................................................................... 80
Figure 5-11: Impacts of Disruption Bound for the uniformly distributed PUs............................. 81
Figure 5-12: Impacts of Disruption Bound for the Poisson distributed PUs ................................ 81
xiv

Figure 6-1: System implementation overview .............................................................................. 87
Figure 6-2: Physical topology of the primary and secondary networks ....................................... 89
Figure 6-3: Experimental whitespace distributions. (a) Uniform with 20% variation, (b) Low rate
video stream, and (c) Multi- modal data stream ........................................................................... 91
Figure 6-4: PU packets deferred for 120 ppm uniform traffic ...................................................... 94
Figure 6-5: PU packets deferred for 120 ppm uniform traffic in Log scale ................................. 94
Figure 6-6: RTT for 120 ppm uniform traffic ............................................................................... 95
Figure 6-7: MST for 120 ppm uniform traffic .............................................................................. 95
Figure 6-8: Simulation results for 120 ppm uniform traffic ......................................................... 96
Figure 6-9: Simulation results for various uniform traffic rates ................................................... 97
Figure 6-10: PU packets deferred for video stream based traffic ................................................. 98
Figure 6-11: MST for video stream based traffic ......................................................................... 98
Figure 6-12: Simulation results for video stream based traffic..................................................... 99
Figure 6-13: PU packets deferred for multi-modal based traffic ................................................ 100
Figure 6-14: MST for multi-modal based traffic ........................................................................ 100
Figure 6-15: Simulation results for multi-modal based traffic ................................................... 101
Figure 7-1: DSTS transmission opportunities............................................................................. 105
Figure 7-2: Functional diagram of the SU access strategy ......................................................... 106
Figure 7-3: TBV creation via sub-optimal solution .................................................................... 111
Figure 7-4: TBV of DSTS for different traffic profiles .............................................................. 112
Figure 7-5: Primary user topologies and the traffic flows .......................................................... 116
Figure 7-6: Impacts of PU topology with Uniform, Poisson, TCP and Video traffic on whitespace
distributions. The graphs (a), (b), (d), and (e) show the pdf, along with (c) and (f) presenting the
corresponding cdf for the interval 0 ms to 1 ms ......................................................................... 117
xv

Figure 7-7: Performance for uniformly distributed PU traffic.................................................... 118
Figure 7-8: Performance for Poisson distributed PU traffic ....................................................... 120
Figure 7-9: Impacts of the SU packet size on EST and PTD...................................................... 121
Figure 7-10: Performance with time-varying w(n) for 802.11 ................................................... 122
Figure 7-11: Experimental whitespace distributions. (a) Uniform with 20% variation, (b) Low
rate video stream, and (c) Multi- modal data stream .................................................................. 124
Figure 7-12: Performance with varying SU traffic load under Uniform traffic ......................... 125
Figure 7-13: Performance with Multi-modal whitespace pdf ..................................................... 126
Figure 7-14: Performance with Video Stream whitespace pdf ................................................... 127
Figure 7-15: Performance with time-varying w(n) for CSMA MAC ......................................... 128
Figure 8-1: TBV with the corresponding TO structure .............................................................. 131
Figure 8-2: TO Access in TBVAP .............................................................................................. 132
Figure 8-3: Delay analysis for TBVAP ...................................................................................... 133
Figure 8-4: Multiple SUs and PUs within the same collision domain ........................................ 135
Figure 8-5: Impacts of 1, 2, and 4 Poisson primary users on whitespace distributions. ............. 137
Figure 8-6: Impacts of 6, 8, and 10 Poisson primary users on whitespace distributions. ........... 137
Figure 8-7: Impacts of 1, 2, and 4 Uniform primary users on whitespace distributions. ........... 137
Figure 8-8: Impacts of 6, 8, and 10 Uniform primary users on whitespace distributions. ......... 138
Figure 8-9: Idle time fraction for various Poisson PU inter-packet arrivals ............................... 139
Figure 8-10: Idle time fraction for various Uniform PU inter-packet arrivals ........................... 139
Figure 8-11: Performance for Poisson distributed PU traffic ..................................................... 140
Figure 8-12: Performance for uniformly distributed PU traffic.................................................. 141
Figure 8-13: Performance for multiple SUs under Poisson PU traffic and 4 PUs...................... 142
xvi

Figure 8-14: Performance for multiple SUs under Uniform PU traffic and 4 PUs .................... 142
Figure 8-15: Performance for multiple SUs under Poisson PU traffic and 1 PUs...................... 143
Figure 8-16: Performance for multiple SUs under Uniform PU traffic and 1 PUs .................... 144
Figure 8-17: Delay performance for 20 SUs under Poisson traffic with 5 PUs ......................... 145
Figure 8-18: Delay performance for 20 SUs under Uniform traffic with 10 PUs ...................... 145
Figure 9-1: Example of Multiple Whitespace Domains ............................................................. 151
Figure A-1: Overview of PUN implementation.......................................................................... 155
Figure A-2: Message structure of mote ...................................................................................... 156
Figure A-3: PC side overview..................................................................................................... 158
Figure A-4: Primary user control interface ................................................................................. 159
Figure A-5: Primary User communication interface configuration screen ................................. 161
Figure A-6: Measurement data structure .................................................................................... 162
Figure A-7: Traffic Simulator control screen ............................................................................. 163
Figure A-8: PDF of the first multimodal scheme ....................................................................... 164
Figure A-9: PDF of the second multimodal scheme ................................................................... 165
Figure A-10: PDF of the RTT distribution of packets ................................................................ 166
Figure A-11: Example of Traffic Generator delays .................................................................... 168
Figure A-12: PDF of scheduled packets at 1 pps with a variance of 20%.................................. 169
Figure A-13: PDF of sent packets at 1 pps with a variance of 20% ........................................... 169
Figure A-14: PDF of scheduled packets at 10 pps with a variance of 20%................................ 169
Figure A-15: PDF of sent packets at 10 pps with a variance of 20% ......................................... 170
Figure A-16: Pseudo-code of RSSI-Sampling ............................................................................ 171
Figure A-17: RSSI Measurement example ................................................................................. 172
xvii

Figure A-18: RSSI register values and actual RF level [101] .................................................... 173
Figure A-19: RSSI measurement of 2 pps and a 1 kHz sampling rate ....................................... 174
Figure A-20: RSSI measurement of 2 pps with 20% variation and a 1 kHz sampling rate........ 175
Figure A-21: RSSI measurement of Uniform distribution with 7 pps with 20% variation ........ 176
Figure A-22: RSSI measurement of Video Stream profile ......................................................... 176
Figure A-23: RSSI measurement of Multi-model traffic profile ................................................ 177
Figure A-24: Overview of SUN implementation........................................................................ 178
Figure A-25: Pseudo code of packet generation ......................................................................... 179
Figure A-26: Flow chart of CSTS implementation..................................................................... 182
Figure A-27: Flow chart of DSTS implementation .................................................................... 184
Figure A-28: Secondary user control interface ........................................................................... 186

xviii

Chapter 1: Introduction
1.1

Prioritized Device Coexistence in Wireless Networks
Recent research on Dynamic Spectrum Access (DSA) has paved the way for a set of

Secondary Users (SU) to access underutilized spectrum between Primary Users’ (PU)
transmissions in space, time and frequency. It has been shown [1-3] that such SU access can be
feasible through studies of the primary user spectrum usage and the protocols that govern them.
While this is particularly true for the license bands, it also applies to the unlicensed bands due to
the recent proliferation and the subsequent overcrowding of wireless consumer technologies such
as Bluetooth, ZigBee, WLAN, cordless phones, and related applications.
Of particular interest in DSA, is the issue of prioritized coexistence of various devices
utilizing unlicensed bands [4-6]. An example of such coexistence is the sharing of the 2.4 GHz
Industry Scientific and Medical (ISM) band among different device types including sensors and
actuators, Voice-over-IP (VoIP) handsets (e.g. Skype phones), cordless phones, and data
terminals such as laptops and data-enabled 3G/4G phones [7]. A suitable primary-secondary
relationship among those co-existing devices are imposed based on their respective traffic
priorities. For example, consider a VoIP primary user network utilizing a 2.4 GHz band. Along
with the high-priority primary user devices there may coexist with low priority secondary user
devices (e.g. sensors, laptops etc.) using the same band. The requirement here is to design a
strategy to allow the secondary users to access the channel and utilize the bandwidth without
disrupting or restricting the high-priority primary user devices.

1

1.2

General Definition of Primary and Secondary Users
In this work, a solution for multiple networked devices coexisting in a single wireless

channel is investigated. Within such a network, a priority structure is defined among the devices,
which are grouped into two types namely, Primary Users (PU) and Secondary Users (SU).
A PU is given exclusive access rights to the channel. These rights can be given by a
network designer or government agencies such as Federal Communications Commission (FCC).
For example, the FCC can determine that only police networks can use a particular wireless
channel. The traffic properties of the PU network can vary greatly depending on the topology,
traffic pattern, and data rates. As a PU, when there is data to send, channel access is immediate,
unless there is another user with equal rights currently using the channel.
SUs are given lower access rights to the channel, which requires SUs to defer channel
access when a PU is present. The overall goal of the SU is to minimize disruptions caused by its
presence on the channel to a PU network. In opposition to this goal, the SU also wants to
maximize its own throughput on the channel. This requires a smart strategy for the SUs to access
the channel.
The cooperation between PUs and SUs can vary greatly. On one extreme, PUs and SU
work closely together to access the channel. This can manifest in terms of off-line and on-line
scenarios. In the off-line case, the PUs and SUs can divide the available bandwidth in terms of
frequency or temporal, such as in TDMA protocols. For the online case, PUs can send
notification messages to the SUs to indicate their arrival on the channel. Once the PU returns,
SUs can just vacate the channel within a short delay.
In general, a secondary device may or may not be of the same type as the primary device.
Additionally, the SU device may not understand protocol messages from the primary devices,
2

which limits the amount of possible cooperation between the PUs and SUs. The goal of this
thesis is to develop methods for SUs to access various types of PU networks. Therefore, PU and
SUs are assumed completely independent of each other in-terms of sending and receiving
packets. In other words, the PUs and the SUs cannot communicate with each other. However, the
SUs have the ability to detect the primary user signal exclusively through the periodic
measurement of the shared channel. This approach allows a number of general access solutions
to be developed.
1.3

Role of MAC Layer to support Device Coexistence
Supporting device coexistence in wireless networks can be achieved using the Medium

Access Control (MAC) layer through defining a suitable structure to control device access to the
available network bandwidth. Coexistence between multiple systems can be achieved through
spatial, frequency or temporal separation at the MAC layer, as reported in [7] and [8].
In spatial-based device separation, nodes define their range of communication i.e.
transmission range, such that their transmissions do not interfere with other nodes’ transmissions.
For example, transmission power control can be used to adjust a nodes transmission power to
define a suitable non-interfering transmission range [9]. SUs can set their transmission ranges at
a minimal range that still allows communication but reduces the amount of disruptions to the PU
traffic [10].
Frequency-based separation involves distributing nodes into multiple channels so that their
transmissions do not interfere with each other. In this case, lower priority nodes would first
identify unoccupied channels, then freely access the unused channels [2, 11]. Additionally
adaptive frequency hopping techniques can be used.

3

Another promising mechanism for traffic prioritization is to utilize the temporal separation
between a primary and secondary device accessing the same radio frequency (RF). In this
approach, a secondary user models the channel behavior of the primary user network. Then
access the channel at an advantageous time that does not interfere with the primary user. In this
thesis, a secondary user access strategy is presented to utilize the temporal separation between
primary and secondary devices accessing the same RF spectral segments.
1.4

Bandwidth Scavenging for Device Coexistence
To support prioritized device coexistence, a secondary device or user can utilize temporal

based access. The secondary users’ goal, therefore, is to access the idle times between the
transmissions of primary users. These idle times or whitespaces typically last for short durations
(i.e. order of milliseconds) in between the active data transmissions of PUs. From an SU’s
standpoint, once a whitespace is identified, the next step is to send a packet in that whitespace,
only if the estimated chance of completing the transmission before the whitespace ends is high.
The quality of whitespace access, that is bandwidth scavenging in this context, is determined by
the resulting secondary user throughput which should be maximized, and the primary user
disruption, which should be minimized for a given whitespace profile.

4

Figure 1-1: Different whitespace profiles and their CDFs
The SU whitespace access strategy depends solely on the statistical profile of the
whitespace, which in turn, is determined by the topology, traffic and routing characteristics of
the primary user network. Depending on these factors, the available whitespace can have widely
varying characteristics. For example, the neighborhood of an SU may experience a large number
of whitespace occurrences but each whitespace lasts only for a short duration (see Profile-1 in
Figure 1-1.). Conversely, there can be only a few whitespaces, but each with very long durations,
as shown in Profile-2 in Figure 1-1. The cumulative distribution functions (CDF) of the two
whitespace profiles are also shown in the bottom part of the figure. These two different
whitespace profiles indicate that even when the average white space length are the same, for a
given target primary user interference, profile-1 is able to support less number of fixed length SU
packet transmissions due to a higher level of capacity fragmentation during the ends of the
whitespaces. Therefore, the SU throughput for profile-1 is less than that for profile-2, for the
same level of target primary user interference.
5

With non-deterministic primary traffic patterns, an SU is not able to deterministically
predict when a detected whitespace will end. As a result, the SU access strategy cannot be
deterministic. A reasonable solution is for the SU to access a given whitespace based on a
previously observed statistical whitespace profile. Once an SU identifies a whitespace, it sends a
packet in that whitespace only if the estimated chance of completing the transmission before the
whitespace will have ended is reasonably high. An aggressive access may result in higher
secondary user throughput, but can also increase primary user disruptions, which happens when
an SU’s transmission (the last packet transmission in a whitespace) does not end before the
completion of the whitespace in question. A conservative access strategy may produce a reverse
effect, meaning a low PU disruption, but at the expense of an SU throughput that is lower than
what could be achieved for a given whitespace profile. Therefore, the objective for an SU is to be
able to make a transmission decision based on previously observed whitespace statistics, such
that by prudently transmitting packets, it is able to maximize the achievable throughput and
minimize the disruptions. We term this opportunistic access during ultra-short and nondeterministic whitespaces as bandwidth scavenging by the secondary users. In other words, the
SUs scavenge the channel capacity left over by the PUs. In this thesis, a solution to prioritized
device coexistence is proposed by designing secondary user channel access strategies for
efficient bandwidth scavenging.
1.5

State of the Art
In this section, an overview of the current research on device coexistence is presented.

6

1.5.1

Traffic Models of Primary Users
The traffic profiles of packet based primary users studied in literature have varied greatly

[12-15]. The authors in [15] model the primary users’ packet arrival intervals and packet
durations as two separate Poisson processes. Similarly, [13] uses an alternating ON-OFF pattern,
where the duration of the idle and busy states follows an exponential distribution with means v
and l respectively. In [12], a Markov chain with known transition probabilities is used to model
primary user behavior. Lastly in [14], a Markov model with Exponential, Uniform, Weibull and
Generalized Pareto distributions are considered for PUs arrival times. In line with these efforts in
the literature, this thesis investigates SU access mechanisms for a variety of PU traffic profiles.
1.5.2

Random Access Based Primary User Networks
Networks with primary users running 802.11 MAC protocol have recently been

investigated for possible access by secondary users. The authors in [16, 17] develop a
methodology for formally analyzing the whitespace available within 802.11 primary traffic in
infrastructure mode. The key idea is to model the whitespace as a semi-Markov process that
relies on the underlying 802.11 state model involving DIFS, SIFS, DATA, and ACK
transactions. The model describes the whitespace profile in terms of holding times of the idle and
busy states of the channel.
The authors in [6] develop an WLAN access strategy for secondary users in which the SUs
utilize packet size slots for the channel access. At the beginning of each slot, an SU senses the
channel and if the channel is free then it transmits with a specified probability that is calculated
from previous measurement. A greedy version also uses probabilistic analysis to decide whether
or not to transmit in a given slot. The objective is to minimize PU disruption and maximize SU
throughput. The use of time-slotting in their approach requires time synchronization across the
7

secondary users. In the access mechanism proposed in this thesis, the need for such inter-SU
time synchronization is avoided via asynchronous whitespace access based on a stochastic
whitespace modeling approach.
Prioritized device coexistence within CSMA based WLANs can be achieved [18, 19] by
using different inter frame spacing (IFS) periods for different user and/or traffic classes. MAC
protocol 802.11e [18, 20], for instance, uses different Arbitration IFS (AIFS) periods to provide
CSMA based prioritized access among different device and/or traffic classes. When a channel is
found free, a node waits for a specific AIFS periods depending on the device or traffic class,
before it attempts to send a packet. For higher priority primary traffic, a node waits for a smaller
AIFS period. This ensures when multiple nodes contends for the channel, the primary users’
nodes (with smallest AIFS) wins.
While providing reasonable access differentiation, these approaches rely only on the
instantaneous channel status (i.e. free or busy) for granting access. This leads to undesirable
disruptions to the PU traffic as follows. Consider a situation in which an SU intends to transmit a
packet and it does so after finding the channel free (i.e. a whitespace) for the AIFS specified for
the SUs. Now, in the middle of this SU’s packet transmission if an PU in the vicinity intends to
send a packet, it needs to wait until the current SU transmission is over. This causes an
undesirable delay for the PU traffic, which in turn will affect the PUs’ application performance.
Since this is mainly a result of the SUs’ reliance only on the instantaneous channel state, a more
robust approach for the SUs would be to also consider the long term whitespace model. The
following DSA approaches attempt to accomplish that.
The topic of device coexistence has also been studied in the context of Bluetooth and
WLAN coexistence. The authors in [5] use OLA (OverLap Avoidance) mechanisms, which use

8

simple traffic scheduling techniques at the MAC layer. In the proposed scheme, overlap in time
between the Bluetooth traffic and the 802.11 data packets is avoided by performing a proper
scheduling of the traffic transmissions at the WLAN stations. In a Bluetooth network, each link
occupies the channel slots according to a deterministic pattern. Therefore, an 802.11 station
begins transmitting when the Bluetooth channel is idle and adjust the length of the WLAN
packet so that it fits between two successive Bluetooth transmissions.
1.5.3

TDMA based Primary User Networks
Time Division Multiple Access (TDMA) protocols have been investigated as a possible

method to assign priority to different traffic classes. TDMA works by dividing a channel into
fixed sized time slots, which are organized into periodic frames. Each node in a network is
assigned a slot or multiple slots to be used during each periodic frame. To institute priority
amongst the nodes, higher priority nodes are assigned a larger portion of the available slots in a
frame over the lower priority nodes. This process ensures that the higher priority nodes utilizes
the channel with higher frequency and allows dedicated bandwidth to the higher priority traffic.
Additionally, conflict free access is supported, which means that the high priority traffic should
not be disrupted by the lower priority traffic during channel access. However, when the higher
priority nodes are not utilizing the allocated bandwidth, lower priority nodes cannot use the
unused bandwidth. This can reduce the efficiency of the channel usage.
To overcome this issue, the wireless sensor network protocol presented in [21] divides the
TDMA frames into a broadcast period and scheduled access period. The scheduled access
sections are further divided into priority groups with multiple transmission slots in each group.
Within each frame, a node broadcasts their traffic priority during the broadcast period, which
reserves their transmissions for the specified priority groups. During the scheduled period, nodes
9

send packets using one of the available slots within its priority group section. Another protocol
named PMAC [22] proposes a TDMA based prioritization MAC utilizing a wide range of prefix
reservation and inter-slot intervals.
In [12, 23], the proposed access strategies rely on specific types of PU networks. Protocols
are developed in [12] for a slotted PU network where the SUs are synchronized with the PU’s
slots. In this case, if an PU does not use a slot then the SU can transmit after a small duration of
time in the same slot. The PUs in [23] are assumed to use a specific combination of time and
frequency multiplexing to prevent inter-PU access collisions.
TDMA based traffic prioritization provides the benefit of absolute priority for different
traffic classes. A drawback of the protocol is the requirement of tight time synchronization
needed to implement time slotting. Additionally, TDMA protocols can suffer from reduced
bandwidth usage due to unused scheduled slots during variable bit rate traffic.
In this thesis, we formulate the PU network more generally by not assuming any specific
PU MAC layer such as the slotted ones in [12]. The target here is to develop an SU access
mechanism, which can deal with unslotted as well as stochastic PU MAC layers such as 802.11,
CSMA and ALOHA protocols. This assumption of a general PU access protocol broadens the
applicability of the SU access approach proposed in this thesis.
1.5.4

Other Secondary User Access Strategies
Game theoretic mechanisms have been introduced in [24-32] for spectrum access for the

SUs. The PUs in this framework cooperates with the SUs through pricing strategies and
subscription fees to facilitate spectrum sharing. These approaches require the PUs to have
knowledge of the SUs presence. In contrast, the proposed mechanism in this thesis, no such
cooperation is assumed, thus no changes in the PUs behavior is needed. The idea here is to
10

develop an access mechanism for the SUs for scavenging the PUs’ leftover bandwidth, without
the PUs being aware of such scavenging. For example, data-enabled hand-held devices should be
able to scavenge bandwidth in a WLAN without the primary VoIP handsets being aware of the
scavenging process. There are also other approaches to access strategies [33-38].
1.5.5

Access Strategies Considering Multiple Secondary Users
Functionality and performance of MAC layer protocols in the presence of multiple

secondary users have mostly been studied in the context of multiple channels [6, 39-42]. The
authors in [39] utilizes a common control channel between the SUs. When a SU wants to send a
packet it first switches to the control channel and sends an RTS. Once a CTS is received from the
destination node, the two nodes sense every available PU channel then transmit on a jointly
chosen channel. MAC layer protocols for multiple SUs in a single channel context have not
received equal attention. The authors in [43] present a random access MAC protocol for
secondary users in a single channel. Their proposed protocol utilizes a frame structure where at
the beginning of each frame the SUs sense the channel for the presence of PUs. If the channel is
sensed free, the SUs continuously send packets until the next frame begins. It is assumed that the
whitespace will remain free for the entire duration of the frame. This protocol may not work well
when PU traffic pattern are highly dynamic as in packet based primary user networks. The
presented solutions in this thesis avoid this limitation through dynamically allocating SU
transmissions based on the statistics of the PU traffic. Cooperative sensing have also been
investigated as an approach to support multiple secondary users [44, 45].

11

1.6

Case Study: Coexistence in 802.11 Networks
The proposed bandwidth scavenging concept is applied to scenarios, namely, coexistence

802.11 in networks and traffic protection in CSMA networks.
1.6.1

Motivation and Applications
The recent proliferation and the subsequent overcrowding of wireless consumer

technologies such as Bluetooth, ZigBee, WLAN, cordless phones, and related applications in
unlicensed bands is leading to a need for prioritized coexistence of various devices [4-6]. An
example of such coexistence is the sharing of the 2.4 GHz Industry Scientific and Medical (ISM)
band among different device types. The ISM bands are intended for the operation of unlicensed
devices, many of which are consumer level devices. Communication devices using the ISM
bands must tolerate any interference from other ISM equipment. Therefore, these devices must
co-exist within the band. Many home devices such as cordless telephones and car alarms utilize
the 2.4 Ghz band for operation. Packet based communication device also share the ISM band.
Wireless Personal Area Network (WPAN) and Wireless Local Area Networks (WLAN)
applications such as Bluetooth, Zigbee, and 802.11 utilize the ISM bands for communication [7].
A suitable primary-secondary relationship among those co-existing devices are imposed
based on their respective traffic priorities. For example, consider a VoIP primary user network
utilizing a 2.4 GHz band. Along with the high-priority primary user devices there are coexisting
low priority secondary user devices (e.g. sensors, laptops etc.) using the same band. The
objective here is to design a strategy to allow the secondary users to access the channel and
utilize the bandwidth without disrupting or restricting the high-priority primary user devices.

12

1.6.2

Problem Definition
The secondary users’ goal is to access the idle times between the transmissions of primary

users. These idle times or whitespaces typically last for short durations (i.e. order of
milliseconds) in between the active data transmissions of PUs. The primary challenge in
designing an access strategy for the SUs is how to minimize the disruption to the primary users
while maximizing the secondary user throughput without deterministic knowledge about the
future occurrences of whitespaces and their durations. With non-deterministic primary traffic
patterns, an SU is not able to deterministically predict when whitespaces will occur and when a
detected whitespace will end. As a result, the SU access strategy cannot be deterministic. A
reasonable solution is for the SUs to access a given whitespace based on the statistical profile of
the previously observed whitespaces. This requires a study of the whitespace characteristics of an
802.11 primary network so that an appropriate statistical model can be developed. Additionally,
efficient access strategies will need to be developed based on the model.
1.7
1.7.1

Case Study: Traffic Protection in CSMA based Sensor Networks
Motivation and Applications
Applications utilizing Wireless Sensor Networks (WSNs) can be categorized into real time

and non real time. Real time applications such as event surveillance typically require constant or
variable rate data streams to be sent over the sensor network. Depending on the specific
application, the rate of this data can be of the order of a few packets per second to tens of packets
per second. The objective is to transport such real-time application data packets with minimum
delivery delay and packet losses. Non-real time applications such as environmental monitoring
need to measure levels of temperature, moisture and other ambient parameters in a time driven or
13

event driven manner. Such applications typically require lower data rates compared to the realtime data streams and have less stringent packet delay and loss requirements.
Applications utilizing WSNs can be categorized into two classes, namely, real time
applications and non real time applications. Real time applications typically require some
constant rate of data to be sent over the sensor network. Depending on the application, the
generation rate of this data can be on the order a few packets per second or hundreds of packets
per second. For example, an environmental monitoring WSN may measure the temperature or
moisture level every few seconds. Since the temperature and moisture levels do not change with
large time scales, sending a few packets per second is a sufficient data rate for real time
measurement of the required information. In battlefield monitoring, target surveillance, and
audio or video applications data rate requirements may be higher.
Non-real time applications may be also supported by a WSN. Event driven monitoring are
typical non real time applications were sensor nodes react when an event occurs within the
region that is currently being monitored. Some examples of no-real time applications are
structural integrity monitoring, home automation, and animal monitoring. For example, consider
a sensor node that is triggered to start sending data when animal movement is detected or when
the noise level of an environment reaches a certain threshold. Once the event occurs, the sensor
nodes will send data about the event to the collection node or sink.
1.7.2

Problem Definition
A heterogeneous WSN may often require supporting both real time and non real time

applications. In this type of networks, nodes that support real time applications must coexist with
the nodes that are running non-real time application. In order to protect the real-time traffic from

14

large channel access delay caused due to interruptions from the non real-time traffic, an access
priority structure is usually needed.

15

Chapter 2: Dissertation Scope and Contributions
2.1

Introduction
The principal goal of a secondary user utilizing bandwidth scavenging is to access

spectrum resources efficiently while avoiding disruptions to the primary user traffic. Since the
SUs and PUs are completely independent of each other, primary traffic patterns are nondeterministic. This means that an SU is not able to deterministically predict when the spectrum
will be free. As a result, an SU will need an access strategy that can adjust to the dynamics of the
primary user traffics. A reasonable solution is for the SU to access a given idle times or
whitespace based on a previously observed statistical whitespace profile. Once an SU identifies a
whitespace, it sends a packet in that whitespace only if the estimated chance of completing the
transmission before the whitespace will have ended is reasonably high. This opportunistic access
during ultra-short and non-deterministic whitespaces is termed as Bandwidth Scavenging by the
secondary users. In other words, the SUs scavenge the channel capacity left over by the PUs.

16

SU
Application

Secondary User

Channel
Measurement
& Modeling

Access
Strategy
Parameter
Computation

Access
Strategy
Parameters

Sensing

SU
Access
Module
Data

Sensing Packets

Wireless Channel
PU1

…

PU2

PUn

Primary User Network
Figure 2-1: Architectural components of Bandwidth Scavenging
2.2

Bandwidth Scavenging Components
Figure 2-1 shows a functional diagram of the architectural components in the Bandwidth

Scavenging Concept (BSC). The BSC operates at the Physical (PHY) and Medium Access
Control (MAC) layers of the network stack. The SU continuously senses the shared channel
using the Channel Measurement and Modeling module. This module creates a statistical model
of PU traffic characteristic. Using the resulting model, the Access Strategy Parameter
Computation module executes the access strategy algorithm and generates access parameters.
The access parameters are then fed into the SU Access Module, which, in addition to channel
sensing, transmits SU user packets into the channel using a designated process.

