2016 Rajab Samer Dissertation

UNIVERSITY OF OKLAHOMA
GRADUATE COLLEGE
TOWARD ENHANCED WIRELESS COEXISTENCE IN THE 2.4GHZ ISM BAND
VIA TEMPORAL CHARACTERIZATION AND EMPIRICAL MODELING OF
802.11B/G/N NETWORKS
A DISSERTATION
SUBMITTED TO THE GRADUATE FACULTY
in partial fulfillment of the requirements for the
Degree of
DOCTOR OF PHILOSOPHY
By
SAMER RAJAB
Norman, Oklahoma
2016
TOWARD ENHANCED WIRELESS COEXISTENCE IN THE 2.4GHZ ISM BAND
VIA TEMPORAL CHARACTERIZATION AND EMPIRICAL MODELING OF
802.11B/G/N NETWORKS
A DISSERTATION APPROVED FOR THE

SCHOOL OF ELECTRICAL AND COMPUTER ENGINEERING
BY
________________________________
Dr. Hazem H. Refai, Chair
________________________________
Dr. Curt Adams
________________________________
Dr. John Dyer
________________________________
Dr. Ali Imran
________________________________
Dr. Thordur Runolfsson
Copyright by SAMER RAJAB 2016
All Rights Reserved.
To the soul of my Father, my beloved mother, siblings and wife, and my little yet to be
born bliss
Acknowledgements
The work in this dissertation wouldnt have been possible without the
continuous support of individuals whom Id like to take this opportunity to
acknowledge.
My guiding beacon and first idol, Dr. Adel Rajab may your soul rest in peace.
Love and appreciation goes to my mother Amal, Khaled, Reem, Majd, my wife
Mayssaa and all of my family for their unconditional love and support.
Special thanks go to my advisor Dr. Hazem Refai for his invaluable guidance,
and support. His help and directions have been the major reason for this dissertation
success.
My PhD committee; Dr. John Dyer, Dr. Thordur Runolfsson, Dr. Ali Imran and
Dr. Curt Adams are highly appreciated for their valuable support.
Acknowledgment goes to my friends whose help and support was a major
motive for me to succeed.
Michelle Farabough is acknowledged for her instrumental help in editing this
dissertation.
The University of Oklahoma is very well appreciated and acknowledged for
support and embracement
iv
Table of Contents
Chapter 1: Introduction..................................................................................................... 1
Chapter 2: Literature Review ........................................................................................... 9
Chapter3: Experimental and Simulation Setup .............................................................. 36
Chapter 4: Spectrum Characterization 802.11 Throughput and Duty Cycle............... 52
Chapter 5: Spectrum Characterization 802.11 Experimental Time Distributions ....... 78
Chapter 6: 802.11 Time Distributions Empirical Modeling ........................................... 94
Chapter 7: Applications and Case Studies.................................................................... 117
Chapter 8: Conclusion and Future Work ...................................................................... 137
References .................................................................................................................... 142
APPENDIX A .............................................................................................................. 148
v
List of Tables
Table 1. 802.11b timing parameters................................................................................ 29
Table 2. CCK phase parameters...................................................................................... 29
Table 3. 802.11g timing parameters................................................................................ 30
Table 4. 802.11n timing parameters................................................................................ 33
Table 5. 802.11g simulation parameters values ............................................................. 49
Table 6. Comparison between DC Statistics in time domain and frequency domain ..... 59
Table 7. Comparison between DC Statistics in time domain and frequency domain ..... 71
Table 8. 802.11n/g saturation throughput....................................................................... 85
Table 9. Notations .......................................................................................................... 96
Table 10. Complete distibutions emperical modelling DKL results for 802.11 networks
...................................................................................................................................... 110
Table 11. Above minimum contention window emperical modelling for 802.11
networks ....................................................................................................................... 114
Table 12. ZigBee adaptive packet sizes ....................................................................... 120
Table 13. Features used for wireless standards identification ...................................... 131
Table 14. Features used for number of transmitters identification ............................. 131
Table 15. Homogeneous networks idle time distributions empirical modelling
distributions parameters................................................................................................ 148
Table 15. Heterogeneous networks idle time distributions empirical modelling
distributions parameters................................................................................................ 151
vi
List of Figures
Figure 2-1. Saturation throughput vs. frame size for different FER values [31] ............. 23
Figure 2-2. 2.4GHz ISM band frequency channels. ........................................................ 26
Figure 2-3. DCF functionality........................................................................................ 28
Figure 2-4. CCK functionality ........................................................................................ 30
Figure 2-5. OFDM functionality ..................................................................................... 31
Figure 2-6. Aggregated MAC service data unit .............................................................. 34
Figure 2-7. Aggregated MAC protocol data unit ............................................................ 34
Figure 3-1. Example of 802.11 test setup ....................................................................... 37
Figure 3-2. Mikrotik 802.11 router boards and miniPCI network adapters .................... 37
Figure 3-3. Graphical representation of test setup with separation dimensions .............. 40
Figure 3-4. Time domain (TD) algorithm pseudo-code. ................................................. 42
Figure 3-5. Frequency domain algorithm pseudo-code................................................... 43
Figure 3-6. Noise floor lab measurement. ....................................................................... 45
Figure 3-7. Spectrogram for 802.11g at 25Mbps. ........................................................... 46
Figure 3-8. Spectrogram for 802.11g at 5Mbps. ............................................................. 46
Figure 4-1. 802.11b/g/n single-pair DC vs. throughput. ................................................. 52
Figure 4-2. Set throughput vs. achieved throughput for 802.11b two-pair network. ...... 54
Figure 4-3. DC results for 802.11b two-pair network. .................................................... 54
Figure 4-4. Set throughput vs. achieved throughput for 802.11g two-pair network. ...... 55
Figure 4-5. DC results for 802.11g two-pair network. .................................................... 55
Figure 4-6. Set throughput vs. achieved throughput for 802.11n two-pair network. ...... 56
Figure 4-7. DC results for 802.11n two-pair network. .................................................... 56
Figure 4-8. Set throughput vs. achieved throughput for 802.11b three-pair network. .... 57
Figure 4-9. DC results for 802.11b three-pair network. .................................................. 57
Figure 4-10. Set throughput vs. achieved throughput for 802.11g three-pair network. .. 58
Figure 4-11. DC results for 802.11g three-pair network. ................................................ 58
Figure 4-12. Set throughput vs. achieved throughput for 802.11n three-pair network. .. 58
Figure 4-13. Set throughput vs. achieved throughput for 802.11n three-pair network. .. 58
Figure 4-14. Comparison between DC measurements in time domain and frequency
domain. ........................................................................................................................... 60
Figure 4-15. Linear regression calculated between time domain and frequency domain.
........................................................................................................................................ 61
Figure 4-16. Set throughput vs. achieved throughput for two heterogeneous pairs:
802.11gn (GN). ............................................................................................................... 63
Figure 4-17. DC value for two heterogeneous pairs: 802.11gn. ..................................... 63
802.11bn (BN). ............................................................................................................... 64
Figure 4-19. DC value for two heterogeneous pairs: 802.11bn. ..................................... 64
802.11bg (BG). ............................................................................................................... 64
Figure 4-21. DC value for two heterogeneous pairs: 802.11bg. ..................................... 64
Figure 4-22. Set throughput vs. achieved throughput for three heterogeneous pairs:
802.11bbg (BBG). .......................................................................................................... 65
Figure 4-23. DC value for three heterogeneous pairs, 802.11bbg................................... 65
vii
802.11ggb (GGB). .......................................................................................................... 66
Figure 4-25. DC value for three heterogeneous pairs: 802.11ggb (GGB)....................... 66
802.11bbn (BBN). .......................................................................................................... 67
Figure 4-27. DC value for three heterogeneous pairs: 802.11bbn (BBN). ...................... 67
802.11nnb (NNB). .......................................................................................................... 67
Figure 4-29. DC value for two heterogeneous pairs: 802.11nnb (NNB). ....................... 68
802.11ggn (GGN). .......................................................................................................... 68
Figure 4-31. DC value for three heterogeneous pairs: 802.11ggn (GGN). ..................... 68
Figure 4-32. Set throughput vs. achieved throughput for three heterogeneous pairs,
802.11nng (NNG). .......................................................................................................... 69
Figure 4-33. DC value for three heterogeneous pairs, 802.11nng (NNG). ..................... 69
802.11g, 802.11b, and 802.11n (GBN). .......................................................................... 70
Figure 4-35. DC value for three heterogeneous pairs: 802.11g, 802.11b, and 802.11n
(GBN). ............................................................................................................................ 70
Figure 4-36. Comparison between DC measurements in time domain and frequency
domain. ........................................................................................................................... 71
Figure 4-37. Linear Regression time domain vs. frequency domain for heterogeneous
network. .......................................................................................................................... 72
Figure 4-38. Simulation and experimental DC and throughput comparison for a below
saturation 802.11g one-pair network............................................................................... 73
Figure 4-39. Simulation and experimental DC and throughput comparison for 802.11g
one-, two- and three- pair saturated networks. ................................................................ 74
Figure 4-40. 802.11g simulation extrapolated throughput and DC values for a large
number of transmitters .................................................................................................... 75
Figure 4-41. Per transmitter throughput for 802.11g/n shared medium for various
numbers of 802.11n aggregated frames. ......................................................................... 76
Figure 4-42. Medium access distributions for 802.11g/n shared medium at various
numbers of 802.11n aggregated frames .......................................................................... 76
Figure 5-1. Idle-time distributions for: a) 802.11b, b) 802.11n and c) 802.11g one pair81
Figure 5-2. 802.11b idle time distributions for two and three pairs ............................... 82
Figure 5-3. 802.11g idle time distributions for two and three pairs ............................... 82
Figure 5-4. 802.11n idle time distributions for two and three pairs ............................... 83
Figure 5-5. 802.11b saturation activity distributions for one- two- and three pairs ....... 84
Figure 5-6. 802.11g saturation activity distributions for one- two- and three pairs ....... 84
Figure 5-7. 802.11n saturation activity distributions for one- two- and three pairs ...... 84
Figure 5-8. 802.11n one pair activity distribution ......................................................... 86
Figure 5-9. Two pairs heterogeneous networks idle time distribution .......................... 87
Figure 5-10. Three pairs heterogeneous networks idle time distribution ....................... 88
Figure 5-11. Three pairs bgn heterogeneous networks idle time distribution ................ 89
Figure 5-12. Two pairs heterogeneous networks activity distribution ........................... 90
Figure 5-13. Three pairs heterogeneous networks activity distribution ......................... 90
viii
Figure 5-14. Simulation vs. experimental idle time distributions .................................. 91
Figure 6-1. Process for analyzing 1.4GHz ISM band 802.11 idle time distributions .... 95
Figure 6-2. 802.11n models evaluation metrics a) DB for various throughput values a)
DKL for various throughput values c) Mean DB d) Mean DKL ...................................... 102
Figure 6-3. 802.11n models vs. empirical distribution for different throughput levels
...................................................................................................................................... 103
Figure 6-4. 802.11nn models evaluation metrics a) DB for various throughput values, b)
DKL for various throughput values, c) Mean DB d) Mean DKL ..................................... 104
Figure 6-5. 802.11nn models vs. empirical distribution for different throughput levels
...................................................................................................................................... 104
Figure 6-6. 802.11nnn models evaluation metrics a) DB for various throughput values,
b) DKL for various throughput values, c) Mean DB d) Mean DKL ................................. 105
Figure 6-7. 802.11nnn models vs. empirical distribution for different throughput levels
...................................................................................................................................... 105
Figure 6-8. 802.11gn models evaluation metrics a) DB for various throughput values, b)
DKL for various throughput values, c) Mean DB d) Mean DKL ..................................... 106
Figure 6-9. 802.11gn models vs. empirical distribution for different throughput levels
...................................................................................................................................... 107
Figure 6-10. 802.11bn models evaluation metrics a) DB for various throughput values,
Figure 6-11. 802.11bn models vs. empirical distribution for different throughput levels.
...................................................................................................................................... 108
Figure 6-12. 802.11gbn models evaluation metrics a) DB for various throughput values,
Figure 6-13. 802.11gbn models vs. empirical distribution. .......................................... 109
Figure 6-14. Above minimum contention window802.11nnn models vs. empirical
distribution .................................................................................................................... 111
Figure 6-15. Above minimum contention window 802.11bn models vs. empirical
distribution .................................................................................................................... 111
Figure 6-16. Above minimum contention window 802.11gn models vs. empirical
distribution .................................................................................................................... 112
Figure 6-17. Above minimum contention window 802.11gbn models vs. empirical
distribution .................................................................................................................... 112
Figure 7-1. 802.15.4 ZigBee and 802.11g coexistence experimental setup ................. 119
Figure 7-2. 802.15.4 ZigBee PER simulation vs. experimental results ....................... 119
Figure 7-3. ZigBee throughput for fixed and adaptive packet sizes ............................ 122
Figure 7-4. ZigBee PER for fixed and adaptive packet sizes ....................................... 122
Figure 7-5. Wireless technology identification stage ................................................... 126
Figure 7-6. Overlapping idle time distribution features ............................................... 130
Figure 7-7. Idle time distribution features with improved technology separation ....... 130
Figure 7-8. Homogeneous networks identification accuracy ....................................... 132
Figure 7-9. Distribution of identified observations for homogeneous networks.......... 132
Figure 7-10. Number of transmitters identification accuracy ..................................... 133
Figure 7-11. Per-standard, two stages, number of transmitters identification accuracy
...................................................................................................................................... 133
Figure 7-12. Heterogeneous networks identification accuracy .................................... 134
ix
Figure 7-13. Distribution of identified observations for heterogeneous networks ....... 135
Figure 8-1. Proposed hidden node test setup ................................................................ 139
Figure 8-2. Proposed exposed node test setup.............................................................. 140
x
Abstract
This dissertation presents an extensive experimental characterization and empirical
modelling of 802.11 temporal behavior. A detailed characterization of 802.11b/g/n
homogeneous and heterogeneous network traffic patterns is featured, including idle time
distribution and channel utilization.
Duty cycle serves as a measure for spectrum busyness. Higher duty cycle levels directly
impact transceivers using the spectrum, which either refrain from transmission or suffer
from increased errors. Duty cycle results are provided for 802.11b, g and n Wi-Fi
technologies at various throughput levels. Lower values are observed for 802.11b and g
networks. Spectrum occupancy measurements are essential for wireless networks
planning and deployment.
Detailed characterization of 802.11g/n homogeneous and heterogeneous network traffic
patterns, including activity and idle time distribution are presented. Distributions were
obtained from time domain measurements and represent time fragment distributions for
active and inactive periods during a specific test. This information can assist other
wireless technologies in using the crowded ISM band more efficiently and achieve
enhanced wireless coexistence.
Empirical models of 802.11 networks in the 2.4 GHz Industrial, Scientific, and Medical
(ISM) band are also presented. This information can assist other wireless technologies
aiming to utilize the crowded ISM band more efficiently and achieve enhanced wireless
coexistence. In this work models are derived for both homogeneous and heterogeneous
802.11 network idle time distribution.
xi
Additionally, two applications of 802.11 networks temporal characterization are
presented. The first application investigates a novel method for identifying wireless
technologies through the use of simple energy detection techniques to measure the
channel temporal characteristics including activity and idle time probability distributions.
In this work, a wireless technology identification algorithm was assessed experimentally.
Temporal traffic pattern for 802.11b/g/n homogeneous and heterogeneous networks were
measured and used as algorithm input. Identification accuracies of up to 96.83% and
85.9% are achieved for homogeneous and heterogeneous networks, respectively. The
second application provides a case study using 802.15.4 ZigBee transmitter packet size
on-line adjustments is also presented. Packet size is adaptively modified based on channel
idle time distribution obtained using simple channel power measurements. Results
demonstrate improved ZigBee performance and significant enhancement in throughput
as a result of using adaptive packet size transmissions.
xii
Chapter 1: Introduction
With estimated annual Wi-Fi devices transactions reaching nearly $315 million dollars in
2014, it is clear that Wi-Fi has become a major contributor to todays global
communications economy [1]. Demand for wireless data transfer has increased
significantly in recent years, rendering wireless spectrum an expensive and scarce
resource. The ability to operate in the free, unlicensed ISM (Industrial Scientific and
Medical) band has prompted cost effective Wi-Fi chip production. Hence, a large number
of modern day gadgets, such as laptops, smartphones, printers, cameras and wearable
devices, are now equipped with Wi-Fi wireless communications technologies, making Wi-
Fi one of todays prevailing wireless communications technologies. New bandwidth-
hungry and distributed applications like HD video transfer, social media, Internet of
Things, and cloud computing have exacerbated the problem. Recent telecommunication
research has focused on developing more efficient means for exploiting available, yet
limited, frequency spectrum.
Various wireless technologies are uniquely characterized according to time, frequency,
and power domains. Wireless technology characterization and identification provides a
wireless device attempting to access the medium with valuable information, including
anticipated transmission periodicity, frequency bandwidth, and transmit power, among
other parameters. Awareness about these characteristics is useful, especially in
heterogeneous frequency bands like the ISM band.
Spectrum scarcity has risen significantly in the past few years, triggering intensified
research in multiple areas, such as spectrum occupancy measurements and cognitive radio.
1
An ever-increasing reliance on wireless communications makes frequency spectrum an
expensive resource. Although a number of spectrum occupancy measurement campaigns
have been presented in literature, as discussed in following sections, most target to study
underutilized spectrum and multiple bands scanning. Limited work has investigated
2.4GHz ISM band spectrum occupancy.
Spectrum scarcity makes heterogeneous spectrum access a necessity in current wireless
networks primarily because the scarcity forces wireless technologies to share the same
frequency band [2]. As a consequence, wireless coexistence among various wireless
technologies sharing the same medium is gaining increased attention as an integral feature
for reliable operation. Coexistence is defined as the ability of one system to perform a
task in a given shared environment where other systems have an ability to perform their
tasks and may or may not be using the same set of rules [3]. Knowledge of both
anticipated and current spectrum occupancy levels enables efficient planning and
parameter adaption for coexisting wireless technologies, thus enhancing transmission
efficiency.
The ISM band is unlicensed, making it an attractive solution for wireless device
manufacturers. Several protocols (e.g., Wi-Fi, Bluetooth, Bluetooth Low Energy (BLE),
and ZigBee) have been developed to operate in the ISM band. Wireless chips for these