17

Coexistence in 802.11
Networks
NS2 Implementation

Simulation based
RSSI Measurement
and Analysis

Access Strategy
Simulation
Evaluation

Device Coexistence using
Bandwidth Scavenging
Access
Strategy
Algorithm

Traffic Protection in CSMA
Networks
Sensor Network
Testbed
Implementation

Algorithm
Framework

Experimental
RSSI
Measurements

Algorithm
Analysis

Access strategy
Prototype and
Performance
Evaluation

Generalization of Bandwidth Scavenging for Wireless Networks
Dynamic Access Strategy

Multiple Secondary Users Support

Optimal Solution

MAC Protocol for SUs

Sub-Optimal Solution

Delay Analysis

Figure 2-2: Application of the Bandwidth Scavenging Concept. For interpretation of the
reference to color in this and all other figures, the reader is referred to the electronic version of
this dissertation.
2.3

Scope
The Bandwidth Scavenging Concept (BSC) has been applied as a solution to device

coexistence in wireless networks. Figure 2-2 shows a visual summary of the application of BSC.
The concept has been applied to support prioritized device coexistence in 802.11 networks,
traffic protection in CSMA networks and lastly general dynamic spectrum access networks.

18

2.4
2.4.1

Contributions
Coexistence in 802.11 Networks
The BSC is applied to support device coexistence in 802.11 networks. Three key

contributions are made in this case study. First, an extensive ns2 based simulation is performed.
In this simulation, a 98 PU and SU node network is developed to investigate the interaction
between PUs and SUs. Additionally, a number of PU topologies, traffic patterns, and data rates
are implemented to measure the whitespace characteristics of a primary network. Second, an
access strategy is developed that facilitates efficient SU channel access in an 802.11 based
primary user network. Lastly, an extensive simulation and theoretical evaluation is provided
using whitespace traces from the simulation.
2.4.2

Traffic Protection in CSMA based Sensor Networks
The BSC is also utilized to facilitate coexistence between real time nodes and non-real

time nodes in a sensor network. Two key contributions are made in this case study. First, an
TinyOS based testbed of a priority sensor node network is implemented. The testbed is able to
measure the RSSI of a Zigbee channel and report this information. Second, SU access strategies
are implemented and evaluated on the testbed. The testbed is able to preliminarily validate BSC
on real hardware.
2.4.3

Generalization of Bandwidth Scavenging Concept for Wireless Networks
In this work, the BSC is generalized to support dynamic spectrum access in wireless

networks. The BSC is specifically extended to support a generalized access strategy for any type
of primary network. Additionally, the protocol is extended to support multiple secondary users.

19

Chapter 3: 802.11 Whitespace Analysis
3.1

Introduction
Discovery of appropriate transmission opportunities requires the secondary users to first

measure and then model the occurrences and durations of whitespace, resulting from the primary
users’ transmissions. It follows that the whitespace patterns are a function of the physical
locations, topology, traffic profile, and routing protocols used by the primary users. This chapter
aims to measure, model and analyze the whitespace proprieties in an Ad-hoc 802.11 network.
The chapter is organized as follows. First a description of the simulation setup used for analysis
is given. Then an analysis of an Ad-hoc 802.11 network channel characteristics is presented.
Finally, a whitespace model is developed and is extensively characterized.
3.1.1

Motivation
Fundamentally, the secondary users’ goal is to access the idle times between the

transmissions of the primary users. In the context of packet based networks, these idle times or
whitespaces typically last for milliseconds in between the active data transmissions by the PUs.
Therefore, the secondary user needs to first analyze and model the whitespace characteristics. A
model of the whitespace will provide a framework that will allow the secondary user to develop
access strategies that can adapt to the changing characteristics of the channel.
3.1.2

Problem Definition
With 802.11 primary traffic, the whitespaces typically last for milliseconds in between the

active data transmissions by the PUs. This is in contrast to long durations of whitespaces
experienced in TV bands [46]. In that context, the primary challenge for an SU is to discover
20

unoccupied channels within the frequency domain. When an SU attempts to access TV bands,
disruption to the PUs manifests in the form of ‘Co-Channel’ and ‘Adjacent Channel’ interference
[11, 46]. In contrast, in accessing packet-based networks such as 802.11, the disruption is
defined as when an SU transmits during a whitespace, and the time needed to successfully
complete the SU transmission is more than the current whitespace duration purely from the PU
traffic standpoint. Therefore, the first step is analyzing the whitespace characteristics from a
temporal perspective. PU traffics can exhibit many different properties depending on the
topology, traffic pattern, and data rates. These different properties require an SU to define a
proper model of the whitespace that can adjust to the different characteristics. This chapter aims
to present that model.
3.1.3

Related work
Networks with primary users running 802.11 MAC protocol have recently been

investigated for possible access by secondary users. The authors in [16, 17] develop a
methodology for formally analyzing the whitespace available within 802.11 primary traffic in
infrastructure mode. The key idea is to model the whitespace as a semi-Markov process that
relies on the underlying 802.11 state model involving DIFS, SIFS, DATA, and ACK
transactions. The model describes the whitespace profile in terms of holding times of the idle and
busy states of the channel.
The traffic profiles of packet based primary users studied in literature have varied greatly
[12-15]. The authors in [15] model the primary users’ packet arrival intervals and packet
durations as two separate Poisson processes. Similarly, [13] uses an alternating ON-OFF pattern,
where the duration of the idle and busy states follows an exponential distribution with means v
and l respectively. In [12], a Markov chain with known transition probabilities is used to model
21

primary user behavior. Lastly in [14], a Markov model with Exponential, Uniform, Weibull and
Generalized Pareto distributions are considered for PUs arrival times. In line with these efforts in
the literature, the proposed SU access mechanism in this thesis is evaluated for a variety of
higher layer PU traffic profiles. However, in contrast to these works, we investigate a
measurement-and–model based approach for modeling PUs’ traffic. This approach captures the
impacts of network dynamics, which may not be captured in the model-only (e.g. 2-state Markov
[12, 14]) approaches in the literature. Other analysis of PU networks have been investigated [4754].
3.2

Simulation Setup
The goal of the secondary users is the characterization of the whitespace left over from

primary users transmissions. A method for measuring the available whitespace is through
detecting the received signal strength (RSSI) at a given channel frequency [16]. We implemented
an 802.11 RSSI measurement scheme in the network simulator ns-2 in order to capture the
durations of whitespaces between PUs’ transmissions.
3.2.1

NS2 Implementation
Ns-2 is an event-driven C++ and TCL language based simulator. It is open source and can

be extended to implement many wired and wireless network protocols. Ns2 is used to provide a
framework for the PU and SU topology, 802.11 traffic profiles, and the capturing of whitespace
durations.
3.2.1.1 Primary and Secondary Users Topologies
In order to control all the characteristics of the PUs and SUs network a grid topology is
used. Figure 3-1 shows topology of the PUs and SUs. There are 98 wireless nodes within an area
22

of 2000 x 2000 meter grid. The nodes designations are split evenly between the secondary users
and primary users. A single PU and a single SU occupy a cell in the grid. PUs are numbered
from 0 to 48 while the SUs are numbered from 49 to 97. PUs are only able to communicate with
other PUs, similarly SUs can only communicate with other SUs.
42, 91

43, 92

44, 93

45, 94

46, 95

47, 96

48, 97

35, 84

36, 85

37, 86

38, 87

39, 88

40, 89

41, 90

28, 77

29, 78

30, 79

31, 80

21, 70

22, 71

23, 72

24, 73 Secondary User node
25, 74
26, 75
IDs are from 49 to 97

27, 76

14, 63

15, 64

16, 65

17, 66

18, 67

19, 68

20, 69

7, 56

8, 57

9, 58

10, 59

11, 60

12, 61

13, 62

0, 49

1, 50

2, 51

3, 52

4, 53

5, 54

6, 55

Primary User node IDs
32, 81
34, 83
are from 0 to33, 82
48

Figure 3-1: Primary and Secondary Users grid topology
42, 91

43, 92

44, 93

45, 94

46, 95

47, 96

48, 97

35, 84

CS Range
36, 85
nodes

37, 86

38, 87

39, 88

40, 89

41, 90

28, 77

29, 78

30, 79

31, 80

32, 81

33, 82

34, 83

21, 70

22, 71

23, 72

24, 73

25, 74

26, 75

27, 76

14, 63

15, 64

16, 65

17, 66

18, 67

19, 68

20, 69

Recv Range
7, 56 nodes 57
8,

9, 58

10, 59

11, 60

12, 61

13, 62

0, 49

2, 51

3, 52

4, 53
Tx node

5, 54

6, 55

1, 50

Figure 3-2: Primary and Secondary Users interference ranges
23

Transmission ranges of the PUs and SUs are set to 250 meters and the carrier sense range
is set to 550 meters. The nodes are located within the grid to control the neighbors of each node
in the network. Figure 3-2 shows an example interference range of nodes 24 (or 73) within the
grid. All nodes, PUs and SUs, share a single channel. Nodes are only allowed to communicate
with their adjacent neighbors in the orthogonal directions (i.e. nodes 17, 23, 25, 31). All other
nodes within the nodes two-hop neighborhood are within the carrier sense range and can only
interfere with the nodes transmissions as shown in Figure 3-2.
Routing for all messages is implemented on the orthogonal paths between the nodes. Paths
can be bi-directional in any orthogonal direction. PUs and SUs routing paths can be completely
independent of each other. Various applications can be sent over both the PUs and SUs nodes.
3.2.1.2 Channel Measurement
Each secondary user has the ability to measure the Received Signal Strength Indicator
(RSSI) of the channel within its Carrier Sense (CS) range. We assume there is interference on a
node if an SU or PU is transmitting within the CS range of the node. The CS range is chosen as
opposed to the reception range, since the SUs within up to the CS range of an PU are affected by
the PU’s signals and vice-versa.
The RSSI measurements are captured within ns-2 in the wireless-phy.cc file. In the file, the
signal strength of all messages that a node receives is reported. If the signal strength is above the
CS range threshold, the value and duration of the signal is output to a file. The RSSI threshold
used in the simulation to determine whitespaces is the RSSI value of -108 dBm, which
corresponds to the Carrier Sense (CS) range defined by the 802.11 standard.

24

3.2.2

Sample Experiments with PUs and SUs topology
In order to validate the simulation setup, few sample topologies were created and

whitespace measurements were captured. The traffic flow was set to UDP traffic with a 200 byte
packet sizes. The packet arrival interval time for the PU flows was set to 0.07 ms. Whitespace
measurements are captured from secondary nodes in the network.
42, 91

43, 92

44, 93

45, 94

46, 95

47, 96

48, 97

35, 84

36, 85

37, 86

38, 87

39, 88

40, 89

41, 90

28, 77

29, 78

30, 79

31, 80

32, 81

33, 82

34, 83

21, 70

22, 71

23, 72

24, 73

25, 74

26, 75

27, 76

14, 63

15, 64

16, 65

17, 66

18, 67

19, 68

20, 69

7, 56

8, 57

9, 58

10, 59

11, 60

12, 61

13, 62

0, 49

1, 50

2, 51

3, 52

4, 53

5, 54

6, 55

Figure 3-3: Primary and Secondary Users interference ranges for three 2-hop PU horizontal
connections
Figure 3-4 shows the whitespace measurements from the experiment. In the figure, a 1
value on the y-axis represents the noise floor where there is no signal at that measurement time
and a 3 represents a signal on the channel. Observe that the whitespace is captured in between
PU transmissions over time by both of the nodes.
Figure 3-5 shows the average whitespace available to all the secondary users in the
network. Observe that the SUs close to the PU traffic have reduced available whitespace since
the channel is occupied by the PU. With the current PU traffic load, approximately half of the
25

channel is being utilized. This means that 50% of the whitespace is left over in the channel.
Figure 3-6 and Figure 3-7 shows the average whitespace available to SUs when there are five 4hop horizontal connections. Horizontal connections are referred to as connections from the far
left nodes in the grid to the right nodes in each row as indicated by the block in the figure.
Observe that the larger the amounts of PU connections are affecting a higher amount of SU
nodes. SUs that are near the outside PUs nodes transmissions gradually gain more whitespace
since they are being affected by a lower amount of PUs. Similarly, Figure 3-8 and Figure 3-9
shows the whitespace for connections spanning the entire PU network. In this case, all SUs are
affected with the exception of the SUs near the right edge. Finally, Figure 3-10 and Figure 3-11
shows the connections for two 6-hop PU horizontal connections. Here only the SUs within the
transmission ranges of the PUs are affected by the PUs. The presented figures validate the
experimental setup ability to capture the whitespace of PUs during the simulation. In addition,
the figures show that the SUs proximity to PUs affects the whitespace that is available for usage.

26

Node 49

Node 57
3.5

RSSI (1 = off, 3 = on)

RSSI (1 = off, 3 = on)

3.5
3
2.5
2
1.5
1
0.5
0

3
2.5
2
1.5
1
0.5
0

0.04 0.46 0.88 1.30 1.72 2.14 2.56 2.98

0.01 0.52 0.94 1.40 1.89 2.32 2.77

Time (ms)

Time (ms)

Figure 3-4: Secondary Users whitespace measurements for three 2-hop PU horizontal
connections

Figure 3-5: Secondary Users whitespace measurements over entire network for three 2-hop PU
horizontal connections
27

42, 91

43, 92

44, 93

45, 94

46, 95

47, 96

48, 97

35, 84

36, 85

37, 86

38, 87

39, 88

40, 89

41, 90

28, 77

29, 78

30, 79

31, 80

32, 81

33, 82

34, 83

21, 70

22, 71

23, 72

24, 73

25, 74

26, 75

27, 76

14, 63

15, 64

16, 65

17, 66

18, 67

19, 68

20, 69

7, 56

8, 57

9, 58

10, 59

11, 60

12, 61

13, 62

0, 49

1, 50

2, 51

3, 52

4, 53

5, 54

6, 55

Figure 3-6: Primary and Secondary Users interference ranges for five 4-hop PU horizontal
connections

Figure 3-7: Secondary Users whitespace measurements over entire network for five 4-hop PU
horizontal connections
28

42, 91

43, 92

44, 93

45, 94

46, 95

47, 96

48, 97

35, 84

36, 85

37, 86

38, 87

39, 88

40, 89

41, 90

28, 77

29, 78

30, 79

31, 80

32, 81

33, 82

34, 83

21, 70

22, 71

23, 72

24, 73

25, 74

26, 75

27, 76

14, 63

15, 64

16, 65

17, 66

18, 67

19, 68

20, 69

7, 56

8, 57

9, 58

10, 59

11, 60

12, 61

13, 62

0, 49

1, 50

2, 51

3, 52

4, 53

5, 54

6, 55

Figure 3-8: Primary and Secondary Users interference ranges for seven 6-hop PU horizontal
connections

Figure 3-9: Secondary Users whitespace measurements over entire network for seven 6-hop PU
horizontal connections
29

42, 91

43, 92

44, 93

45, 94

46, 95

47, 96

48, 97

35, 84

36, 85

37, 86

38, 87

39, 88

40, 89

41, 90

28, 77

29, 78

30, 79

31, 80

32, 81

33, 82

34, 83

21, 70

22, 71

23, 72

24, 73

25, 74

26, 75

27, 76

14, 63

15, 64

16, 65

17, 66

18, 67

19, 68

20, 69

7, 56

8, 57

9, 58

10, 59

11, 60

12, 61

13, 62

0, 49

1, 50

2, 51

3, 52

4, 53

5, 54

6, 55

Figure 3-10: Primary and Secondary Users interference ranges for two 5-hop PU horizontal
connections

Figure 3-11: Secondary Users whitespace measurements over entire network for two 5-hop PU
horizontal connections
30

3.3

Ad-hoc Primary User Network
In the previous section, ns-2 based PU and SU networks simulation were established. The

next step is to apply some specific 802.11 based network topologies and traffic characteristics in
order to capture the whitespace characteristics of an Ad-hoc 802.11 channel.
3.3.1

PU and SU Topologies
Using the PU and SU network shown in Figure 3-1, three scenarios are established. First, a

linear chain PU topology with a single bi-directional multi-hop primary flow is created. Figure
3-12 shows the topology and traffic direction for the chain topology. This scenario is referred to
as TOP1. Note that the arrows show the direction of the traffic. In the second scenario (TOP2)
shown in Figure 3-13, there are two multi-hop bi-directional flows in the network on two
intersecting linear chains. In the last scenario (TOP3) shown in Figure 3-14, there are two multihop flows in a parallel chain network. In each case, the secondary users are within the CS range
of all primary users.

PU1

PU2

PU3

PU4

SU1

PU5

SU2

Figure 3-12: PUs and SUs topology and traffic flow for linear chain. (TOP1)

31

PU6
SU2
PU7
SU1
PU1

PU2

PU3

PU4

PU5

PU8
PU9
Figure 3-13: PUs and SUs topology and traffic flow for intersecting chain. (TOP2)

PU1

PU2

PU3
SU1

PU6

PU7

PU4

PU5

SU2

PU8

PU9

PU10

Figure 3-14: PUs and SUs topology and traffic flow for parallel chain. (TOP3)

3.3.2

Traffic Profiles
The traffic profile and data rate of the primary user directly affect the total amount and

statistical properties of the available whitespace as viewed by the secondary users. We model
three types of PU traffic, namely, bidirectional UDP with a fixed packet size, TCP, and a Video
32

Stream. For the UDP traffic, the packet arrival intervals are modeled as: 1) Uniformly distributed
with a mean of α ms and a variability of ±v ms around the mean, and 2) a Poisson process with
mean packet arrival interval of α ms. The packet arrival intervals for each Uniform and Poisson
flow is changed in order to represent varying intensity of the primary traffic.
3.3.3

Channel Characteristics
In this section, the channel characteristics of the previously mentioned traffic profiles are

analyzed. The channel measurements of SU1 is shown in Figure 3-15 for UDP traffic sent for 0.5
inter-packet time (IPT) over 10 seconds. Figure 3-16 shows measurements with the IPT
decreased to 0.1, which is twice as fast as the previous case. Notice that there are a larger number
of transmissions with the faster PU IPT. Figure 3-17 and Figure 3-18 shows a zoomed in view of
the measurements. Observe that each multi-hop primary user transmission is captured by the SU.
With these measurements, each whitespace can be identified by the secondary user.

33

Figure 3-15: Channel measurements for 0.5 inter-packet time over 10 seconds

Figure 3-16: Channel measurements for 0.1 inter-packet time over 10 seconds

34

Figure 3-17: Channel measurements for 0.1 inter-packet time over 0.3 seconds

Figure 3-18: Channel measurements for 0.1 inter-packet time over 0.05 seconds

35

Control
Whitespace

Figure 3-19: Channel measurements for 0.1 inter-packet time
Figure 3-19 shows the RSSI trace, as observed by SU1, as a result of a single packet flow
from PU1 to PU5. Observe that since SU1 is physically closest to the primary user PU3, the
transmissions from that primary user appears with highest signal strength compared to the other
PUs whose signal strength monotonically reduces with increasing distances. Also, since the PUs
in this experiment use 802.11 as the MAC protocol, each packet transmission creates an RTSCTS-DATA-ACK cycle, resulting in white spaces of multiple durations representing different
802.11 inter frame spacing (i.e. SIFS, DIFS, etc.).
Three types of whitespaces can be observed for this example 802.11 primary traffic. As
pointed in Figure 3-19, the first is the control whitespace caused by the short inter frame spacing
(SIFS) periods between the RTS, CTS, DATA, and ACK transmissions. The second whitespace
36

type, termed as forwarding whitespace, represents the duration between an PU receiving data
and forwarding data. The last whitespace type is the inter-packet whitespace, representing the
inter packet duration which can vary greatly because it is based on the rate of packet
transmission by the PUs.
3.4

Whitespace Characteristics
In the previous section, the channel characteristics were shown for the presented

topologies. The next step is to define the whitespace model. Once the model for a measured
whitespace profile is computed, the SUs rely solely on this function for channel access without
having to worry about the individual PU properties such as topology, traffic pattern, and data
rates.
3.4.1

Whitespace Model
An SU periodically senses the channel to detect the whitespaces. Based on the channel

sampling report, the channel state is interpreted using an ON-OFF (busy-idle) model as shown in
Figure 3-20. The sensing interval of the channel is represented by Tp and the sensing time is
represented by Ts. Figure 3-21 shows an example of the periodic sensing process. We assume
that the sensing time duration is negligible. After each measurement, a busy or idle report is
provided. Consecutive idle measurements are counted (n) and multiplied by Tp to determine the
duration of a whitespace. We model the duration of whitespace using its probability density
function, w(n), which is the probability of a whitespace being nTp time long [55]. This w(n) is
computed periodically to capture any changes in the primary user behavior. Once the w(n) for a
37

measured whitespace is computed, the SUs rely solely on this function for channel access
without having to worry about the individual PU properties such as topology, traffic pattern, and
data rates.
Note that the proposed SU access mechanism in this thesis assumes a certain degree of
stationarity of the distribution w(n) [12, 14, 15, 56]. Meaning, the PU network topology and
traffic behavior remains stationary for long enough duration so that the SUs can make use of
w(n) to make statistically meaningful access decisions.

ON

OFF

Figure 3-20: On-Off whitespace model
Sensing Time, Ts

Sensing Period, Tp

ON
OFF
0

1

0

1

1

0

0

Figure 3-21: Sensing period
3.4.2

Whitespace Statistics
Utilizing the whitespace pdf w(n), varying topologies, traffic rates, and traffic distributions

are shown in this section. For these whitespace measurements, the secondary user’s channel
sensing interval Tp is kept smaller than the shortest duration of possible whitespace in the
channel. Considering that for 802.11 [57] the shortest duration whitespace is the SIFS period at
10 µs, the value of Tp was set to 5µs to capture all possible whitespaces. Note that in the legend

38

used in most of the figures, a convention [topology, traffic profile - packet arrival interval (in
ms)] is used for describing the corresponding scenarios. For example, [TOP1, U90] would
represent a whitespace profile obtained when a packet stream with uniformly distributed interpacket interval (of 90 ms) is sent over a network with TOP1.
3.4.2.1 Uniform and Poisson Distributions
Figure 3-22 shows the w(n) for UDP traffic with 90 ms average inter-packet duration that
is uniformly distributed across ±20% around the average packet arrival interval for each
topology. Observe that the detected whitespace durations from the measured RSSI trace vary a
great deal. For example, in the multi-hop TOP2 case, although all the traffic is generated at about
90 ms intervals, the whitespace durations are distributed over the range from 10 µs to 300 ms.
This variation results from the dynamic nature of the bidirectional multi-hop primary traffic.
Figure 3-23 shows the whitespace for primary traffic with arrival times as an Exponential
distribution (i.e. a Poisson process based packet arrival) with 90 ms inter-packet arrival for each
topology. Similar to the Uniform case the whitespace distribution varies across a wide range of
durations.
Observe that in the pdf figures there is a peak at the on the first millisecond. To highlight
this initial peak in the pdf, Figure 3-24 and Figure 3-25 shows the cumulative density function
(cdf) of the Uniform and Poisson experiments on the interval from 0 ms 1 ms. From the figures,
it is interesting to note that regardless of the traffic profile or topology, the vast majority of the
whitespaces (around 95%) last for less than 1 ms.

39

10-0

Probability

10-1
10-2
10-3
10-4
10-5
Whitespace Duration (nTpms)
Figure 3-22: Impacts of Uniform whitespace pdf distribution

10-0

Probability

10-1
10-2
10-3
10-4
10-5
Whitespace Duration (nTpms)
Figure 3-23: Impacts of Poisson whitespace pdf distribution
40

Whitespace Duration (nTpms)
Figure 3-24: Uniform whitespace cdf distribution within the interval 0 ms to 1 ms

Whitespace Duration (nTpms)
Figure 3-25: Poisson whitespace cdf distribution within the interval 0 ms to 1 msTCP and Video
Distributions
Figure 3-26 and Figure 3-27 present the w(n) for TCP and Video Streams. In the TCP
traffic case, over all the topologies, the whitespace durations are very small. All of the
41

whitespace durations are less than 20 ms long. This would indicate very little opportunity for
secondary user access to the spectrum under TCP traffic. This observation is in line with what
was reported in studies of the impacts of the TCP protocol in DSA networks [58]. The video
stream case shows that there is a larger amount of whitespace available compared to the TCP
case. Figure 3-28 and Figure 3-29 shows cdf of the first millisecond for the TCP and video
stream cases. Observe again that the whitespace properties are similar to the Uniform and
Poisson cases. Where upwards of 95% of the whitespace duration is below 1 ms.
3.4.2.2 Impacts of Data Rates
The impacts of data rates of the primary user traffic on the whitespace distribution are
shown in Figure 3-30 and Figure 3-31. The figures present the packet arrival intervals of 60 ms
and 90 ms for TOP3 with intersecting multi-hop flows, for both Uniform and Poisson
distributions. The figures show that depending on the specific average inter-packet time of the
primary traffic, the whitespace durations will be distributed over a larger range. It follows that a
faster PU data rate will reduce the amount of whitespace available to the SU. Again, observe that
in Figure 3-32, the whitespace trends for all cases from 0 ms to 1 ms have properties that are
similar to the previous figures.

42

10-0

Probability

10-1
10-2
10-3
10-4
10-5
Whitespace Duration (nTpms)
Figure 3-26: Impacts of TCP whitespace pdf distribution

10-0
10-1

Probability

10-2
10-3
10-4
10-5
Whitespace Duration (nTpms)
Figure 3-27: Impacts of video stream whitespace pdf distribution

43

Whitespace Duration (nTpms)
Figure 3-28: TCP whitespace cdf distribution within the interval 0 ms to 1 ms

Whitespace Duration (nTpms)
Figure 3-29: Video Stream whitespace cdf within the interval 0 ms to 1 ms

44

10-0

Probability

10-2

10-4

10-6
Whitespace Duration (nTpms)
Figure 3-30: Impacts of data rate for Uniform pdf distribution

10-0

Probability

10-2

10-4

10-6
Whitespace Duration (nTpms)
Figure 3-31: Impacts of data rate for Poisson pdf distribution

45

Whitespace Duration (nTpms)
Figure 3-32: Impacts of data rate for Uniform and Poisson cdfs within the interval 0 ms to 1 ms

10-0

Probability

10-1
10-2
10-3
10-4
10-5
Whitespace Duration (nTpms)
Figure 3-33: Impacts of sensing period on Uniform whitespace pdf distribution

46

3.4.2.3 Impacts of Finite Sensing Rate
In Figure 3-33 and Figure 3-34, the whitespace pdf w(n) is shown with the value of sensing
period Tp increased to 50 µs, which is 5 times the duration of the smallest possible whitespace
duration for 802.11 primary traffic. This larger sensing period causes some change to the pdf of
the whitespace durations. The reason for this change in probability is that the whitespaces with
smaller duration than the value of Tp may not be detected. This causes a redistribution
(increasing) of the probabilities to whitespace durations greater than Tp. However, as shown in
Figure 3-35, although there is a redistribution of whitespace probabilities, the first millisecond
still shows that a large portion of the whitespaces are still shorter than 1 ms, which was observed
for the faster sensing cases (i.e. Tp set to 5 µs in Figure 3-32).