protocols are now manufactured in large scale, further reducing their prices. Consequently,
the ISM band is widely recognized as one of the most crowded frequency bands and serves
as a timely example of issues related to heterogeneous spectrum access. Spectrum
occupancy measurement serves as a tool to assist wireless device manufacturers to make
cognizant decisions when choosing transmission parameters, including transmission
2
power, frequency, and frame size, among others. This is performed using algorithms to
intelligently characterize the spectrum occupancy and identify wireless technologies
utilizing the spectrum at a particular time.
Wi-Fi is a primary causes of interference in the ISM band due to the technologys high
power and data rate when compared with other technologies sharing the 2.4 GHz ISM
band. Hence, characterizing spectrum occupancy behavior of Wi-Fi is vital for both
frequency channel planning and wireless device development in the 2.4 GHz ISM band.
A means to analyze spectrum occupancy for Wi-Fi is measuring channel duty cycle (DC),
modelling idle time distributions and using this information to identify wireless
technologies sharing the medium. Notably, DC is defined as the fraction of time at which
the received signal strength is above a certain threshold [4].
Wi-Fi networks operating in the ISM band are typically heterogeneous, as they include
802.11b/g/n Wi-Fi standards, which have significant differences in both PHY and MAC
layers, as will be discussed in greater detail in chapters that follow. Networks supporting
any of these standards could exist within close proximity of one another. Thus,
understanding spectrum occupancy for heterogeneous Wi-Fi networks can aid in
improving coexisting network performance. Measuring frequency channel DC is one
method for assessing spectrum occupancy.
This work in part investigates time distributions modelling of 802.11 heterogeneous
networks in the 2.4GHz ISM band, including a detailed characterization of 802.11b/g/n
homogeneous and heterogeneous network traffic patterns. This type of study will facilitate
better understanding of 802.11 temporal behavior, thus enabling improved design for
coexisting wireless technologies. Time distribution modelling focuses on 802.11
3
networks, in particular, primarily due to their high power levels and extensive use. This
makes 802.11 networks the foremost interferer in the exceedingly crowded 2.4GHz ISM
band. This band is of special interest to wireless coexistence for two chief reasons:
1) The relatively limited 80MHz spectrum, and
2) The large number of wireless technologies that share the ISM band.
These aspects cause wireless coexistence difficulties for technologies sharing the ISM
band with 802.11 networks. Consequently, interference between 802.11 and other
technologies (e.g., Bluetooth, ZigBee, and BLE, among others) could have adverse effects
depending on the application of devices sharing the same 802.11 frequency band [5][6][7].
A critical example of the effect of 802.11 interference with other coexisting technologies
can be found in the medical device industry. Many medical device manufacturers
implement wireless technologies in their devices for a number of practical reasons,
including patient convenience. Medical device applications for wireless connectivity
range from transmitting regular telemetry data to patient life-critical applications.
Medical devices must utilize low-power consumption technologies to minimize service
interruption. Several manufacturers employ ISM band wireless technologies as often as
possible, primarily due to global availability and reduced costs due to standardization and
large scale manufacturing. ZigBee, Bluetooth, and BLE are examples of popular wireless
communication technologies currently implemented in medical devices. Devices using
these technologies could suffer severe interference among ISM transmitters caused by
802.11 traffic [8][9], which can jeopardize medical device functionality and impose a risk
to patient safety depending on the specific application of the medical device. Diligent
wireless coexistence risk assessment of wireless medical systems must be performed
4
based on data-rate and delay requirements, as well as anticipated spectrum occupancy.
Consequently, the United States Food and Drug Administration (FDA) has recommended
that radiated coexistence testing should be performed during the wireless medical device
certification process [10]. The experimental wireless technology identification techniques
assessment presented in this work is directly related to the wireless medical devices
example as measurements were performed in the 2.4 GHz ISM band. Note that from
hereafter, wireless technologies sharing frequency bands with 802.11 will be referred to
as coexisting technologies (CT); the 2.4 GHz ISM band is referred to as the ISM band.
Channel awareness in terms of temporal characteristics (e.g., DC, activity distribution, and
idle time distribution) aids CTs intelligent transmission decision-making. Frame size,
channel access frequency, and time (and/or transmission frequency channel) can be
modified for improved performance based on channel temporal traffic patterns.
Various wireless applications operating in different frequency bands will benefit from
wireless technology characterization and identification techniques. For example, cognitive
radio requires spectrum usage awareness for cognitive nodes to operate properly.
Cognitive radio network requires spectrum usability assessment, particularly in licensed
bands, before a cognitive node can access the medium. Threats to wireless coexistence
increase with cognitive radio use regardless of advanced spectrum sensing. Hence,
identification of wireless technologies operating in a specific spectrum band provides a
cognitive node with an estimate of potential interference. A cognitive node can use
gathered information to aid in decision-making about using an intended frequency band.
5
This dissertation presents a detailed temporal characterization of 802.11 homogeneous and
heterogeneous wireless channels in the ISM band based on an energy detection technique.
The contributions of this work include the following:
1. Provide a methodology for wireless channel temporal characterization using an
energy detection technique.
2. Offer an extensive experimental study of traffic patterns and DC for homogeneous
and heterogeneous 802.11 networks and a detailed scheme for DC measurements
for Wi-Fi 802.11b/g/n. The presented study links measured DC to data throughput
level and the number of coexisting networks. Also, a comparison between obtained
results using two developed measurement tools for time domain (TD) and
frequency domain (FD) is presented. DC measurement data is analyzed via
conventional inferential statistical methods (i.e., null-hypothesis tests, linear
regression, and others) to validate DC obtained in TD with DC measured in FD.
3. Provide detailed empirical modelling of 802.11 homogeneous and heterogeneous
networks in the ISM band, including evaluation of idle time devised models
compared with corresponding measured distributions for 802.11b/g/n single and
multi-pair transmissions obtained via energy detection method.
4. Report findings for a study investigating wireless technology identification using
energy detection and machine learning techniques. Wireless technology
identification is possible through received signal demodulation. However, this
method could prove impractical, primarily because a wireless terminal with such
functionality must support many different demodulation schemes. A novel
alternative method for wireless technology identification through the use of simple
6
energy detection techniques is presented. A developed algorithm a) constructs
temporal activity and idle time distributions from measured received signal
strength (RSS), b) extracts distinguishing features from distribution, and c)
provides technology identification decision using machine-learning classifiers.
Experimental assessment of a developed algorithm has been performed, yielding
high identification accuracy: 96.83% for homogeneous Wi-Fi networks and 85.9%
for heterogeneous Wi-Fi networks.
5. Deliver results from a case study investigating CT (ZigBee) performance
improvement when coexisting with an 802.11g network through temporal
characterization. A ZigBee simulation was implemented wherein ZigBee packet
size was changed adaptively according to channel idle time distribution to improve
ZigBee performance under interference. ZigBee throughput and packet error rate
(PER) for both fixed packet sizes and variable packet sizes were tracked and
reported.
This dissertation is organized as follows. Chapter 2 lists related work and provides an
overview of the ISM band technologies investigated in this work. Chapter 3 details
methodology and the experimental setup. Chapters 4 and 5 present 802.11 networks
temporal characterization including; spectrum occupancy analysis and time distributions
of 802.11 networks, respectively. 802.11 idle time distributions empirical modeling work
is presented in Chapter 6. Chapter 7 offers two applications of 802.11 networks temporal
characterization. The first reports adaptive ZigBee frame size based on 802.11g network
for improving performance; the second identifies wireless technology using energy
7
detection and machine learning techniques for devices sharing the medium. Finally, a
conclusion is provided in Chapter 8.
8
Chapter 2: Literature Review
This chapter provides an overview of earlier work related to spectrum occupancy
monitoring. Studies researching energy detection spectrum sensing, spectrum temporal
patterns modelling, throughput-sensing problem, frame aggregation techniques, and
wireless technologies identification are examined in the following subsections. An
overview of wireless technologies investigated in this work is provided, as well.
Energy detection spectrum sensing
Energy detection, also known as transmitter detection or radiometry, is a simple and
widely recognized spectrum sensing techniqueone that has been adopted in many
applications, including cognitive radio [11]. Unlike other spectrum sensing techniques,
those employing an energy detection algorithm do not need prior information about the
channel or received signal. The energy detecting algorithm has three stages. In the first
stage, received signal energy is measured within a frequency span (or bandwidth). In the
second stage, the measured signal is compared with a predefined threshold to separate the
desired signal from background noise. In the third stage, a decision is made as to whether
or not there is an active signal (i.e., if the measured energy value is above or equal to the
threshold). This energy detection technique is characterized by low computational cost
and minimal implementation complexity. However, its performance degrades with a low
signal-to-noise ratio (SNR) [12]. The lower the SNR, the higher the probability of false
detection. This drawback can be overcome by introducing improved energy detection
algorithms, which enhances the performance of the conventional energy detection method
[13][14][15].
9
Presented work in this dissertation details an energy detection method to scan the wireless
spectrum in the ISM band. A variety of tests with varying throughput and number of
transmitters were performed. Idle time and activity distributions were extracted, and
empirical models were devised for a number of 802.11 combinations at varying
throughput. The purpose of these tests is three fold:
1. Describe idle time distribution mathematically with acceptable accuracy to
facilitate a better understanding of 802.11 heterogeneous networks.
2. Characterize temporal spectrum utilization by 802.11 networks to provide channel
occupancy awareness for wireless technologies coexisting with 802.11 networks.
Occupancy awareness results in intelligent and more efficient design.
3. Utilize the activity and idle time distributions obtained via energy detection method
to provide examples of using this approach to enhance the performance of
coexisting technologies.
Spectrum temporal patterns modelling
Following is a summary of related work on spectrum measurement and temporal
patterns modeling.
Earlier work investigating spectrum occupancy measurements in the ISM band is of
particular interest, as it falls within the scope of work presented herein. Limited work has
been found in literature for ISM band spectrum occupancy measurement. In [16], spectrum
occupancy measurements using distributed directional antennas are presented. This work
provides information on spatial dimension influence over duty cycle (DC). ISM band
measurements were conducted in an office environment, and DC for uncontrolled
10
environment was reported. The work adopted conventional energy detection with fixed
threshold for calculating DC. The work was extended in [17], wherein measurements
acquired from two devices with multiple directional antennas were conjoined using
different combining rules. DC average measurement for one day in an office area was
presented.
In [18], a measurement approach in the ISM band using energy detection was proposed.
Measurements were undertaken using frequency domain (FD) sweeping, wherein each
measurement was performed for 30 minutes. DC was presented as a function of frequency.
Notably, the paper focused on measurement tool parameters development. Although other
studies investigated ISM band activity to model idle time distributions, DC results were
not provided [19] [20].
Aforementioned studies focused on spectrum sensing and DC reporting of uncontrolled
transmissions on the spectrum. Work presented in herein is unique in that it investigates
802.11 temporal patterns in a controlled environment for single and multiple homogeneous
and heterogeneous 802.11 transmitters with varying throughput levels. The study also
models 802.11 network idle time distributions, thus capturing temporal behavior of the
main interference sources might exist in the band.
Examples of previous research investigating temporal behavior in a licensed, non-ISM
band can be found in [21] [22]. Work presented in [21] models long term DC and
activity/idle time distributions for measurements obtained from several bands. The paper
provides an accurate model for DC for cellular bands GSM and DCS. Though
measurements were taken for the ISM band, they were not presented nor discussed. The
presented model for idle time distribution is provided for large scale (in order of seconds),
11
whereas presented distributions in this dissertation are for small scale distributions (i.e.,
micro seconds).
Modelling of wireless activity temporal patterns using Continuous-Time semi-Markov
Chain CTMC was presented in [22]. Two measurement devices that provided high and
low time resolution were employed. The longest time resolution achieved and reported
was 128 s, which is inadequate to accurately capture WLAN activity patterns. The work
applied different distributions to various bands. Although the authors in [22] mentioned
ISM measurement of ISM band spectrum utilization, results were not presented.
Throughput-sensing problem
In the ISM band, 802.11, and other Coexisting Technologies (CTs) have equal privilege
to use the unlicensed band. Notably, the high power and data rate of 802.11 have made
this technology the dominant transmitter. The 802.11 interference effect on CT far exceeds
CT interference effect on 802.11. Hence, other CTs must optimize wireless coexistence
with 802.11 to improve their performance. This phenomenon is analogous to cognitive
radio networks where a secondary user (SU) must manage spectrum access to minimize
interference with primary user (PU).
In a cognitive radio (CR) network, the SU frame transmission time is divided into sensing
time and transmission time. The latter is one of the most important parameters effecting
SU throughput [23]. Frame time must be optimized by minimizing sensing time and
maximizing transmission time. Effects of PU traffic on SU throughput has been
investigated in earlier literature [24][25][26][27][28].
12
Researchers in [24] were the first to examine the effect of PU altering transmission state
during SU transmission on CR performance (i.e., throughput). Previous studies considered
a channel to be either busy or available, maintaining only one or the other of the two states
during SU frame transmission. An investigation in [24] proposed an analytical model
wherein one transition from OFF to ON was considered, based on PU throughput.
Researchers did not consider, however, the effect of a changing SU frame size.
In [25], the effect of PU changing its state during an OFF period was also analyzed. In
[26], the possibility of PU changing states one time during the OFF stage was investigated.
CR throughput was maximized based on frame size, assuming that ON/OFF durations and
sensing time maintain an exponential distribution. Work presented in [27] developed an
analytical model to describe and maximize SU throughput based on optimizing SU frame
size and sensing period. The analytical model accounted for up to two PU traffic state
transitions, although this assumption might not be true, depending on PU frame duration
(i.e., acknowledgementACK or data). Authors of [28] implemented a sensing scheme
to estimate the distribution of PU ON/OFF durations and decide whether or not to sense a
channel. PU traffic ON/OFF durations were assumed to have a Gaussian distribution.
Frame aggregation techniques
Frame aggregation has been researched extensively in literature. The research led
eventually to the technology implementation in 802.11n standards MAC layer. Earlier
studies investigating frame aggregation and adaptive frame size are discussed below. This
portion of the literature review will facilitate a better understanding of the frame
aggregation technology in 802.11n networks. Also, it highlights possibilities of
implementing similar approaches for CTs in the ISM band.
13
Adaptive frame aggregation is investigated in [29] [30] [31] [32] [33] [34] [35] [36].
Research in [29] investigated optimal frame size based on Bit Error Rate (BER) estimated
from SNR. The paper assumes saturated nodes with only RTS/CTS scheme. Also,
existence of a feedback mechanism that reports SNR to transmitter is assumed.
[30] introduces an algorithm that dynamically chooses the aggregation scheme to be used
in 802.11s mesh networks. Algorithm uses BER, quantity, and distribution of frame arrival
to the transmitter queue for making a choice regarding the aggregation scheme. This
technique assumes that BER is measured from received frames. Authors of [31] develop
an analytical model to describe impact of frame aggregation on saturation throughput.
Frame error rate is calculated from received block ACK to adaptively change sub-frame
size.
A joint data rate and fragment size adaption based on error probability is investigated in
[32]. In this context, the probability of error is estimated by the transmitter by counting the
number of unsuccessfully received fragments using block ACK whilst not relying on SNR.
Authors of [33] propose a frame size estimation technique based on frame error rate (FER).
Kalman filter is used to estimate optimal frame size using FER.
In [34], an analytical model to optimize frame size based on delay constraints of the node
is proposed. It assumes that the network is in saturated condition. This method also
assumes that all nodes in the network broadcast their delay information wherein delay is
defined as the time required by a node to successfully access the channel. It is assumed
that a node can acquire information about number of transmitting nodes by observing the
network without specifying the mechanism.
14
An algorithm that adaptively changes the aggregated frame size is proposed in [35]. The
algorithm aims at maintaining delay below a specific maximum boundary. In [36] a
dynamic frame aggregation scheme based on channel conditions in 802.11a network is
investigated. The algorithm estimates the probability of error using received ACK signal
strength, and then utilizes a lookup table to adjust the data rate and frame size.
A study presented in [37] precedes frame aggregation technology. Different from
aforementioned studies, researchers investigate variable frame size in terms of
performance of saturated 802.11 DCF in terms of packet size, contending nodes, and
packet error probability. An analytical framework was developed to determine optimal
frame size for maximizing throughput based on BER.
Other research activities studied channel state estimation or frame size change for other
purposes, such as scheduling and technology interoperability (e.g., [38] [39]).
Experimental evaluation of increasing 802.11 frame payload size to 4KB was presented
in [38]. The main contribution of that study was providing a software interface to enable
Linux host to send/receive a frame larger than the Ethernet standard of 1500 Bytes.
However no adaptive frame size variation was presented in the paper. Authors of [39]
propose a scheduling mechanism that takes into account channel state and transmitter
queue size. In this context, access point (AP) is assumed to schedule the user with good
channel state and long queue time to optimize throughput. No variation of frame size was
presented.
Wireless technology and MAC method identification
To the best of the authors knowledge, no previous work has been published that identifies
various 802.11 standards by way of observing power measurements from a medium where
15
standards are operational. Unlike research detailed below, the newly developed method
does not require demodulating the received signal, nor is it necessary to successfully
perform accurate identification based only on raw power measurements.
Bayesian machine learning method and independent component analysis technique have
been employed in [40] and [41] respectively for clustering 2.4 ISM band wireless
networks. These methods investigate blind technology identification (BTI) approach to
group wireless transmissions with similar behavior. Such techniques are useful for
identifying number of wireless transmissions with distinct behavior using the spectrum
band with no prior knowledge about these wireless technologies. Nevertheless, additional
information about the underlying wireless technologies being used by such transmissions
cannot be deduced without a prior knowledge of the corresponding wireless technologies
characteristics.
Authors in [42] aimed at detecting Classic Bluetooth (BT) piconets and Wi-Fi 802.11g
networks. The researchers developed a method using Universal Software Radio Peripheral
(USRP) to collect (4 MHz) bandwidth power measurements in the ISM band. Classic BT
piconets were identified through a time-binning mechanism, which demonstrated that
power bursts falling in the same time bin are originated from a single piconet. An
alternative suggested method requires demodulating sensed bursts, and then obtaining the
72-bit Channel Access Code, which is unique to each piconet. The presence of Wi-Fi
access points is determined by examining distinct equalization sequences included in AP
beacons, and then cross-correlating observations within a beacon time period.
16
Fingerprinting is also heavily studied in literature. In contrast to [42], [43] exploits specific
emitter identification (SEI) second-order cyclic OFDM features for identifying different
hardware emitting 802.11a/g signals. The experiment output is a wireless card model as
opposed to counting the number of nodes or networks existing on the medium. Other
similarly targeted sources in literature can be found by investigating means to identify a
used wireless network card, such as [44] that identifies specific Network Interface Cards
(NICs) among a group of identical NICs. Likewise, [45] promotes the use of transient
features that are manifested when a wireless transmitter is switched on as means for
transmitter identification. The study presented in [46] relies on frame inter-arrival time as
a parameter toward determining a distinct device fingerprint.
Authors of [47] presented a method in which a cognitive radio node identifies the MAC
protocol of other nodes (e.g., primary users of spectrum or other cognitive nodes) among
four types of MACs: TDMA, CSMA/CA, pure ALOHA, and slotted ALOHA. The method
relies on extracting two power features, namely received power mean and variance, and
six time features, including minimum, median, and maximum of both channel idle and
busy durations. This work relies on simulation to generate fictitious date, which is then
input into Support Vector Machine learning-algorithm for identifying a MAC protocol.
Comparison with previous research
This section provides a detailed comparison between work presented in this dissertation
and the most relevant studies presented in the previous sections. Earlier studies on
spectrum temporal patterns modelling focused on idle time distributions in licensed
bands. These studies were expected to be the impetus for a cognitive network solution in
which a secondary user would be privileged to use the spectrum whenever the primary
17
user was inactive. However, several issues were found when employing cognitive radio
in a licensed band (e.g., hidden transmitter problem).
Such problems have spawned alternative solutions to spectrum sensing, such as centralized
databases providing information about primary user activity in licensed bands. Notably,
due to the complexity and heterogeneous nature of this problem, limited work has
investigated spectrum occupancy scanning and modelling of 2.4 GHz ISM band.
Furthermore, no agreed-upon definition exists for an ISM band primary user.
Work presented in this dissertation is focused on overcoming these issues by detailing
temporal distributions modelling in the ISM band. The following assumptions have been
made:
802.11 networks are the main interferer in the ISM band as a results of their high
power and their use as a large data transfer solution, as demonstrated and proven in
earlier literature (e.g. [7][8][48]).
Providing a comprehensive set of experiments to describe 802.11 networks is
important for identifying: 1) Homogeneous and heterogeneous network
combinations; 2) Varying throughput levels from low throughput to saturation; and
3) Varying number of transmitting terminals. Uncontrolled spectrum
measurements are subjective, as indicated by variations in results reported in the
literature. Controlled spectrum measurement narrows the problem with a limited
set of parameters for presenting a reproducible representation of the spectrum.
Of most interest are studies performed by Lopez-Benitez et. al. [21][22]. In [21] spectrum
measurements were performed for the following bands: TETRA-UL, TETRA-DL, E-
18
GSM 900 UL, E-GSM 900 DL, DCS 1800 UL, DCS 1800 DL, DECT, and ISM bands.
Active time, idle time, and DC results were then presented for TETRA, E-GSM, and DCS
bands at a testing site in Barcelona, Spain. Discrete Time Markov-Chain (DTMC)with
two states representing active and idle timewas implemented to model the obtained
spectrum measurements for the three aforementioned bands. Notably, the measurements
were characteristic of a large time scale in the order of seconds. DTMC models were
unable to represent the statistical characteristics of most scanned bands, as these models
diverged from their active and idle time distributions. In an effort to enhance these
models, transition probabilities between the active and idle states were determined using
instantaneous, time variant, DC models rather than total average DC values. In other
words, a time-inhomogeneous, rather than time homogenous, DTMC was implemented.
Models performed well for cellular bands DCS and E-GSM demonstrating a deterministic
DC pattern. TETRA band models did not perform as well, given that this band is more
heterogeneous in nature with random DC behavior. Even though performing ISM band
measurement was mentioned, no modelling results were provided for this band. Also,
time resolution for spectrum measurement activities performed in [21] is very low in
the order of second. As such, the provided measurements fail to capture short-scale timing
behavior demonstrated by the MAC layer of the wireless communication technologies
operating in the studied bands. Alternatively, the work performed in this dissertation
provides a high resolution, extensive spectrum measurement survey for the 2.4GHz ISM
band. A systematic experimental evaluation of the homogeneous and heterogeneous
802.11 networks spectrum activity at wide range of throughput values was performed and
presented, providing a detailed description of DC, active time, and idle time distributions.
19
In [22], authors extended their work by performing empirical modelling for their obtained
active and idle time distributions. The investigated models were Exponential, Generalized
Exponential, Pareto, Generalized Pareto, Log-Normal, Gamma, and Weibull. Short time-
scale measurements were performed using a USRP platform. Models were evaluated and
presented for the following bands: amateur, paging, PMR/PAMR, cellular mobile
communications, and cordless bands. Weibull distribution provided best models for
amateur and PMR/PAMR bands. Paging and cellular mobile communication bands, idle-
time distributions were best modeled using Pareto and generalized exponential
distributions, respectively. However, no short time-scale analysis was performed for the
2.4GHz ISM band. ISM band analysis was only performed for single large-scale
measurement result. Such analysis, as discussed previously, fails to represent temporal
behavior of wireless technologies using the band which in turn is greatly dictated by the
MAC layer implementation. In contrast to previous work, this dissertation provides a
comprehensive experimental analysis of the 802.11 networks temporal behavior on the
spectrum in the ISM band, including empirical modeling of obtained results. Empirical
modeling is performed at a detailed level to capture appropriate distributions and their
parameters that best model different 802.11 combinations at various throughput levels.
Also, this work is performed at high time resolution for capturing the smallest 802.11
MAC layer time duration of 10s. Such high time resolution was not implemented in
previous work, leading to a deficiency in accurately describing temporal behavior of the
2.4GHz ISM bands.
Regarding previous work investigating the throughput-sensing problem, none of these
studies presents a realistic measure of PU activity pattern, due to assumptions regarding
20
ON/OFF duration and frequency [27][28]. It is possible that modelling the ON/OFF
duration with exponential or normal distribution might prove inaccurate, as these
distributions change with protocol type, throughput, and number of nodes in the network.
Experimental identification of a PU traffic pattern and temporal characteristics, including
MAC protocol, can prove useful toward enhancing CT performance. Modelling of this
traffic pattern can then be performed using supporting empirical results. Work presented
in Chapter 5 of this dissertation aims at finding empirical time distribution of
homogeneous and heterogeneous 802.11 networks in the ISM band. Chapter 6 presents a
detailed empirical modelling of idle time distributions in an effort to provide a more
accurate representation of these time distributionsbeyond normal distribution
assumptions made in previous literature.
Regarding adaptive frame size, most aforementioned studies focus on 802.11 networks.
Also, many of these assume existence of a feedback mechanism from the receiver to the
transmitter, which in fact might not exist. On the other hand, variable frame size for CTs
with 802.11 using spectrum temporal pattern provides an opportunity to enhance CTs
performance. CTs with 802.11 networks in the ISM band suffer great degradation in
performance at 802.11 high throughput values (e.g., ZigBee). Temporal characterization
of 802.11 DC and idle time distribution help CTs select a frame size to statistically
minimize frame error rate (FER). For example, [31] presents a technique to change
MPDU size based on the maximum tolerable FER. In that study, FER is estimated from
Block ACK to implement the algorithm. A Bianchi analytical model was constructed to
balance MPDU size based on FER to improve saturation throughput. The model assumes
that contending nodes are using RTS/CTS mechanism to access the medium. However,
21
this assumption might not be realistic, as real networks are heterogeneous and may use
different schemes. Likewise, RTS/CTS scheme is rarely used in real world
implementation due to its low efficiency [49]. Results of the developed analytical model
are depicted in
Figure 2-1. Increasing frame size at low FER value would always result in better
throughput. On the contrary, at higher FER, optimal throughput is demonstrated by a
particular frame size. Increasing frame size beyond that optimal value results in
throughput degradation. Choosing optimal frame size relies heavily on the availability of
a feedback mechanism to obtain an accurate FER estimation. Also, [49] developed an
analytical model for optimizing saturation throughput only while assuming utilization of
RTS/CTS mechanism by all contending transmitters. In comparison, work proposed in
this dissertation uses an empirically found idle time distribution to adaptively regulate
coexisting technology transmitters frame sizes. An extensive experimental campaign is
used to obtain and construct active and idle time distribution for 802.11 networks acting
as an interferer degrading coexisting network performance. Such empirical distribution
provides an accurate representation of medium availability state for both homogeneous
and heterogeneous 802.11 networks. Another advantage of the work presented in this
dissertation is that it does not require a feedback mechanism to obtain Packet Error Rate
(PER) for implementing the adaptive packet size method. Rather, it directly uses
statistical parameters of the obtained idle time distributions to regulate the transmitters
packet size. Nevertheless, PER is acquired and logged for this work to perform analysis
on the obtained results. Idle time distributions themselves can be constructed by any radio
interface measuring received signal strength (RSSI) by simple energy detection method,
22
as will be explained further in the following chapters. Adaptive packet size results
provided in Chapter 8 of this dissertation show a similar trend when compared to
Figure 2-1. Packet size saturates at a specific level for very low spectrum occupancy,
corresponding to low PER, and is limited by the implemented standard, ZigBee,
maximum packet size. This maximum packet size provides highest throughput compared
to smaller packet sizes. Nonetheless, smaller packet sizes demonstrate an improved
throughput compared to larger ones at higher spectrum occupancy levels, corresponding
to higher PER.
Figure 2-1. Saturation throughput vs. frame size for different FER values [31]
Finally, when considering wireless technology identification, work presented in this
dissertation is unique in that it provides a simple, yet effective, time-based feature
detection technique for wireless technology identification. A practical example of this
23
method is presented and validated in chapters that follow. Previous research presented in
[47] is of most interest when compared to the work presented herein, as it relies on features
extracted from spectrum power measurement to perform identification. Authors in [47]
used simulation to identify MAC layer implementation for transmitters using the spectrum.
Support Vector Machine (SVM) algorithm is used to perform the identification process.
Two power featuresreceived power mean and varianceand six time features, namely
minimum, median and maximum idle time, and busy time durations, were used. Notably,
time duration vary considerably in actual implementation based on throughput, number of
transmitters using the spectrum, and the PHY and MAC layers implementation for these
transmitters. As such, time durations obtained from simple MAC technique
implementation, with the assumption of fixed active durations, would not accurately
represent real world implementations. To calculate received power, implemented
simulation assumes a simple Rayleigh fading channel. This assumption constitutes a
simplification that does not represent real world environments with multi-path fading. The
results show the possibility of performing MAC layers identification, Aloha, Slotted
Aloha, CSMA/CA, and TDMA when using the aforementioned features and assuming a
single technology is using the spectrum. Wireless technology identification work
presented in this dissertation is based on extensive experimental results from 802.11
homogeneous and heterogeneous measurements at varying spectrum
occupancy/throughout levels. A set of temporal features extracted from active and idle
time distributions. Compared to previously discussed research, this dissertation presents
and validates a wireless technology identification method, not only MAC layer
implementation, but also using a comprehensive experimental campaign. Notably,
24
temporal-based features are used only to avoid uncertainty associated with power related
features resulting from power fluctuations dependent upon the communication
environment. Wireless technology identification, presented in Chapter 8, demonstrates
identification accuracies of 96.9% and 85.9% were achieved for homogeneous and
heterogeneous 802.11 networks, respectively. Also, number of transmitter identification
was investigated with satisfactory result. Details of the conducted study are offered in
Chapter 8.
2.4 GHz ISM band spectrum overview
The ISM band, spanning over 80 MHz from 2.4 GHz to 2.48 GHz, is the home for many
popular wireless technologies (e.g., Wi-Fi 802.11b/g/n, ZigBee, Bluetooth, Bluetooth Low
Energy (BLE), and others). These technologies share the same spectrum with overlapping
frequency channels. See Figure 2-2 for a visual approximation of frequency channels
belonging to different ISM band technologies.
802.11Wi-Fi has the widest frequency channel size22 MHz, with 11 overlapping
channels, of which 1, 6, and 11 are non-overlapping. 802.15.4 ZigBee has a more limited
frequency channel width of 2 MHz with frequency span of 5 MHz between center
frequencies of adjacent channels. 802.15.4 ZigBee has 16 non-overlapping in the ISM
band. Classic Bluetooth, on the other hand, uses a frequency hopping spread spectrum
access with 79 non-overlapping, 1 MHz channels. Finally, Bluetooth Low Energy (BLE)
has 40 channels, each 2 MHz wide.
802.11 networks, and to some extent 802.15.4 ZigBee, are under investigation in this
work. Consequently, more details on the MAC and physical layers of 802.11b/g/n and
25
802.15.4 ZigBee are provided in the following subsections. A thorough understanding of
these standards will aid in interpreting results in following chapters.
Figure 2-2. 2.4GHz ISM band frequency channels.
802.11b
802.11b was published in 1999 as an amendment to the original 802.11 standard [50]. The
standard supports data rates of up to 11Mbps. This subsection provides information on
802.11b Physical (PHY) and Medium Access Control (MAC) layers. Both have a direct
effect on spectrum temporal characteristics of transceivers implementing the standard.
Distributed coordination function (DCF)
IEEE 802.11b implements Distributed Coordination Function (DCF) as its MAC
technique for which a transmitter follows carrier sense multiple access/collision avoidance
(CSMA/CA) scheme. CSMA/CA uses a technique known as Clear Channel Assessment
(CCA) to ensure that the medium is vacant prior to performing a frame transmission. The
purpose of this procedure is avoiding collisions with other transmitters.
Figure 2-3 illustrates functionality of DCF. A transmitter with a frame in its queue choses
a random back-off value from a specific minimum contention window and continuously
26
senses the medium until it becomes idle. If the medium is occupied, transmitter freezes its
back-off counter. Once the medium becomes idle, transmitter starts decrementing its back-
off counter. This process is repeated until the back-off counter expires (i.e., reaches a value
of 0.) The transmitter then proceeds with frame transmission. Receiver replies with an
Acknowledgment (ACK) after short inter-frame spacing (SIFS) waiting time, provided the
frame was received correctly. Other transmitters should wait for distributed inter-frame
spacing (DIFS) waiting time after transmission ends before continuing to decrement their
back-off counters. If two transmitting nodes choose the same back-off value at the same
time, a collision will occur, and packet transmission will be rendered unsuccessful.
Colliding nodes will then move to a higher back-off stage wherein back-off window is
binary exponentially larger. Details of 802.11b MAC timing parameters can be found in
Table 1.
Notably, a transmitter using DCF will hold off frame transmission until its back-off
counter expires even if the medium remains idle during the entire process. This behavior
results in a low temporal efficiency whenever a single transmitter is attempting to access
the medium. Nevertheless, DCF provides a better chance of coexistence (between 802.11
transmitters strictly) when a large number of these transmitters are using the medium at
the same time.
27
SIFS DIFS
Back-off Back-off
Back-off counter
TX1 Frame counter counter
frozen
frozen decrement
RX1 ACK
Back-off
TX2 Back-off counter frozen counter Frame
decrement
Medium Busy Idle Busy Idle Busy
Figure 2-3. DCF functionality
Two CCA modes are generally used in wide band networks (e.g., 802.11 and 802.15.4
ZigBee) [51][52]:
1. Non-coherent carrier sensing: This method uses a fixed CCA power threshold
to detect spectrum occupancy. It is fast and low power, but prone to errors.
2. Coherent carrier sensing (preamble detection): This method demodulates the
signal and detects the preamble before backing off. It is slower and consumes
more power, but less prone to errors.
Commercially available 802.11 chipsets commonly implement coherent carrier sensing to
take advantage of its robustness. 802.15.4 ZigBee chipsets implement a non-coherent
carrier sensing method to conserve energy. Narrow band networks use energy detection
only.
28
Table 1. 802.11b timing parameters
Parameter
Value
SIFS 10 s
DIFS 50 s
ACK 202 s
MPDU size 1500 Byte
frame duration 1220 s
Complimentary Code Keying (CCK)
802.11b PHY layer supports data rates of 1, 2, 5.5 and 11Mbps. The higher data rates of
5.5 and 11Mbps, which are of interest in this work, use CCK modulation wherein data
ready for transmission is grouped in 8-bits sequences. Each sequence is then used to find
an 8-chips spreading code. Spreading codes are modulated using Differential Quadrature
Phase Shift Keying (DQPSK) and sent over the air at a rate of 11Mchip/s. An 11Mbps
data rate is then achieved, given that an 8-bits sequence was used to find the 8-chip
symbols [53][54].
CCK modulation works by using the aforementioned 8-bits data sequence to choose one
of 64 orthogonal code words. The 8-bits sequence is divided into 2-bits pairs known as
dibits (d0, d1, d2, d3), where d0 represents the two least-significant bits. These dibits are
then used to find phase parameters 1 to 4, as specified in Table 2.
Table 2. CCK phase parameters
Dibit
Phase parameter
00 0
01 /2
10
11 3/2
29
The phase parameters are then substituted in (1) to find the code word to be transmitted.
CCK functionality is depicted in Figure 2-4.
( + + + ) , ( + + ) , ( + + ) ,
= { ( + ) ( + + ) ( + ) } (1)
, , , ( + ) , ( )
Select
8 bits data Find phase DQPSK
spreading CCK code
sequence parameters Modulator
code
Figure 2-4. CCK functionality
802.11g
802.11g standard was published in 2003 [55]. The primary difference between 802.11b
and g resides in the PHY layer. 802.11g supports peak data rate of up to 54Mbps with
Orthogonal frequency-division multiplexing (OFDM) and 64-QAM modulation. Similar
to its 802.11b predecessor, 802.11g implements DCF for its MAC layer with few
differences in timing parameters. Nevertheless, the temporal efficiency effect of DCF is
more pronounced in 802.11g, resulting in average data rates much lower than the
standards peak data rate. This behavior results from the shorter frame duration of 802.11g,
compared to 802.11b due to its higher data rate. Temporal efficiency of the various
standards is discussed in greater details in the following chapters.
Table 3. 802.11g timing parameters
Parameter
Value
SIFS 10 s
DIFS 28s
ACK 30 s
MPDU size 1500 Byte
Frame duration 253 s
30
Orthogonal Frequency Division Multiplexing (OFDM)
OFDM works by modulating data into multiple narrowband adjacent subcarriers, which
are orthogonal in nature. Consequently, sidebands of each subcarrier has limited
interference effect on the other subcarrier. OFDM is characterized by its immunity to
frequency selective fading, as the channel is divided into narrowband subcarriers that are
affected individually[56].
In 802.11g, the 20MHz channel is divided into 52 subcarriers with 312.5 KHz carrier
spacing. Four subcarriers are used as pilot carriers while the remaining 48 are used as data
subcarriers. Data subcarriers are then modulated using 64-Quadrature Amplitude
Modulation (64-QAM) modulation resulting in peak data rate of 54 Mbps when using a
single spatial stream. OFDM functionality is depicted in Figure 2-5 [57]. The scrambler is
used to randomize bit streams to eliminate long strings of 1 and 0. Such long sequences
negatively affect time synchronization at the receiver. The sequences also prevent power
concentration in a particular narrow frequency band. Convolutional encoding is a type of
error-correcting code used in the scheme. Inverse Fast Fourier Transform (IFFT) is
intended to distribute data into different subcarriers. Finally, 64-QAM is a used as a digital
modulation scheme.
Convolutional QAM
Bit stream Scrambler IFFT
encoder modulator
Figure 2-5. OFDM functionality
31
802.11n
802.11n was introduced in 2009 and enabled data rates of up to 600Mbps [58]. The
standard implements several enhancements in both PHY and MAC layer. PHY layer
enhancements include beamforming, channel aggregation, and Multi-Input Multi-Output
(MIMO) data streams. Beamforming provides the transmitter with the capability of
tracking a particular receiver by changing the radiated beam shape, thus taking advantage
of multiple transmitter antennas. This functionality is limited by radio interface
capabilities, number of antennas, and 802.11n radio chip. Most of todays common
802.11n transmitters lack the latter functionality with number of antennas limited to two
for a transceiver.
Channel aggregation allows the choice of either 20 MHz or 40 MHz channel. A channel
of 20MHz is generally chosen in the 2.4 GHz ISM band due to its limited bandwidth of
80 MHz. MIMO allows for up to 4X4 antennas on the transmitter and receiver sides,
enabling higher data rates.
802.11n uses DCF as its MAC layer protocol. Nevertheless, the standard introduces frame
aggregation as MAC layer enhancements. This functionality leads to the foremost effects
on spectrum temporal characteristics.
For testing reported in this work, A-MPDU was used for 802.11n. This enabled a peak
data rate of 65Mbps for a single spatial stream and 64-QAM modulation type. See Table
4 for 802.11n timing parameters in the 2.4 GHz ISM band.
32
Table 4. 802.11n timing parameters.
Parameter
Value
SIFS 10 s
DIFS 28s
RIFS 2 s
Block ACK 36 s
MPDU size 1500 Byte
Variable depending on A-
Frame duration
MPDU size (189 s 8.9ms)
Frame aggregation
Frame aggregation is introduced in 802.11n, wherein multiple sub-frames can be
aggregated into one large aggregated frame [59]. The result is a considerable
improvement in temporal efficiency when compared to earlier standards. However, frame
aggregation increases the likelihood of interference with other non-802.11 CTs. Also,
802.11-b and -g terminals sharing the medium could suffer from performance degradation
resulting from extended 802.11n spectrum occupancy.
Two types of aggregation schemes are defined in 802.11n: 1) aggregated MAC service
data unit (A-MSDU) and 2) aggregated MAC protocol data unit (A-MPDU). A-MSDU
works by aggregating several MSDUs with a single MAC header, a PHY header, and
frame check sequence (FCS). See Figure 2-6.
A-MPDU is formed from multiple MPDUs, each having its own MAC header and FCS.
Each sub-frame has its own FCS and can be retransmitted independently. See Figure 2-7.
A-MSDU is more efficient, as it is characterized with a lower header-to-payload ratio.
However, it is also more prone to erroneous environments. 802.11n introduces block ACK,
33
as well, wherein each sub-frame can be acknowledged individually. This allows
independent retransmission of corrupted frames.
Ethernet frames
ENET ENET
Preamble Data Preamble Data
Hdr Hdr
802.11n headers A-MSDU 1 A-MSDU 2 ... 802.11n