10-0

Probability

10-1
10-2
10-3
10-4
10-5
Whitespace Duration (nTpms)
Figure 3-34: Impacts of sensing period on Poisson whitespace pdf distribution
47

Whitespace Duration (nTpms)
Figure 3-35: Impacts of sensing rate for Uniform and Poisson cdfs within the interval 0 ms
to 1 ms

Available Capacity

0.48
0.46
0.44
0.42

TOP1

0.4

TOP2
TOP3

0.38
0.36
130

120

110

100

90

80

70

60

50

40

Inter-Packet Arrival Time (ms)
Figure 3-36: Available channel capacity in terms of fraction of whitespace for the Uniform
traffic profile with various inter-packet arrival times and topologiesAvailable Whitespace
Capacity
48

The whitespace distributions shown in Section 3.4.2 exhibits very dynamic characteristics
depending on the traffic profile and data rate. A question that may be raised is: Is there any
capacity available to the SUs to exploit in this type of whitespace? In this section, the available
whitespace capacity is investigated. Figure 3-36 shows the available channel capacity in terms of
fraction of whitespace for the Uniform traffic profile with various inter-packet arrival times
(IPT) and topologies. For the highest IPT case (slowest sending rate), the capacity of the channel
is approximately 0.47. This means that the channel is not being utilized nearly half the time. For
the lower IPT the capacity decreases to 0.39. It follows that a faster PU sending rate will
decrease the available channel capacity for the SUs. Observe that there is little difference
between the different topologies. This result occurs because the offered load is similar among all
of the topologies.
Figure 3-37 shows the available channel capacity in terms of fraction of whitespace for the
Poisson traffic profile. The results for this case are similar to the Uniform experiments. Figure
3-38 shows the available channel capacity for the TCP and Video Stream traffic profiles. The
TCP profile has very little available whitespace at approximately 10% for the different
topologies. This was also shown in the pdf of the TCP traffic shown in Figure 3-26. For the
Video Stream profile, there is a larger capacity for the TOP1 than the other two topologies.

49

Available Capacity

0.48
0.46
0.44
0.42

TOP1

0.4

TOP2
TOP3

0.38
0.36
130

120

110

100

90

80

70

60

50

40

Inter-Packet Arrival Time (ms)
Figure 3-37: Available channel capacity in terms of fraction of whitespace for the Poisson traffic
profile with various inter-packet arrival times and topologies
0.4

Available Capacity

0.35
0.3
0.25
0.2

TCP

0.15

Video

0.1
0.05
0
TOP1

TOP2

TOP3

Inter-Packet Arrival Time (ms)
Figure 3-38: Available channel capacity in terms of fraction of whitespace for the TCP and
Video Stream traffic profile

50

3.5

Summary and Conclusions
The distributions presented in Figure 3-22 through Figure 3-38 represent the general

whitespace properties for a wide range of topology and traffic variations that were experimented
using our ns-2 based simulator. A number of key observations can be made from the whitespace
statistics. First, the detected whitespace durations from the measured RSSI trace vary a great deal
across different topologies and traffic patterns. This indicates that a secondary user may
experience vastly different whitespace environments when attempting to access primary
networks. Second, as shown in Figure 3-36 and Figure 3-37, there is available capacity for SUs
to scavenge depending on the traffic profile of the PU.
Finally, as shown in Figure 3-24 and Figure 3-25, regardless of the traffic profile or
topology, the vast majority of the whitespaces (above 90%) last for less than 1 ms. This results
from the large number of small whitespaces created during the RTS-CTS-DATA-ACK cycle for
each packet transmission. The small whitespaces creating the initial peak in the pdfs are the
control and the forwarding whitespaces as defined in Section 3.3.3. For example, in TOP (3)
there are 4 hops for each flow, therefore there will be generally 3 ⋅ 4 + 3 = 15 small whitespaces per
transmission representing 12 control and 3 forwarding whitespaces. While there will be only one
inter-packet whitespace per transmission, resulting in a ratio of 15:1 control and forwarding
whitespace to inter-packet whitespace. Note that this whitespace property is a feature of 802.11
networks, and can be heavily exploited by SUs while accessing this type of networks. The next
chapter presents a secondary user access strategy that exploits this property to maximize SU
throughput and minimize interference to the PUs.

51

Chapter 4: Contiguous SU Transmission Strategy (CSTS)
4.1

Introduction
In this chapter, a secondary user access mechanism, which is termed as Contiguous SU

Transmission Strategy (CSTS) is introduced. This strategy is targeted mainly for whitespace
environments created by Ad-hoc mode 802.11 PU traffic as discussed in Section 3.4.
4.1.1

Motivation
The key challenge in accessing primary networks is that without any prior knowledge of

the starting times and durations of whitespaces, the SUs must access the channel such that the
disruption to the PUs is minimized. Additionally, as a conflicting objective, the throughput of the
SU should be maximized for a given whitespace profile. These objectives should be
accomplished even in the presence of different primary user traffic characteristics.
4.1.2

Problem Definition
From Section 3.4, it is clear that for most 802.11 style primary traffic, due to the inter-

frame spacing at the MAC layer, over 90% of the detected whitespaces are too small to
accommodate the secondary packets. Additionally, as shown in Section 0, there is capacity
available to the SUs in some types of PU profiles. Therefore, in order to facilitate efficient
scavenging of the channel capacity left over by the PUs, a secondary user access strategy that
can exploit these properties to maximize SU throughput and minimize interference to the PUs is
needed.

52

4.1.3

Assumptions
The key assumptions are as follows. First, no cooperation is assumed between the PUs and

SUs. Second, the SUs are assumed to have the ability to differentiate between the PUs’ and the
SUs’ transmissions through physical layer waveform detection, which is feasible through
cyclostationary feature detection [59] or waveform signature detection [60]. Third, an SU is able
to detect collisions with PUs when an PU returns to the channel (i.e. begins transmission) during
a whitespace. Although it is not the focus of this thesis, this capability can be realized by channel
sensing after each packet transmission is completed by an SU. If the SUs’ packet size is smaller
than that of the PUs’, a collision can be inferred by an SU by detecting an PU’s signal after the
SU’s transmission is completed [14]. Finally, for the purpose of protocol analysis, it is assumed
that the primary user traffic pattern does not change appreciably as a result of secondary
transmissions [12, 14, 15, 56].
4.1.4

Related Work
Prioritized device coexistence within CSMA based WLANs can be achieved [18, 19] by

using different inter frame spacing (IFS) periods for different user and/or traffic classes. MAC
protocol 802.11e [18], for instance, uses different Arbitration IFS (AIFS) periods to provide
CSMA based prioritized access among different device and/or traffic classes. When a channel is
found free, a node waits for a specific AIFS periods depending on the device or traffic class,
before it attempts to send a packet. For higher priority primary traffic, a node waits for a smaller
AIFS period. This ensures when multiple nodes contends for the channel, the primary users’
nodes (with smallest AIFS) wins.
While providing reasonable access differentiation, these approaches rely only on the
instantaneous channel status (i.e. free or busy) for granting access. This leads to undesirable
53

disruptions to the PU traffic as follows. Consider a situation in which an SU intends to transmit a
packet and it does so after finding the channel free (i.e. a whitespace) for the AIFS specified for
the SUs. Now, in the middle of this SU’s packet transmission if an PU in the vicinity intends to
send a packet, it needs to wait until the current SU transmission is over. This causes an
undesirable delay for the PU traffic, which in turn will affect the PUs’ application performance.
Since this is mainly a result of the SUs’ reliance only on the instantaneous channel state, a more
robust approach for the SUs would be to also consider the long term whitespace model. The
following DSA approaches attempt to accomplish that.
The authors in [6] develop an WLAN access strategy for secondary users in which the SUs
utilize packet size slots for the channel access. At the beginning of each slot, an SU senses the
channel and if the channel is free then it transmits with a specified probability that is calculated
from previous measurement. The objective is to minimize PU disruption and maximize SU
throughput. The use of time-slotting in their approach requires time synchronization across the
secondary users. In the proposed access mechanism in this thesis, the need for such inter-SU
time synchronization is avoided via asynchronous whitespace access based on a stochastic
whitespace modeling approach.
In [12, 23], the proposed access strategies rely on specific types of PU networks. Protocols
are developed in [12] for a slotted PU network where the SUs are synchronized with the PU’s
slots. In this case, if an PU does not use a slot then the SU can transmit after a small duration of
time in the same slot. The PUs in [23] are assumed to use a specific combination of time and
frequency multiplexing to prevent inter-PU access collisions. In this thesis, we formulate the PU
network more generally by not assuming any specific PU MAC layer such as the slotted ones in
[12]. The target here is to develop an SU access mechanism, which can deal with unslotted as

54

well as stochastic PU MAC layers such as 802.11, CSMA and ALOHA protocols. This
assumption of a general PU access protocol broadens the applicability of the SU access approach
proposed in this thesis.
The authors in [61, 62] do not restrict PUs to a slotted MAC protocol although their access
strategy requires a slotted SU network. During the beginning of each slot the SU senses the
channel and if the channel is free then the SU transmits. A greedy version also uses probabilistic
analysis to decide whether or not to transmit in a given slot. This framework requires time
synchronization across the SUs. In our approach, SUs use a stochastic mechanism and are not
required to maintain time synchronization.
Game theoretic mechanisms have been introduced in [24, 25] for spectrum access for the
SUs. The PUs in this framework cooperates with the SUs through pricing strategies and
subscription fees to facilitate spectrum sharing. These approaches require the PUs to have
knowledge of the SUs presence. In contrast, the proposed mechanism in this thesis, no such
cooperation is assumed, thus no changes in the PUs behavior is needed. The idea here is to
develop an access mechanism for the SUs for scavenging the PUs’ leftover bandwidth, without
the PUs being aware of such scavenging. For example, data-enabled hand-held devices should be
able to scavenge bandwidth in a WLAN without the primary VoIP handsets being aware of the
scavenging process.
4.2

CSTS Access Algorithm Overview
From Section 3.4, it is clear that for most 802.11 style primary traffic, due to the inter-

frame spacing at the MAC layer, over 90% of the detected whitespaces are too small to
accommodate the secondary packets. Since the actual duration of any whitespace is not known apriori to the SUs, if an SU transmits at the beginning of each whitespace then a vast majority of
55

such transmissions will end up in primary traffic disruptions (PTDs). Primary Traffic Disruption
(PTD) is defined as the probability that an SU will cause disruption to PUs as a result of
transmitting at the end of a whitespace. If an SU defers transmission attempts, however for a
suitably chosen wait-threshold duration µ, such PTDs can be potentially reduced without
sacrificing the Effective Secondary User Throughput (EST). In line with this logic, a secondary
user access strategy is developed that utilizes a minimum wait-threshold duration µ that is
dimensioned based on the statistical properties of the Ad-hoc mode 802.11-based primary
networks.

JmaxS

µ
Stage 2

Stage 1

Stage 3

End of whitespace
(Unknown to SU)

Start of
whitespace

Figure 4-1: Different stages of CSTS access
As shown in Figure 4-1, in CSTS access there are three distinct stages. In the first stage, an
SU initially waits for the channel to become idle. Once the channel becomes idle, the SU waits
for an additional wait threshold duration µ before it attempts to transmit its first packet of
duration S. At this point, the access protocol enters into stage-2. At this stage, if the SU has more
packets to send, it continues sending until the measured value of є (time elapsed since the
beginning of the current whitespace) becomes so large that the expected PTD for new
transmissions become equal to or larger than a pre-determined Disruption Bound (DB). Let us
define a term Jmax, which denotes the maximum number of packets that the secondary user can

56

send before the expected disruption reaches the pre-determined DB. The quantity Jmax is
computed by the SU based on the observed whitespace statistics. After Jmax packets are sent, the
secondary user abandons transmission attempts and starts channel sensing for detecting the next
whitespace.
If, however, a whitespace ends before all Jmax packets are sent by the SU, it vacates or
exits the channel after detecting that a primary user has started transmitting. Note that in spite of
the SU vacating the channel, disruption to the PUs’ traffic can still occur since the PU must defer
its transmission until the channel is vacated by the SU. This second stage defines the contiguous
block of time that an SU transmits in any whitespace.
Finally, the third stage of CSTS corresponds to the situation in which a whitespace
duration is greater than the end time of the Jmax

th

packet from the SU. No PTD will be caused in

this stage because the SU will exit the whitespace after Jmax transmissions. The channel access
strategy is summarized in the form of the pseudo code in Algorithm 1 in Figure 4-2.

57

Figure 4-2: CSTS access algorithm
4.3

Functional Overview
Figure 4-3 shows a functional diagram of the architectural components in CSTS. The user

tunable control inputs into the system are secondary user packet size (S), primary disruption
bound (DB), and the sensing interval (Tp), which is a property of the secondary user hardware.
The SU continuously computes the whitespace pdf w(n) with a channel-sensing interval of Tp.
The access parameters are then fed into the access module, which, in addition to channel sensing,
transmits SU user packets into the channel using the process described in Section V.B. The
objective is to satisfy a user specified primary disruption bound DB.

58

Secondary User
S

Tp

SU
Application

DB

Access
Channel
Strategy
W(n)
Measurement
Parameter
& Modeling
Computation
Sensing

µ

SU
Access
Module
Data

Sensing Packets

Wireless Channel
PU1

…

PU2

PUn

Primary User Network
Figure 4-3: Functional diagram of the SU access strategy
4.4
4.4.1

Algorithm Analysis
Primary Traffic Disruption
For a given whitespace profile w(n), the PTD is defined by the probability that the

whitespace duration is in the range between µ and µ + J max
packet size and

γ

⋅ S + ( J max − 1) ⋅ γ

where S is

is the inter-packet spacing for the SUs. Therefore, an SU accessing a

whitespace characterized by pdf w(n) will create primary user disruption (PTD):

µ + J max ⋅S + ( J max −1)⋅γ

∑ w(n)
µ

PTD =
59

(4-1)

4.4.2

Effective Secondary Throughput
EST is defined as the average number of SU packets successfully transmitted per

whitespace, normalized by the number of packets that could have been transmitted per
whitespace for a given w(n). EST represents the efficiency of channel usage by the SU. To model
EST, we first define Sj, the probability of sending exactly j packets in a whitespace. This
corresponds to the event in which the j

th

packet from an SU in a whitespace has disrupted the

PUs, i.e. the whitespace had ended during the transmission of the j

th

packet. The quantity Sj can

be written as:

µ + j ⋅S + ( j −1)⋅γ

Sj =

∑( wγ()n)
µ (
)

(4-2)

+ j −1 ⋅ S +

For a whitespace that lasts between

µ

and µ

+ ( J max − 1) ⋅ (S + γ ) ,

packets sent by an SU is j. For whitespace duration between
and

∞,

the number of

µ + J max ⋅ S + ( J max − 1) ⋅ γ

the number of packets sent by an SU is Jmax, since the secondary user vacates the

whitespace after sending Jmax packets. Therefore, throughput, the expected number of secondary
packets sent for a given whitespace, can be expressed as:

J max −1

Throughput =

∑ j⋅S
j =1

∞

j

+

∑J
j = J max

60

max

⋅Sj

(4-3)

The throughput equation is valid only when the secondary user captures all available
whitespace using perfect sensing (i.e. Tp set to zero). However, when Tp is non-zero, there is a
probability that a whitespace smaller than Tp will not be detected and thus will not be accessed
by the SU. In that case, the probability of missing a whitespace of length w is equal to (Tp -w)/
Tp. We account for this loss of throughput using this equation:
T p 
 S 





Loss =

∑
j =1


T p 

 S −

 j⋅S 
j

T p 

 S





j





(4-4)

Although non-zero Tp values can affect the whitespace access, the impacts are alleviated
by requiring an SU to wait at least Tp duration before accessing any whitespace.
The effective capacity available to the SU can be expressed as the average number of
packets that is sent per whitespace. This quantity can be written as (with µ and γ equal to zero):

∞

Capacity = ∑ j ⋅ S j
j =1

(4-5)

Therefore, the Effective Secondary user Throughput (EST) can be expressed as:

EST =

Throughput − Loss
Capacity

which is computed from Eqns. 3, 4, and 5.
61

(4-6)

4.4.3

Dimensioning Wait-threshold µ
The selection of wait-threshold µ is critical since it affects both PTD and EST as shown in

Equations 4-1 and 4-6. A large

µ

can reduce primary traffic disruption by preventing the SUs

from transmitting during very short whitespaces. A large

µ , on the other hand, can bring down

the EST, since some portions of large whitespaces will be lost due to this conservatively chosen
large wait period. The goal is to pre-dimension the parameter

µ,

based on the measured pdf

function w(n), in order to strike a desirable balance between the PTD and EST. Note that

µ

is

parameterized by Tp and the lowest value of µ is a least one Tp period, since at least one sensing
interval is needed to determine if the channel is free. In CSTS, the wait-threshold period
determines the PTD caused by the first SU packet during a whitespace. For a wait-threshold

tx,

such PTD can be expressed as:

tx +S

f (t x ) = ∑ w(n)
tx

(4-7)

Now, the minimum and maximum values of tx, are zero and 2S, where S is the secondary
packet duration. The maximum wait-threshold is 2S, since waiting for a duration that is greater
than or equal to 2S would mean missing out sufficient whitespace that could have been used by
the secondary user to send at least one packet. Therefore, for a given whitespace with pdf w(n),

62

the optimal wait-threshold

µ

is chosen as the smallest

tx, over the range 0 to 2.S, such that the

quantity f(tx) in Equation 4-7 is minimized.

4.4.4

Jmax Computation
In CSTS, the number of packets the SU is allowed to transmit in a given whitespace is

Jmax. This quantity is determined by finding the maximum j that satisfies the inequality:

∞

1−

(

∑)(w(γn ) ≤ DB
)

µ + J max −1 ⋅ S + + S

(4-8)

The left side of the inequality represents the PTD when the SU sends Jmax packets in a
given whitespace. Therefore, the inequality in Eqn. 8 gives the Jmax that will reduce disruption
to the pre-defined bound DB. By plugging in this value of Jmax in Equation 4-1, we find the
overall PTD caused by the CSTS access strategy.

4.4.5

Impacts of wait threshold µ
As discussed in the preceding sections, larger

µ

values can significantly reduce PTD by

avoiding SUs’ transmissions in short whitespaces. Figure 4-4 presents theoretical results for the
CSTS mechanism when the wait-threshold µ is varied from 0 ms to 2 ms. The results correspond
to topologies TOP1 and TOP2 in Fig. 1 with both uniform and Poisson distributed primary
63

traffic. All traffic flows correspond to a mean of 80 ms inter-packet arrival time. The SU packet
duration S was chosen to be 1.2 ms and the sensing interval Tp is set to 5 µs.
Observe in Figure 4-4 that as

µ

increases, the PTD decreases for all combinations of

network topology, traffic profile, and data rate. For

µ

= 0, representing the case when the SU

transmits immediately at the start of a whitespace, the amount of PTD is very high for all cases.
This is because as shown in Section 3.4, most of the whitespaces (above 90%) are of very short
duration, consisting of control and forwarding whitespaces in 802.11 primary traffic. As

µ

increases, the PTD first decreases abruptly and then relatively moderately, but in a monotonic
manner.
The figure also shows the effects of

µ

on EST. Recall that EST is the efficiency of

whitespace usage by the SU. With µ set to zero, since the SU attempts packet transmission in all
available whitespaces, the throughput is expected to be the maximum. As

µ

increases, the

throughput decreases since the SU waits for a longer duration before transmitting in a given
whitespace. The values of

µ , larger than the threshold found in Equation 4-7, cause the EST to

reduce drastically by making the access strategy overly conservative.

64

Figure 4-4: Evaluation of the impact of µ in CSTS
4.5

Summary and Conclusion
In this chapter, a secondary users access strategy is presented that access an Ad hoc 802.11

PU whitespace profile. The access strategy exploits the properties of the whitespace by avoiding
transmissions during the first millisecond of a whitespace. Once a whitespace is accessed, the
parameter Jmax limits the amount of SU packets sent within that whitespace. Under this process,
the secondary users maximize its own throughput while keeping the primary user disruptions to a
predefined bound.

65

Chapter 5: Performance Evaluation for
Coexistence in 802.11 Network
5.1

Introduction
In this chapter, the performance of the CSTS strategy is evaluated experimentally and

theoretically in terms of the Primary Traffic Disruption (PTD) and the Effective Secondary
Throughput (EST).
5.1.1

Comparison Protocols
We compare CSTS with a sense-and-transmit based protocol, called VX scheme [56], and

a benchmark scheme (Benchmark).
5.1.1.1 VX Scheme
The VX (Virtual-transmit-if-Busy) Scheme has a similar strategy as CSMA where the SU
will sense the channel and transmit when idle [56]. Initially the SU senses the channel, when the
channel becomes idle a packet of length Ls is transmitted. Then, the SU starts a vacation from
the channel of length Vs. After the vacation, if the channel is busy, the SU starts a virtual
transmission stage and then enters into another vacation stage afterwards. Here, a virtual
transmission means that the SU does not actually transmit the packet but waits for a time interval
which is equal to the packet length Ls. After the vacation, the SU starts the channel sensing
again. In contrast to the CSTS scheme, the VX scheme does not use statistical data but relies on
short term sensing for its transmission decisions. Algorithmic analysis for this scheme has been
66

presented in [56].

5.1.1.2 Benchmark
The Benchmark protocol assumes full knowledge of the starting and ending times of every
whitespace. Therefore, it transmits SU packets in every available whitespace and stops when the
next SU transmission will cause disruption to the PUs. The Benchmark protocol can be run only
offline to find the best SU performance. This protocol essentially determines the highest
throughput an SU can expect without causing any disruption to the PUs.

PU2

PU1

PU3

PU4

SU1

PU5

SU2

Multi-hop: TOP1
Traffic Flow Direction

PU6
SU2
SU1
PU1

PU2

PU3

Intersecting
TOP2

PU7
PU4
PU8

PU1

PU2

PU3
SU1

PU5
PU6

PU7

PU8

Parallel
TOP3

PU9

Figure 5-1: Primary user topologies and the traffic flows

67

PU4

PU5

SU2
PU9

PU10

5.1.2

Simulation Implementation
The access strategies are evaluated through simulation. There are many steps involved in

the creation of the simulator to evaluate the protocols as shown in Table 5-1. First, the network
simulator ns-2 was used to create the network topologies described in Section 3.3.1 and shown in
Figure 5-1. The primary users use ad hoc mode 802.11 as the MAC layer, and the secondary
users in Figure 5-1 observes whitespace profiles similar to those shown in Section 3.4.2. These
whitespace profiles including uniform and Poisson distributions, a video stream, and TCP traffics
are defined in the ns-2 TCL simulator file in Step 2. Then in Step 3, using AWK based scripts,
simulations are created to run ns-2 for the various traffic profiles and topologies. With the
combinations of the different topologies, traffic profiles, and data rates there are a large number
of simulations that are needed to run.
Step #

Implementation Step

1

Creation of primary and secondary user grid topology

2

Setting of traffic profiles for secondary users

3

Use AWK scripts to create the required large number of simulations

4

Use the wireless-phy.cc file to capture RSSI measurements

5

Use the TCL scripts to parse RSSI outputs to create whitespace durations

6

Import whitespace durations into Matlab

7

Implement access strategies and various performance characteristics

8

Run experiments
Table 5-1: Simulator implementation steps

Next in Step 4, RSSI measurement traces are collected from the SUs for each traffic profile
through the modification of the wireless-phy.cc file. The duration and starting times of every
68

whitespace are created in Step 5 by parsing the output RSSI file from Step 4. With the
whitespace durations files, the entire primary user transmission behavior is captured. Note that it
is assumed that the primary user traffic profile does not change as a result of the activity of the
secondary users [12, 14, 15, 56]. Afterwards in Step 6, the whitespace durations files are
imported into Matlab. Within Matlab whitespace statistics are gathered such as the whitespace
pdf w(n). Lastly, the various access strategies are implemented and evaluated using Matlab and
the extracted whitespace durations from the ns-2 simulations.
5.2

CSTS Evaluation
In this section, protocol performance is evaluated using multiple variables including PU

loads, SU packet size, and disruption bound. For the presented results, the SU packets are
generated back-to-back, meaning packets are always available to send within an SU’s buffer.
The user-defined disruption bound (DB) is set to 0.05 [6], which indicates that only 5% of the
PUs’ traffic is allowed to experience disruptions from the SUs’ transmissions. The sensing
interval Tp is set to 5 µs. For the VX scheme, Vs, the channel wait time, is set to 0.5 ms and the
Virtual Xmit duration is set to 16 ms, which was experimentally shown to deliver the best
performance for that access protocol.
5.2.1

Primary User Load
Figure 5-2 demonstrates the impacts of varying PU load on the SU throughput and PU

disruption for TOP2 in Figure 5-1. The PU inter-packet arrival duration is varied from 50 ms to
120 ms for simulating different loading conditions. This is represented in the figure as 8 to 20
packets per second (pps). For all the primary packet rates, a ±20% uniformly distributed inter-

69

packet arrival variation was introduced. The SU packet duration was set to 1.2 ms, which is half
the PU packet size.
From Figure 5-2, observe that the CSTS access protocol is able to maintain the required
primary disruptions within the specified Disruption Bound (DB) across a wide range of PU data
rates and topologies. This is because the µ and Jmax values are set according to the observed
whitespace statistics. As the PU data rate increases, the duration of usable whitespace in the
channel reduces. This occurs because the primary user traffic is utilizing a larger portion of the
channel for its transmissions. With increasing PU rate, the number of distinct whitespaces also
increases, since the whitespaces become more fragmented. Fragmentation of whitespaces
happens because the increase in bidirectional traffic prevents the creation of longer duration
whitespaces. In addition, a larger amount of small control and forwarding whitespaces are
created in every multi-hop transmission. Since the SU access strategy is executed on a perwhitespace basis, with the increased number of distinct whitespaces, the SU attempts to execute
its PU protection mechanisms more often. This causes the SUs’ channel access efficiency or EST
to decrease due to increased PU load.

70

Vx Scheme
Benchmark
CSTS

Vx Scheme
MBAS

Figure 5-2: Performance for uniformly distributed PU traffic
For all the experimented topologies, the throughput for MBAS remains close to that of the
Benchmark protocol. Using the full prior knowledge of the whitespace, the Benchmark protocol
is able to evaluate the maximum possible EST with zero PTD. This shows that CSTS is able to
maximize throughput while maintain PTD bounds by decreasing the Jmax as necessary due to
the increasing PU traffic load.
For the VX scheme, disruption to the PU traffic is observed to be higher. Utilizing the
Virtual Xmit wait times, the SU avoids sending during busy times in a channel. However, once
the SU starts accessing a whitespace, it stops sending only if an PU returns during a Vs wait time.
Thus, there remains a high probability of causing disruption during the SU transmissions. The
VX scheme has a relatively lower throughput since it waits for a Vs time after every
71

transmission, which is not the case of CSTS mechanisms. These results suggest that by utilizing
the history of the channel, it is indeed possible to improve the SU throughput while maintaining
a bound of disruption on the primary networks.
Figure 5-3 shows results for the same setup, but with Poisson distributed PU traffic. Note that the
performance trends for this scenario are very similar to those for the Uniform case as shown in
Figure 5-2. The throughput is slightly decreased because the Poisson traffic creates a larger
amount of fragmented whitespaces, which makes CSTS reduce its transmissions to maintain the
required DB. Similar results were also observed for video stream PU traffic for which the pdf
was shown in Figure 3-27.
We have also experimented with the TCP primary traffic. Since TCP opportunistically
utilizes available bandwidth in the channel, the amount of resulting whitespace for TCP traffic is
small. As a result, the achievable SU throughput remains severely restricted when the primary
disruption bounds are set at low values [17].

72

Vx Scheme
Benchmark

CSTS
Vx Scheme

MBAS

Figure 5-3: Performance for Poisson distributed PU traffic
5.2.2

Secondary User Packet Size

Figure 5-4 shows the performance of CSTS under a varying SU packet size from 625 µs to 2 ms,
as done in [14]. The PU rate is set to 10 pps with uniformly distributed traffic on topology TOP1.
Observe that the throughput decreases as the SU packet size increases. In order to maintain the
DB with larger packet sizes, the access strategy must reduce the transmission probabilities since
a larger secondary packet has a higher probability of causing disruption to the primary at the end
of a whitespace. Similar results are shown in Figure 5-5 and Figure 5-6 for topologies TOP2 and
TOP3 respectively. Figure 5-7 shows the performance with the video stream traffic profile.
Observe that the EST is lower in this case since the duration of whitespace is generally smaller
with the video stream profile as shown in Figure 3-27.
73

Additionally the whitespace is fragmented more with the larger amount of traffic.
Figure 5-4 through Figure 5-7 presents both experimentally measured (via ns2 simulation)
and theoretical values (using Equations 4-1 through 4-8), which are marked as EXP and THY
respectively. It is evident that the experimental results match very well with the theoretical

Primary User Interference (PUI)

0.06
0.055
0.05
0.045
0.04
0.035
0.03

Effective Secondary User Throughput (EST)

results.