MAC ENET ENET
Preamble Hdr Data Data FCS
Hdr Hdr Hdr
Figure 2-6. Aggregated MAC service data unit
A-MPDU 1 A-MPDU 2 A-MPDU n

MAC MAC MAC
Preamble Hdr
Hdr
Data FCS Preamble Hdr
Hdr
Data FCS . Preamble Hdr
Hdr
Data FCS
RIFS
Figure 2-7. Aggregated MAC protocol data unit
802.15.4 ZigBee
802.15.4 ZigBee is designed as a low power, low data rate technology, making it suitable
for wireless sensor network (WSN) and wireless medical devices applications [60]. The
standard supports a data rate of 250 kbps in the ISM band by implementing Direct-
Sequence Spread Spectrum (DSSS) coding and Orthogonal Quadrature Phase-Shift
Keying (OQPSK) digital modulation. ZigBee technology has a maximum packet size of
127 bytes, including header, and uses a CSMA/CA for regulating medium access while
operating in ad hoc mode [61].
A simulation using 802.11g/n and 802.15.4 heterogeneous networks MAC layer was
developed in Matlab. The simulation was intended to validate experimental results
presented herein. The setup supported multiple nodes so that each node can be configured
with its own 802.11 or 802.15.4 standard, frame size, and number of aggregated frames.
34
The same simulation was used for a ZigBee adaptive packet size case study, which is
presented later in this dissertation.
35
Chapter3: Experimental and Simulation Setup
This chapter provides a detailed description of both experimental and simulation setups
implemented in this work. For the experimental setup, 802.11 networks, including their
equipment and setup, are first described. Second, measurement equipment and their tasks
are presented. Finally, a description of algorithms implemented to analyze collected data
is provided. For computer simulation setup, simulation approach and parameters are
described for both 802.11 and 802.15.4 networks.
Experimental setup
The spectrum temporal characterization, empirical modelling and wireless technology
identification work presented in this dissertation was performed using a proper
experiments campaign that implemented an energy detection technique. This method can
be easily applied to any radio interface measuring received signal strength (RSS) values.
Power measurements were first performed on 802.11 networks using external
measurement devices. Detected signal was then processed to extract DC values and
temporal distributions. Next, this data was used to perform empirical modelling of 802.11
idle time distributions. Extracted distributions were also used in wireless technology
identification using machine-learning methods. The following subsections detail network
and measurement equipment and their layout.
Networks under test
The experimental setup utilized one-, two-, or three-pair 802.11 networks, each with an
access point acting as a transmitter (TX) and a station acting as a receiver (RX) positioned
36
at a prefixed separation distance of one meter. Equipment was deployed on wooden tables
at an elevation of one meter. See Figure 3-1.
Figure 3-1. Example of 802.11 test setup
Tests were conducted at the University of Oklahoma-Tulsa Wireless and Electromagnetic
Compliance and Designer (WECAD) center. 802.11 nodes were Mikrotik router boards
433UAH [62] that were interfaced with a R52Hn miniPCI network adapter supporting
802.11b/g/n networks using Atheros chipsets. See Figure 3-2.
Figure 3-2. Mikrotik 802.11 router boards and miniPCI network adapters
802.11 networks were each configured to operate using the standards under investigation
(i.e., 802.11b, 802.11g, and 802.11n). 802.11 transmitters operated using a Unix-based
37
operating system with a graphical user interface capable of controlling all of the 802.11
chip functionalities. The operating system included test software that was used for
generating traffic at a particular throughput [63]. Requested throughput values were
incremented for consecutive tests, starting from low throughput to saturation. Throughput
increment values were varied based on the 802.11 standard used, number of pairs, and
the standards combination under test.
A one-pair 802.11 network was used for first testing stage. Another pair was added to
second testing stage, and a third pair was included for third testing stage. All pairs were
configured to use the same 802.11 standard for each homogeneous networks test. The
following combinations were used for both 2-pair and 3-pair heterogeneous tests: 1)
802.11_b-g, b-n, and g-n; and 2) 802.11_b-b-g, b-b-n, b-g-g, g-g-n, b-g-n, b-n-n, and g-n-
n. Achieved throughput values and other 802.11 configuration parameters were logged
for each test.
Terminals were set to maintain a transmit power of 16dBm for homogeneous networks
tests. Such transmission power proved to be high, as 802.11 receiver RF front end might
be saturated and cause instability in the recorded throughput. Consequently, some of the
homogeneous network tests were repeated until stable throughput was achieved to ensure
consistent output. To avoid this problem in heterogeneous networks tests, transmission
power was dropped to 8dBm. This transmission power still ensures a signal to noise ratio
(SNR) high enough for distances considered in the test setup, making noise effect
negligible.
38
Measurement setup
Two NI Vector Signal Analyzers (VSAs) were configured to collect time and frequency
domain measurements [64]. The used devices are VSA are PXIe-5644R and PXIe-5663
for time domain and frequency domain measurements, respectively. Both devices were
running NI LabVIEW software. For frequency domain measurements, a channel width
of BW=20MHz and resolution bandwidth of RBW=400 kHz was used. Notably,
frequency domain measurements were used merely for DC comparison and validation of
the time domain measurements.
For time domain measurements, the VSAs local oscillator was pre-tuned to the center
frequency of channel in use. The selected I/Q rate determine the sampling rate for this
channel. A 10MS/s rate was used for measurement, and a moving averageserved as a
low pass filterwas implemented to smooth instantaneous power values. Note that the
smallest time parameter defined in all 802.11b, g and n standards MAC layers is the Short
Inter-Frame Spacing (SIFS). SIFS for 802.11b, g and n in the 2.4 GHz ISM band is 10s.
Assuming a sampling rate four times the Nyquist rate of (2*f), f being the highest
frequency component in the message signal, a sampling rate of 0.8 MS/s should suffice
for providing a satisfactory representation of the sampled signal. Time domain
measurements were used for all analysis and modelling work presented in this
dissertation.
VSAs were used to perform spectrum activity measurements of 802.11 networks using
energy detection method. Two VSAs were deployed adjacent to one another at a prefixed
distance from 802.11 transmitters. Distance between 802.11 transmitters and VSAs was
39
calibrated so that average power received from all transmitters was equal. Figure 3-3
shows a graphical representation of the test setup, including separation distances.
Figure 3-3. Graphical representation of test setup with separation dimensions
Energy detection algorithms
Two different measurement methods were employed in this research: time domain (TD)
and frequency domain (FD) [65]. In TD measurement, received power values from a
specific frequency band were recorded continuously at a predetermined sampling rate.
VSA swept a set of frequencies, collecting and storing received power values in FD
measurements.
Time domain algorithm
Time domain measurements were conducted at two phases. In phase 1, a VSA was tuned
to the center frequency of the 802.11 channel under use (e.g., channel 6, 2.437GHz).
Amplitude values were streamed as data file-to-disk during testing at a sampling rate of
10MS/s. In phase 2, a time domain algorithm processed collected data to extract DC
values, active time distributions, and idle time distributions.
The algorithm commences by extracting the power threshold separating channel active
and idle times. Next, the program analyzes the entire recording and compares power
40
averages with the found threshold. Power values above the threshold are assigned a 1,
representing activity; power values below the threshold are assigned a 0, representing
an inactive channel. Strings of consecutive 1s and 0s are then used to build active and
idle time distributions. These strings are also used test data to determine duty cycle.
The time domain analysis algorithm process is depicted in Figure 3-4. The following
provides a step-by-step description of the algorithms process:
1) Calculate the instantaneous power values.
2) Smooth the signal, using a moving average algorithm that serves as a low pass filter.
3) Calculate the power signal threshold by i) determining the power histogram for a set
of the data recordings, ii) fitting a Gaussian distribution to the lowest power hill in the
histogram, and iii) setting the threshold to five standard deviations from the peak of the
lowest power hill to ensure ample separation between 802.11 signal and noise floor
[66].
4) Analyze entire data recording by comparing calculated power averages to calculated
threshold (i.e., 1 represents channel activity and is assigned if the average power value
is above the threshold; 0 represents channel idle state and is assigned in case average
power values is below the threshold). As such, data recording is converted to 1s and
0s, representing active and idle periods.
6) Calculate consecutive active and idle time durations.
5) Calculate and save DC value.
41
Figure 3-4. Time domain (TD) algorithm pseudo-code.
Frequency domain algorithm
For frequency domain measurements, VSA first scans the medium and measures the
noise floor. A threshold value at least 5dB above the noise floor is set. VSA then samples
802.11 frequency channel. Average power value is constantly compared with the
reference threshold. Given that average power value is above threshold, an activity state
1 is logged, meaning the channel is occupied/utilized for this particular sample.
Conversely, if average power value is below threshold, logging a 0 indicates an idle
time state. Duty cycle indicates the level of channel utilization and is calculated by taking
the number of 1s over the total number of logged 0s and 1s (e.g., DC = [1s / (1s
+ 0s)]) for both time and frequency domain measurements [65]. Frequency domain
measurements aim solely at validating time domain duty cycle readings.
42
Frequency domain DC measurements were conducted in real-time. The algorithm
process included the following steps:
1) Set a VSA to sweep the 802.11 frequency channel bandwidth (e.g., 22MHz) under
test (e.g., channel 6).
2) Scan the medium and calculate the noise-floor peak power value.
3) Set the threshold value at least 5dB above the noise-floor to minimize false detection
and to distinguish signal from noise.
4) Run the VSA to sweep the selected 802.11 frequency channel.
5) Compare each power value in every sweep with the reference threshold.
6) Calculate DC, indicating the level of channel utilization in frequency domain [65]
by calculating the number of 1s over the total number of combined 0s and 1s.
7) Use cumulative moving average to assimilate DC values calculated in each sweep.
Figure 3-5 illustrates a pseudo-code of the used frequency domain algorithm.
FD measurements Algorithm
Input: Spectrum, Bandwidth, Channel-Band, Bin-Size
Output: Duty-cycle
Initialization:
1: Spectrum=2.40~2.48GHz; Bandwidth=80MHz;
Bin-Size = 100KHz; Channel-Band = 22MHz
2: Read Power Spectrum
3: CALC/Set Threshold value
Duty-Cycle:
1: Read Power Spectrum
2: for each Power-value in a Channel-Band do
if Power-value >= Threshold then
Activity-States = +1
end if
States-Number = +1
end for
3: Duty-cycle = Activity-States / States-Number
4: do Moving Average Duty-cycle
Figure 3-5. Frequency domain algorithm pseudo-code.
43
Test methodology
The following steps describe the testing protocol implemented to collect
aforementioned measurements of 802.11 networks:
- A pre-scan for the medium was conducted prior to testing to ensure no
interference from adjacent buildings in the ISM band and to measure noise floor.
Noise floor was measured less than -105dBm. An example of measured noise
floor is depicted in Figure 3-6.
- 802.11 network terminals were configured on the standard under study (802.11b,
802.11g and 802.11n), and 802.11 channel 6 was chosen as transmit channel.
- Starting from low throughput to saturation, pre-defined throughput values were
incremented in steps for consecutive tests. The increment step was defined based
on the 802.11 protocol under test. Figure 3-7 and Figure 3-8 depict examples of
spectrograms for 802.11g 25Mbps and 5Mbps throughputs, respectively. These
figures illustrate the effect of high throughput vs. low throughput on the spectrum
occupancy at a high level. As expected, higher throughput leads to higher detected
power levels for longer times on the spectrum. This results in higher measured DC
values. A detailed analysis investigating DC levels corresponding to various
802.11 networks and varying throughput levels is presented in the following
chapters.
44
Figure 3-6. Noise floor lab measurement.
- Achieved throughput values and other 802.11 configuration parameters were
logged for each test.
- Two NI VSAs were configured to collect and log time-domain and frequency-
domain measurements, as described in previous sections.
45
Figure 3-7. Spectrogram for 802.11g at 25Mbps.
Figure 3-8. Spectrogram for 802.11g at 5Mbps.
46
Simulation setup
MATLAB simulation was developed for both 802.11g and 802.15.4 ZigBee MAC layer
transmission. The purpose of these simulations is threefold:
1) Comparing simulation and experimental measurements achieved DC and
inactivity distributions for 802.11g networks, as well as highlighting some
behavior of large 802.11 networks that might be difficult to investigate through
experimental effort.
2) Creating an interfering medium to test for 802.15.4 ZigBee packet error rate (PER)
variation based on 802.11g traffic, highlighting sever effect of 802.11 networks on
CTs in the ISM band.
3) Assessing 802.15.4 ZigBee throughput enhancement when implementing variable
frame size based on 802.11g idle time distributions, which demonstrates possible
enhancements to CTs performance that might be obtained by 802.11 temporal
characterization.
802.11g simulation
Simulation implements the 802.11 MAC layer, DCF (Distributed Coordination Function),
in a MATLAB environment and defines multiple 802.11g nodes that can contend over
the medium. A node can be either saturated or Poisson. A saturated node has a full frame
queue at all times, as the name implies. In a Poisson node, frames arrive to the transmission
queue following a Poisson distribution with average arrival rate of . Varying results in
variable achieved throughput values. The following parameters can be controlled in the
47
developed simulation: number of transmitters, frame size, minimum contention window,
number of back-off stages, and average arrival rate.
According to current configuration, simulation provides total network throughput, channel
utilization, idle distribution, probability of collision, per transmitter back-off value, and
per transmitter back-off stages. Comparing simulation and experimental results is
beneficial for validating both approaches.
The following steps describe the 802.11g simulation process:
- Number of nodes and their types are selected (e.g., Poisson, saturated), and the
simulation parameters are chosen (e.g., arrival rates for Poisson nodes, contention
windows sizes).
- Simulation duration, including warm-up duration, is chosen.
- The simulation commences. 802.11 measurements are not logged during the
warm-up period.
- A node with a packet in its queue attempts to access the medium using DCF
functionality, described in Chapter 2, with the following three results:
o Successfully access the medium and perform transmission after waiting for
randomly selected back-off time and Distributed Inter Frame Spacing
(DIFS) value. 802.11g first back-off stage consists of 16 values ranging
from 0 to 15 time slots, where each time slot is 9s. A transmitting node
will wait for a DIFS value after an end-of-transmission prior to choosing a
random back-off value, where DIFS value is 28s.
o If the medium becomes occupied with other 802.11 transmissions during
the back-off period, the 802.11 transmitter waits for those to finish.
48
o If two transmitting nodes choose the same back-off value, a collision will
occur and packet transmission is rendered unsuccessful. Colliding nodes
will jump to a higher back-off stage for which the back-off window is
exponentially larger.
- The following data is logged for the post warm-up simulation duration, based on
the 802.11g transmission:
o Achieved throughput, which might be different from requested throughput
due to possible medium congestion and collisions, based on the number of
simulated nodes.
o Duty cycle.
o Active and idle time distributions.
o Medium occupancy durations, which are an array of durations at which the
medium was occupied with 802.11g transmission for later use in the
802.15.4 ZigBee simulation.
Table 5 provides timing parameters used in 802.11g simulation.
Table 5. 802.11g simulation parameters values
Parameter Value
SIFS 10 s
DIFS 28 s
Minimum contention window 0 144 s
ACK 30 s
802.11 MPDU size 1500 Byte
The following equations provide parameters contributing to calculated simulation timing
output. Note that timing outputs apply for both 802.11g and 802.15.4 simulations:
49
o Transmission time = PHY header + MAC header + packet payload +
SIFS + 2*propagation delay + ACK + DIFS
o Channel utilization time during Tx = PHY header + MAC header +
packet payload + ACK
o Collision time = PHY header + MAC header + packet payload + DIFS +
propagation delay
o Channel utilization time during collision = PHY header + MAC header +
packet payload
802.15.4 ZigBee simulation
802.15.4 ZigBee simulation implements 802.15.4 ZigBee MAC mechanism. ZigBee uses
DCF for distributed medium access with several variations from 802.11. To save energy,
ZigBee transmitter does not perform carrier sensing while decrementing its back-off
counter. Thus, it does not freeze the back-off counter in the event of a busy medium.
Rather, back-off counter is decremented until it reaches the penultimate back-off slot
where carrier is sensed for the last two time slots. In the event of occupied medium,
transmission is deferred and ZigBee terminal jumps to a higher back-off stage. In this
simulation a node can either have frame arrival resulting in a saturated queue or Poisson-
distributed frame arrival with average arrival rate of . Note that ZigBee has data rate of
250Kbps and time slot duration of 320s.
Notably, 802.11 nodes perform coherent clear channel assessment wherein they back off
only to other 802.11 transmitters. 802.15.4 ZigBee nodes perform non-coherent clear
channel assessment wherein they perform back off merely based on the Received Signal
50
Strength (RSS) value. The purpose of this behavior is to conserve energy. Both coherent
and non-coherent clear channel assessment methods are described in Chapter 2.
Developed simulation assesses 802.15.4 ZigBee coexistence in the presence of interfering
transmission on the same frequency channel (e.g. 802.11 transmissions). The simulation
utilizes medium occupancy information, which can be obtained from 802.11g simulation
to determine 802.15.4 ZigBee throughput and probability of collision. It is assumed that
802.11g is using coherent carrier sensing and, thus, does not back off for ZigBee
transmission [67]. Also, any collision between ZigBee and 802.11g transmission results
in a corrupted ZigBee packet.
The following parameters can be controlled in the developed simulation:
- Number of transmitters.
- Minimum contention window.
- Number of back-off stages.
- Packet size.
- Frame arrival rate.
- Medium occupancy durations.
Simulation output includes total network throughput, channel utilization, and packet error
rate. Simulation provides several other parameters unrelated to this work, thus, are not
described.
51
Chapter 4: Spectrum Characterization 802.11 Throughput and Duty
Cycle
This chapter presents experimental spectrum occupancy results obtained in this research,
including DC and throughput for single-pair and multi-pair homogeneous and
heterogeneous 802.11 networks [68][69]. Experimental and simulation results are
compared, and then presented in an effort to provide a more detailed analysis of DC results
and to extrapolate for a higher number of communicating 802.11 terminals. DC
measurements are validated by comparing time-domain and frequency-domain spectrum
occupancy results.
Homogeneous networks DC
Each of the three standards was assessed individually for a single-pair network. In this
configuration, an access point acts as a transmitter by sending UDP data at multiple
throughput values to a station that acts as a receiver.
Throughput and duty cycle results for one-pair networks

DC results per throughput for single-pair 802.11b, g, and n networks are shown in Figure
4-1.
Figure 4-1. 802.11b/g/n single-pair DC vs. throughput.
52
802.11g achieved maximum DC of 67.29% at throughput of 28.1 Mbps. The relatively
low DC observed for 802.11g results from overhead imposed by DCF standard. In this
scenario, the transmitter must perform a back-off at each single-data frame transmission.
802.11b achieves a higher maximum DC of 84.95%, at throughput of 7.1 Mbps. Although
both 802.11 b and g have similar back-off functionality, 802.11b transmits at the lower
data rate of 11Mbps, which results in a longer frame duration. Consequently, 802.11b
records higher DC for a much lower throughput.
802.11n achieves the highest DC of 97.97% with throughput reaching 56.4 Mbps. High
DC demonstrated by 802.11n is a direct result of the frame aggregation mechanism
implemented in the standard, rendering it an extremely time-efficient technology.
Throughput and duty cycle results for two-pairs networks
Spectrum occupancy for multiple pairs was investigated next. Figure 4-2 and Figure 4-3
depict throughput and DC for two-pair 802.11b. Test number labeling used in this chapter
refers to the standard being investigated, as well as the throughput value (e.g., G05N10
refers to two pairs: 802.11g with set throughput value of 5Mbit/s and 802.11n with set
throughput value of 10Mbit/s). Figure 4-2 shows that achieved throughput is divided
equally among contending pairs when offered traffic for both pairs reaches saturation
point.
53
Figure 4-2. Set throughput vs. achieved throughput for 802.11b two-pair network.
Figure 4-3. DC results for 802.11b two-pair network.
Total network throughput reaches 7.8Mbps at saturation, which is slightly higher than the
one-pair scenario. Recorded DC for two pairs reaches up to 92.26% for saturation. This
phenomenon is a consequence of a more efficient spectrum use for two pairs. Whenever a
one-pair transmitter is performing back-off, the second pair has a chance to proceed with
transmission.
Results for two-pair 802.11g, including throughput and DC, are shown in Figure 4-4
and Figure 4-5. Similar to 802.11b, saturation DC and total achieved throughput recorded
an increase for two-pair 802.11g. Saturation DC was 77.86%, indicating a 10% increase
over the single pair.
54
Figure 4-4. Set throughput vs. achieved throughput for 802.11g two-pair network.
Figure 4-5. DC results for 802.11g two-pair network.
Two-pair spectrum occupancy results for 802.11n are presented in Figure 4-6 and
Figure 4-7. A marginal increase of less than 1% in DC saturation is observed for two pair
when compared to a single pair. However, saturation aggregate throughput dropped to
52.2Mbps. 802.11n initially achieves DC of more than 97% for single-pair scenario.
Adding more transmitters did not significantly impact temporal efficiency. Nevertheless,
additional transmitters increased the chances of erroneous transmissions, which in turn led
to a decrease in total achieved throughput.
55
Figure 4-6. Set throughput vs. achieved throughput for 802.11n two-pair network.
Figure 4-7. DC results for 802.11n two-pair network.
Throughput and duty cycle results for three-pair networks
802.11 b, g, and n three-pair spectrum occupancy results are depicted in Figure 4-8 through
Figure 4-13. Figure 4-9 shows that maximum achievable DC of 92.48% was recorded for
802.11b, which is comparable to results for two pair. Maximum saturation aggregate
throughput drops to 7.6Mbps when compared to 7.8Mbps achieved for two pair as can be
seen in Figure 4-8.
802.11g three pair achieves maximum DC saturation of 80.65%. The highest aggregate
throughput for 802.11g was comparable to two pair. 802.11n three-pair DC remains at
approximately 98% and demonstrates a marginal increase over two pair. Maximum
achieved throughput dropped to 51.5Mbps.
56
Figure 4-8. Set throughput vs. achieved throughput for 802.11b three-pair network.
Figure 4-9. DC results for 802.11b three-pair network.
All transmitters achieved their requested throughput for tests characterized as below
saturation. DC for two pair with a specific aggregate throughput is comparable to DC
value at the same throughput level for single pair at a difference not exceeding 4%, in
most scenarios. For example, 802.11g two-pair test with 10Mbps throughput for both
pair-A and pair-B yielded a DC of 50%, which is comparable to DC of 48.48% achieved
for single pair with 20Mbps achieved throughput. Another example is 802.11n two pair
with achieved throughput of 10Mbps for pair-A and 20Mbps for pair-B. This test recorded
a DC of 69.81%, which is comparable to DC of 65.48% achieved for single pair with
30Mbps throughput. Similar observations apply when comparing three-pair DC to two-
pair.
57
Figure 4-10. Set throughput vs. achieved throughput for 802.11g three-pair network.
Figure 4-11. DC results for 802.11g three-pair network.
Figure 4-12. Set throughput vs. achieved throughput for 802.11n three-pair network.
Figure 4-13. Set throughput vs. achieved throughput for 802.11n three-pair network.
58
Duty cycle error analysis
This subsection provides error analysis for DC results obtained from time-domain and
frequency-domain measurements. Error analysis aims at validating both DC
measurement techniques. Time-domain DC measurements were obtained via offline
analysis, while frequency-domain DC measurements were calculated in real-time. Hence,
this analysis allows the flexibility to choose the technique that provides the best results
given resources available. Two 6.6 GHz VSAs (NI PXIe-5663E) were tested side by side.
The primary difference in measurements was an algorithm running on VSA-1 for
calculating DC in time domain and an algorithm running on VSA-2 for calculating DC
in frequency domain. Three validation methods were conducted to compare DC
measurements obtained from each VSA.
Comparing the mean and Standard Deviation:
Comparing standard deviation and mean values of time domain and frequency domain
measurement groups demonstrated identical variation from the mean, which indicates
statistical confirmation for both groups. See Table 6. Figure 4-14 details a test-by-test
comparison between DC measurements obtained in time domain and frequency domain.
Clearly, both approaches achieved similar results.
Table 6. Comparison between DC Statistics in time domain and frequency domain
Methodology Number Mean Standard Deviation

of tests
DC- time domain 128 75.282 24.462
DC- frequency domain 128 74.983 24.223
Differences 0.299 0.239
59
Figure 4-14. Comparison between DC measurements in time domain and frequency domain.
Linear Regression Analysis (LRA):
Linear Regression Analysis (LRA) was used to model the relationship and correlation
between TD and FD measurements, as shown in Figure 4-15. Given two equal size sets of
values; X and Y, LRA fits the group of observations to a line. This line represents the
best-fit line in a sense that the error between real Y values and corresponding predicted
values using fitted line equation is minimized [70]. Then a value r2 is calculated as a
measure of linear relationship between X and Y. The r2 is a fraction that assumes a value
between 0 and 1. It is computed using the following equation:
SSreg
r2 = 1 (2)
SStot
Where SSreg represents the sum of deviations between fitted line and (x, y) observations
and SStot is the sum of deviations between null hypothesis and (x, y) observations [71].
Null hypothesis is a horizontal line that passes through the mean of all Y observations. An
r2 value of 0 means that X provides no information on Y whereas a value of 1 means that
all (x, y) observations lie on the fitted line perfectly. An r of 0.9979 was calculated for
60
homogeneous DC tests representing a strong correlation between the two datasets of
measurements and confirming a trivial difference between the two groups.
Figure 4-15. Linear regression calculated between time domain and frequency domain.
Root Mean Square Deviation (RMSD):
Root Mean Square Deviation (RMSD), in other words Root Mean Square Error (RMSE),
was adopted to find the average difference between the two DC measurement datasets.
See (3).
=1(, , )2
RMSD = (3)