0.6
0.7
0.8
0.9
1
Secondary User Packet Size (ms)

1
0.95
0.9
0.85
0.8
0.75
0.7
0.65
0.6
0.55
0.5

0.6
0.7
0.8
0.9
1
Secondary User Packet Size (ms)

Figure 5-4: Impacts of the SU packet size on EST and PTD on TOP1

74

0.055
0.05
0.045
0.04
0.035
0.03

0.6
0.7
0.8
0.9
1
Secondary User Packet Size (ms)

Effective Secondary User Throughput (EST)

Primary User Interference (PUI)

0.06

1
0.95
0.9
0.85
0.8
0.75
0.7
0.65
0.6
0.55
0.5 0.6
0.7
0.8
0.9
1
Secondary User Packet Size (ms)

Primary User Interference (PUI)

0.06
0.055
0.05
0.045
0.04
0.035
0.03
0.6
0.7
0.8
0.9
1
Secondary User Packet Size (ms)

Effective Secondary User Throughput (EST)

Figure 5-5: Impacts of the SU packet size on EST and PTD on TOP2
1
0.95
0.9
0.85
0.8
0.75
0.7
0.65
0.6
0.55
0.5

0.6
0.7
0.8
0.9
1
Secondary User Packet Size (ms)

Figure 5-6: Impacts of the SU packet size on EST and PTD on TOP3

75

0.04

0.035

0.03

Effective Secondary User Throughput (EST)

Primary User Interference (PUI)

0.045

0.6 0.7 0.8 0.9
1
Secondary User Packet Size (ms)

1
0.95
0.9
0.85
0.8
0.75
0.7
0.65
0.6
0.55
0.5 0.6 0.7 0.8 0.9
1
Secondary User Packet Size (ms)

Figure 5-7: Impacts of the SU packet size on EST and PTD for Video Stream
5.2.3

Dynamic PU Traffic Patterns
Figure 5-8 reports the protocol performance from an experiment in which the PU traffic

pattern was dynamically changed over time for TOP2 and TOP3. For TOP 2, initially the PU
traffic was generated with 60 ms inter-packet time (IPT) with Uniform profile. The pattern was
changed after 60 seconds when the IPT was increased to 120 ms. Finally, at 100 seconds the
traffic was changed to Poisson distribution with 90 ms IPT. The PUI traffic pattern for TOP 3
was started at 80 ms IPT with Poisson profile, and then after 60 seconds the IPT was changed to
50 ms and finally at 100 seconds it was changed to uniform distribution with 90 ms IPT. The
whitespace w(n) is continually computed by the access protocol after collecting every 1000
whitespace samples.

76

Observe in Figure 5-8 that while the primary user traffic pattern changes over time, the
PTDs resulting from CSTS stay consistently within the vicinity of the pre-set disruption bound of
0.05. Also, note that there is very little PU traffic disruption during the transitions of w(n) as
indicated in both the frames in Figure 5-8. This is primarily because the whitespace pdf for ad
hoc mode 802.11 traffic maintains very similar properties across different traffic patterns, i.e.
above 90% of the whitespaces are below 1 ms. Similar results were shown in Figure 5-9 for
different traffic patterns and topologies.
Note that the SU throughput increases after an PU traffic pattern transition when the
useable whitespace increases due to the increase in inter-packet interval of the PU traffic.
Additionally, the SU dynamically throttles back its transmissions to protect the PUs when the
available whitespace reduces during a transition. These results demonstrate the robustness of the
CSTS access strategy against varying data rates over time. Additionally, the results suggest that
by periodically measuring the channel state the SUs are able to dynamically model the channel
whitespace and adjust their access parameters accordingly to minimize the PU disruptions and to
maximize the SU throughput.

77

Primary Traffic Disruption (PTD)
0.07

0.06

0
20

40 60 80
Time (s)

0.05

0.04

SU Throughput (pps)

PU Traffic Change 2
PU Traffic Change 2

0.03

40 60 80 100 120
Time (s)

0.09

0.05

0.04

100 120

78

350

300

250

100
0

450

200

150

0.03

100
0

CSTS, TOP(1)
CSTS, TOP(3)

20

20

40 60 80
Time (s)

Figure 5-9: Performance with time-varying w(n) for TOP1 and TOP3

PU Traffic Change 2
PU Traffic Change 2

PU Traffic Change
PU Traffic Change 11

400

PU Traffic Change
PU Traffic Change 2 2

0.08

20
450

PU Traffic Change
PU Traffic Change 1 1

0.06

SU Throughput (pps)

0.07

PU Traffic Change 1
PU Traffic Change 1

0.08

PU Traffic Change
PU Traffic Change 22

0

PU Traffic Change 11
PU Traffic Change

Primary Traffic Disruption (PTD)
0.09

200

150
40 60 80 100 120
Time (s)

Figure 5-8: Performance with time-varying w(n) for TOP2 and TOP3
CSTS, TOP(1)
CSTS, TOP(3)

400

350

300

250

100 120

5.2.4

Channel Sensing Rate
The sensing interval Tp can vary across different hardware and can impact the access

protocol performance since it influences the whitespace pdf w(n). In this Section we investigate
such performance impacts over a practical sensing interval range from 5 µs to 1 ms [63, 64] for
various wireless network interfaces. Figure 5-10 shows such impacts for different topologies
with Uniform and Poisson distributed PU traffic and 60 and 70 ms inter-packet arrival times for
the CSTS protocol. Both experimentally measured (via ns-2 simulation) and theoretical values
(using Equations 4-1 through 4-8), marked as EXP and THY, are presented. The experimental
results match well with the theoretical results, indicating the validity of the experimental
observations. Observe that as Tp increases, the EST and PTD remain relatively constant since
with Tp within the range of 5 µs to 1 ms, the SUs can detect most of the whitespaces that are
smaller than the SUs’ packet size, which is 0.6 ms in this case. In the CSTS protocols, an SU
does not access a whitespace until the whitespace lasts for at least the Tp duration. In other
words, with larger Tp, the SUs behave more conservatively, leading to higher EST and lower
PTD. These results indicate that the proposed access strategies can work under varying channel
sensing rates that are well within the practical range for the currently available 802.11 hardware
[63].

79

Figure 5-10: Impacts of channel sensing interval Tp

5.2.5

Disruption Bound
Depending on the specific PU applications, the SU access strategies may need to run with

different disruption bounds for the PUs. Figure 5-11 and Figure 5-12 presents the performance
with varying pre-defined disruption bound or DB values for Uniform and Poisson distributions
respectively. PU topologies TOP1-3 in Figure 5-1 and inter-packet arrival rates of 60 and 90 ms
are evaluated for the CSTS protocol. The left graph in Figure 5-11 validates that the protocol is
able to maintain the PTD within the pre-set DB for a wide range of DB values. The right graph
depicts that by allowing larger DBs it is possible for the SUs to scavenge more bandwidth from
the channel. Another interesting trend in this graph is that the EST rises quite sharply with higher
disruption bounds. A practical implication of this large slope is that if the PUs can tolerate a very
small amount of disruption (e.g. ≈ 5%), reasonably high SU throughput can be achieved. This is
in contrast to the very low SU throughputs when the PUs cannot tolerate any disruptions.

80

Figure 5-11: Impacts of Disruption Bound for the uniformly distributed PUs

Figure 5-12: Impacts of Disruption Bound for the Poisson distributed PUs
81

5.3

Summary and Conclusions
This chapter presented an extensive set of results to evaluate the CSTS access strategy. The

simulation-based experiments provided a method to explore the impacts of different parameters
on CSTS. The primary user load results showed that the protocol is able to maintain high SU
throughput under different load conditions. Results from the SU packet size experiments showed
that a smaller SU packet provides better throughput for the SUs. Varying PU traffic pattern
results showed that CSTS could dynamically adjust to the PU traffic over time. Additionally,
periodic sampling of the whitespace provides enough information to maintain good SU
throughput while maintain PU disruption bounds. The channel sensing interval results showed
that as long as the sensing period is smaller than usable whitespaces, there was enough
information to operate the CSTS protocol. Finally, the experiments on the effects of varying
disruption bound shows that PUs allowing a small disruption can facilitate more bandwidth
scavenging by the SUs.

82

Chapter 6: Traffic Protection in a Prototype
CSMA based Sensor Networks
6.1

Introduction
In this chapter, the Bandwidth Scavenging Concept (BCS) is applied to traffic protection in

CSMA based sensor networks. The BSC is implemented in an experimental testbed using the
Crossbow TelosB sensor motes [65]. The previously presented 802.11-based simulated network
allows explorations of the effects of various traffic and topology conditions while the
experimental setup discussed in this chapter provides insights into real world measurement
issues. The chapter is organized as follows. First, an overview is given of the TelosB platform.
Then the PU network is described in detail. Finally, the SU network is discussed. Software
details about the prototype is added in Appendix A.
6.1.1

Motivation
Applications utilizing Wireless Sensor Networks (WSNs) can be categorized into real time

and non real time. Real time applications such as event surveillance typically require constant or
variable rate data streams to be sent over the sensor network. Depending on the specific
application, the rate of this data can be of the order of a few packets per second to tens of packets
per second. The objective is to transport such real-time application data packets with minimum
delivery delay and packet losses [66]. Non-real time applications such as environmental
monitoring need to measure levels of temperature, moisture and other ambient parameters in a
time driven or event driven manner. Such applications typically require lower data rates

83

compared to the real-time data streams and have less stringent packet delay and loss
requirements.
6.1.2

Problem Definition
A heterogeneous WSN may often require supporting both real time and non real time

applications. In this type of networks, nodes that support real time applications must coexist with
the nodes that are running non-real time application. In order to protect the real-time traffic from
large channel access delay caused due to interruptions from the non real-time traffic, an access
priority structure is usually needed.
Now, if a network designer wants to ensure that the requirements of the real time traffic
network are met in this heterogeneous WSN, a traffic priority structure can be used. In this case,
the real time traffic is given priority access to channel over the non-real time traffic. To address
this type of traffic prioritization problem in general, some issues that would need to be addressed
are:
•

Maintaining a priority structure: Support is needed to facilitate different priority
levels for the various sets of network devices

•

Supporting higher priority requirements: Data requirements of the higher priority
traffic must be met

•

Limiting the effects of delaying the higher priority traffic: When lower priority traffic
has data to send, the impacts on the higher priority traffic must be minimized

•

Supporting different high priority traffic loads: As the high priority traffic changes,
the lower priority traffic should adjust its usage of bandwidth accordingly

•

Maximizing the throughput of the lower priority traffic: Lower priority traffic
throughput need to be maximized given the available channel bandwidth.
84

Channel history-based mechanisms, namely the Bandwidth Scavenging Concept, may
provide a suitable solution due to its ability to dynamically adjust lower priority traffic access to
protect higher priority traffic.
6.1.3

Related Work
In baseline 802.11 access [57] when a message arrives from the upper layer, the MAC

layer accesses the channel with a DIFS (DCF Inter-Frame Spacing) and Contention Window
(CW) duration irrespective of the traffic class. Traffic priority has been introduced in 802.11e
[18], by using class specific (referred to as Access Categories or ACs) DIFS and CW values. The
quantity DIFS in 802.11e is parameterized as arbitration inter-frame space with specified access
classes (AC). Similarly, the CW is redefined with CWmin[AC] and CWmax[AC] allowing for
different priority traffic to be assigned with different overlapping back-off times. The addition of
class-specific AIFS and CW values allow for prioritizations since nodes with smaller AIFS
periods and lower CW ranges will access the channel first and more often. This core 802.11e
prioritization concept has been refined in [19, 67] which propose mechanisms for choosing the
class-specific AIFS and CW values dynamically under different network conditions.
As discussed in Section I.B, although 802.11e (based) protocol introduces priority to
network traffic, there is no guarantee that a higher priority node’s transmission will not be
deferred if it arrives at the MAC layer when a lower priority node is in the middle of its packet
transmission. Our proposed protocol attempts to reduce such high priority traffic deferrals by
allowing low priority nodes to attempt channel access only when it estimates the likelihood of
such disruption is low.

85

A number of papers [21, 22, 68] have proposed traffic prioritization schemes specifically
designed for WSNs. The mechanism in [68] is similar to 802.11e in using variable size
contention windows and different inter-frame spacing for different types of sensing events. To
further differentiate the priority events, the authors improve upon their original design by adding
a period where each higher priority event node randomly sends a burst of pulse just after their
respective AIFS periods to notify lower priority nodes to backoff until the transmission is
finished. While being customized specifically for WSNs, this protocol [11] can suffer from the
same 802.11e issues as discussed in Section I.B.
The works in [22] and [21] use time slots similar to TDMA to assign priority to traffic
classes. The protocol in [21] divides MAC frames into random access and scheduled access
sections. The scheduled access sections are further divided into priority groups with multiple
transmission slots in each group. A node broadcasts their traffic priority during the random
access period, which reserves their transmissions for specified priority groups in the scheduled
section. The protocol PMAC [22] presents another TDMA based prioritization MAC utilizing a
wide range of prefix reservation and inter-slot intervals. Both of these TDMA protocol require
tight time synchronization to implement priority. In contrast, our approach does not require time
synchronization, thus offering a more practical solution that is suitable for ad hoc deployment.
Other dynamic spectrum access testbeds has also been developed [69-76].
6.2

Testbed Setup
In this section, an overview of the experiment testbed is presented. The specific

implementation details of the testbed are presented in the Appendix. Figure 6-1 shows an
overview on the implementation. In the Bandwidth Scavenging Concept, a secondary user device
attempts to access a wireless channel that is occupied by a primary user device. We implement
86

these two separate networks as two independent systems, namely the Primary User Network
(PUN) and Secondary User Network (SUN). The purpose of PUN is to generate various traffic
patterns on a channel. While the purpose of the SUN is to first perform measurements of the PU
channel activity then utilize an access strategy to send messages in between the PU
transmissions.
Primary User Network

Secondary User Network

Figure 6-1: System implementation overview
As shown in Figure 6-1, the PUN and SUN is completely independent of each other,
although they share an IEEE 802.15.4 Zigbee channel. Crossbox Telosb motes [65] are used as
the primary and secondary devices. TelosB was used since the TinyOS port on it provides
enough driver level APIs for implementing a custom MAC layer (i.e. CSMA in this case), which
is not easily possible for majority of the commercially available 802.11 cards. The
communication and programming of the motes are done using the TinyOS 2.1 operating system.
For the purposes of traffic generation and statistics gathering, the motes are connected to
personal computers running Ubuntu Linux 8.04 LTS. Separate computers are used for the PUN

87

and SUN. The implementation of the computer interfaces is done using java with NetBeans as
the IDE.
The communication flow of the experimental network is done using two closed paths. In
the PUN path, an PU traffic pattern is generated on the primary user computer and then sent to
an PU mote. This mote will send the traffic to another PU mote or multiple motes. When the
packets reach the last mote in the path, the packets are forwarded back to the originating PC. The
SUN is designed in a similar fashion, by performing a second path using the SU motes and SU
computer. Performance metrics for the experimental testbed are gathered on each PUN and SUN
computers.
6.2.1

Crossbow TelosB Motes
The Crossbow TelosB Mote hardware consists of a small circuit board with several

microcontrollers, three LED’s, two physical buttons, a USB connector, connectors for sensors
board and a battery pack. This platform was developed by a research community at UC Berkeley
and is now sold by several manufacturers.
The motes are running TinyOS as the operating system. TinyOS provides a modular
operating system for Wireless Sensor Networks that supports several sensor platforms. TinyOS
is maintained and developed by the TinyOS Alliance. This Alliance is structured in working
groups consisting of researchers from several universities and companies.
6.2.2

Hardware Components and Network Topology
The testbed is currently setup using two personal computers and 4 TelosB motes. To

ensure undistorted results in the measurements, the implementation of the PU and SU networks
has been installed on separate computers.
88

Both computers are running:
• Ubuntu Linux 8.04 LTS, RealTime kernel rt-2.6
• TinyOS 2.1.0 with gedit and gcc programs
• Java 1.6
• NetBeans IDE 6.8
The PUN computer has a dual core Intel processor with 2 GB of ram and the SUN is an Intel
Celeron with 768 MB ram.
The topology of the test-bed is kept as simple as possible to keep unwanted effects as low
as possible. Figure 6-2 shows the physical layout of the test-bed. On the left side are two Primary
User motes. The top left node (PU Sim) is sending the test traffic pattern to the bottom left node.
While these motes are sending the bottom right mote (USB0 SUTX) accesses the channel and
sends messages to the top right mote. The top right mote is called Basesation because it forwards
all messages to the PC to create a full log of the tests.

Figure 6-2: Physical topology of the primary and secondary networks
6.3

Primary User Traffic Profiles
To explore the characteristics of different PU traffic, various traffic profiles are sent by a

PU and subsequently measured by an SU. A PU transmitter sends messages to a PU receiver
89

according to the given traffic profile. Concurrently, an SU measures the channel using an RSSI
sampling program implemented on the motes. The channel is measured at 1 kHz rate with RSSI
values being reported at each time step [77]. Consecutive measured RSSI values below a predefined threshold (-50 dBm) are considered as whitespaces.
In order to represent PU traffic diversity, three traffic scenarios are established i.e. sent
over the PU network:
• A uniformly distributed traffic that is sent at 7 packets per second (pps). Each packet has
an added ±20% inter-packet time to vary the arrival time on the channel. This profile
corresponds to TelosB based sensor network applications involving constant bit rate
sensing.
• A low rate video file streamed over the channel, which represents video surveillance
applications.
• A multi-modal traffic with two “peaks” of whitespaces in the pdf w(n). In general, a
stream with multiple peaks in its whitespace can be formed by merging multiple constant
packet rate streams. This is representative of a multi-application primary traffic at
different traffic generation rates.
The packet size of the PU node is set to 120 bytes, which corresponds to approximately 5
ms channel access duration given the bandwidth of the channel (250 kbps data rate). Given that
the SU senses the channel at Tp = 1 ms intervals, it is able to determine the approximate start and
ending times of PU packets (i.e. ON-OFF durations). Figure 6-3a-c shows the whitespace pdfs
w(n) measured and computed by an SU Transmitter, which is within the transmission range of
the PU Transmitter that is generating the traffic profiles as described above. In Figure 6-3a-c, the
pdf is plotted as a function of the sensing interval time, i.e. nTp.
90

0.015

0.08

0.02

Probability

0.025

0.06

0.015

0.01

0.04

0.01
0.005
0

0.005

0.02

0
0
50
100 150
200
Whitespace Duration (nTpms)
a)

0 100 200 300 400 500
Whitespace Duration (nTpms)
b)

0

0 100 200 300 400 500
Whitespace Duration (nTpms)
c)

Figure 6-3: Experimental whitespace distributions. (a) Uniform with 20% variation, (b) Low rate
video stream, and (c) Multi- modal data stream
The following observations can be made from the whitespace statistics shown in Figure
6-3. In the uniform traffic case of Figure 6-3a, all the whitespaces are clustered between 70 ms
and 180 ms, which is the result of the 20% variation on the 7 pps traffic generation rate. Observe
that the peak does not form a perfect uniform shape as expected. This result comes from the
combined effects of the TinyOS operating system’s scheduling jitter and the asymmetry in the
hardware delays in the transmission times of the packets. For the pdf of the video stream in
Figure 6-3b, the large peak corresponds to fixed rate video packets. The other small peaks
correspond to embedded audio packets and uncorrelated video packets, which are transmitted at
a slower rate. For the multi-modal pdf case in Figure 6-3c, the whitespace is distributed within
two separate ranges. The two peaks correspond to applications that generate a traffic rate of 4
pps with 25% spread, and 2 pps with 75% spread respectively.
The analysis of whitespace in the experimental network highlights the various whitespace
environments that an SU may experience when attempting to access unused spectrum.

91

6.4

Performance Evaluation of CSTS on Prototype Testbed
In this section, the performance of the CSTS strategy is evaluated experimentally using the

testbed presented in this chapter and using a Matlab based simulator discussed in Section 5.1.2.
6.4.1

Compared Protocols
We compare CSTS with an inter-frame spacing based protocol (AIFS). The AIFS is based

on IEEE 802.11e [18] where priority is given between classes of traffic by waiting class-specific
inter-frame spacing (IFS) before sending a packet. When the SU has a packet to send, using
AIFS, it waits for the specified IFS time for the channel to remain free before transmitting. In our
implementation, a zero IFS was chosen for the PUs and a non-zero IFS was chosen for the SUs.
Like in 802.11e, this arrangement gives rise to priority channel access to the PUs. The IFS for
the SU is chosen as half the SU packet size, which provided the best performance for the AIFS.
In the simulation results, a benchmark protocol is also plotted. The Benchmark protocol
assumes full knowledge of the starting and ending times of every whitespace. Therefore, it
transmits SU packets in every available whitespace and stops when the next SU transmission will
cause disruption to the PUs. The Benchmark protocol can be run only offline to find the best SU
performance. This protocol essentially determines the highest throughput an SU can expect
without causing any disruption to the PUs.
6.4.2

Experiment Implementation
The evaluation of the access strategies on different traffic profiles and data rates required

a large number of experiments. In contrast to a simulation environment, testbed experiments are
run in real time, which causes the runtime to be significantly high. In the experiments, 5000
primary user sample-packets were used. The time necessary to create a single graph point is over
92

80 min. Hence the creation of a full graph can take up to over 13 hours of runtime ((5000 packets
* 5 points in the graph * 2 Access strategies) / (60 packets/min)). This condition required the
automation of the experiments. First, all of the Java based GUI elements were removed and
replaced with a command line interface. Then AWK batch scripts where used to set all of the
necessary experimental parameters. Next, the two Primary and Secondary computers must be
synced up together during the running of an experiment. This task was accomplished through
networking the computers together with a shared folder. In the folder, files were passed between
the computers to signal the starting, ending, and data collection periods of the experiments. This
setup allowed experiments to be run over night.
During the experiments, the following values are captured: the Round Trip Time (RTT) of
each PU packet, the PU packet losses and a full packet log. From these values, the packets
deferred are calculated by counting the packets with a RTT > 100 ms plus the packets lost. The
indicator for the secondary user performance is the Measured Secondary User Throughput
(MST), which is calculated from the full packet log. In the figures, the AIFS protocol is referred
to as priority based and the CSTS is referred to as History based.
6.5

Performance with Uniform PU traffic Profile
Figure 6-4 and Figure 6-5 show the percentage of PU packets that where deferred because

of the AIFS and CSTS protocols (referred to as Priority and History based respectively). The PU
traffic profile is a uniform distribution at 120 packets per minute (ppm) similar to the distribution
shown in Figure 6-3. The traffic load generated by the SU is varied in order to evaluate the
impacts of increasing non-priority traffic on the priority traffic. Observe that CSTS has a low
percentage of deferred PU packets across all SU rates, while AIFS suffers from greater
disruptions. Figure 6-4 shows the actual percentage and Figure 6-5 show the same data on a Log93

Log scale. The disruptions to the PU traffic are quite low for the CSTS strategy. AIFS suffers
from greater disruptions since the SU will always transmit after waiting for an IFS. Even after an
IFS period, there is some probability that the PU will return to the channel during the SU
transmission, causing disruptions for the PU traffic. Figure 6-6 shows similar results through the
Round Trip Time.

PU Packets Deferred

30%

20%
Priority Based
History Based

10%

0%
0

10

20
SU PPM

30

40

Figure 6-4: PU packets deferred for 120 ppm uniform traffic
100%

PU Packets Deferred

1

10

100

10%
1%

Priority Based
History Based

0.1%
0%
SU PPM

Figure 6-5: PU packets deferred for 120 ppm uniform traffic in Log scale

94

Round Trip Time (ms)

500
400
300
Priority Based

200

History Based

100
0
0

10

20
SU PPM

30

40

Figure 6-6: RTT for 120 ppm uniform traffic
Figure 6-7 shows the Measured SU throughput (MST) results for the experiment. Observe
that the as the SU ppm increases the throughput increases. Note that due to the limitations of the
TelosB motes, the channel cannot be completely filled by either primary or secondary user
traffic. This limitation was discussed in [78]. Therefore to capture the effects of the completely
filling the channel simulation based experiments were performed.
90k

MST

70k
50k

Priority Based
History Based

30k
10k
0

0

10

20
SU PPM

30

40

Figure 6-7: MST for 120 ppm uniform traffic

95

Figure 6-8 show the simulation results for the 120 ppm uniform scenario. This simulation
result was created by using the RSSI trace files from the testbed. Observe that CSTS maintain
the required disruption bounds (indicated as DBs in the figure) across all SU rates, while AIFS
suffers from greater priority traffic disruptions, which is above 10% for the higher SU rates. In
terms of throughput, the access strategies are able to support low SU generation rates. However,
once the specified DB is reached, CSTS throttle the SU transmissions in order to protect the PU
traffic. Notice that AIFS is able to achieve higher SU throughput by always sending after waiting
for an IFS, but at the expense of higher disruptions to the PU traffic. This demonstrates how
CSTS are able to protect the high priority traffic by limiting the low priority traffic throughput
whenever appropriate.

Figure 6-8: Simulation results for 120 ppm uniform traffic
We show the impacts of PU traffic rates in Figure 6-9. PU traffic with the uniform profile
is varied from 1 packet per second (pps) to 7 pps. In the presented results, the SU generation rate
is set to a high value, which translates to the SU always having packets buffered to send for the
96

entire duration of a whitespace. Observe that as the PU traffic rate increases, CSTS is able to
maintain the specified PU disruption bounds while the PTD values for the AIFS protocol are
quite high. This is because CSTS create different parameters (Jmax) for each PU traffic rate to
maintain the desired bound. As expected, as the PU rate increases, there are less whitespaces for
the SU to access the channel, hence lowering the SU throughput.

Figure 6-9: Simulation results for various uniform traffic rates
6.6

Performance with PU Traffic with Low-rate Video
The protocol performance with PU traffic from a low-rate video stream (as shown the pdf

in Figure 6-3b) is shown in Figure 6-10 through Figure 6-12. Note that the relative performance
of the three evaluated protocols in this case is very similar to those observed for the uniform
traffic case in the previous section. As in the uniform traffic case, the CSTS protocols are able to
maintain the specified DB at the expense of limiting the SU throughput at higher SU load levels.
In terms of throughput, the values are smaller that the uniform case. This results from different
availability of whitespace in the video stream profile. The simulation results show similar trends.
97

PU Packets Deferred

20%
16%
12%
Priority Based
8%

History Based

4%
0%
0

5

10

15

20

Secondary User Packets Per Second

Figure 6-10: PU packets deferred for video stream based traffic
16k

MST

12k
8k

Priority Based
History Based

4k
0
0

5

10

15

20

SU PPM

Figure 6-11: MST for video stream based traffic

98

Figure 6-12: Simulation results for video stream based traffic
6.7

Performance with Multi-modal PU Traffic
Similar results to the video stream profile are demonstrated in the multi-modal case (Figure

6-3c) presented in Figure 6-13 through Figure 6-15. Again the CSTS protocol maintains the
required PU traffic disruption bounds while the maximizing the SU throughput.