Xobs is the observed value and Xmodel is the modelled value at time/place i. Results indicated
RMSD = 1.18219, which represents the mean deviation between time-domain and
frequency-domain DC measurements in terms of percentage (i.e., 1.18219%), which is
negligible.
Heterogeneous networks DC
This section provides spectrum occupancy results, including DC and throughput for mutli-
pair heterogeneous 802.11 networks in the 2.4 GHz ISM band.
61
Throughput and duty cycle results for two-pairs
The setup included two access points, transmitters (TX),sending UDP data to two stations
acting as receivers (RX). Various throughput values were set for each node in the
network, ranging from low throughput to saturation.
Figure 4-16 depicts set throughput vs. achieved throughput for two heterogeneous pairs.
One pair was setup as 802.11g, and the other as 802.11n. The figure demonstrates that
802.11n dominates 802.11g and achieves a much higher percentage of peak data rate than
802.11g in the scenario of saturation. This behavior is a result of frame aggregation
implemented in 802.11n. Thus, an 802.11n device will transmit multiple frames after
accessing the medium, whereas 802.11g releases frequency channel after each single-
frame transmission. Both 802.11 g and n achieve desired throughput when below
saturation. Figure 4-17 shows measured DC for 802.11gn combination. A maximum DC
of 97.88% was recorded, which compares to results obtained from 802.11n single pair.
In addition, total achieved DC for below saturation is comparable to aggregate DC of
corresponding single pair tests with the same throughput. For example, two-pair tests
using 10Mbps achieved throughput for both 802.11g, and 802.11n recorded a DC of
47.1%. This compares to 47.38% aggregate DC for both 802.11g and n single-pair tests
with 10Mbps achievable throughput.
Figure 4-18 demonstrates set vs. achieved throughput for 802.11bn. Similar to 802.11gn,
the transmitter using 802.11n dominates the spectrum, leaving little chance for 802.11b
communication. Figure 4-19 depicts DC results with maximum achievable DC of
98.06%. Results are similar to 802.11g-n for below saturation in that measured DC is
62
comparable to single-pair aggregate DC resulting from the two corresponding tests with
achieved throughput values equal to the two-pair scenario.
Set throughput vs. achieved throughput results for 802.11g-b tests are shown in Figure
4-20. 802.11b achieved a higher percentage of its peak data rate when compared to
802.11g under saturation. Because of 802.11b lower data rate, its frame duration is much
longer than that of 802.11g. Consequently, 802.11b occupies the medium for much longer
durations compared to 802.11g, thus resulting in the higher percentage of achieved
throughput to peak data rate. Figure 4-21 depicts achieved DC for 802.11g-b test.
Maximum achieved DC reached 88.17%. This constitutes a slight increase over the
802.11b single-pair scenario.
Figure 4-16. Set throughput vs. achieved throughput for two heterogeneous pairs: 802.11gn (GN).
Figure 4-17. DC value for two heterogeneous pairs: 802.11gn.
63
Figure 4-18. Set throughput vs. achieved throughput for two heterogeneous pairs: 802.11bn (BN).
Figure 4-19. DC value for two heterogeneous pairs: 802.11bn.
Figure 4-20. Set throughput vs. achieved throughput for two heterogeneous pairs: 802.11bg (BG).
Figure 4-21. DC value for two heterogeneous pairs: 802.11bg.
Throughput and duty cycle results for three-pairs
In this subsection, spectrum utilization for 802.11b, g, and n three-pair heterogeneous
networks combinations were investigated. Figure 4-22 and Figure 4-23 depict achieved
64
throughput and measured DC for 802.11b-b-g combination. Pairs A and B operated on
802.11b, whereas pair C operated on 802.11g. The effect of 802.11b for three pairs on
802.11g transmitter was similar to two-pair scenario where 802.11g pair achieved only
5Mbps of the approximately 30Mbps maximum achievable throughput. Two 802.11b
pairs achieved analogous throughput values of 3.3 and 2.9Mbps. Maximum saturation DC
of 87.89% was recorded, which is comparable to 802.11bg two-pair scenario. Achieved
throughput and DC for 802.11g-g-b combination are shown in Figure 4-24 and Figure
4-25. Although three-pair combinations achieved similar throughput values, 802.11g has
a peak data rate five times greater than 802.11b. Maximum saturation DC of 87.94% was
recorded.
Figure 4-22. Set throughput vs. achieved throughput for three heterogeneous pairs: 802.11bbg (BBG).
Figure 4-23. DC value for three heterogeneous pairs, 802.11bbg.
65
Figure 4-24. Set throughput vs. achieved throughput for three heterogeneous pairs: 802.11ggb (GGB).
Figure 4-25. DC value for three heterogeneous pairs: 802.11ggb (GGB).
Figure 4-26 through Figure 4-29 depict achieved throughput and DC for 802.11bbn and
802.11nnb combinations. Examining 802.11b and n three-pair combinations, it can be seen
that 802.11n pairs dominate the spectrum with higher achieved throughput to peak data
rate. 802.11b pairs were unable to achieve throughput higher than 1.2Mbps for either one
of the two combinations at saturation. Saturation DC reached 96.21% and 98.65% for
802.11bbn and 802.11nnb, respectively.
66
Figure 4-26. Set throughput vs. achieved throughput for three heterogeneous pairs: 802.11bbn (BBN).
Figure 4-27. DC value for three heterogeneous pairs: 802.11bbn (BBN).
Figure 4-28. Set throughput vs. achieved throughput for three heterogeneous pairs: 802.11nnb (NNB).
67
Figure 4-29. DC value for two heterogeneous pairs: 802.11nnb (NNB).
Set throughput vs. achieved throughput and DC results for 802.11ggn and 802.11nng
combinations are shown in Figure 4-30 through Figure 4-33. 802.11g performed poorly in
both combinations, which is a similar behavior to 802.11gn two-pair scenario. This is a
result of 802.11n frame aggregation depriving 802.11g transmitters from frequently
accessing the medium. This is reflected in achieved DC, which reached up to 97.2%,
primarily for 802.11n transmissions.
Figure 4-30. Set throughput vs. achieved throughput for three heterogeneous pairs: 802.11ggn (GGN).
Figure 4-31. DC value for three heterogeneous pairs: 802.11ggn (GGN).
68
Figure 4-32. Set throughput vs. achieved throughput for three heterogeneous pairs, 802.11nng (NNG).
Figure 4-33. DC value for three heterogeneous pairs, 802.11nng (NNG).
Results of 802.11b-g-n combination are presented in Figure 4-34 and Figure 4-35 and
confirm previous findings. At saturation throughput, 802.11n transmitter scored the
highest percentage of achieved throughput-to-peak data rate, achieving 60.46% of its peak
data rate. 802.11b and 802.11g achieved 7.81% and 3.89%, respectively, of their peak data
rates.
Duty cycle error analysis
This subsection provides error analysis for DC results obtained from time-domain and
frequency-domain measurements. The two aforementioned VSAs were used to collect
time-domain and frequency-domain measurements. Similar to the homogeneous scenario,
69
the algorithm operating on VSA-1 calculated DC in time domain, and the algorithm
operating on VSA-2 calculated DC in frequency domain. Three validation methods were
conducted to compare DC measurements obtained from each VSA.
Figure 4-34. Set throughput vs. achieved throughput for three heterogeneous pairs: 802.11g, 802.11b, and
802.11n (GBN).
Figure 4-35. DC value for three heterogeneous pairs: 802.11g, 802.11b, and 802.11n (GBN).
Comparing the mean and Standard Deviation:
Standard deviation and mean values of time-domain and frequency-domain measurement
groups show similar values, as shown in Table 7. This suggests that statistically both
groups have similar distributions.
70
Table 7. Comparison between DC Statistics in time domain and frequency domain
Methodology Number of tests Mean Standard Deviation
DC- time domain 343 88.3162 11.97395
DC- frequency domain 343 87.5784 11.92431
Differences 0.7378 0.04964
Figure 4-36 depicts a test-by-test comparison between DC measurements obtained in
time domain and frequency domain. Both methods achieved similar results.
Figure 4-36. Comparison between DC measurements in time domain and frequency domain.
Linear Regression Analysis (LRA):
LRA was also used to model the relationship and estimate the correlation between time-
domain and frequency-domain 802.11 heterogeneous networks measurements, as shown
in Figure 4-37. The value R=0.9925 was calculated and represents strong goodness of fit
(i.e., correlation) between the two dataset measurements. This result confirms a negligible
difference between the two groups.
71
Root Mean Square Deviation (RMSD):
RMSD was also used to find the average difference between the two time-domain and
frequency-domain DC measurements for the heterogeneous 802.11 networks. Results
provided an RMSD value of 1.37717. As explained earlier, this value represents the mean
deviation between the measurement sets in percentage. This value is negligible.
Figure 4-37. Linear Regression time domain vs. frequency domain for heterogeneous network.
Simulation results
This subsection provides a comparison for DC and throughput results between
simulation and experimental tests. Figure 4-38 shows DC at incrementing throughput
values for a single 802.11g node. Although results are nearly identical, experimental
measurements achieved slightly higher DC than simulation for all throughput values.
72
50
11g Experimental test
45 11g Simulation
40
35
30
DC (%)
25
20
15
10
5
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Throughput (Mbps)
Figure 4-38. Simulation and experimental DC and throughput comparison for a below saturation 802.11g
one-pair network.
Figure 4-39 compares maximum achieved DC and throughput for one, two, and three
802.11g nodes. Contrary to the below saturation scenario, simulation DC for the
saturation scenario was slightly higher than the experimentally measured DC. The most
significant difference for a single-pair saturation test was 5.63%. Differences did not
exceed 2.08% for all other saturation and below-saturation tests. All experimental and
simulation results followed the same trend.
Saturation results for both experimental and simulation showed that as the number of
transmitter increases, DC increases. Notably, the rate of DC increase is lower for three pair
when compared to two pair. Total network throughput increased for two pair, and then
dropped for three pair and resulted from an increased number of collisions reported in the
simulation.
73
Figure 4-39. Simulation and experimental DC and throughput comparison for 802.11g one-, two- and
three- pair saturated networks.
Figure 4-40 presents extrapolated throughput and DC results for a large number of
transmitters using 802.11g simulation. Highest achieved throughput was recorded for two
transmitters. Total throughput starts dropping after populating the network with three or
more transmitters. Nevertheless, DC continues to increase as the network is populated with
more transmitters. Rate of increase becomes low when populating the network with more
transmitters.
74
802.11g simulation
100
90
80
Throughput / DC (%)
70
60
50
Throughput
40 Duty cycle
30
20
10
0
0 10 20 30 40 50 60 70 80 90 100
Number of nodes
Figure 4-40. 802.11g simulation extrapolated throughput and DC values for a large number of transmitters
Aggregated frame size has a significant effect on heterogeneous 802.11 network
throughput and wireless coexistence, as has been discussed earlier. An experiment was
conducted using simulation where one 802.11g node was coexisting with nine 802.11n
nodes. Two 802.11n aggregated frame sizes were used10 aggregated frames and 42
aggregated frames.
Figure 4-41 shows achieved per node throughput. 802.11g node achieves lower throughput
as the number of 802.11n aggregated frames increases. Nonetheless, all nodes, including
the nine 802.11n and one 802.11g nodes, have uniform medium access, as can be seen in
Figure 4-42. Even though the opportunity for medium access is fairly distributed, 802.11g
nodes send only one frame per medium access whereas 802.11n nodes send multiple
frames per medium access. This phenomenon deprives 802.11g from medium access for
longer periods of time and causes sever degradation in 802.11g node performance when
coexisting in the same vicinity with 802.11n nodes.
75
Figure 4-41. Per transmitter throughput for 802.11g/n shared medium for various numbers of 802.11n
aggregated frames.
Figure 4-42. Medium access distributions for 802.11g/n shared medium at various numbers of 802.11n
aggregated frames
76
Conclusion
This chapter presented a comprehensive study of 802.11b, g and n spectrum occupancy.
DC and throughput levels were presented and discussed for one, two, and three
communicating pairs. Both 802.11 homogeneous and heterogeneous networks were
investigated. Tools developed for time domain and frequency domain were used to obtain
DC measurements. Error analysis between time domain and frequency domain DC
measurements was presented, as well. Analyses included four statistical methods for
confirming equivalence between two methods for obtaining DC measurements.
77
Chapter 5: Spectrum Characterization 802.11 Experimental Time
Distributions
This chapter provides 802.11 time distribution results obtained from test performed for
the work presented in this dissertation. Results detail activity and idle time distributions
for both homogeneous and heterogeneous 802.11 networks in the 2.4 GHz ISM band.
Distributions were obtained from time domain measurements and represent time
fragment distributions for active and inactive periods during a specific test. Active periods
are times during which a channel is occupied with a transmission; idle periods are times
during which the channel is available. Along with duty cycle (DC) values, these
distributions provide a detailed analysis of channel traffic patterns. Results for one-, two-
, and three-pair 802.11 networks are presented and discussed below. Details in this
chapter highlight temporal efficiency of various 802.11 standards, as well as their
aggressiveness towards other 802.11 standards. Idle time distribution simulation results
for an 802.11g network are also presented and compared with empirical distribution
results.
Homogeneous networks
This section offers results obtained from homogeneous experimental tests. The 802.11
channel was populated with networks composed of one-, two-, or three-transceiver pairs.
Networks were configured to operate on 802.11b, g, or n for any given test.
78
Idle time distributions
Idle time distribution provides valuable information on channel availability to the CT,
which can be used to estimate appropriate frame size and time of channel access. The aim
is to minimize the probability of collision with 802.11 transmissions.
Various behaviors can be observed in three distinct idle-time duration regions for all idle
time distributions presented in this work. The first region represents idle-time fragment
durations equal to or below the SIFS value and corresponds with up to 50% of total idle-
time fragments. They represent inter-frame spacing before ACK transmission. The
second idle-time region represents the DCF standard minimum contention window.
Distribution in this region depends on the number of transmitters and offered traffic. It
also assumes an exponential shape when more than one transmitter utilizes the network.
Idle-time fragment distribution is uniform when the network is populated with only one
saturated transmitter. The third idle-time region includes idle-time fragments with
duration longer than the minimum contention window. Distribution in this region depends
mainly on offered traffic and throughput distribution between contending transmitters, as
discussed below.
Figure 5-1 illustrates one pair idle-time CDFs for 802.11b, g and n. SIFSs account for
approximately 50% of total idle-time fragments presented. Remaining time fragments are
distributed between DCF minimum contention window (i.e., second region) and higher
idle-time durations (i.e., third region), depending on throughput. Time fragments
resulting from minimum contention window account for a higher percentage of the total
79
idle time for higher throughput. Figure 5-1 also indicates that idle-time fragments in the
second region are uniformly distributed. This is a direct result of the DCF functionality,
as discussed earlier. Notably, the second idle-time region for 802.11b is larger than that
of 802.11g/n, primarily due to the wider minimum contention window of 802.11b with
32 time slots as compared to 802.11g and n with 16 time slots.
Regarding the third region, idle-time distribution is primarily dependent on frame arrival
rate and transmitter queue state. Figure 5-1 shows that 802.11g and n 10Mbps are nearly
the same for idle-time CDFs, duty cycle, and exhibited behavior. 802.11g and n 20Mbps
distributions exhibit similar behavior. The reason for the observed similarity is that peak
data rate for 802.11n is 65Mbps whereas peak data rate for 802.11g is 54Mbps, thus both
standards will have comparable frame arrival rate for throughput values much lower than
saturation. CDFs for the two standards diverge at higher throughput values.
Figure 5-1 also shows that 802.11g saturates at 67% DC at the point where nearly 99%
of idle-time fragments are either SIFS or back-off values resulting from minimum
contention window. This means that even with a saturated queue, channel is available
almost 33% of the time when a single 802.11g pair is occupying the medium. However,
802.11n saturates at 98% DC resulting from frame aggregation. Hence, the standard is
extremely time efficient. Time fragments resulting from SIFSs and minimum contention
window equal approximately 95% of total idle time fragments, given 802.11n saturation.
802.11b saturates at DC of 85%.
80
1 1 1
0.9 0.9 0.9
0.8 0.8 0.8
0.7 0.7 0.7
0.6 0.6 0.6