99

14%
10%
Priority Based
6%

History Based

2%
0%
0

5

10

15

20

SU PPM

Figure 6-13: PU packets deferred for multi-modal based traffic
40k
30k
MST

PU Packets Deferred

18%

20k

Priority Based
History Based

10k
0
0

5

10

15

20

SU PPM

Figure 6-14: MST for multi-modal based traffic

100

Figure 6-15: Simulation results for multi-modal based traffic
6.8

Summary and Conclusions
This chapter presented a TinyOS based testbed implementation of the Bandwidth

Scavenging Concept. A Primary User Network (PUN) was implemented with the purpose of
generating various traffic patterns on a channel. Along with a Secondary User Network (SUN)
was created to perform measurements of the PU channel activity. Then utilize an access strategy
to send messages in between the PU transmissions. The PUN was shown to be able to create
multiple traffic patterns using a PC and TelosB mote configurations. The SUN was shown to be
able to measure the PU transmissions on the channel. Additionally, multiple SU access strategies
were implemented on the motes.
The access strategies were then evaluated using the testbed that was implemented using
TelosB motes. From the experimental results, it was shown that the access strategy was able to
support the offered SU load. Due to the limitations of the TelosB hardware, additional
simulations were run in Matlab. The simulation results showed that the CSTS protocol would
101

throttle back the transmissions of the SUs to protect the PU traffic. Similar to the 802.11 results
the Bandwidth Scavenging Concept was shown to minimize PU throughput while maximizing
SU throughput.

102

Chapter 7: Divided SU Transmission Strategy (DSTS)
7.1
7.1.1

Introduction
Motivation
The whitespace profile of a general PU network can be a function of an individual PU

properties such as topology, traffic pattern, and data rates. This creates a major challenge for SUs
attempting to access the available whitespace. Therefore, an SU access strategy that can adapt to
the various whitespace profiles is needed.
7.1.2

Problem Definition
The CSTS protocol that was used in Chapter 4 through Chapter 6 was designed to take

advantage of the specific whitespace characteristics in an Ad hoc mode 802.11 network. As a
result, in networks with non-802.11 traffic, it may miss transmission opportunities because of an
over-conservative sending limit that is imposed based on the assumption of small whitespaces
corresponding to ad hoc mode 802.11 control packets. The proposed Divided Secondary user
Transmission Strategy (DSTS) in this chapter removes this limitation of CSTS.
7.1.3

Related Work
The authors in [16, 17] developed a methodology for formally analyzing the whitespace

available within 802.11 primary traffic in infrastructure mode. The key idea in their methodology
is to model the whitespace as a semi-Markov process that relies on the underlying 802.11 state
model involving DIFS, SIFS, DATA, and ACK transactions. Their semi-Markov model
describes the whitespace in terms of holding times of the idle and busy states of the channel.
Building on this model, in [6] the authors further develop an WLAN access strategy for
103

secondary users in which the SUs utilize packet-size slots for the channel access. At the
beginning of each slot, an SU senses the channel and if the channel is free then it transmits with a
specified probability that is calculated from previous measurement. In our proposed access
mechanism, a SU utilizes whitespaces based on a stochastic model that minimize PU disruption
and maximize SU throughput.
The authors in [14] developed an optimal access strategy for secondary users. This work
has certain similarities with ours in that both approaches attempt to maximize the SU throughput
while keeping the PU disruption below a pre-determined bound. However, there are a number of
key differences as listed below. First, we formulate the problem in a 0-1 knapsack framework
and solve it using dynamic programming whereas in [14], a threshold based mechanism is used.
Second, we evaluate our access strategy in a realistic scenario where multiple SUs exist within a
single collision domain whereas in [14], the authors consider two isolated and independent SUs
in separate collision domains. Finally, the characterization of whitespace profiles under a diverse
set of traffic profiles using ns-2 simulations and experimental testbed is a unique feature of our
work compared to [14].
7.2

Access Based on Transmission Opportunities
A Transmission Opportunity (TO) within a whitespace is defined as a period of time that is

large enough to send a single secondary user packet. Clearly, there may be multiple TOs for
sending multiple SU packets within a whitespace. However, since the duration of a whitespace is
unknown in advance, there is a non-zero probability for an SU transmission to spill over the endtime of the whitespace, causing a PU disruption. The objective is to reduce the occurrences of
such transmissions within a whitespace so that the chance of PU traffic disruption is minimized.

104

Conversely, that reduction needs to be selective so that the SU throughput can be maximized by
judiciously scavenging for the legitimate TOs within the whitespace.

TOs
t1,b1 t2,b2

…

S

tn,bn

Figure 7-1: DSTS transmission opportunities
Consider the sample whitespace profile shown in Figure 7-1. Observe that the probability
that a whitespace ends between [0-200] ms or [300-400] ms is zero. This means that an SU can
access these whitespaces without disrupting the PUs. However, utilizing whitespace durations
within the ranges of [200-300] ms or [400-550] ms may cause disruption to the PUs. This is
because there is a non-zero probability that a whitespace ends between these ranges. The key
idea behind the DSTS strategy for an SU is to determine the best useable TOs based on the
recently observed whitespace statistics.
7.3

Transmission Bitmap Vector
In DSTS, a whitespace is divided into S duration TOs, where S is the duration of an SU

packet plus SU control packets. As shown in Figure 7-1, the TOs are referred to as t1, t2, …,tm,
where the quantity tmS represents the largest whitespace observed by the SU. Let the SU packet

105

transmission decision in those TOs be represented by the bits b1, b2, …, bm, where bit b0
indicates if the SU should transmit a packet (if available) in TO t0. These m-bits (1: transmit, and
0: not transmit) form a Transmission Bitmap Vector (TBV). If the bit bi is set to 1, then the
access module in an SU transmits a packet in TO ti only if a channel-sensing finds the channel to
be free during that TO. In other words, a transmission can only occur if the whitespace lasted
until the TO ti. If the channel is found busy (meaning the whitespace has not lasted until the TO
ti) no transmission is made even though the bi was set to be 1. In addition, if bit bi was set to be 0
in the TBV, no transmission is made during the TO ti. Once a whitespace has ended, the
procedure resets for the next whitespace. Note that each bit in the TBV is calculated using the
w(n) resulting from the RSSI measurements similar to those shown in Chapters 3, 6, and 8.

Secondary User
S

Tp

SU
Application

DB

Channel
Access
W(n)
Measurement
Parameter
& Modeling
Computation
Sensing

TBV

TBV
Access
Protocol
Data

Sensing Packets

Wireless Channel
PU1

PU2

…

PUn

Primary User Network
Figure 7-2: Functional diagram of the SU access strategy
106

7.4

Functional Overview
Figure 7-2 shows a functional diagram of the architectural components in DSTS. The user

tunable control inputs into the system are secondary user packet size (S), primary disruption
bound (DB), and the sensing interval (Tp), which is a property of the secondary user hardware.
The SU continuously computes the whitespace pdf w(n) with a channel-sensing interval of Tp.
The access parameters are then fed into the access module, which, in addition to channel sensing,
transmits SU user packets into the channel. The objective is to satisfy a user specified primary
disruption bound DB.
7.5

PU Disruptions and SU Throughput
A method must now be defined to properly select the best TOs that the SU should utilize.

Ideally, the SU would like to use every TO to maximize its throughput but disruptions can occur
since the duration of a whitespace is unknown. Therefore, each transmission opportunity has a
certain gain and loss if it is chosen for use. A gain is represented by throughput for the SU and a
loss is represented by a possibility of disruption to the PU. In this section, the throughput and
disruptions are formulated for each TO.
7.5.1

Probability of TO Existence
An SU packet transmission during a TO ti will be made only if the whitespace lasts until

that TO. The probability of the existence of TO ti can be expressed as the probability that the
whitespace will last at least for iS duration, and can be written as:

107

Pi

existence

 iS 
=1−W  
T 
 p

(7-1)

where W(n) is the cumulative distribution function (cdf) of the whitespace pdf w(n), which is
expressed as a multiple of the channel sensing interval Tp. When the bit bi is set to 1 in the TBV,
the probability Pi

or not. The

existence

Pi existence

will determine if an SU transmission in TO ti will actually be possible

value for a given TO directly represent the throughput an SU will achieve

if it sends in that TO. It follows that a TO with high existence probability will occur with greater
frequency, thus resulting in higher SU throughput.
7.5.2

Probability of Disruption

An SU packet transmission during a TO ti will cause a PU traffic disruption if the current
whitespace does not last for a full S duration since the start of the transmission. The probability
of disruption is expressed as the probability that a whitespace lasts anywhere between (i-1)S to iS
duration, and can be written as:

Pi

disrupt

 iS 


  − W  (i − 1) S 
=W
T 
 T 
p
 p



Now consider a situation in which a specific TBV,
all m bits in

ˆ
T

is selected. Let

(7-2)

δ

represent a subset (of

ˆ
T ) of bits that are 1. The rest of the bits are set to 0. During a whitespace, after an

SU transmits based on

ˆ
T , the resulting Primary Traffic Disruption (PTD) can be expressed as:
108

PTD =

Pi disrupt
∑

(7-3)

∀i∈δ

PTD represents the overall probability that a Primary Traffic Disruption will occur at the
end of a whitespace. If each whitespace corresponds to a PU packet transmission (i.e. when no
back-to-back PU packets are transmitted), PTD represents the probability that a PU packet will
be disrupted due to an SU packet transmission. With a specific subset δ , the Effective
Secondary user Throughput (EST) can be expressed as:

EST =

Pi existence
∑

(7-4)

∀i∈δ
7.6

Computing TBV via Solving Knapsack Problem
For a given whitespace pdf w(n), the problem is to compute a Transmission Bitmap Vector

ˆ
T such that the EST is maximized while limiting the PTD to a pre-defined bound DB. We

propose to solve this problem using the 0-1 knapsack problem solution, which is summarized as
follows. In the knapsack problem, there are m objects (j=1,2,...,m) and the j

th

object has a weight

wj and volume vj. The goal is to select a subset of the objects in a knapsack so that the total
weight is maximized while the volume of the knapsack does not exceed a predefined bound.
The TBV computation is posed as the knapsack problem by mapping the TO ti to m

objects. The TO existence probability

Pjexistence
109

from Equation 8-1 and the disruption

probability

Pjdisrupt

from Equation 8-2 are mapped to wj and vj respectively. In addition, the

PU disruption bound DB is mapped as the knapsack volume bound, V. The mapping and solution
process is formalized as follows.
The problem can be defined as:

m

maximizeW = ∑ w j x j
j =1

(7-5)

subject to:

m

∑v x
j

j

≤V,

x j ∈ {0 ,1}

j =1

(7-6)

To solve this problem, we can use dynamic programming [79] by defining s[j,v] to be the
solution for objects 1,2,…j with maximum volume v. Then define s[j,v] recursively as:

0,
if j or v = 0

s[ j , v] = s[ j − 1, v],
if v < v j

max{w j + s j − 1, v − v j , s[ j − 1, v]}, if v j ≤ v

[

]

(7-7)

The optimal solution can then be found by calculating s[j,V]. Under the assumption that all
whitespace can be detected, the

Pjexistence

and

Pjdisrupt

values quantify the SUs’ throughput

and PU disruption respectively for a given TO tj. Hence, mapping the knapsack solution to TBV
can give an optimal secondary user transmission strategy.

110

7.7

Sub-optimal TBV Computation
Although providing optimal solutions, dynamic programming can be computationally

intensive and may not be feasible for resource-constrained SU hardware. We provide a
computationally feasible heuristic solution as shown in Figure 7-3.

Figure 7-3: TBV creation via sub-optimal solution
disrupt

existence
Initially, we compute the Density ( Pj
/ Pj

) of all m objects (i.e. TO) and

then sort them in descending order of their density. Afterwards, we assign a 1 in the
Transmission Bitmap Vector

ˆ
T for

the highest density TOs, representing good transmission

opportunities. We stop assigning 1s when the Disruption Bound is reached, ensuring that the
probability of disruption will meet the pre-defined bound. This scheme provides a relatively fast
and simple heuristic solution to the problem. After this algorithm is completed, the

ˆ
T

for the

MS-DSTS access strategy is created. At any given time in a whitespace, if an SU has a packet to
111

send, it will send the packet in the next TO indicated in

ˆ
T , only if the whitespace lasts until that

TO. An SU following this strategy will maintain the specified DB while maximizing its
throughput.
7.8

TBV Solution Example

Figure 7-4: TBV of DSTS for different traffic profiles
Figure 7-4 presents an example Transmission Bitmap Vectors (TBV) solution computed by
DSTS for three experiments shown in Section 6.3 corresponding to uniform, video stream and
multi-modal traffic profiles. The black and while vertical bars in the diagram correspond to the
ones and zeroes in the TBV. Observe how the ones in TBV, representing useable transmission
opportunities for the SUs, occupy different part of the whitespace spectrum for different traffic
profiles.
7.9

Performance Evaluation of the DSTS Protocol
In the following sections, the performance of the DSTS strategy is evaluated

experimentally and theoretically in terms of the Primary Traffic Disruption (PTD) and the
Effective Secondary Throughput (EST). The whitespace profiles characterized in Sections 3.4.2,
and 6.3, i.e. Ad hoc 802.11 and test bed measured CSMA respectively are used to evaluate
DSTS.
112

7.9.1

Comparison Protocols
A comparison of DSTS between multiple protocols is provided. The compared protocols

are a sense-and-transmit based protocol, called VX scheme [56], a threshold based access
strategy (TBAS) [14], the CSTS protocol presented in Chapter 4 and lastly a benchmark scheme
(Benchmark).
7.9.1.1 VX Scheme
The VX (Virtual-transmit-if-Busy) Scheme has a similar strategy as CSMA where the SU
will sense the channel and transmit when idle [56]. Initially the SU senses the channel, when the
channel becomes idle a packet of length Ls is transmitted. Then, the SU starts a vacation from
the channel of length Vs. After the vacation, if the channel is busy, the SU starts a virtual
transmission stage and then enters into another vacation stage afterwards. Here, a virtual
transmission means that the SU does not actually transmit the packet but waits for a time interval
which is equal to the packet length Ls. After the vacation, the SU starts the channel sensing
again. In contrast to the CSTS scheme, the VX scheme does not use statistical data but relies on
short term sensing for its transmission decisions. Algorithmic analysis for this scheme has been
presented in [56].
7.9.1.2 TBAS Protocol
The TBAS scheme determines an optimal transmission strategy for the SUs using an
analytical model [14]. In this protocol, first a time related decision metric g(t) is introduced. A
greater value of g(t) represents a smaller probability of disruption. Next, based on a given
disruption bound and g(t), an optimal threshold γ * is calculated by evaluating all possible
113

combinations of g(t) and γ. Finally, using the metric g(t) and threshold γ * , a policy q(t) is
proposed to determine if the SU should transmit at a given time t within a whitespace. Under the
assumptions of continuous time and very small SU packet size, it is shown that the policy is
optimal. Additionally, for more practical scenarios with finite packet size, the authors propose a
heuristic solution. First, the minimum g(t) value is found within the length of time of a SU
packet duration. Then using the minimum g(t) and the corresponding threshold γ, a sub-optimal
q(t) is computed.
7.9.1.3 CSTS Access Strategy
Contiguous SU Transmit Strategy (CSTS) protocol operates based on the channel usage
history, and attempts to bound the amount of PU disruption and maximize SU throughput. In
STL, there are two distinct stages. In the first stage, an SU initially waits for the channel to
become idle. Once the channel becomes idle, the SU attempts to transmit its first packet of
duration S. If the SU has more packets to send, it continues sending until the measured value of є
(time elapsed since the beginning of the current whitespace) becomes so large that the expected
PTD for new transmissions become equal to or larger than a pre-determined Disruption Bound
(DB). A term Jmax is defined, which denotes the maximum number of packets that the SU can
send before the expected disruption reaches the pre-determined DB. The quantity Jmax is
computed by the SU based on the observed whitespace statistics. After Jmax packets are sent, the
low priority node abandons transmission attempts and starts channel sensing for detecting the
next whitespace.

114

If, however, a whitespace ends before all Jmax packets are sent by the SU, it exits the
channel after detecting that a PU has started transmitting. Note that in spite of the SU vacating
the channel, disruption to the PUs’ traffic can still occur since the PU must defer its transmission
until the channel is vacated by the SU. This stage essentially specifies a block of time in every
whitespace that the SU is allowed to transmit. This block is bounded by the start of a whitespace
to the ending of the Jmax

th

packet.

The second stage of STL corresponds to the situation in which a whitespace duration is
greater than the end time of the Jmax

th

packet from the SU. No PTD will be caused in this stage

because the SU will exit the whitespace after Jmax transmissions.

7.9.1.4 Benchmark Protocol
The Benchmark protocol assumes full knowledge of the starting and ending times of every
whitespace. Therefore, it t0ransmits SU packets in every available whitespace and stops when
the next SU transmission will cause disruption to the PUs. The Benchmark protocol can be run
only offline to find the best SU performance. This protocol essentially determines the highest
throughput an SU can expect without causing any disruption to the PUs.
7.9.2

Ad hoc 802.11 Whitespace

7.9.2.1 802.11 Whitespace profile
In order to represent different 802.11 based Ad-hoc network topologies and traffic
characteristics, three scenarios are established (see Figure 7-5). (TOP1) is a linear chain topology
with a single multi-hop primary flow representing a simple PU topology. (TOP2) is two multi115

hop flows in a network on two intersecting linear chains, which highlights the effects on
whitespace of competing primary flows. (TOP3) is two multi-hop flows in a parallel chain
network, which emphasize the effects on whitespace of disjoint primary flows. In each case, the
secondary users are within the CS range of all primary users.

PU2

PU1

PU3

PU4

SU1

PU5

SU2

Multi-hop: TOP1
Traffic Flow Direction

PU6
SU2
SU1
PU1

PU2

PU3

Intersecting
TOP2

PU7
PU4
PU8

PU1

PU2

PU3
SU1

PU5
PU6

PU7

PU8

PU4

PU5

SU2
PU9

PU10

Parallel
TOP3

PU9

Figure 7-5: Primary user topologies and the traffic flows
Figure 7-6 shows the w(n) for UDP traffic with 90 ms average inter-packet duration that is
uniformly distributed across ±20% around the average packet arrival interval for each topology.
The distributions presented in Figure 7-6 represent the general whitespace properties for a wide
range of topology and traffic variations of the Ad hoc 802.11 whitespace.

116

10-0

Probability

10-1
10-2
10-3
10-4
10-5
Whitespace Duration (nTpms)
Figure 7-6: Impacts of PU topology with Uniform traffic on whitespace distributions.
7.9.2.2 Impacts of Primary User Load
Figure 7-7 demonstrates the impacts of varying PU load on the SU throughput and PU
disruption for TOP2 in Figure 7-5. The PU inter-packet arrival duration is varied from 50 ms to
120 ms for simulating different loading conditions. For all the primary packet rates, a ±20%
uniformly distributed inter-packet arrival variation was introduced. The SU packet duration was
fixed to 1.2 ms, which is half the PU packet size. For the presented results, the SU packets are
generated back-to-back, meaning packets are always available to send within an SU’s buffer.
The user-defined disruption bound (DB) is set to 0.05 [6], which indicates that only 5% of the
PUs’ traffic is allowed to experience disruptions from the SUs’ transmissions. The values of µ,
Jmax, and TBV are calculated dynamically depending on the current PU traffic statistics. The
sensing interval Tp is set to 5 µs. For the VX scheme, Vs, the channel wait time, is set to 0.5 ms

117

and the Virtual Xmit duration is set to 16 ms, which was shown (via simulation) to deliver the
best performance for that access protocol.

Figure 7-7: Performance for uniformly distributed PU traffic
From Figure 7-7, observe that both CSTS and DSTS access protocols are able to maintain
the required primary disruptions within the specified Disruption Bound (DB) across a wide range
of PU data rates. As the PU data rate increases, the duration of usable whitespace in the channel
reduces. Also, the number of distinct whitespaces increases with the PU data rate, since the
whitespaces become more fragmented. Since the SU access strategies are executed on a perwhitespace basis, with the increased number of distinct whitespaces, the SU attempts to execute
its PU protection mechanisms (e.g. TBV in DSTS) more often. This causes the SUs’ channel
access efficiency or EST to decrease due to increased PU load.
In Figure 7-7, it should be also observed that the EST for DSTS, CSTS and TBAS access
strategies remain close to that of the offline Benchmark protocol, while the target DB is
maintained. Our simulation experiments comparing the performance gap between the optimal
and sub-optimal DSTS solutions showed minimal difference. Additionally, the EST results for
118

sub-optimal DSTS, as shown in Figure 7-7, are only within approximately 3% of the Benchmark
performance. Since the Benchmark represents the best possible EST, this 3% difference in fact
represents the upper-bound of the performance gap between the optimal and sub-optimal DSTS
solutions.
Note that DSTS achieves slightly higher EST over the TBAS. This difference arises in the
method used to accommodate non-zero SU packet sizes in DSTS and TBAS. DSTS uses the SU
packet size as an input into the knapsack problem while TBAS uses the minimum g(t) value over
the range of the packet size.
For the VX scheme, disruption to the PU traffic can be observed to be higher. Utilizing the
Virtual Xmit wait times, the SU avoids sending during busy times in a channel. However, once
the SU starts accessing a whitespace, it stops sending only if a PU returns during a Vs wait time.
Thus, there remains a high probability of causing disruption during the SU transmissions. The
VX scheme has a relatively lower throughput since it waits for a Vs time after every
transmission, which is not the case of DSTS, CSTS and TBAS mechanisms. These results
suggest that by utilizing the history of the channel, it is indeed possible to improve the SU
throughput while maintaining a bound of disruption on the primary networks.

119

Figure 7-8: Performance for Poisson distributed PU traffic
Figure 7-8 shows results for the same setup, but with Poisson distributed PU traffic. Note
that the performance trends for this scenario are very similar to those for the Uniform case as
shown in Figure 7-7. Similar results were also observed for video stream PU traffic for which the
pdf was shown in Figure 7-6e. We have also simulated TCP primary traffic. Since TCP
opportunistically utilizes available bandwidth in the channel, the amount of resulting whitespace
for TCP traffic is small. As a result, the achievable SU throughput remains severely restricted
when the primary disruption bounds are set at low values [58].

120

Figure 7-9: Impacts of the SU packet size on EST and PTD
Figure 7-9 shows the impacts of varying the SU packet size from 625 µs to 2 ms, as done
in [14]. The PU rate is set to 10 pps with Uniformly distributed Traffic. Observe that the
throughput decreases as the SU packet size increases. In order to maintain the DB with larger
packet sizes, the access strategies must reduce their transmission probabilities since a larger
secondary packet has a higher probability of causing disruption to the primary at the end of a
whitespace. Additionally note that with increasing packet size, DSTS is able to achieve better
performance than the Threshold based scheme TBAS. This is because unlike TBAS, which
works with the assumption of very small SU packet size, DSTS takes the actual SU packet length
into consideration while setting its parameters.
7.9.2.3 Performance under Dynamic PU Traffic Patterns
Figure 7-10 reports the protocol performance from a simulation experiment in which the
PU traffic pattern was dynamically changed over time for TOP3. Initially, the PU traffic is
generated with 60 ms inter-packet time (IPT) with the Uniform profile. The pattern changes after
60 seconds when the IPT is increased to 120 ms. Finally, at 100 seconds the IPT is changed to
121

Poisson distribution with 90 ms average. The whitespace w(n) is continually computed by the

PU Traffic Change 2

PU Traffic Change 1

PU Traffic Change 2

PU Traffic Change 1

access protocol after collecting every 1000 whitespace samples.

Figure 7-10: Performance with time-varying w(n) for 802.11
Observe in Figure 7-10 that while the primary user traffic pattern changes over time, the
PTDs resulting from the CSTS and DSTS access strategies stay consistently within the vicinity
of the pre-set disruption bound of 0.05. Also, note that there is very little PU traffic disruption
during the transitions of w(n) as indicated in both the frames in Figure 7-10. This is primarily
because the whitespace pdf for ad hoc mode 802.11 traffic maintains very similar properties
across different traffic patterns, i.e. above 90% of the whitespaces are below 1 ms (see Figure
7-6).
Note that the SU throughput increases after a PU traffic pattern transition when the useable
whitespace increases due to the increase in inter-packet interval of the PU traffic. Additionally,
the SU dynamically throttles back its transmissions to protect the PUs when the available
whitespace reduces during a transition. These results demonstrate the robustness of the CSTS

122

and DSTS access strategies against varying data rates over time. Additionally, the results suggest
that by periodically measuring the channel state the SUs are able to dynamically model the
channel whitespace and adjust their access parameters accordingly to minimize the PU
disruptions and to maximize the SU throughput.
7.9.3

CSMA Profile Using Test Bed Measured Whitespace
In this section, we evaluate the DSTS strategies using the extracted whitespace pdf w(n)

from a TelosB based prototype test-bed for a single SU.
7.9.3.1 CSMA Whitespace Profile
In order to represent PU traffic diversity, three traffic scenarios are established i.e. sent
over the PU network:
•

A uniformly distributed traffic that is sent at 7 packets per second (pps). Each packet
has an added ±20% inter-packet time to vary the arrival time on the channel. This
profile corresponds to TelosB based sensor network applications involving constant
bit rate sensing.

•

A low rate video file streamed over the channel, which represents video surveillance
applications.

•

A multi-modal traffic with two “peaks” of whitespaces in the pdf w(n). In general, a
stream with multiple peaks in its whitespace can be formed by merging multiple
constant packet rate streams. This is representative of a multi-application primary
traffic at different traffic generation rates.

123

Figure 7-11a-c shows the whitespace pdfs w(n) measured and computed by an SU
Transmitter, which is within the transmission range of the PU Transmitter that is generating the
traffic profiles as described above.
0.08

0.02

Probability

0.025

0.06

0.015

0.015
0.01

0.04

0.01
0.005
0

0.005

0.02

0
0
50
100 150
200
Whitespace Duration (nTpms)
a)

0 100 200 300 400 500
Whitespace Duration (nTpms)
b)

0

0 100 200 300 400 500
Whitespace Duration (nTpms)
c)

Figure 7-11: Experimental whitespace distributions. (a) Uniform with 20% variation, (b) Low
rate video stream, and (c) Multi- modal data stream
7.9.3.2 Uniformly Distributed Primary Traffic
Figure 7-12 shows the PTD and EST with varying SU traffic load when the PU traffic
profile is set to be uniformly distributed. The corresponding whitespace pattern observed by the
SUs is shown in Figure 7-11a. In Figure 7-12, observe that both CSTS and DSTS are able to
maintain the required disruption bounds across all the experimented SU rates. In terms of
throughput, the access strategies are able to support low SU generation rates until the primary
user disruption bound is reached. However, once the specified DB is reached, for both CSTS and
DSTS the SU transmissions are throttled back in order to protect the PU traffic.

124

Figure 7-12: Performance with varying SU traffic load under Uniform traffic
Notice that in Figure 7-12 both CSTS and DSTS have very similar maximum SU
throughput values. This performance similarity is a direct result of the uniform PU traffic and the
resulting whitespace. Since all the whitespace is within a very small range (70 ms and 180 ms),
in order to maintain the required DB, both access strategies send packets until approximately the
same point in a whitespace. This explains the similar SU throughput numbers while guaranteeing
the same DB.

125

Figure 7-13: Performance with Multi-modal whitespace pdf
7.9.3.3 Multi-modal PU Traffic
Protocol performance for PU traffic with multi-modal whitespace profile (as shown the pdf
in Fig. Figure 7-11c) is shown in Figure 7-13. The relative performance of CSTS, DSTS, and
Benchmark in this case is very similar to those observed for the uniform traffic case in Figure
7-12. The primary difference here is the ability of the protocol DSTS to support higher
sustainable SU throughput compared to that for CSTS. As in the uniform traffic case, both CSTS
and DSTS are able to maintain the specified DB at the expense of limiting the SU throughput at
higher SU load levels. However, observe that DSTS is able to support 56.0% and 60.3% higher
throughput compared to CSTS when the DB is specified at 5% and 3% respectively. This large
increase in throughput over CSTS results from DSTS’s ability in identifying the best
transmission opportunities for sending an SU packet with minimal disruption.
Since the first peak of whitespace pdf around 250 ms (in Fig. Figure 7-11c) contains 25%
of all whitespaces, the Jmax value in CSTS is expected to be set near the start of the first peak.
This makes CSTS to miss transmission opportunities corresponding to the second peak in the
126

whitespace pdf, which the mechanism DSTS is able to access. Also note that in both DSTS and
CSTS, the throughputs corresponding to DB = 3% are slightly lower than that in the DB = 5%
case. This occurs because in order to maintain a lower DB, a lower number of SU packets are
sent. As shown in Figure 7-14, results for the video stream case showed similar results to this
multi-modal case.

Figure 7-14: Performance with Video Stream whitespace pdf
7.9.3.4 Dynamic PU Traffic Profile
The performance of CSTS and DSTS under a time-varying whitespace profile is shown in
Figure 7-15. The disruption bound DB is set to 0.05. The PU traffic pattern was changed using
the three pdf types in Figure 7-11. Initially, the PU traffic is generated with a uniformly
distributed rate of 7 pps. The pattern changes after 60 seconds when the traffic profile is changed
to multi-modal. Finally, at 130 seconds, the PU traffic is changed to a video stream.

127

PU Traffic Change 2

PU Traffic Change 1

PU Traffic Change 2

PU Traffic Change 1

Figure 7-15: Performance with time-varying w(n) for CSMA MAC
Observe that at steady state, the PTD remains close to the defined DB for both CSTS and
DSTS. However, during the w(n) transitions, the PTD increases past the disruption bound. In
contrast to the dynamic 802.11 based PU traffic case (see Section 7.9.2.3 and Figure 7-10), the
changes in w(n) during traffic transitions for CSMA traffic is quite significant. This explains the
relatively larger (compared to that in Figure 7-10 for the 802.11 case) transient peaks in the PTD
during the w(n) transitions in Figure 7-15. In terms of throughput, the protocol DSTS is able to
achieve higher throughput compared to CSTS when the traffic changes to a video stream and to
one that produces multi-modal whitespace.
7.10 Summary and Conclusion
In this chapter, a secondary user access strategy named Divided SU Transmission Strategy
(DSTS) is presented that provides a method to access a general PU whitespace profile. The
access strategy utilizes multiple transmission opportunities (TO) within a whitespace to find the
best times to transmit SU packets. These TOs are consolidated into a Transmission Bitmap
Vector (TBV) that defined when SUs should transmit within a whitespace to maximize SU
128

throughput while limiting PU disruptions to a predefined bound. Optimal and Sub-optimal
protocols are provided to select the best transmissions opportunities. Evaluation of DSTS under
different traffic profiles showed that the protocol outperforms comparison protocols.

129

Chapter 8: DSTS With Support for Multiple Secondary Users
8.1
8.1.1

Introduction
Motivation
A challenge in prioritized device coexistence is how to efficiently allow multiple SUs to

access the channel while minimizing the disruptions to the primary user traffic. The access
protocol presented in this chapter aims to provide a solution to that problem.
8.1.2

Problem Definition
An access protocol for multiple secondary users should support efficient SU access while

minimizing disruptions to the primary users. The protocol should also support fairness amongst
the secondary users. As discussed in Section 7.3, DSTS defines a Transmission Bitmap Vector
(TBV) that specifies the best Transmission Opportunities (TOs) within a whitespace. The key
idea of the access protocol, termed Transmission Bitmap Vector Access Protocol (TBVAP), is to
share a TO among the secondary users. Each SU would independently measure, model, and
calculate the TBV. From a functional standpoint, TBVAP operates in the SU Access Module in
Figure 7-2. Once the identification of the TOs within the whitespaces left by the PUs is done in
the Access Parameter Computation Module, TBVAP is used for sending SU data. Utilizing the
features of the DSTS access strategy, the proposed TBVAP aims to satisfy these SU access and
fairness requirements.

130

Whitespace
TBV
1 1

PU

TO1

0

1

… 1

…

TO2

PU

TOM

TO1

…
Time

RES

SU DATA

ACK

TR

TD

TA

TS
Figure 8-1: TBV with the corresponding TO structure
8.2

TBVAP Structure
An idle duration within the channel (whitespace) can be modeled as a sequence of

transmission opportunities for the secondary users. The number of utilized TOs is determined by
the duration of the whitespace that the SUs are currently observing. It is clear that the number of
TOs varies greatly with the dynamic nature of the PU traffic. In TBV calculation, the number of
TOs is set to the length of the longest duration observed whitespace divided by the size of a TO.
This ensures that with varying whitespace duration, a portion of the TBV will be utilized. Figure
8-1 shows the structure of the TBVAP. Depending on the length of a whitespace, each
whitespace consists of j number of TOs. Since the SUs are within the same whitespace domain,
the beginning of a whitespace is common to all SUs. Therefore, at the start of a whitespace, each
SU dynamically creates 1, 2,…j TOs of length Ts until the whitespace ends.

131

Figure 8-2: TO Access in TBVAP
Within a TO, there are three different time periods, namely Reservation, Data, and
Acknowledgment. The Reservation period is used to support fairness among the SUs. As shown
in Figure 8-2, first, at the start of the period, each SU selects a uniformly distributed number
between [1, TR/TSL] where TR is the length of the reservation period and TSL is the minimum
duration needed to detect a nodes transmission. The node with the lowest value gains access to
the TO. This process reduces inter-SU collisions and supports fairness among the SU nodes.
Next, the successful node sends a packet within TD seconds during the Data period. Note that the
length of TR+TD is fixed and the lengths of TR and TD can be adjusted depending on the number
of the SUs and the desired size of a SU packet. Lastly, a positive acknowledgment packet is sent
to the sender SU during the Acknowledgment period within TA seconds. Let TPD be the
propagation delay, thus the TO length, TS is given by TR+TD+TA+2TPD. Note that regardless
132

of the number of SUs, the access protocol guarantees the disruption bound on the primary user
traffic.

P1
P(ws≥k1)

1-1/N

P(ws<k2)

P(ws<k1)

1/N

P2

1/N

k2

1-1/N

S

1/N
P(ws<k3)
P(ws<M)

k3

D

kM

P3

:
.

k1

1/N

PM
Figure 8-3: Delay analysis for TBVAP
8.3

Delay Analysis
The average delay of the proposed access protocol is computed in this section. As shown

in Figure 8-3, we first define a state S where a SU will attempt to send a packet in a whitespace.
Let ki be the index of the i

th

1 in the TBV (i.e. locations where packet transmissions actually

happens). With probability P1, the whitespace will last for k1 duration to allow a SU packet to be
successfully transmitted. Assuming that there is N SUs attempting to access TO1, there is a 1/N
probability of success for a SU to deliver its packet. A successfully delivered packet, denoted as
state D, would then have a delay of k1. If the SU did not gain access to the TO slot, it will
133

attempt to transmit in the next slot with probability P2. If the whitespace does not last long
enough to send in the second TO, the SU will go back to state S. This process continues until the
packet is sent.
We denote the expected delay for a SU packet sent in a whitespace as Ews. Thus, we have:

M

E ws = ∑ kxk P

1
k N

k =1
where Pk is the

(1 − )

1 f (k )
N

(8-1)

Pkexistence , M is the maximum number of TOs, and f(k) is a function that gives

the number of 1’s in a TBV1…k. The term xk value is 1 when there is a 1 in TBVk and 0
otherwise.
Note that within a whitespace, at most M transmissions can be accommodated. Therefore,
sometimes a SU packet may be deferred until the next whitespace. To calculate this, we define ρ
as the probability that a packet can be sent in a whitespace. ρ is given by:

M

ρ = ∑x P

1
k k N

(1 − )

1 f (k )
N

k =1

(8-2)

Thus, the expected number of whitespaces before a secondary user packet can be successfully
sent is 1/ρ. Using Eqns. (9) and (10) we denote the average delay as:

Delay = Ews +
where L is the average whitespace length.
134

L

ρ

(8-3)

8.4

Whitespace Analysis for Multiple PUs
When an SU attempts to measure the whitespace in a general network, the number of PUs

can affect the whitespace characteristics. In this section, an investigation of the effects of
multiple primary users on the whitespace characteristics is presented.
8.4.1

Network Setup
To investigate a general wireless network, we consider a network with SUs existing within

a single collision domain with the PUs. Thus, each secondary user’s whitespace is equally
affected by transmissions of the primary users. Figure 8-4 shows the setup of the PUs and SUs.
In this network, each secondary user can hear the transmissions of all of the primary users.

PU1

SU3

SU5

PU3

PU2

SU6

SU8

Collision Domain

SU7

PU4
SU4

SU1

SU2

PU5

Figure 8-4: Multiple SUs and PUs within the same collision domain
The above network scenario is chosen to gain controlled insight to the interactions between
multiple primary traffic flows, and their effects on the resulting whitespace as perceived by the
secondary users.

135

8.4.2

Primary User Traffic Profile
The number of primary users and traffic data rate directly affect the total amount and

statistical properties of the available whitespace as viewed by the secondary users. We model
two PU traffic profiles, namely, Poisson and Uniform. The Poisson profile is modeled as a
Poisson process with mean packet arrival interval of α ms. The Uniform profile consists of
uniformly distributed packet arrival intervals with a mean of α ms and a variability of ±v ms
around the mean. Each PU independently transmits according to the specified traffic profile. The
packet arrival intervals for each Uniform and Poisson along with the number of PUs flow are
changed in order to represent varying intensity of the primary traffic.
8.4.3

Whitespace Analysis
The whitespace pdf w(n), with varying number of PUs in a fully connected network are

shown in Figure 8-5 through Figure 8-8. Figure 8-5 shows the w(n) for Poisson traffic with α =
200 ms average inter-packet duration. Observe that with a single PU, the whitespace follows an
exponential distribution with an average whitespace duration equal to 197 ms ≈ α. This is
expected, as the PU represents a single Poisson process. When the number of PUs is increased to
two, the whitespace duration is exponentially distributed with the average reduced to 98.3 ms ≈
α/2. Note that the average whitespace durations are slightly reduced by the PU packet time and
collisions between the PU traffic. The above trend for distribution of whitespace holds for a
higher number of PUs shown in Figure 8-6. In general, with n PUs, the average whitespace
duration is ≈ α/n.

136

Whitespace Duration (nTpms)

Whitespace Duration (nTpms)

Whitespace Duration (nTpms)

Figure 8-5: Impacts of 1, 2, and 4 Poisson primary users on whitespace distributions.

Whitespace Duration (nTpms)

Whitespace Duration (nTpms)

Whitespace Duration (nTpms)

Figure 8-6: Impacts of 6, 8, and 10 Poisson primary users on whitespace distributions.
Figure 8-7 shows the whitespace for primary user Uniform traffic with arrival times
uniformly distributed across ±20% around a 200 ms average packet arrival interval for each PU.
With a single PU, the whitespace follows a uniform distribution between [160, 240] ms. As the
number of PUs increases as shown in Figure 8-8, the shape of w(n) approaches that of an
exponential distribution similar to the Poisson Case.

Whitespace Duration (nTpms)

Whitespace Duration (nTpms)

Whitespace Duration (nTpms)

Figure 8-7: Impacts of 1, 2, and 4 Uniform primary users on whitespace distributions.

Whitespace Duration (nTpms)

Whitespace Duration (nTpms)

137

Whitespace Duration (nTpms)

Figure 8-8: Impacts of 6, 8, and 10 Uniform primary users on whitespace distributions.
With the uniformly random traffic generation rate, if we quantize time in terms of
transmission slots, at any given slot there is a probability γ that a packet is sent by one of the
primary users. With a larger number of primary users in the network, this probability will be
higher. Given that a whitespace is a sequence of slots without any packet, the probability that
there is a whitespace with length l slot is equal to:

P( ws = l ) = (1 − γ )l γ

(8-4)

From Equation 9-4, it can be seen that with the discrete uniform traffic of multiple PUs,
the number of empty slots before next transmission is geometrically distributed. With the large
number of traffic generated by multiple PUs, the geometric distribution approaches the
exponential distribution. This is an interesting result given that each primary user is sending at
fixed interval with uniformly distributed variance. This leads to an important claim namely, the
whitespace profile of the channel follows an exponential distribution if the primary users have a
sufficiently large traffic generation rate.
8.4.4

Whitespace Capacity
Figure 8-9 shows the fraction of channel idle time that is left over from the PUs traffic for

various numbers of primary users and packet arrival rates for Poisson traffic. Observe that the
fraction of idle time is large when there is a smaller number of PUs. As the number of PUs
increase the available capacity to the SUs decreases. Moreover, with faster transmission times,
i.e. smaller packet arrival intervals, the capacity reduces at a higher rate. For example, with 200
ms inter-packet arrival time and 10 PUs the fraction of idle time is 0.89 compared to 0.73 for the
80 ms case. These capacity trends are also similar for the Uniform case shown in Figure 8-10.
138

1

Idle Time Fraction

0.95
0.9
0.85
0.8

P400
P200
P100
P80

0.75
0.7

1

2

3

4
5
6
Number of PUs

7

8

9

10

Figure 8-9: Idle time fraction for various Poisson PU inter-packet arrivals
1

Idle Time Fraction

0.95
0.9
0.85
0.8
U400
U200
U100
U80

0.75
0.7

1

2

3

4
5
6
Number of PUs

7

8

9

10

Figure 8-10: Idle time fraction for various Uniform PU inter-packet arrivals

139

8.5
8.5.1

Performance Evaluation of TBVAP
Impacts of Primary User Load
Figure 8-11 demonstrates the impacts of varying PU load on the SU throughput and PU

disruption. The PU inter-packet arrival duration is varied from 80 ms to 400 ms for simulating
different loading conditions. The SU packet duration i.e. TO slot, was fixed to 1 ms. The number
of SUs is set to 15, with each SU generating packets with a 100 ms Poisson inter-arrival time.
The sensing interval Tp is set to 5 µs. The user-defined disruption bound (DB) is set to 0.05 [6],
which indicates that only 5% of the PUs’ traffic is allowed to experience disruptions from the
SUs’ transmissions. The value of TBV is calculated dynamically depending on the current PU
traffic statistics.

Figure 8-11: Performance for Poisson distributed PU traffic
From Figure 8-11, observe that TBVAP is able to maintain the required primary
disruptions within the specified Disruption Bound (DB) across a wide range of PU numbers and
data rates. As the number of PUs increases, the duration of usable whitespace in the channel
reduces. Moreover, the number of distinct whitespaces increases with the PU data rate, since the
140

whitespaces become more fragmented. Given that the SU access strategy is executed on a perwhitespace basis, with the increased number of distinct whitespaces, the SU attempts to execute
its PU protection mechanisms (e.g. TBV in DSTS) more often. This causes the SUs’ throughput,
presented in terms of packets per second (pps), to decrease due to increased number of PUs.
When the PU data rate (i.e. inter-packet arrival time) is decreased from 400 ms to 80 ms, the
resulting throughput also decreases.

Figure 8-12: Performance for uniformly distributed PU traffic
Figure 8-12 shows results for the same setup, but with a ±20% uniformly distributed interpacket arrival variation PU traffic. Note that the performance trends for this scenario are very
similar to those for the Poisson case as shown in Figure 8-11.

141

Figure 8-13: Performance for multiple SUs under Poisson PU traffic and 4 PUs

Figure 8-14: Performance for multiple SUs under Uniform PU traffic and 4 PUs
8.5.2

Impacts of Secondary User Load
Figure 8-13 and Figure 8-14 show the impacts of varying number of SUs for both Poisson

and Uniform traffic respectively. In the Poisson and Uniform case, the number of PUs is set to 4
and the inter-arrival time is 100 ms. The Secondary user Inter-packet time (SI) is varied from
200 ms to 1 ms, to represent increasing SU load. Each SU independently generate packets
according to the specified data rate. Observe that with a single SU and high data rate, the
142

maximum allowable throughput for the SUs is reached. Moreover, the disruption bound is
reached. As the number of SUs increases, the throughput decreases. This occurs, because of the
increased number of collisions between the larger numbers of SUs. It follows that the collisions
within assigned TOs are increasing with larger numbers of SUs. When the SU traffic rate is
lowered (i.e. 100 ms) the amount of PTD increases until it reaches the disruption bound, DB.
This occurs because the channel is able to support the SU load. Furthermore, the maximum
network throughput is achieved with different numbers of SUs depending on the SU traffic load.
For example, with SI = 50 ms, the maximum throughput is reached around 5 SUs. In contrast,
with SI = 20 ms, it is reached around 13 SUs. After the maximum throughput is reached,
increasing the number of SUs reduces the SU throughput due to increased collisions.

Figure 8-15: Performance for multiple SUs under Poisson PU traffic and 1 PUs

143

Figure 8-16: Performance for multiple SUs under Uniform PU traffic and 1 PUs
Figure 8-15 and Figure 8-16 present the results for a single PU with Poisson and Uniform
distributions respectively. Both traffic profiles have a 100 ms inter-packet arrival time. Observe
that the channel with Uniform traffic can support over 5 times the amount of SU throughput with
respect to the Poisson traffic. This interesting result comes from the properties of the whitespace.
Notice the whitespace pdf for 1 PU in Figure 8-7, the uniform distribution creates a large
whitespace duration for lasting for ≈ 160 ms. This type of whitespace provides SUs with a large
amount of channel time to utilize. In contrast, the Poisson distribution with 1 PU in Figure 8-5,
creates highly dynamic whitespace durations. Therefore, the SUs must be more conservative to
protect the PU traffic. As shown in Figure 8-15 and Figure 8-16, when the number of PUs is
increased the whitespace characteristics are similar between the Uniform and Poisson profiles.
Therefore, the SU performance is similar. These results follow from the discussion in Section
8.4.3.

144

Figure 8-17: Delay performance for 20 SUs under Poisson traffic with 5 PUs

Figure 8-18: Delay performance for 20 SUs under Uniform traffic with 10 PUs
8.5.3

Delay Performance
As discussed is Section 8.2, a feature of the TBVAP is supporting fairness amongst the

SUs. Figure 8-17 shows the packet delay in terms of TO slots for 20 secondary users. The
primary user traffic is Poisson with 80 ms inter-packet time and 5 PUs. Observe that the delay
range for all SUs is clustered around 11 slots signifying that each SU is given equal access to the
channel. This trend is also shown in Figure 8-18 where the primary user traffic is uniformly
distributed with 80 ms inter-packet time and 10 PUs. These results show that with the dynamic
nature of the PU traffic the DSTS protocol is able to provide fairness amongst the SUs.

145

8.6

Summary and Conclusion
In this chapter, the Transmission Bitmap Vector Access Protocol (TBVAP) was introduced

to support access for multiple SUs in a wireless channel. The protocol provides fairness in that
each SU is given an equal chance to utilize a Transmission Opportunity. These features were
demonstrated using results from a multiple PU CSMA based whitespace profile.

146

Chapter 9: Summary and Conclusions
9.1

Contributions
In this thesis, we have developed access strategies that allow secondary user (SU) devices

to coexist with primary user (PU) devices through utilizing unused spectrum or whitespace found
between PUs’ transmissions. A secondary user continually monitors its surrounding whitespace,
models it, and then attempts to access the available spectrum so that the effective secondary user
throughput is maximized while the resulting disruption to the primary users is kept limited to a
pre-defined bound. This opportunistic access during ultra-short and non-deterministic
whitespaces is termed as bandwidth scavenging by the secondary users. In other words, the SUs
scavenge the channel capacity left over by the PUs.
Chapter 3 presented the whitespace properties for an Ad hoc 802.11 PU network under a
wide range of topology and traffic variations that were experimented with using an ns-2 based
simulator. A number of key observations were made from the whitespace statistics. First, the
detected whitespace durations from the measured Received Signal Strength Indicator (RSSI)
trace vary a great deal across different topologies and traffic patterns. This indicates that a
secondary user may experience vastly different whitespace environments when attempting to
access primary networks. Second, it was shown that there is available capacity for SUs to
scavenge depending on the traffic profile of the PU. Third, regardless of the traffic profile or
topology, the vast majority of the whitespaces (above 90%) last for less than 1 ms. This results
from the large number of small whitespaces created during the RTS-CTS-DATA-ACK cycle for
each packet transmission.

147

In Chapter 4, a secondary user access strategy called the Contiguous SU Transmission
Strategy (CSTS), is described. CSTS is used to access an Ad hoc 802.11 PU whitespace profile.
CSTS exploits the properties of the whitespace, analyzed in Chapter 3, by avoiding transmissions
during the first millisecond of a whitespace. Once a whitespace is accessed, the parameter Jmax
limits the amount of SU packets sent within that whitespace. Under this process, the secondary
users maximize its own throughput while keeping the primary user disruptions to a predefined
bound.
In Chapter 5, an extensive set of results to evaluate the CSTS access strategy is presented.
Simulation-based experiments provided a method to explore the impacts of different parameters
on CSTS. The primary user load results showed that the protocol is able to maintain high SU
throughput under different load conditions. Results from the SU packet size experiments showed
that a smaller SU packet provides better throughput for the SUs. Varying PU traffic pattern
results showed that CSTS could dynamically adjust to the PU traffic over time. Additionally,
periodic sampling of the whitespace provides enough information to maintain good SU
throughput while maintaining PU disruption bounds. The channel sensing interval results showed
that as long as the sensing period is smaller than usable whitespaces, there is enough information
to operate the CSTS protocol. Analytical expressions for the secondary throughput and primary
disruption resulting from the proposed schemes were also developed for validating the
simulation results. Finally, the results from the effects of varying disruption bound showed that
PUs allowing a small disruption could facilitate more bandwidth scavenging by the SUs.
Chapter 6 presented a TinyOS based testbed implementation of the Bandwidth Scavenging
Concept (BSC). The BSC was applied to traffic protection in heterogeneous networks. A
Primary User Network (PUN) was implemented with the purpose of generating various traffic
148

patterns on a channel. Along with a Secondary User Network (SUN) was created to perform
measurements of the PU channel activity. Then access strategies is utilized to send messages in
between the PU transmissions. The PUN was shown to be able to create multiple traffic patterns
using a PC and TelosB mote configurations. The SUN was shown to be able to measure the PU
transmissions on the channel. Additionally, multiple SU access strategies were implemented on
the motes.
The access strategies were then evaluated using the testbed that was implemented using
TelosB motes. From the experimental results, it was shown that the access strategy was able to
support the offered SU load. Due to the limitations of the TelosB hardware, additional
simulations were run in Matlab. The simulation results showed that the CSTS protocol would
throttle back the transmissions of the SUs to protect the PU traffic. Similar to the Ad hoc 802.11
results the BSC was shown to minimize PU throughput while maximizing SU throughput.
In Chapter 7, a new and novel secondary user access strategy named Divided SU
Transmission Strategy (DSTS) was presented that provided a method to access a general PU
whitespace profile. The access strategy utilizes multiple transmission opportunities within a
whitespace to find the best times to transmit SU packets. These TOs are consolidated into a
Transmission Bitmap Vector (TBV) that defined when SUs should transmit within a whitespace
to maximize SU throughput while limiting PU disruptions to a predefined bound. Optimal and
Sub-optimal protocols are developed to select the best transmissions opportunities. Evaluation of
DSTS under different traffic profiles showed that the protocol outperforms comparison
protocols.
Chapter 8 a protocol named the Transmission Bitmap Vector Access Protocol (TBVAP)
was introduced to support access for multiple SUs in a wireless channel. The protocol provides

149

fairness in that each SU is given an equal chance to utilize a Transmission Opportunity identified
by the DSTS protocol. These features were demonstrated using results from a multiple PU
CSMA based whitespace profile.
9.2

Future Work
The channel analysis and access strategies presented in this thesis were developed for a

single wireless channel. The single channel case presented many challenges as highlighted
throughout the thesis. Another challenge that arises in this paradigm is the multiple whitespace
domains (MWD) protection problem. The MWD problem arises when there are multiple PUs
and multiple SUs spatially distributed within a given area. Research in this area has been mostly
focused in the context of multiple channels [40-42, 80-100]. In this approach, SUs have access to
two or more channels, which may or may not have PUs using the channels. Typically, the SU
solution to avoid PU disruptions is to switch channels when a PU is occupying the channel. With
PUs occupying different channels, separate whitespace domains may exist on each channel.

150

WD = Whitespace Domain

WD1

Collision Domain

SU2

SU5

SU3

WD3

WD5

PU2

PU1
SU6

WD2

SU8

SU4
SU7

SU1

WD4

PU3
Possible
PU Disruptions

Figure 9-1: Example of Multiple Whitespace Domains
Even in a single channel environment, multiple protection zones can be created for each
PU. As shown in Figure 9-1, PU1, PU2 and PU3 have individual collision domains that must be
protected from SU traffic. Consequently, this means that protecting the PUs becomes more
difficult and some additional issues arise. SUs affected by a single primary user, such as SU3,
SU5, and SU6 in whitespace domain (WD1), would have a single access strategy (i.e. TBV) to
protect the PU. When a SU is within the collision domains of multiple PUs, such as SU1, SU4,
or SU7, (WD2, WD3, and WD4 respectively) there is a different solution for each domain. Given
that SUs may be separated by multiple hops, they may not be able to hear the transmission of
151

other SUs. Thus, the combined effects of these two solutions can create more disruptions on a
PU than allowed. The objective is to develop a secondary user access strategy that allows the
protection of primary users within multiple whitespace domains.

152

APPENDIX

153

Appendix A: TelosB Based TestBed Implementation
A.1 Introduction
With Bandwidth Scavenging, a secondary user device attempts to access a wireless
channel that is occupied by a primary user device. In this appendix, the details of a TelosB based
tested is presented. The testbed is implemented as a Primary User Network (PUN) and a
Secondary User Network (SUN). The purpose of the PUN is to generate various traffic patterns
on a channel. While the purpose of the SUN is to first perform measurements of the PU channel
activity then utilize an access strategy to send messages in between the PU transmissions.
A.2 Primary User Network (PUN)
This section will discuss the details of the Primary User Network (PUN). The PUN system
components are implemented on a PC and on the motes. Recall that the purpose of PUN is to
generate various traffic patterns on a channel. The implementation provides a flexible creation of
different traffic patterns on the channel:
•

Uniform Traffic Distributions

•

Multi-Modal Traffic Patterns

•

Trace based Traffic Patterns

These traffic patterns can be created at various data rates with different variances. Figure A-1
shows an overview of the communication flow of the PUN system. On the PC side, the Traffic
Generator creates packets according to the desired traffic pattern and sends the packets to the
measurement class. The measurement class keeps track of the timing of each generated packet
and sends them to the communication interface. This interface takes care of the communication
between the PC and motes. Once a mote receives the packet from the communication interface,
154

the packet is sent on the channel. Then it is received by another mote and sent back to through
the PC-mote communication interface. The communication interface then forwards the message
back to the measurement class. From there an output will be generated detailing the statistics of
the packets. In the example figure, there are 6 traffic flows with three motes.

Traffic Generator
12 21 13 31 23

Output
32

Measurement
Communication Interface

PC Side

Com 1
MoteIF
/Source

Mote Side

Com 2
MoteIF
/Source

Com 3
MoteIF
/Source

Mote2
Mote1

Zigbee

Mote3

Figure A-1: Overview of PUN implementation
A.2.1

Mote Side Implementation
The TinyOS implementation on the mote is mostly used for forwarding messages from the

PC to the mote radio and vice versa. First, the PC sends a command to the mote, which includes
the destination mote and the payload. From there the message will be put into a queue on the
mote and dispatched as soon as possible. On the return path, the mote will receive a message,
check whether it has the correct destination and then put the message into a second queue. From
there it will be sent to the pc as soon as possible.