CDF
0.5 0.5 0.5
0.4 0.4 0.4

11b, 1Mbps, DC=13% 11g, 5Mbps, DC=12%
11b, 3Mbps, DC=35% 11g, 10Mbps, DC=24% 11n, 20Mbps, DC=45%
0.3 0.3 0.3
11g, 15Mbps, DC=36% 11n, 50Mbps, DC=87%
11b, 5Mbps, DC=58%
11g, 20Mbps, DC=48% 11n, 10Mbps, DC=23%
0.2 11b, 6Mbps, DC=69% 0.2 0.2
11g, 25Mbps, DC=62% 11n, 1Mbps, DC =3%
11b, 7Mbps, DC=81%
11g, Sat, DC=67% 11n, Sat, DC=98%
0.1 11b, 8Mbps, DC=85% 0.1 0.1
0 0 0 0 1 2 3 4
0 0
1 2 3 4 5 1 2 3 4
10 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10
(a) (b) (c)
Inactivity time duration (us)
Figure 5-1. Idle-time distributions for: a) 802.11b, b) 802.11n and c) 802.11g one pair
Figure 5-2, Figure 5-3, and Figure 5-4 depict two- and three-pair idle-time CDFs for
802.11b, g and -n, respectively. Test runs for identical standards with comparable
throughput were shown to have similar DC. However, idle-time CDFs had dissimilar
third region patterns, depending on throughput distribution among transmitters. Idle-time
fragments were found to spread over a wide range of time durations when throughput was
distributed more equally among transmitters. Nevertheless, idle-time fragments become
more concentrated around a certain value whenever one transmitter dominates network
throughput. See dashed curves in the figures below. Idle-time fragments in the second
region have an exponential distribution when two or more pairs at high throughput value
occupy the medium when compared with uniform distribution observed for one-pair
scenario. This scenario is a direct result of the transmitter back-off counter freezing when
the medium is sensed busy during the transmission of another node. This leads to further
fragmentation of second region idle-time intervals.
81
1
0.9
0.8
0.7
0.6 B: 1M, B: 1M, DC=25%

B: 1M, B: 3M, DC=49%
CDF
0.5 B: 1M, B: 5M, DC=10%

B: 2M, B: 2M, DC=55%
0.4 B: 2M, B: 5M, DC=81%
B: 3M, B: 3M, DC=73%
0.3 B: 3M, B: 5M, DC=90%
B: Sat, B: Sat, DC=92%
0.2 B: 1M, B: 1M, B: 1M, DC=38%
B: 1M, B: 1M, B: 5M, DC=85%
0.1 B: 5M, B: 5M, B: 1M, DC=88%
B: Sat, B: Sat, B: Sat, DC=93%
0
0 1 2 3 4 5
10 10 10 10 10 10
Figure 5-2. 802.11b idle time distributions for two and three pairs
0.9
0.8
0.7 G: 5M, G: 1M, DC=16%

G: 5M, G: 5M, DC=26%
0.6
G: 10M, G: 1M, DC=28%
CDF
0.5 G: 15M, G: 1M, DC=43%

G: 5M, G: 15M, DC=51%
0.4 G: 20M, G: 1M, DC=52%
G: 5M, G: 20M, DC=70%
0.3 G: Sat, G: Sat, DC= 78%
G: 5M, G: 5M, G: 5M, DC=37%
0.2
G: 5M, G: 5M, G: 10M, DC=50%
G: 10M, G: 10M, G: 10M, DC=71%
0.1
G: Sat, G: Sat, G: Sat, DC=77%
0
0 1 2 3 4 5
10 10 10 10 10 10
Figure 5-3. 802.11g idle time distributions for two and three pairs
82
1
0.9
0.8
0.7
0.6
CDF
N: 10M, N: 1M, DC=27%

0.5
N: 10M, N: 10M, DC=47%
0.4 N: 20M, N: 1M, DC=48%
N: 30M, N: 1M, DC=69%
0.3 N: 10M, N: 20M, DC=70%
N: 50Mbps, N: 1Mbps, DC=89%
0.2 N: Sat, N: Sat, DC=98%
N: 10M, N: 10M, N: 10M, DC=75%
0.1 N: Sat, N: Sat, N: Sat, DC=99%
0
0 1 2 3 4 5
10 10 10 10 10 10
Figure 5-4. 802.11n idle time distributions for two and three pairs
Activity distributions
802.11b, g and n activity distributions for one-, two-, and three-saturation tests are
depicted in Figure 5-5, Figure 5-6 and Figure 5-7, respectively. Note that these
distributions are not normalized and that they represent time duration, count of ACK
frames, and count of data frames. The resulting frame and ACK durations are
straightforward for both 802.11b and g, as frame size is fixed. Thus, most frame durations
are fixed around a certain value because they primarily depend on data rate. As for
802.11n, activity distribution depends on the number of aggregated frames, which is in
turn is affected by offered traffic, assuming a large enough queue. This behavior is clearly
demonstrated from activity distributions as 802.11n throughput is increased. See Figure
5-8 for 802.11n one-pair activity distributions.
83
4
x 10
10
One pair 11b, Sat
Two pairs 11b, Sat
8 Three pairs 11b, Sat
Count
4
0
0 1 2 3 4
10 10 10 10 10
Activity time duration (us)
Figure 5-5. 802.11b saturation activity distributions for one- two- and three pairs
5
x 10
3.5
One pair 11g, Sat
Two pairs 11g, Sat
3 Three pairs 11g, Sat
2.5
2
Count
1.5
0.5
0
0 1 2 3 4
10 10 10 10 10
Figure 5-6. 802.11g saturation activity distributions for one- two- and three pairs
4
x 10
3
One pair 11n, Sat
Two pairs 11n, Sat
2.5 Three pairs 11n, Sat
2
Count
1.5
0.5
0
0 1 2 3 4
10 10 10 10 10
Figure 5-7. 802.11n saturation activity distributions for one- two- and three pairs
84
An interesting observation for all three standards is that the number of detected frames
and ACKs were larger for two pairs when compared with both the one-pair and the three-
pair cases. The reason for this behavior is the following: 1) for the one-pair case,
requested throughput is achieved without any risk of collision. However, the spectrum
resource is not fully utilized due to the DCF procedure; 2) transmitters achieve a more
efficient use of the frequency spectrum when more data frames are sent with a very low
potential probability of collision in the two-pair case; 3) given three pair scenario,
transmitters occasionally send frames at lower data rate due to the increased number of
errors when compared with two- and one-pair cases. This in turn results in fewer
transmitted frames per unit time. See table IV for 802.11g/n saturation achieved
throughput.
Table 8. 802.11n/g saturation throughput
standard Number of Achieved Throughput

pairs (Mbps)
802.11g One 28.1
Two 29.4
Three 28.3
802.11n One 56.4
Two 52.2
Three 51.5
802.11n one-pair activity CDFs while increasing throughput from 1Mbps to saturation
are depicted in Figure 5-8. Clearly, the size of aggregated frames becomes larger as
throughput increases. This in turn leads to a lower number of aggregated frames
transmitted on the channel after a certain throughput level is achieved. As a result, the
transmitted frames become larger.
85
5
x 10
3
1Mbps
10Mbps
2.5
20Mbps
30Mbps
2 40Mbps
50Mbps
Count
Sat
1.5
0.5
0
0 1 2 3 4
10 10 10 10 10
Figure 5-8. 802.11n one pair activity distribution
Heterogeneous networks
This section presents experimental results of heterogeneous networks tests. Distributions
include a combination of two- and three-pair tests running different standards where each
pair can be 802.11b, g or n. Results aid in understanding temporal behavior of mixed
environments that contain multiple networks operating on different 802.11 standards
(e.g., apartment buildings and complexes, shopping areas, downtown buildings, and other
like settings).
Idle time distributions
Two- and three-pair heterogeneous idle-time distributions for 802.11g and n (and not
802.11b) are presented in this chapter to avoid redundancies. 802.11bgn three-pair
combination results are presented as well. Figure 5-9 shows idle-time CDF for two-pair
heterogeneous network tests, including an 802.11gn combination. Throughput for each
transmitter was incremented from low values to saturation. Select representative tests are
shown in Figure 5-9. Results for two-pair heterogeneous networks idle-time distribution
were similar to homogeneous networks in that idle-time distributions are divided into
86
three regions. Two tests (see dashed curves in Figure 5-9) demonstrated variable behavior
when SIFS accounted for less than 50% of total idle-time count. The decrease in SIFS
percentage resulted from missing ACKs, as indicated by dashed curves in Figure 5-12
below802.11g/n activity distributions.
0.9
0.8
0.7
0.6 G: 5M, N: 10M, DC=35%

G: 10M, N: 10M, DC=47%
CDF
0.5 G: 5M, N: 20M, DC=57%

G: 20M, N: 10M, DC=64%
0.4
G: Sat, N: 10M, DC=66%
0.3 G: 5M, N: 30M, DC=70%
G: 10M, N: 20M, DC=71%
0.2 G: 20M, N: 20M, DC=76%
G: Sat, N: 30M, DC=79%
0.1 G: Sat, N: Sat, DC=98%
0
0 1 2 3 4 5
10 10 10 10 10 10
Figure 5-9. Two pairs heterogeneous networks idle time distribution
Figure 5-10 depicts idle-time CDFs for three pair 802.11g and n heterogeneous networks,
including ggn and gnn combinations. Second and third regions merge when there is no
clear separating margin observed between the two regions. This results from further
fragmentation of idle-time durations in both regions.
87
1
0.9
0.8
0.7
0.6
CDF
0.5 G: 5M, G: 5M, N: 10M, DC=54%

G: 5M, G: 5M, N: 20M, DC=70%
0.4 G: 5M, G: 10M, N: 10M, DC=71%
G: 10M, G: 10M, N: 20M, DC=89%
0.3 G: Sat, G: Sat, N: Sat, DC=97%
N: 10M, N: 10M, G: 5M, DC=66%
0.2 N: 20M, N: 30M, G: 10M, DC=95%
N: Sat, N: Sat, G: Sat, DC=96%
0.1
0
0 1 2 3 4 5
10 10 10 10 10 10
Figure 5-10. Three pairs heterogeneous networks idle time distribution
Figure 5-11 depicts idle time-CDFs for three-pair 802.11bgn combination. The effect of
802.11b on idle-time distribution is minimal, as both 802.11g and n have smaller minimum
contention window that results in significantly more frequent medium access.
Consequently, the second idle-time region appears to be similar to that of 802.11gn two-
pair CDFs. The third-idle time region did not exceed 2000s of idle time-fragment for
offered throughput values. This behavior is an outcome of extended spectrum occupancy
resulting from 802.11b and n traffic. Also, 802.11b and g, lacking frame aggregation
capability, would likely attempt to access the spectrum at a higher frequency to achieve
their requested throughput when compared with 802.11n.
88
11g/11b/11n CDFs for three pairs
1
0.9
0.8 G: Sat, B: Sat, N: Sat, DC=98%

G: 5M, B: 1M, N: 20M, DC=72%
0.7 G: 5M, B: 1M, N: 30M, DC=86%
G: 5M, B: 1M, N: 40M, DC=95%
0.6 G: 5M, B: 5M, N: 10M, DC=84%
G: 5M, B: 5M, N: 20M, DC=91%
CDF
0.5
G: 5M, B: 5M, N: Sat, DC=98%
0.4 G: 5M, B: Sat, N: Sat, DC=98%
G: 10M, B: 1M, N: 10M, DC=61%
0.3 G: 10M, B: 5M, N: 10M, DC=87%
G: 10M, B: 5M, N: 20M, DC=90%
0.2 G: 10M, B: 5M, N: 40M, DC=96%
G: 20M, B: 1M, N: 10M, DC=78%
0.1 G: 20M, B: 1M, N: 40M, DC=96%
G: 20M, B: 1M, N: Sat, DC=98%
0
0 1 2 3 4 5
10 10 10 10 10 10
Figure 5-11. Three pairs bgn heterogeneous networks idle time distribution
Activity distributions
Activity-time fragment distributions for heterogeneous networks are of interest, as they
aid in standard identification through energy detection, which is presented as an
application in Chapter 8 of this document. Time duration of activity fragments represents
transmitted frame duration, which is related to frame size, data rate, and frame
aggregation. Detecting frame duration through energy detection provides an effective
way for standard identification primarily because it does not require demodulation of
received frames.
Figure 5-12 depicts 802.11g/n two-pairs heterogeneous networks activity CDFs for select
tests. Frames belonging to the two standards can be distinguished by their time duration.
802.11b and g frame durations are generally fixed due to fixed frame size. A small
number of these frames might have a specifically longer duration when an 802.11g
terminal transmits at certain data rate that is lower than 54Mbps.
89
802.11n frame durations are variable and result from frame aggregation, which depends
upon offered traffic, number of datagrams in the transmitter queue, and queue size.
Similar behavior can be observed in Figure 5-13 for three pair 802.11b/g/n combination.
5
x 10
3
G: 10M, N: 10M
G: 5M, N: 20M
2.5 G: 20M, N: 10M
G: 5M, N: 30M
G: 10M, N: 20M
2 G: 20M, N: 20M
G: 8M, N: 30M
Count
1.5 G: 10M, N: 30M

G: Sat, N: Sat
0.5
0
0 1 2 3 4
10 10 10 10 10
Figure 5-12. Two pairs heterogeneous networks activity distribution
5
x 10
3
G: 5M, B: 1M, N: 20
G: 5M, B: 1M, N: 30
G: 5M, B: 1M, N: 40
2.5
G: 10M, B: 1M, N: 10
G: 10M, B: 5M, N: 10
G: 10M, B: 5M, N: 30
2 G: 10M, B: 5M, N: 40
G: 20M, B: 1M, N: 10
G: 20M, B: 1M, N: 20
Count
1.5 G: 20M, B: 1M, N: 8

G: 20M, B: 5M, N: 20
G: 20M, B: 5M, N: 30
1 G: 8M, B: 1M, N: 20
G: 8M, B: 1M, N: 40
G: 8M, B: 5M, N: 10
G: 8M, B: 5M, N: 30
0.5
G: 8M, B: 8M, N: 20
G: 8M, B: 8M, N: 8
0
0 1 2 3 4
10 10 10 10 10
Figure 5-13. Three pairs heterogeneous networks activity distribution
90
Simulation results
This section discusses results obtained from simulation for 802.11g and compares them
with experimental results. As previously discussed, simulation for an 802.11g network
was developed for time distribution and DC comparison purposes and to assess possible
enhancement to technologies coexisting with 802.11 networks in the ISM band. Figure
5-14 demonstrates similarities between experimental and simulation DC for 802.11g in
both unsaturated and saturated network.
By examining CDF idle time in Figure 5-14, one can observe that idle-time distributions
obtained from the experimental study and the simulation are comparable for saturated
networks. Notably, dissimilarity becomes evident at time durations equal to or higher
than the first back-off stage for unsaturated network. This results from the assumption of
Poisson distributed frame arrival in the simulation, which is explained in more detail
below. Saturation idle-time fragments result primarily from SIFS and DCF back-off
durations because the transmitter queue is always saturated. Nevertheless, frame arrival
to the transmitter queue had a higher impact on idle-time distribution at lower traffic.
1
0.9
0.8
0.7
0.6
CDF
0.5
0.4 11g Simulation, Sat, DC = 72.55%
11g Experimental, Sat, DC = 66.92%
0.3 11g Simulation, 15Mbps, DC = 35.59%
0.2 11g Experimental,15Mbps, DC = 36.25%
11g Simulation, 5Mbps, DC = 11.8%
0.1 11g Experimental, 5Mbps. DC = 12.38%
0 0
1 2 3 4 5
10 10 10 10 10 10
Figure 5-14. Simulation vs. experimental idle time distributions
91
In the simulation, frame arrival is modeled using a Poisson distribution, which is a popular
and typical mathematical tool for modeling frame arrival into queue.
In presented experiments, Poisson distribution provided accurate results for modeling DC
experimental measurement, as observed in Chapter 4. Given idle-time distribution, the
simulation shows a steady increase in frequency for periods longer than the minimum
contention window. Experimental idle-time distribution exhibits a sudden rise in percent
of idle time fragments for the aforementioned idle-time region. This rise indicates that
the specific transmitter application provided a deterministic frame arrival rate at a specific
value. This phenomenon explains dissimilarity observed in Figure 5-14. Hence, 802.11
channel traffic patterns in an idle-time region higher than the minimum contention
window is directly impacted by frame arrival to the queue.
In summary, traffic patterns are directly affected by both MAC layer standard and the
queue frame arrival. Results demonstrated that an applied method for assessing traffic
patterns on a wireless channel is necessary for both wireless coexistence and cognitive
radio applications. An empirical modelling of 802.11 time distributions would provide a
mathematical approach to capture realistic 802.11 networks temporal behavior.
Conclusion
This chapter presented an extensive temporal characterization of 802.11b, g and n traffic
patterns. Presented results provided information about active and idle time distribution.
These distribution illustrated behavior of 802.11 homogeneous and heterogeneous
networks for one-, two- and pairs. Investigation were performed for throughput ranging
92
from low values to saturation and delivered critical information regarding wireless
coexistence potential with 802.11 networks in the 2.4GHz ISM band.
93
Chapter 6: 802.11 Time Distributions Empirical Modeling
This chapter presents empirical modeling methodology and results for 802.11
homogeneous and heterogeneous networks. Empirical idle-time distributions obtained
and discussed in previous chapters were used as a basis for this study. Modeling analysis
was performed to accurately describe best-fit models for 802.11 empirical idle-time
distributions. Doing so is essential for determining the appropriate design of wireless
technologies able to coexist with 802.11 networks. Furthermore, the models serve as a
foundation for enabling cognitive transmission on the ISM band. Idle-time distributions
are of special interest, as they can be used to design a superior wireless network and
improve performance of other coexisting wireless devices. Several approaches can be
utilized to model idle-time distributions.
Earlier research focused on passive energy scanning of activity in various bands to build
traffic pattern distribution and model idle-time distributions. Though this approach is
valid for licensed bands, such practice is inadequate for the unsilenced ISM band. Traffic
patterns in the heterogeneous ISM band are random in nature, as they depend on multiple
parameters such as desired throughput, 802.11 networks combination, and number of
transmitters in the network. To overcome this limitation, a comprehensive set of
experiments was designed to encompass a wide variety of 802.11 combinations and
traffic levels. The experimental setup was presented in Chapter 3 described in earlier
chapters. Energy detection method was used to sense 802.11 activity and construct idle-
time distributions. Later, idle-time distributions were modeled based on a number of
mathematical distributions. Two metrics are used to evaluate constructed models. Figure
6-1 provides an overview of the work flow. A detailed description of the process is
94
provided in the following subsections to help the research community 1) reproduce the
experiment; and 2) accurately utilize models in future research.
Experimental setup Idle time