155

A.2.1.1

Application Layer

As discussed in the previous section, the program on the mote bridges messages from the
PC to the radio and vice versa. This bidirectional bridging behaves similar to a base station
application, where one mote receives messages in order to forward them to the PC.
To minimize the workload generated by the PC-mote communication only message type,
source, destination, id and a randomly generated payload are transferred from the pc. Figure A-2
shows the message structure on the mote. Each of these variables uses two bytes. The message
type currently not used and is kept in the implementation for potential future use. The mote can
store the message data in a queue and start dispatching from the queue if it is not already active.
The queue is implemented as a ring buffer, which has a FIFO behavior.
USB Data
Message Type

Source

Destination

Radio Data
Message Type

Source

Destination

Figure A-2: Message structure of mote
The ringbuffer consists of an array of data structures and two pointers. The data structs
contain the actual data of the message, while the pointers point data-in and data-out positions in
the array. The pointers increment and point to the different structs on the array and start again
from the beginning when they reach the end of the array.
As soon as data sending is started, it takes the first data set from the queue and generate a
new AM-TYPE message. The message contains the desired data and a dummy payload. The
length of the dummy payload can be set on compile time. In the experiments, the total payload
length was set to 60 bytes. The message id and the random generated payload use 4 bytes of
156

these 60 bytes, while the source and destination are found in the message header. The AM-TYPE
message adds one more additional byte of payload to the message after being sent to the lower
layer.
If the sending to the lower layers returns anything except “success”, the application
schedules a resend. This improves the sending reliability but does not guarantee a successful
transmission because it only handles internal resource failures but does not have a mechanism
like ACK packets.
When the lower layer signals the application layer that a message has been received from
the radio, the application layer extracts the same variables from the message that were stored in
the message previously. These variables are stored in a queue to be sent to the pc. Similar to the
path from the mote, the dispatch of the messages to the pc are activated if it is not active yet.
A.2.1.2

MAC Layer

There are two MAC layer implementations for the PUN, which take care of the dispatch of
the messages to the Physical layer. These two protocols are non-persistent CSMA with backoffs
and a simplified CSMA-Mechanism with backoffs turned off.
The non-persistent CSMA is the standard MAC-Layer implementation of the TinyOS with
the low power listening turned off. In this implementation, first the mote attempts to access the
channel and senses if it is free. To sense whether the channel is free or not, the mote takes two
clear channel assessment samples separated by a 250 µs waiting period. This waiting period is to
make sure an ACK-Turnaround period is not sampled. If the channel is busy the mote performs a
congestion back off to allow the other mote to finish their transmissions. After that time, the
mote checks again, if the channel is free. The congestion is randomized and increased for each
retry. If the mote cannot detect a free channel for several retries, it returns fail.
157

The simplified CSMA is a modified version of the non-persistent CSMA with the back
offs turned off. The simplified CSMA is implemented in an independent application platform
called telsobPU. Since the backoffs are disabled, the mote waits until 2 consecutive samples to
indicate that the channel is free. The number of required consecutive samples can be set on
compile time. If the channel is found to be free, the mote sends its message.
A.2.2

PC Side Implementation
The PC side implementation is done using Java. The implementation is done across several

functional classes as show in Figure A-3. As shown in the figure, there is a Traffic Generator
Class, a Measurement Class, a Communication Interface Class and an Output Class.
Additionally, there are classes for a graphical interface and batch processing of measurements.

Traffic Generator
12 21 13 31 23

Output
32

Measurement
Communication Interface
Com 1
MoteIF
/Source

Com 2
MoteIF
/Source

Com 3
MoteIF
/Source

Figure A-3: PC side overview
The Control Interface is shown in Figure A-4. The screen is split up into four major logical
parts. On the top left corner, there is a listing of the current configuration of the traffic profile. In
the example case shown, there are two flows sending 2000 packets per minute with a uniform
distribution and 20% variance. One flow is from mote A to B and the other one is from mote B

158

to A. The number of packets sent is split evenly between the two flows, which leave each of
them with 1000 packets per minute.
In the lower left hand corner, the current measurement statistics are shown. In this
example, the average round trip time (RTT) for a message was 81 ms and the minimum RTT has
been 14 ms while the maximum has been 1 second. On the top right field, system messages are
displayed.

Figure A-4: Primary user control interface
The lower right hand corner allows the user to control the other classes. The button “open
sources” tells the communication interface to set up the mote communication and prepare the
necessary configurations for the test run. The Start and Stop buttons control the packet
generation in the traffic generator. “Sim 10k pkgs” schedules 10,000 packets to be sent by the
traffic simulator. After a test has run, clicking the PDF button prints out the measured data. The
menu bar allows the user to access further settings like com-port setup or traffic profiles. The

159

modular design with several functional classes provides a flexible and reliable implementation.
In the following sections, a detailed description of the implemented classes are given.
A.2.2.1

Communication Interface

The Communication Interface (CI) is the central access point for data to and from the
motes. It provides functions for other classes to send data to the mote and sends received data
from the motes to the desired destination classes.
The CI provides a flexible implementation, which can access up to 12 motes. Figure A-5
shows the CI configuration screen for the Primary User. The inputs given to the CI are the
currently used COM-Ports and the flow directions that should be set. The possible options are
sending only, receiving only, and sending and receiving. From the given information, the
program calculates which flows the Traffic Generator shall set up and access the necessary
COM-ports. A list of flows and accessible COM-ports can be accessed from other classes. To
allow all the motes to be accessed concurrently and independently, a thread is generated for each
mote. Each thread accesses the mote through a mote interface provided by the TinyOS toolchain. The mote interface itself accesses a “phoenix source” which accesses a “packet source”.
The packet source access the actual comport. The reason for this layering is that with each layer
more functionality is implemented. Filtering for specific messages and signaling that a specific
message type is received are some of the features implemented by these layers.

160

Figure A-5: Primary User communication interface configuration screen
A.2.2.2

Measurement Class Implementation

The Measurement Class (MC) is responsible for keeping track of each packet and for
generating a log of the measurement runs. To provide this functionality, it keeps track of the
messages that are being sent. To keep track of these messages, the MC provides a send and
receive interface for other classes.
The MC is above the Communication Interface. The idea here is to capture the timestamps
for each packet as close to the actual dispatch time as possible. In the MC, a list of all messages
that have been sent and not returned yet is maintained. Figure A-6 shows the MC data struct. The
161

start timestamp, message type, source, destination, message ID and a two-byte random payload
are saved in the struct. The message id works as a unique identifier for the message.
Measurement class
Start Timestamp

Stop Timestamp

Difference

Data Correct

USB Data
Message Type

Source

Destination

Radio Data
MSG ID

Data

Figure A-6: Measurement data structure
The figure also shows that a larger amount of data is stored on the PC side than what is
passed down to the mote. The mote handles the messages as discussed in Section A.2.1.1. Upon
reception in the mote, the message is passed on to the output handling. A java timer task is
scheduled which regularly checks for messages. If a message is overdue, the timer task marks
them as timed out.
A.2.2.3

Traffic Generator Class

The Traffic Generator Class (TGC) creates messages according to a desired traffic profile.
The Traffic Generator is implemented in two classes, a control class and a message creation
class. The message generation can be controlled by the input form, which is displayed in Figure
A-7. Different rates, variance and probability distributions can be set. In the current
implementation, the probability distributions are uniform distribution, two multi modal versions
and trace based traffic.

162

Figure A-7: Traffic Simulator control screen
After the variables are set, the control class reads the configuration and creates a thread for
each flow. The thread is in a loop while a flag called MessageGenerationON is set to true. The
underlying control class controls this flag. Within this loop, it checks whether it is time to send
by comparing the timestamp for the next send to the current system time. If it is time to send, a
packet is generated and sent to the Measurement Class described above. After sending the
packet, the time for the next packet is calculated. Then the loop waits again until the next packet
is ready. To wait the thread is sent to sleep for one nanosecond. This is shown to be the most
efficient implementation for the java and Linux scheduler. The calculation of the next times to
send is based on the used access strategy. For the uniform traffic probability distribution, the
next sending time is calculated by calculating the average inter-packet time plus or minus the
variance percentage and adding this value to the last timestamp.
163

In the multi-modal case, there are two different implementations. The first implementation
works like the uniform probability distribution, though only a certain percentage of messages are
actually sent. The corresponding Probability Density Distribution can be seen in Figure A-8. The
other implementation is a dual version of the multi modal protocol. It works like the first version
though an “else” statement is added to the sending condition. The else statement calculates the
next value like in the uniform case and then send the message immediately. This creates a twopeaked PDF as shown in Figure A-9. This peak has a uniform distribution while the second peak
has the shape of a triangle.

PDF of Sending Schedule

Probability

1.4%
1.0%
0.6%
0.2%
0%
0

200

400

600

800

Interpacket time (ms)

Figure A-8: PDF of the first multimodal scheme

164

1000

PDF of Sending Schedule

Probability

1.4%
1.0%
0.6%
0.2%
0%
0

50

100

150

200

250

300

350

Interpacket time (ms)

Figure A-9: PDF of the second multimodal scheme
The last implemented traffic profile is a trace based packet generation. This approach takes
an actual traffic pattern as an input. This input can be derived from a video or audio stream or
any other application. These traces are then post processed and converted into inter-packet times.
These times are stored in a file, which is imported by the java program. Next, a thread is created
and the next time to send is calculated according to the next value in the tracefile. This
implementation provides a flexible and accurate execution for the generation of different traffic
patterns.
A.2.2.4

Output Class

The output class is designed to capture and pre-process the output date. The data is sent to
the output class in the format described in Section A.2.1.1. From this data, the following
indictors are created:
•

A full log of the messages for debugging and verifying purposes

•

The minimum value

•

The maximum value
165

•

The average Round Trip Time

•

Number of messages that timed out

•

Total Messages received

•

Total Messages sent

•

Messages with a Round Trip Time over 100 ms. This indicator is chosen as most
messages are within the range of 20 to 40 ms.

•
A.2.3

Probability Density Function for the Round Trip Time from 0 to 100 ms.
Traffic Generation Evaluation
In the evaluation of the system, it has proven to work reliably under most scenarios. Figure

A-10 shows a typical example of a Round Trip Time (RTT) PDF. The PDF was created with an
average inter packet time of 100 ms and a variance of 20%.
Packet loss and values over 100ms are extremely rare cases so this implementation builds a
solid base line for measurements with the secondary user. However, some issues found during
the evaluation of the implementation have to be addressed.
90

Probability

70
50
30
10
0
0

20

40
60
80
Packet RTT (ms)

100

Figure A-10: PDF of the RTT distribution of packets
166

120

While stressing the system and using multiple flows, the system shows some unwanted
behavior. Figure A-11 shows that the actual dispatch differs from the desired timings. The
different flows seem to block each other. On the top left, the scheduled messages of the first flow
are listed. The delays shown in the top right red box are all within an acceptable limits. The first
column shows the time difference between the scheduled time and the time when it is to be sent
to the mote. The second column shows the time difference between the scheduled time and the
time the java has completed the dispatch to the mote. The third column shows the delay from
before the sending to the mote to the time when the dispatch is completed. The second box on
the top shows the inter packet time of the first flow. While all these values look good, the first
flow blocks the second flow, which should be sending in between. This causes the delays to add
up to one second inter-packet time. Afterwards the second flow tries to catch up and send its
scheduled packets, so it is sending the messages to the mote as fast as it can. This leads to an
inter packet time of 15 ms instead of 100 ms +-20% variance. Additionally, the first flow gets
blocked and stops sending adding up messages in its queue.

167

Figure A-11: Example of Traffic Generator delays
The displayed case is a worst-case scenario, though it is a good example to address the
problem of the scheduler in windows and the Java Virtual Machine. Both of these schedulers
together determine the system response time. Unfortunately, these schedulers are optimized for
throughput instead of real-time applications. Several tweaks in the program code have
significantly improved the programs behavior however there is no guarantee that this does not
happen in multi flow cases.
Another issue is that the total precision of actual sending times. Figure A-12 and Figure A13 show that the scheduler has some kind of grid, which gets laid over the originally scheduled
messages. The comparison shows that the messages are only dispatched at distinct peaks with a
constant interval of roughly 10-15 ms. Figure A-14 and Figure A-15 shows the pdf of scheduled
and sent packets with a variance of 20% respectively.

168

Number of Packets

4
3
2
1
0
400

600

800

1000

1200

1400

1600

Packet RTT (ms)

Number of Packets

Figure A-12: PDF of scheduled packets at 1 pps with a variance of 20%
12
10
8
6
4
2
0
400

600

Packet RTT1000
(ms)
800
Packet RTT (ms)

1200

1400

1600

Number of Packets

Figure A-13: PDF of sent packets at 1 pps with a variance of 20%
40
30
20
10
0
0

20

40

60

80

100

120

140

Packet RTT (ms)

Figure A-14: PDF of scheduled packets at 10 pps with a variance of 20%

169

160

Number of Packets

300
200
100
0

0

20

40

60
80
100
Packet RTT (ms)

120

140

160

Figure A-15: PDF of sent packets at 10 pps with a variance of 20%
A.3 Secondary User Network (SUN)
This section discusses the details of the Secondary User Network (SUN). The SUN is
implemented on the PC and on the motes. Recall that the purpose of the SUN is to perform
measurements of the PU channel activity. Then utilize an access strategy to send messages in
between the PU transmissions.
A.3.1

RSSI Measurement
A crucial part of the proposed secondary user access strategy is the gathering of

information about the channel occupancy. In this work, the method used to collect PU channel
activity information is done via Received Signal Strength Indicator (RSSI) based sampling. The
channel is sampled with a TelosB mote running an RSSI-Sample application. The application
was contributed by Stanford University to the TinyOS contribution repositories [77]. This
application is capable of sensing the channel at a rate of up to 1.3 kHz.
Figure A-16 shows the main part of the RSSI-Sample application. When a timer interrupt
fires, it calls a “Resource.request();” command and then schedules a new alarm at

170

ALARM_PERIOD (sampling interval). This procedure causes a loop, which is repeated until the
desired number of samples is gathered.

async event void Alarm0.fired()
{
call Resource.request();
call Alarm0.start(ALARM_PERIOD);
}
Figure A-16: Pseudo-code of RSSI-Sampling
The Resource.request command produces a callback that accesses the CC2420 registers.
When this happens, the application reads the RSSI value from the register and store it in the flash
memory. The RSSI value register is being continuously updated and always provides the average
RSSI value over the last 8 symbol periods (125µs) [101]. The application sends all samples to
the PC when the sampling is done. In the proposed implementation, 1,048,576 samples can be
collected at rates up to 1.3 kHz. Table A-1 provides a brief overview about the key stats of two
proposed sampling rates. Given that, the payload length is 60 or 110 bytes while sampling the
channel, both sampling rates prove to give us valid results. The payloads correspond to 2.5 ms
and 4 ms channel occupation.
Sampling frequency

1.0 kHz

1.3 kHz

Test runtime

17:30 min

13:06 min

Sample duration

8 Symbols / 125 µs

8 Symbols / 125 µs

Inter sample time

1000 µs / 875 µs

750 µs / 525 µs

Table A-1 RSSI sampling specifications
The collected data gets post processed in Matlab, to produce a PDF of the whitespace
durations. First, each sample is determined as a blackspace or a whitespace by comparing the

171

RSSI value to a reference threshold. Then the time between each blackspace is being calculated.
Finally, the PDF and CDF of the whitespace are plotted.
40

dB

20
0
-20
-40
-60
0

500

1000
1500
Time (ms)

2000

2500

Figure A-17: RSSI Measurement example
An example of a whitespace trace is shown in Figure A-17. The peaks at +20 dB
represent the measured RSSI values for transmitted packets while the lower values at around -50
dB represent the noise floor. Note that the values of the samples are misleading as the register
values offered by the Mote have a linear offset. Figure 36 shows the relation between the RSSIregister value and the RF Level on the channel. The RSSI offset is approximately -45 dB, which
was found empirically during the development of the system [101]. For the whitespace
calculations, a comparison threshold of -40 dB has lead to a reliable detection of packets without
interpreting the noise floor or interference as packets.

172

Figure A-18: RSSI register values and actual RF level [101]
A.3.2

Evaluation of the RSSI Sampling
During the evaluation of the RSSI-sampling some minor flaws were detected. Figure A-18

shows the sampling results from packets sent at a constant inter packet time (IPT) of 500 ms. As
the IPT is constant, the single spike is consistent with the traffic we generated. To get the
whitespace duration from an IPT we have to deduct it by the time the packet occupies the
channel. In this case, the packet is 128 Bytes long and occupies the channel for 4ms. This creates
a mismatch in the timing of roughly 10%.

173

PDF of Whitespace
0.4
0.35

Probability

0.3
0.25
0.2
0.15
0.1
0.05
0
0

200
400
600
800
Whitespace Duration (ms)

1000

Figure A-19: RSSI measurement of 2 pps and a 1 kHz sampling rate
Figure A-19 shows the sampling result of a PU uniform traffic pattern that sends 2
packets per second with an variance of 20%. Similar to the previous example, an offset of 10%
can be detected. The starting and endpoint of the big block should be from 400 to 600 minus the
packet time instead of ~360ms to ~540ms. The measurement values of 800 ms to 1000 ms
suggest that there are long whitespaces, though these whitespaces had never been on the channel.
Knowing that these whitespaces never happened leads to the conclusion, that these whitespaces
was not detected by the sensing application (a similar effect can be seen in Figure A-20 with a
tiny spike at 900ms). In theory, every packet lasts 4 ms on the channel and therefore should
generate at least 3 positive samples, yet we have these results.

174

PDF of Whitespace
0.012
0.01

Probability

0.008
0.006
0.004
0.002
0
0

200

400
600
800
Whitespace Duration (ms)

1000

1200

Figure A-20: RSSI measurement of 2 pps with 20% variation and a 1 kHz sampling rate
Figure A-21 through Figure A-23 shows examples of different traffic patterns, including
uniform, video stream, and multi-modal that were measured by the RSSI-Sampling application.
In conclusion, the sampling application verifies the scheduling issues that were detected
during the development of the PUN. Additionally extensive test were made to verify the
behavior of the RSSI-Sampling application. This implementation provides reliable data about the
channel occupation characteristics and the identified flaws can be taken into account while
processing the data.

175

Figure A-21: RSSI measurement of Uniform distribution with 7 pps with 20% variation

Figure A-22: RSSI measurement of Video Stream profile

176

Figure A-23: RSSI measurement of Multi-model traffic profile
A.3.3

Mote Side Implementation
In this section, the SUN implementation is provided. The SUN implementation on the

motes requires heavy modifications in the TinyOS platform. Flexible control of the motes for
multiple access strategies requires tight synchronization between the application and network
layers.
The application layer processes four main tasks:
•

Control-Interface

•

Packet generation

•

Packet queuing

•

Access strategy (Packet scheduling according to desired MAC-Layer)

The network layer of the TinyOS platform performs the actual sending of packets. This network
layer is kept very simple as many MAC layer functionalities were moved up to the application
layer within the mote.
Figure A-24 shows the overall information flow in the mote. The packet generator
schedules packets into the queue. Then the access strategy takes the messages from the queue
177

and sends it. All three steps are configured and controlled by the mote control interface, which
communicates with the Java Control Interface on the PC. Observe that this process is different
from the process used in the PUN. The purpose of this configuration is because the SUN must
support and access strategy, while the PUN only needs to send messages.

λ-Package generation

Packet Queue

Access Strategy

Mote Control Interface

Application Layer

PC Side
Java
Control
Interface

Access Strategy

Network Layer
Figure A-24: Overview of SUN implementation
A.3.3.1

Application Layer

The procedures on the application layer are handled by two-timer loops. The first timer
loop is the packet generation. During the startup of the mote, the SchedulerAlarm is started while
messageGenerationActive is set to false. Figure A-25 shows the code. The Alarm fires although
nothing happens until messageGenerationActive is set to true by the control interface.

178

Async event void SchedulerAlarm.fired (){
uint16_t time;
if (messageGenerationActive){
atomic{
messageQueueC++;
if
(messageQueueC>PACKAGEGENERATION_MAXQUEUESIZE){
messageQueueC=PACKAGEGENERATION_MAXQUEUESIZE;
setMessageGenerationActive(FALSE);
}
}
}
time = messageGenerationRate;
//time = call Random.rand16();
call SchedulerAlarm.start(time);
}

Figure A-25: Pseudo code of packet generation
Once messageGenerationActive is set, the program starts adding messages to the packet
queue. Unlike regular queues, this queue does not store actual messages, but the number of
messages in the queue. This slim implementation is sufficient, as actual data is not transmitted.
In case of too many packets are being scheduled, there is a queue overflow check comparing the
message

queue

messageQueueC

to

a

maximum

queue

size

PACKETGENERATION_MAXQUEUESIZE.
The queue, which is filled by the packet generator, is the common interface between the
packet generation and the access strategy that is being tested. The queue is implemented as a
simple variable integer, representing the number of messages in the queue. Each time the mote
sends a message it decreases the message queue counter. The second timer loop takes care of the
scheduling of the message on the channel. The timer loop continuously checks whether there are
messages in the queue that should be sent. Two core principles handle the behavior of the mote’s
access strategy. One is the timer mentioned and the other is the m_state variable. The m_state
variable defines what has to be done when the second timer is fired.
179

Table A-2 shows an overview of the states and the access strategies. Currently the
Arbitrary InterFrame Spacing (AIFS) and the CSTS protocols are implemented. The state of the
mote is also the interface for the control interface. By changing the state, the desired access
strategy can be activated. The actual change does not happen immediately, but on the next call of
the handleStates procedure.
Access strategy

States

Idle, do not Access

S_READY

AIFS

S_SEND_EPSILON

CSTS

S_SEND_MU, S_SEND_J,
S_SENSE_END_OF_WHITESPACE_HIST
ORY_ONE
Table A-2: Access strategies and states

A.3.3.1.1 AIFS Access Strategy
The Arbitrary Inter-Frame Spacing (AIFS) access strategy is inspired by the 802.11e
prioritization scheme [18]. The prioritization of the traffic is implemented through using
different AIFS. In the implementation, the waiting period is called epsilon and it is assumed that
there is always a message to send. If there is no message to send, the mote waits for a short
period and check again if a message has been generated. Before the sending of each message, the
mote checks whether the channel is free for a period of epsilon. As soon as this occurs, it
transmits the message.

180

A.3.3.1.2 CSTS Access Strategy
The CSTS access strategy accesses the channel according to the primary user traffic
pattern. Therefore, the access strategy starts when a whitespace is detected. Figure A-26 shows a
flow chart of CSTS with a filled queue. The code also checks if there is a message to send before
each sending command.
In CSTS, at the start of a whitespace, transmissions are delayed for a µ period. The
duration of µ is based on the statistics of the channel. After the channel is free for a µ duration
the first packet is sent. Afterwards, the next packet is sent if the duration in the whitespace, t is
smaller than tmax. tmax is calculated using the process as Jmax detailed in Chapter 4 and then
converted into packet numbers. Therefore, the actual check is done against the number of packets
that have been sent into that whitespace rather than comparing actual time values. Before the
packet is sent, the mote checks whether the channel is free for an epsilon period. Once this
message is sent the program returns to the t<tmax check as shown in Figure A-26.

181

Figure A-26: Flow chart of CSTS implementation
Both the AIFS and CSTS access strategies require sensing of the channel to detect when
it is free for an µ or epsilon period. The implementation of the waiting for the beginning of the
next whitespace is done by a procedure, which is sampling the Clear Channel Assessment pin of
the CC2420 chip. If the CCA-pin indicates that the channel is empty, the mote continues waiting
until the PU returns. If the CCA-pin indicates that, the channel is busy and has indicated that it
was busy for the last X samples, the primary user is assumed to have returned. Therefore, the
beginning of a new whitespace is expected. The requirement for having multiple positive
182

samples is because the CCA-pin detects any energy on the channel and interference shall not
trigger the Secondary User to send his messages. These functions are implemented in the
Network Layer and is described in the next section.
A.3.3.1.3 DSTS Access Strategy
The DSTS access strategy is able to access more complex whitespace traces as discussed
in Chapter 7:. Similar to the CSTS access strategy, DSTS starts at the beginning of a whitespace.
Since time is divided into timeslots, DSTS first checks if the current whitespace-slot is a transmit
slot based on the Transmission Bitmap Vector (TBV). The mote calculates whether to send in the
current timeslot by accessing file that contains the TBV assignments for each slot. If the current
slot is not used for sending a packet, the mote waits for the next timeslot. After the program has
sent a packet or not, it monitors the channel in a similar fashion as the sensing for the end of a
whitespace protocol described above. The duration of a timeslot can be set on runtime, though a
slot time of 20ms was used for all test runs as it works reliably and is sufficiently slow.
If the primary user returns to the channel, the mote waits for the beginning of the next
whitespace. Otherwise, it checks if the current timeslot is larger than the maximum timeslot.
Figure A-27 shows the flowchart of the DSTS implementation.

183

Figure A-27: Flow chart of DSTS implementation
A.3.3.2

Network Layer
The modifications to the Network layer are implemented on the two lowest layers on top

of the Hardware Abstraction Architecture (HAA). The application layer sends the Network
Layer arguments as to how it should process a message. Network access has been simplified
from its original version to grant the upper layers as fast access to the channel as possible. The
original non-persistent CSMA-CA network access has been deactivated and two CSMA
algorithms have been implemented. The original CSMA-CA code was the original file for the
network access. Many parts of the original code have been kept, especially the parts that handle
concurrency, write into the CC2420 mote’s registers and the mote’s HAA interfaces.

184

The first CSMA version is used by the µ and epsilon packets of the previously described
access strategies. Initially, it writes the message on to the mote and prepares it for sending. After
the message is on the mote and ready to send it samples the channel to ensure, that it is free for a
certain amount of time (epsilon or µ). If one of samples returns busy, the counter is reset. This
procedure continues until the message is sent on the channel by strobing a pin connected to the
CC2420.
The j-packets of CSTS also use the second version of the CSMA algorithm. It is identical
to the first one except for the sampling procedure. If a sample returns busy the sending process is
stopped and a fail signal is returned to the upper layers to signal that the primary user has
returned.
A.3.4

PC Side Implementation
The destination SU mote receives all messages from the PU and the SU motes. All

messages are then forwarded to the PC. A modified version of the BaseStation application
contributed to the TinyOS is used [102]. The application has been modified to have a longer
queue and to support a longer payload. The reason for forwarding all messages instead of just SU
messages is to create a full log of all messages. This log is maintained by the Java control
interface described below along with the computation of the packets loss and the SU throughput.
A.3.4.1

SU Control Interface

The Secondary User Control interface provides a convenient way of scheduling, measuring
and reporting test runs of the access strategies. Figure A-28 shows the Secondary User Control
Interface. In the top left corner, the buttons are the buttons to control the access strategy that is
used by the mote. Right below that, the report controls are grouped. On the top right corner is a
185

text field, which displays the report. In the bottom left corner are the control buttons for the
packet scheduling and in the bottom right corner are the controls for the access strategy settings.

Figure A-28: Secondary user control interface
All command buttons, which changes the behavior of the mote, send a command message
over the serial port to the SUTX mote. The changes is set on the mote runtime as soon as the
message is received. The access strategy controls send a command message to change the
m_state variable of the mote. The change of m_state forces the mote to go into the loop of the
desired access strategy. The commands for the packet scheduling accesses the loop timing of the
access strategy and the Boolean value, which decides whether packets are being scheduled or
not. The controls on the bottom allow changing the parameters of the access strategies. The

186

button Set Jmax is the corresponding variable for the calculated time tmax in subsequently sent
packets.
The report functions are only processed on the mote. A click on the “Request Report”
button generates a full report from the stored data and display it in the textbox. Indicators like SU
throughput, PU throughput, scheduled SU messages and a full log of all messages sent during the
experiments are part of this report. The full report in the window and manually added comments
can be saved to a file by clicking the Save report button.
A.3.4.2

Data Collection
The data collection is implemented by directly accessing the packet source. The SURX

forwards all messages to the PC and the PC stores all these messages are in an ArrayList during
an experiment. Every time the Java interface receives a message, it updates the internal counters
for the measured data.
The application distinguishes between Primary User Network and Secondary User
Network by their AM_TYPE. The measured Secondary User throughput is calculated by
counting the messages and analyzing the messageDispatchCounter. This counter increments for
each message that is scheduled to be sent by the SUTX. The number of missing packages is
derived from missing “increments” in the messageDispatchCounter. After each experiment, the
results can be validated by comparing the number of received and missing packages to the
number of packets that had been scheduled by the traffic generator.

187

Appendix B: LIST OF PUBLICATIONS
Peer Reviewed Journals:
1. A. Plummer, M. Taghizadeh, and S. Biswas, “Measurement based Bandwidth
Scavenging in Wireless Networks”, In Press, IEEE Transactions on Mobile Computing,
2011.
2. A. Plummer, and S. Biswas, “Distributed Spectrum Assignment for Cognitive Networks
with Heterogeneous Spectrum Opportunities”, In Press, Wiley Journal of Wireless
Communications and Mobile Computing, 2011.
3. A. Plummer, M. Taghizadeh, and S. Biswas, “Model Based Bandwidth Scavenging for
Device Coexistence in Wireless LANs”, Submitted, Elsevier Pervasive and Mobile
Computing Journal, Invited based on the paper presented in ICDCN 2011.
4. A. Plummer, M. Taghizadeh and S. Biswas, “Supporting Prioritized Device Coexistence
in the Presence of Multiple Secondary Users”, under preparation, IEEE
Communications.
Peer Reviewed Book Chapter:
5. A. Plummer, M. Taghizadeh, and S. Biswas, “Traffic Protection in Heterogeneous
Sensor Networks”, In Press, Wireless Multi-Access Environments and Quality of Service
Provisioning: Solutions and Application. IGI Global.
Peer Reviewed Conferences:
6. A. Plummer, M. Taghizadeh, and S. Biswas, “Model Based Bandwidth Scavenging for
th
Device Coexistence in Wireless LANs”, In proceedings of the 11 International
Conference on Distributed Computing and Networking (ICDCN), January 2011, Kolkata,
India.
7. A. Plummer, M. Taghizadeh and S. Biswas, “Statistical Bandwidth Scavenging for
th
Prioritized Device Coexistence”, In proceedings of the 29 IEEE International
Performance Computing and Communications Conference (IPCCC), December 2010,
Albuquerque, New Mexico.
8. A. Plummer M. Schmidt, M. Taghizadeh, and S. Biswas, “Traffic Protection via
rd
Bandwidth Scavenging in Heterogeneous Sensor Networks”, In proceedings of the 53
IEEE Global Communications Conference (Globecom), December 2010, Miami, Florida.
9. M. Taghizadeh, A. Plummer, and S. Biswas, “Towards Optimal Cooperative Caching in
rd
Social Wireless Networks”, In proceedings of the 53 IEEE Global Communications
Conference (Globecom), December 2010, Miami, Florida.
188

10. M. Taghizadeh, A. Plummer, and S. Biswas, “Cooperative Caching for Improving
th
Availability in Social Wireless Networks”, In proceedings of the 7 IEEE International
Conference on Mobile Ad-hoc and Sensor Systems (MASS), November 2010, San
Francisco, California.
11. A. Plummer, M. Taghizadeh, and S. Biswas, “Measurement based Capacity Scavenging
nd
via Whitespace Modeling in Wireless Networks” In proceedings of the 52 IEEE Global
Communications Conference (Globecom), December 2009, Honolulu, Hawaii, (Won
Best Paper Award).
12. A. Plummer, and S. Biswas, “Capacity Scavenging in Wireless Networks: A
th
Comparative Study”, In proceedings of the 28 IEEE Military Communications
Conference (MILCOM), October 2009, Boston, Massachusetts.
13. M. Quwaider, A. Plummer, J. Rao, M. Taghizadeh, and S. Biswas, “Real-time Posture
th
Detection using Body Area Sensor Networks”, In proceedings of the 13 International
Symposium on Wearable Computers (ISWC), September 2009, Linz, Austria.
14. A. Plummer, T. Wu, and S. Biswas, “A Localized and Distributed Channel Assignment
Framework for Cognitive Radio Networks”, In proceedings of the ACM International
Workshop on Cognitive Wireless Networks (CWNets), August 2007, Vancouver,
Canada.
15. A. Plummer, T. Wu, and S. Biswas, “A Cognitive Spectrum Assignment Protocol using
th
Distributed Conflict Graph Construction”, In proceedings of the 26 IEEE Military
Communications Conference (MILCOM), October 2007, Orlando, Florida.
16. Q. Huo, S. Biswas and A. Plummer, “Towards a Pulse Switching Protocol for Event and
th
Target Tracking using Ultra Wide Band Impulse Radio”, under review, 8 IEEE Sensor,
Mesh and Ad Hoc Communications and Networks (SECON), June 2011, Salt Lake City,
Utah.

189

BIBLIOGRAPHY

190

Bibliography

[1]

S. Geirhofer, L. Tong, and B. M. Sadler, "Dynamic Spectrum Access in WLAN
Channels: Empirical Model and Its Stochastic Analysis," in ACM TAPAS, 2006.

[2]

V. R. Petty, R. Rajbanshi, D. Datla, W. Frederick, and et al., "Feasibility of Dynamic
Spectrum Access in Underutilized Television Bands," in Symposium on Dynamic
Spectrum Access Networks (DySPAN), 2007.

[3]

I. F. Akyildiz, W. Lee, M. C. Vuran, and S. Mohanty, "Next generation/dynamic
spectrum access/cognitive radio wireless networks: A survey," Computer Networks
Journal (Elsevier), 2006.

[4]

N. Golmie, N. Chevrollier, and O. Rebala, "Bluetooth and WLAN Coexistence:
Challenges and Solutions," IEEE Trans. Wireless Communication, vol. 10, pp. 22-29,
Dec. 2003.

[5]

C. F. Chiasserini and R. R. Rao, "Coexistence mechanisms for interference mitigation
between IEEE 802.11 WLANs and Bluetooth," in INFOCOM, 2002.

[6]

S. Geirhofer, L. Tong, and B. M. Sadler, "Cognitive Medium Access: Constraining
Interference Based on Experimental Models," IEEE Jrnl on Selected Areas in
Communications, vol. 26, no. 1, Jan. 2008.

[7]

C. F. Chiasserini and R. R. Rao, "Coexistence mechanisms for interference mitigation in
the 2.4-GHz ISM band " IEEE Transactions on Wireless Communications, vol. 2, no. 5,
2003.

[8]

L. Jingli, L. Xiangqian, and A. Swami, "Collision analysis for coexistence of multiple
bluetooth piconets and WLAN with dual channel transmission," IEEE Transactions on
Communications, vol. 57, pp. 1129-1139, 2009.

[9]

I. Stojmenovic, S. Giordano, M. Conti, and S. Basagni, "Mobile Ad Hoc Networking,"
IEEE, 2005, pp. 175 -203.

[10]

J. Mietzner, L. Lampe, and R. Schober, "Distributed transmit power allocation for
multihop cognitive-radio systems," IEEE Transactions on Wireless Communications, vol.
8, pp. 5187-5201, 2009.

[11]

R. Ahuja, R. Corke, and A. Bok, "Cognitive Radio System using IEEE 802.11a over
UHF TVWS," in Symposium on Dynamic Spectrum Access Networks (DySPAN), 2008,
pp. 1-9.
191

[12]

Y. Chen, Z. Qing, and A. Swami, "Joint Design and Separation Principle for
Opportunistic Spectrum Access in the Presence of Sensing Errors," IEEE Transactions
on Information Theory, vol. 54, no. 5, May 2008.

[13]

S. Huang, X. Liu, and Z. Ding, "On Optimal Sensing and Transmission Strategies for
Dynamic Spectrum Access," in New Frontiers in Dynamic Spectrum Access Networks.
DySPAN 2008., Oct. 2008.

[14]

S. Huang, X. Liu, and Z. Ding, "Optimal Transmission Strategies for Dynamic Spectrum
Access in Cognitive Radio Networks," IEEE Transactions on Mobile Computing, vol. 8,
pp. 1636-1648, 2009.

[15]

B. Wang, Z. Ji, and K. J. R. Liu, "Primary-Prioritized Markov Approach for Dynamic
Spectrum Access," in Dynamic Spectrum Access Networks (DySPAN), 2007, pp. 507515.

[16]

S. Geirhofer, L. Tong, and B. M. Sadler, "Dynamic Spectrum Access in WLAN
Channels: Empirical Model and Its Stochastic Analysis," in ACM Workshop on
Technology and Policy for Accessing Spectrum (TAPAS), August 2006.

[17]

S. Geirhofer, T. Lang, and B. M. Sadler, "A Measurement-Based Model for Dynamic
Spectrum Access in WLAN Channels," in Military Communications Conference, 2006.
MILCOM 2006. IEEE, 2006, pp. 1-7.

[18]

S. Mangold, C. Sunghyun, G. R. Hiertz, O. Klein, and B. Walke, "Analysis of IEEE
802.11e for QoS support in wireless LANs," Wireless Communications, IEEE, vol. 10,
pp. 40-50, 2003.

[19]

M. Mishra and A. Sahoo, "A Contention Window Based Differentiation Mechanism for
providing QoS in Wireless LANs," in 9th International Conference on Information
Technology, (ICIT '06), 2006, pp. 72-76.

[20]

C. R. Stevenson, W. K. C. Wireless, G. Chouinard, W. Hu, S. J. Shellhammer, W.
Caldwell, and F. T. Group, "IEEE 802.22: The first cognitive radio wireless regional area
network standard," IEEE Communications Magazine, vol. Vol. 47, 1, , 2009.

[21]

L. Hongjun, L. Xun, J. Shen, and M. Hongxu, "A Fair Multi-priority MAC Protocol
Design of Wireless Sensor Networks," in Networks Security, Wireless Communications
and Trusted Computing, (NSWCTC '09), 2009, pp. 455-458.

[22]

P. Lei, W. Hongyi, and T. Nian-Feng, "An Efficient and Scalable Prioritized MAC
Protocol (PMAC) for Backbone Communication in Wireless Sensor Networks," in
Sensor Technologies and Applications (SENSORCOMM '09.), 2009, pp. 508-513.

192

[23]

E. Jung and L. Xin, "Opportunistic Spectrum Access in Heterogeneous User
Environments," in New Frontiers in Dynamic Spectrum Access Networks. DySPAN
2008., Oct. 2008.

[24]

I. Stanojev, O. Simeone, Y. Bar-Ness, and T. Yu, "Spectrum Leasing via Distributed
Cooperation in Cognitive Radio," in IEEE International Conference on Communications.
ICC '08. , May 2008.

[25]

A. O. Ercan, L. Jiwoong, S. Pollin, and J. M. Rabaey, "A Revenue Enhancing
Stackelberg Game for Owners in Opportunistic Spectrum Access," in New Frontiers in
Dynamic Spectrum Access Networks, (DySPAN 2008), Oct. 2008.

[26]

Z. Qu, Z. Qin, J. Wang, L. Luo, and Z. Wei, "A cooperative game theory approach to
resource allocation in cognitive radio networks," in Conference on Information
Management and Engineering (ICIME), 2010.

[27]

L. Yi-Bing, Y. Rui, and Y. Fang, "Non-cooperative spectrum allocation based on game
theory in cognitive radio networks," in Conference on Bio-Inspired Computing: Theories
and Applications (BIC-TA), 2010, pp. 1134-1137.

[28]

E. Del Re, G. Gorni, L. Ronga, and R. Suffritti, "A power allocation strategy using Game
Theory in Cognitive Radio networks," in Game Theory for Networks, 2009, pp. 117-123.

[29]

Y. B. Reddy and C. Bullmaster, "Application of Game Theory for Cross-layer Design in
Cognitive Wireless Networks," in Information Technology: New Generations, 2009, pp.
510-515.

[30]

L. Yun, P. Qi-Cong, S. Huai-Zong, C. Xing-Feng, and W. Ling, "Power control
algorithm based on game theory in cognitive radio networks," in Apperceiving
Computing and Intelligence Analysis (ICACIA), 2010, pp. 164-168.

[31]

C.-g. Yang, J.-d. Li, and W.-y. Li, "Joint rate and power control based on game theory in
cognitive radio networks," in Communications and Networking in China (ChinaCOM),
2009, pp. 1-5.

[32]

L. You-En, L. Kun-Hsing, and H. Hung-Yun, "Design of Power Control Protocols for
Spectrum Sharing in Cognitive Radio Networks: A Game-Theoretic Perspective," in
International Conference on Communications, 2010, pp. 1-6.

[33]

Y. Ge, Y. Sun, S. Lu, and E. Dutkiewicz, "ADSD: An Automatic Distributed Spectrum
Decision method in Cognitive Radio networks," in First International Conference on
Future Information Networks, 2009.

[34]

D. Teo, K. Zhong, and B. C. Ng, "An Iterative Threshold Selection Algorithm for
Cooperative Sensing in a Cognitive Radio Network," in Dynamic Spectrum Access
Networks (DySPAN), 2010, pp. 1-8.
193

[35]

R. Urgaonkar, S. Member, M. J. Neely, and S. Member, "Opportunistic Scheduling with
Reliability Guarantees in Cognitive Radio Networks," Optimization, vol. 8, pp. 766-777,
2009.

[36]

Y. C. Liang, Y. Zeng, E. Peh, and A. T. Hoang, "Sensing-Throughput Tradeoff for
Cognitive Radio Networks," in IEEE International Conference on Communications,
2007, pp. 5330-5335.

[37]

S. Huang, X. Liu, and Z. Ding, "Short Paper: On Optimal Sensing and Transmission
Strategies for Dynamic Spectrum Access," in Dynamic Spectrum Access Networks
(DySPAN), 2008, pp. 8-12.

[38]

L. Cao, H. Zheng, and S. Barbara, "Stable and Efficient Spectrum Access in Next
Generation Dynamic Spectrum Networks," in INFOCOM, 2008, pp. 1543-1551.

[39]

J. Jia, Q. Zhang, and X. Shen, "HC-MAC: A Hardware-Constrained Cognitive MAC for
Efficient Spectrum Management," IEEE Journal on Selected Areas in Communications,
vol. 26, 2008.

[40]

M. Thoppian, S. Venkatesan, R. Prakash, and R. Chandrasekaran, "MAC-Layer
Scheduling in Cognitive Radio based Multi-Hop Wireless Networks," in WoWMoM '06,
2006.

[41]

S. Eljack, A. Igbal, and W. Furong, "A Multi Channel Cognitive MAC Protocol with
Efficient Channel Reservation and Collision Avoidance Method," in Conference on
Multimedia Information Networking and Security (MINES), 2009.

[42]

N. Baldo, A. Asterjadhi, and M. Zorzi, "Multi-channel medium access using a virtual
network coded control channel," in International Wireless Communications and Mobile
Computing Conference (IWCMC), 2009.

[43]

Q. Chen, Y. Liang, M. Motani, and W. Wong, "CR-CSMA: A Random Access MAC
Protocol for Cognitive Radio Networks," in IEEE 20th International Symposium on
Personal, Indoor and Mobile Radio Communications, 2009.

[44]

E. C. Y. Peh, Y.-C. Liang, Y. L. Guan, and Y. Zeng, "Optimization of Cooperative
Sensing in Cognitive Radio Networks: A Sensing-Throughput Tradeoff View," IEEE
Transactions on Vehicular Technology, vol. 58, pp. 5294-5299, 2009.

[45]

O. Fatemieh, R. Chandra, and C. a. Gunter, "Secure Collaborative Sensing for Crowd
Sourcing Spectrum Data in White Space Networks," Dynamic Spectrum Access Networks
(DySPAN), pp. 1-12, 2010.

194

[46]

V. R. Petty, R. Rajbanshi, D. Datla, W. Frederick, D. Daniel, and et al., "Feasibility of
Dynamic Spectrum Access in Underutilized Television Bands," in New Frontiers in
Dynamic Spectrum Access Networks. DySPAN 2007., Apr. 2007.

[47]

G. J. Minden, J. B. Evans, L. S. Searl, D. Depardo, R. Rajbanshi, J. Guffey, T. R.
Newman, V. R. Petty, F. Weidling, M. Peck, B. Cordill, and D. Datla, "An Agile Radio
for Wireless Innovation," IEEE Communications Magazine, pp. 113-121, 2007.

[48]

H. Harada, H. Fujii, T. Furuno, S. Miura, and T. Ohya, "Iterative Cyclostationarity-Based
Feature Detection of Multiple Primary Signals for Spectrum Sharing Scenarios," in
Dynamic Spectrum Access Networks (DySPAN), 2010, pp. 1-8.

[49]

H. Li, "Learning the Spectrum via Collaborative Filtering in Cognitive Radio Networks,"
in Dynamic Spectrum Access Networks (DySPAN), 2010, pp. 1-12.

[50]

J. Nasreddine, J. Riihijarvi, and P. Mahonen, "Location-Based Adaptive Detection
Threshold for Dynamic Spectrum Access," in Dynamic Spectrum Access Networks
(DySPAN), 2010, pp. 1-10.

[51]

L. Yu, W. Zhang, S. J. Shellhammer, and B. D. Rao, "Noncoherent Diversity Combining
for Spectrum Sensing," in Dynamic Spectrum Access Networks (DySPAN), 2010, pp. 1-7.
O. In, "Primary User Behavior in Cellular Networks and Implications for Dynamic
Spectrum Access," IEEE Communications Magazine, vol. 47, 3,, pp. 88-95, 2009.

[52]
[53]

T. V. Nguyen and F. Baccelli, "A Probabilistic Model of Carrier Sensing Based
Cognitive Radio," in Dynamic Spectrum Access Networks (DySPAN), 2010, pp. 1-12.

[54]

R. Tandra, A. Sahai, and V. V. Veeravalli, "Space-Time Metrics for Spectrum Sensing,"
in Dynamic Spectrum Access Networks (DySPAN), 2010, pp. 1-12.

[55]

H. Kim and K. G. Shin, "Fast Discovery of Spectrum Opportunities in Cognitive Radio
Networks," in New Frontiers in Dynamic Spectrum Access Networks. DySPAN 2008.,
Oct. 2008.

[56]

S. Huang, X. Liu, and Z. Ding, "Opportunistic Spectrum Access in Cognitive Radio
Networks," in IEEE INFOCOM 2008, April 2008.

[57]

"IEEE std. 802.11 - 2007: Wireless LAN Medium Access Control (MAC) and
Physical
Layer
(PHY)
specifications,"
http://standards.ieee.org/getieee802/download/802.11-2007.pdf.

[58]

A. M. R. Slingerland, P. Pawelczak, R. V. Prasad, A. Lo, and R. Hekmat, "Performance
of Transport Control Protocol Over Dynamic Spectrum Access Links," in DySPAN,
2007, pp. 486-495.

195

[59]

D. Cabric, S. M. Mishra, and R. W. Brodersen, "Implementation Issues in Spectrum
Sensing for Cognitive Radios," in Conference on Signals, Systems and Computers, Nov.
2004.

[60]

P. D. Sutton, B. Ozgul, K. E. Nolan, and L. E. Doyle, "Bandwidth-Adaptive Waveforms
for Dynamic Spectrum Access Networks," in IEEE Dynamic Spectrum Access Networks
(DySPAN), 2008, pp. 1-7.

[61]

Q. Zhao, S. Geirhofer, L. Tong, and B. M. Sadler, "Optimal Spectrum Access via
Periodic Channel Sensing," in IEEE Wireless Communications and Networking
Conference (WCNC), March 2007.

[62]

S. Geirhofer, L. Tong, and B. M. Sadler, "Cognitive Medium Access: Constraining
Interference Based on Experimental Models," IEEE Jrnl on Selected Areas in
Communications, vol. 26, no. 1, Jan. 2007.

[63]

Maxim,
"2.4GHz
802.11b
Zero-IF
ic.com/en/ds/MAX2820-MAX2821.pdf.

[64]

Intersil,
"Direct
Sequence
Spread
Spectrum
Baseband
http://www.datasheetarchive.com/datasheet-pdf/012/DSA00211974.html.

[65]

Crossbox,
"TelosB
mote
http://www.xbow.com/Products/productdetails.aspx?sid=252.

[66]

Z. Karakehayov, "Adaptive Medium Access for Wireless Sensor Networks," in IEEE
Workshop on Intelligent Data Acquisition and Advanced Computing Systems:
Technology and Applications, 2007.

[67]

M. Jims, N. Sarma, and S. Nandi, "Priority Based Fairness Provisioning QoS-Aware
MAC Protocol," in International Conference on Advanced Computing and
Communications, 2007, pp. 593-598.

[68]

A. M. Firoze, L. Y. Ju, and L. M. Kwong, "PR-MAC A Priority Reservation MAC
Protocol For Wireless Sensor Networks," in International Conference on Electrical
Engineering, (ICEE '07), 2007, pp. 1-6.

[69]

G. J. Minden, J. B. Evans, L. Searl, D. DePardo, V. R. Petty, R. Rajbanshi, T. Newman,
Q. Chen, F. Weidling, J. Guffey, D. Datla, B. Barker, M. Peck, B. Cordill, a. M.
Wyglinski, and a. Agah, "KUAR: A Flexible Software-Defined Radio Development
Platform," in Dynamic Spectrum Access Networks (DySPAN): Ieee, 2007, pp. 428-439.

[70]

M. B. H. Weiss, S. Delaere, and W. H. Lehr, "Sensing as a Service: An Exploration into
Practical Implementations of DSA," in Dynamic Spectrum Access Networks (DySPAN).
vol. 15260, 2010, pp. 1-8.

196

Transceivers.,"

http://pdfserv.maximProcessor,"
platform,"

[71]

H. Harada, "A small-size software defined cognitive radio prototype," in Symposium on
Personal, Indoor and Mobile Radio Communications, 2008, pp. 1-5.

[72]

H. Harada, "Software defined radio prototype toward cognitive radio communication
systems," in Dynamic Spectrum Access Networks (DySPAN), 2005, pp. 539-547.

[73]

Z. Yan, Z. Ma, H. Cao, G. Li, and W. Wang, "Spectrum Sensing, Access and
Coexistence Testbed for Cognitive Radio using USRP," in Conference on Circuits and
Systems for Communications, 2008, pp. 270-274.

[74]

R. Dhar, G. George, A. Malani, and P. Steenkiste, "Supporting Integrated MAC and PHY
Software Development for the USRP SDR," in IEEE Workshop on Networking
Technologies for Software Defined Radio Networks, 2006, pp. 68-77.

[75]

Z. Miljanic, I. Seskar, K. Le, and D. Raychaudhuri, "The WINLAB Network Centric
Cognitive Radio Hardware Platform, WiNC2R," in Conference on Cognitive Radio
Oriented Wireless Networks and Communications, 2007, pp. 155-160.

[76]

Z. Tong, M. S. Arifianto, and C. F. Liau, "Wireless transmission using universal software
radio peripheral," in Conference on Space Science and Communication, 2009, pp. 19-23.

[77]

Stanford SING, "SourceForge.net Repository, Retrieved October 21st, 2009, from
http://tinyos.cvs.sourceforge.net/viewvc/tinyos/tinyos-2.x-contrib/stanford-sing/," 2009.

[78]

F. Österlind and A. Dunkels, "Approaching the maximum 802.15.4 multi-hop
throughput," in HotEmNets, 2008.

[79]

P. Toth, "Dynamic programming algorithms for the Zero-One Knapsack Problem,"
Computing, vol. 25, pp. 29-45, 1980.

[80]

X. Liu, B. Krishnamachari, and H. Liu, "Channel Selection in Multi-Channel
Opportunistic Spectrum Access Networks with Perfect Sensing," in Dynamic Spectrum
Access (DySPAN), 2010.

[81]

K. Ghaboosi, A. B. MacKenzie, L. a. DaSilva, A. S. Abdallah, and M. Latva-Aho, "A
Channel Selection Mechanism based on Incumbent Appearance Expectation for
Cognitive Networks," in IEEE Wireless Communications and Networking Conference,
2009.

[82]

C. Cordeiro and K. Challapali, "C-MAC: A Cognitive MAC Protocol for Multi-Channel
Wireless Networks," in Dynamic Spectrum Access Networks (DySPAN), 2007.

[83]

Y. R. Kondareddy, P. Agrawal, and K. Sivalingam, "Cognitive Radio Network setup
without a Common Control Channel," in Military Communications Conference
(MILCOM), 2008.

197

[84]

K.-l. A. Yau, P. Komisarczuk, and P. D. Teal, "A Context-aware and Intelligent Dynamic
Channel Selection Scheme for Cognitive Radio Networks," in Cognitive Radio Oriented
Wireless Networks and Communications (CROWNCOM), 2009.

[85]

H. Su and X. Zhang, "CREAM-MAC: An Efficient Cognitive Radio-EnAbled MultiChannelMAC Protocol forWireless Networks," in World of Wireless, Mobile and
Multimedia Networks (WoWMoM), 2008.

[86]

H. Su and X. Zhang, "Design and analysis of a multi-channel cognitive MAC protocol for
dynamic access spectrum networks," in Military Communications Conference
(MILCOM), 2008.

[87]

S. Zheng, Y. Liang, C. Tham, and P. Kam, "Design of MAC with cooperative spectrum
sensing in ad hoc cognitive radio networks," in IEEE 20th International Symposium on
Personal, Indoor and Mobile Radio Communications, 2009.

[88]

M. Wellens, J. Riihijarvi, and P. Mahonen, "Evaluation of Adaptive MAC-Layer Sensing
in Realistic Spectrum Occupancy Scenarios," in Dynamic Spectrum Access Networks
(DySPAN), 2010.

[89]

L. Cao, L. Yang, and H. Zheng, "The Impact of Frequency-Agility on Dynamic Spectrum
Sharing," in Dynamic Spectrum Access Network (DySPAN), 2010.

[90]

Y. Gai, B. Krishnamachari, and R. Jain, "Learning Multiuser Channel Allocations in
Cognitive Radio Networks: A Combinatorial Multi-Armed Bandit Formulation," in
Dynamic Spectrum Access Networks (DySPAN), 2010, pp. 1-9.

[91]

Y. Ke, "MCR-MAC: Multi-channel cognitive radio MAC protocol for cooperative
incumbent system protection in wireless ad-hoc network," in Ubiquitous and Future
Networks Conference, 2009.

[92]

T. Shu, S. Cui, and M. Krunz, "Medium Access Control for Multi-Channel Parallel
Transmission in Cognitive Radio Networks," in Globecom, 2006, pp. 1-5.

[93]

F. Wang and M. Krunz, "Multi-channel spectrum-agile MAC protocol with adaptive load
control," in World of Wireless, Mobile and Multimedia Networks, 2009.

[94]

M. Sun, V. Le, and Z. Feng, "A Multi-channel Transmission Scheme in Cognitive Radio
Networks," in Communications and Networking in China (ChinaCOM), 2009.

[95]

B. R. Tamma, N. Baldo, B. S. Manoj, and R. R. Rao, "Multi-Channel Wireless Traffic
Sensing and Characterization for Cognitive Networking," in IEEE Communications
Conference, 2009.

198

[96]

A. K.-L. Yau, P. Komisarczuk, and P. D. Teal, "On Multi-Channel MAC Protocols in
Cognitive Radio Networks," in Australasian Telecommunication Networks and
Applications Conference, 2008.

[97]

N. B. Chang and M. Liu, "Optimal Channel Probing and Transmission Scheduling for
Opportunistic Spectrum Access," IEEE/ACM Transactions on Networking, vol. 17, pp.
1805-1818, 2009.

[98]

H. M. Almasaeid and A. E. Kamal, "Receiver-Based Channel Allocation for Wireless
Cognitive Radio Mesh Networks," in Dynamic Spectrum Access Networks (DySPAN),
2010.

[99]

S. Eljack, B. Huang, L. Tu, and P. Zhang, "Synchronized Multi-Channel Cognitive MAC
Protocol with Efficient Solutions for Second Spectrum Access," in Symposia and
Workshops on Ubiquitous, Autonomic and Trusted Computing, 2009.

[100] J. Park, P. Pawelczak, and D. Cabric, "To Buffer or to Switch: Design of Multichannel
MAC for OSA Ad Hoc Networks," in Dynamic Spectrum Access Networks (DySPAN),
2010, pp. 1-10.
[101] "CC2420: 2.4 GHz IEEE 802.15.4 / ZigBee-ready
http://focus.ti.com/docs/prod/folders/print/cc2420.html.
[102] TinyOS, "Index of
2.x/apps/BaseStation/.

/tinyos-2.x/apps/BaseStation,"

199

RF

Transceiver

",

http://www.tinyos.net/tinyos-