Data analysis and
and data collection distributions
distributions
using energy modelling and
costruction
detection evaluation
Figure 6-1. Process for analyzing 1.4GHz ISM band 802.11 idle time distributions
A number of candidate distributions (Exponential, Weibull, Log-Normal, Generalized
Pareto, and Gamma) are selected and described based on previous literature work
[22][20][72][19]. These candidate distributions are used as a starting point for deriving
empirical 802.11 network idle-time Cumulative Distribution Functions (CDF). Two
metrics, namely Kullback-Leibler divergence ( ) and Bhattacharyya distance, are
described and discussed in the following section. These metrics are used to evaluate
models with various distributions against 802.11 empirical idle-time distributions.
Empirical modelling results for a variety of 802.11 combinations are then presented and
evaluated. Finally, a set of best performing models that describe 802.11 combinations are
presented.
Candidate distribution for 802.11 modelling
This section describes candidate distributions that were identified in a comprehensive
search of the literature and utilized for empirical modeling reported in this chapter. The
distributions serve as preliminary candidates to model 802.11 idle-time distributions in
the ISM band. Table 9 provides notations used for distribution employed in this work.
95
Table 9. Notations
Notation Description
Exponential distribution rate
1/ parameter
Scale
Shape
Log mean
Log standard deviation
Threshold parameter
The following provides a description of corresponding distributions [73]:
Exponential distribution
The commonly known Exponential distribution has been widely used for modeling
various physical phenomena, especially in networking applications. Exponential
distribution describes the time between consecutive events (i.e., packet arrivals) in a
Poisson process. Probability Density Function (PDF) and Cumulative Density Function
(CDF) are given in (4) and (5):
1
(; ) = (4)

(; ) = 1 (5)
The following equations provide moments of the Exponential distribution:
() =
(6)
() = 2
(7)
Weibull distribution
Weibull distribution is extensively used in reliability engineering, as well as other
applications, given its simplicity and adaptability. The distribution is drastically affected
96
by its scale and shape parameters, c and k respectively. The flexibility of Weibull
distribution is a primary reason for implementation in this investigation, as well as in
other time-dimension spectrum modeling research reported in literature. The following
equations provide PDF and CDF of Weibull distribution:
1 ( )
(; , ) = ( ) (8)

( ) (9)
(; , ) = 1
Moments of Weibull distributions are provided in (10) and (11):
1
() = . ( + 1) (10)

2
2
2 1
() = { ( + 1) ( + 1) } (11)

where (. ) is the complete Gamma function.
Log-normal distribution
Log-normal distribution corresponding variable is the product of a large number of
independent and identically distributed (i.i.d), variables. Log-normal distribution is often
used in wireless communications to describe received power fluctuations around a mean
value. It has also been used for modelling spectrum occupancy CDFs for various licensed
frequency bands. The following equations provide PDF and CDF of log-normal
distribution:
1 1 ln 2
( )
(; , ) = 2 (12)
2
1 1
(; , ) = + ( ) (13)
2 2 . 2
97
Moments of Log-normal distribution provided in (14) and (15):
2
+ (14)
() = 2
2 2
() = 2+ ( 1) (15)
Generalized Pareto distribution
Generalized Pareto distribution is often used to model tails of other distributions. This
particular distribution is of special interest as it has shown plausible results when used in
studies reported in literature for spectrum temporal distributions modeling. PDF and CDF
of Generalized Pareto distribution are provided in the following:
1
1 ( ) 1 (16)
(; , , ) = (1 + )

1
( ) (17)
(; , , ) = 1 (1 + )

Equations 18 and 19 describe moments of Generalized Pareto distribution:

() = + (18)
1
2
() = (19)
(1 )2 (1 2)
Gamma distribution
Gamma distribution is generally used to model the sum of exponentially distributed
random variables. As such, it has been included in this work as a candidate distribution.
Gamma distribution has shown favorable results for modeling Terrestrial Trunked Radio
98
(TETRA) idle-time distributions in literature [22]. The following formulas provide PDF
and CDF of Gamma distribution:
1
1
(; , ) = (20)
()
1
(; , ) = (, ) (21)
()
where; (. ) is the lower incomplete Gamma function.
Moments of Gamma distribution are described in (22) and (23):
() =
(22)
() = 2
(23)
Models evaluation metrics
This section describes the metrics and methods used to evaluate the obtained 802.11 idle-
time distribution models. Two metrics, namely Kullback-Leibler divergence and
Bhattacharyya distance, were employed in this research to evaluate idle-time distribution
models. The following subsections provide a brief description of each metric.
Kullback-Leibler divergence
Kullback-Leibler divergence ( ) represents the relative entropy between two
distributions [74]. is commonly used as a measure to evaluate whether or not a set of
data follows a particular model. The symmetric is described in (24):

( ) ( )
= ( ) ln( ) + ( ) ln( ) (24)
( ) ( )
=1 =1
99
where; ( ) is the evaluated model and ( ) represents idle-time distribution. The
smaller the value is, the more representative the evaluated model of the idle-time
distribution.
Bhattacharyya distance
The Bhattacharyya distance ( ) is used as a measure of divergence between two
distributions [75]. is provided in (25):
= ln( ( ). ( )) (25)
=1
In this work, DB evaluates distribution F(t n ) with mean f and standard deviation f ,
modeling idle-time distribution ( ) with mean and standard deviation . can
alternatively be described using (26):
2
1 1 2 2 1 ( )
= ln ( ( 2 + 2 + 2)) + ( 2 ) (26)
4 4 4 + 2
As can be seen from (26), the first term in represents the separation between the two
variances belonging to the two distributions ( ) and ( ). The second term in
represents the distance between the mean values of the two distributions.
Generally speaking, provides a stronger measure to assess target models when
compared to , as it evaluates information divergence between two distributions, rather
than the distance between the two distributions moments. Therefore, was used to
provide conclusions in the following section regarding which models best describe
100
802.11 empirical distributions. Nevertheless, both and have been implemented in
this study for the sake of comparison.
Empirical modelling results
This section presents and evaluates idle-time models when compared with corresponding
empirical distributions for 802.11b, g, and n single- and multi-pair transmissions. Minor
fluctuation can be observed when examining and for different models at varying
throughput values. Given a particular 802.11 combination, a single model was shown to
outperform all other investigated models for most throughput value experiments for a
corresponding combination.
Furthermore, it is possible that different 802.11 combinations and various number of pairs
are best modeled by different distributions. Figure 6-2 to Figure 6-13 provide 802.11n,
nn, nnn, gn, bn, and bgn combinations models, along with the results of the models
evaluation metrics serving as examples. These include DB and DKL metrics calculated for
different distributions modeling empirical idle time CDFs at different throughput values.
Mean DB and DKL of all throughput values for different 802.11 combinations were shown
in the figures to highlight various distribution performances. An optimal match between
a particular distribution and its model is achieve when mean DB and DKL asymptote 0.
As mean DB and DKL increase, the investigated model becomes less representative of the
distribution. In this work, mean DB and DKL are used as relative measures to find
particular models that best describe empirical distributions compared to others. The
following notations are used to describe the studied distributionsE: Exponential, W:
Weibull, LN: Log-Normal, GP: Generalized Pareto, and G: Gamma.
101
Figure 6-2 indicates that, according to DKL divergence metric, Gamma distribution
outperforms others in describing 802.11n idle-time distribution. Gamma distribution
ranks second with minimal difference over Log-Normal distribution, according to
DB distance. DKL divergence and DB distance metrics provided comparable conclusions
for most 802.11 combinations. Both metrics are shown in the following graph. Notably,
the analysis included only divergence to avoid unnecessary reiteration. The reason
for adopting divergence was previously discussed in detail.
802.11n
0.9 6.5 E
E
W
0.8 W 6
LN
LN
5.5 GP
0.7 GP
DKL
G
DB
G 5
0.6
4.5
0.5 4
0.4 3.5
0.2 0.4 0.6 0.8 1 0.2 0.4 0.6 0.8 1
Throughput Throughput
(a) (b)
0.8 6
5
0.6
KL
4
Mean DB
Mean D
0.4 3
2
0.2
1
0 0
E W LN GP G E W LN GP G
Distribution Distribution
(c) (d)
Figure 6-2. 802.11n models evaluation metrics a) DB for various throughput values a) DKL for various
throughput values c) Mean DB d) Mean DKL
Figure 6-3 depicts 802.11n models overlaying their corresponding empirical distribution.
Clearly, all models diverge from the empirical distribution for very low idle-time
durations, which correspond to DCF minimum contention window. Note the logarithmic
scale for idle-time duration in the presented graphs. Shorter idle-time duration
distributions may be better described with discrete geometric distribution. Nevertheless
presented models converge to the empirical distribution for higher idle-time duration. A
more in-depth discussion of this phenomenon is provided later in this section to highlight
102
the significance of longer idle-time duration when compared with shorter durations, as
well as to emphasize the need for accurate descriptions of their distribution.
(a) Throughput: 0.35461 802.11n (b) Throughput: 0.53191

1 1
0.8 0.8
CDF 0.6 Emperical 0.6 Emperical
CDF
E E
0.4 W 0.4 W
LN LN
0.2 GP 0.2 GP
G G
0 0
0 5 0 2 4
10 Idle time (us) 10 10 10 10
Idle time (us)
(c) Throughput: 0.70745 (d) Throughput: 1
1 1
0.8 0.8
0.6 Emperical 0.6 Emperical

CDF
CDF
E E
0.4 W 0.4 W
LN LN
0.2 GP 0.2 GP
G G
0 0
0 5 0 5
10 Idle time (us) 10 10 Idle time (us) 10
Figure 6-3. 802.11n models vs. empirical distribution for different throughput levels
Figure 6-4 demonstrates that Weibull distribution best describes 802.11nn combination,
slightly outperforming Gamma and Generalized Pareto distributions. Figure 6-6 shows
that Weibull distribution best describes 802.11nnn combination, as well. Gamma,
Generalized Pareto, and Exponential distributions for 802.11nnn combination provided
relatively similar results.
Figure 6-5 depicts empirical distribution overlaid with 802.11nn models and shows
more satisfactory results for low idle-time duration when compared with the 802.11n
combination. Similarly, Figure 6-7, Figure 6-9, Figure 6-11, and Figure 6-13 depict
802.11nnn, gn, bn and gbn combination models, respectively with corresponding
empirical distribution. Corresponding DKL and DB results for these combinations are
depicted in Figure 6-6, Figure 6-8, Figure 6-10, and Figure 6-12 Best results were found
103
for the 802.11nnn combination; poorest results were observed for the 802.11gbn
combination.
802.11nn
0.6 5 E
E W
0.5 W 4 LN
LN GP
0.4 GP 3
DKL
G
DB
G
0.3 2
0.2 1
0.1 0
0.4 0.6 0.8 1 0.4 0.6 0.8 1
(a) (b)
0.4 2.5
Mean DKL
0.3
Mean DB
1.5
0.2
1
0.1
0.5
0 0
(c) (d)
Figure 6-4. 802.11nn models evaluation metrics a) DB for various throughput values, b) DKL for various
throughput values, c) Mean DB d) Mean DKL
(a) Throughput: 0.37175 802.11nn (b) Throughput: 0.55762

1 1
0.8 0.8

CDF
CDF
E E
0.4 W 0.4 W
LN LN
0.2 GP 0.2 GP
G G
0 0
0 2 4 0 2 4
10 10 10 10 10 10
Idle time (us) Idle time (us)
1 1
0.8 0.8

CDF
CDF
E E
0.4 W 0.4 W
LN LN
0.2 GP 0.2 GP
G G
0 0
0 2 4 0 2 4
10 10 10 10 10 10
Figure 6-5. 802.11nn models vs. empirical distribution for different throughput levels
104
802.11nnn
0.35 2.5 E
E
W
W
0.3 LN
LN
2 GP
GP
G
DKL
B
0.25 G
D
1.5
0.2
1
0.6 0.7 0.8 0.9 1 0.6 0.7 0.8 0.9 1
(a) (b)
0.25 2
0.2
1.5
B
KL
Mean D
0.15
Mean D
1
0.1
0.5
0.05
0 0
(c) (d)
Figure 6-6. 802.11nnn models evaluation metrics a) DB for various throughput values, b) DKL for various
(a) Throughput: 0.5767 802.11nnn (b) Throughput: 0.5767

1 1
0.8 0.8

CDF
CDF
E E
0.4 W 0.4 W
LN LN
0.2 GP 0.2 GP
G G
0 0
0 2 4 0 2 4
10 10 10 10 10 10
1 1
0.8 0.8

CDF
CDF
E E
0.4 W 0.4 W
LN LN
0.2 GP 0.2 GP
G G
0 0
0 2 4 0 1 2 3
10 10 10 10 10 10 10
Figure 6-7. 802.11nnn models vs. empirical distribution for different throughput levels
As for heterogeneous 802.11 combinations, Figure 6-8 demonstrates that Generalized
Pareto outperforms other distributions for 802.11gn combination. See Figure 6-10 and
Figure 6-12 for 802.11 802.11bn and bgn combinations DKL and DB . Weibull distribution
provided best results for 802.11bn combination and 802.11ggn combinations. 802.11bgn
105
and 802.11nng combinations were best modeled by a Log-normal distribution.
Generalized Pareto outperformed the other studied models for all other 802.11
heterogeneous combinations. Table 10 presents mean DKL for complete distribution
models corresponding to all studied 802.11 combinations.
802.11gn E
1.8 E 12 W
W LN
1.6 LN GP
10
GP G
1.4
G
DKL
DB
8
1.2
6
1
0.8 4
0.4 0.6 0.8 1 0.4 0.6 0.8 1
(a) (b)
1.5 10
8
1
Mean DKL
Mean DB
4
0.5
2
0 0
(c) (d)
Figure 6-8. 802.11gn models evaluation metrics a) DB for various throughput values, b) DKL for various
106
(a) Throughput: 0.38095 802.11gn (b) Throughput: 0.57143
1 Emperical 1
E
0.8 0.8
W
LN
0.6 0.6 Emperical
CDF
CDF
GP
E
G
0.4 0.4 W
LN
0.2 0.2 GP
G
0 0
0 1 2 3 0 2 4
10 10 10 10 10 10 10
1 1
0.8 0.8

CDF
CDF
E E
0.4 W 0.4 W
LN LN
0.2 0.2
GP GP
G G
0 0
0 2 4 0 2 4
10 10 10 10 10 10
Figure 6-9. 802.11gn models vs. empirical distribution for different throughput levels
802.11bn
1.6 12 E
E
W W
1.4 LN 10 LN
GP GP
G
DKL
G
DB
1.2 8
1 6
0.8 4
0.4 0.6 0.8 1 0.4 0.6 0.8 1
(a) (b)
1.5 8
6
KL
1
Mean DB
Mean D
0.5
2
0 0
(c) (d)
Figure 6-10. 802.11bn models evaluation metrics a) DB for various throughput values, b) DKL for various
107
(a) Throughput: 0.36688 802.11bn (b) Throughput: 0.54784
1 1
Emperical Emperical
E E
0.8 0.8
W W
LN LN
0.6 0.6
CDF
CDF
GP GP
G G
0.4 0.4
0.2 0.2
0 0
0 2 4 0 2 4
10 10 10 10 10 10
1 1
Emperical
E
0.8 0.8
W
LN
0.6 0.6
CDF
CDF
Emperical GP
E G
0.4 0.4
W
LN
0.2 0.2
GP
G
0 0
0 2 4 0 2 4
10 10 10 10 10 10
Figure 6-11. 802.11bn models vs. empirical distribution for different throughput levels.
E 802.11gbn
1.8 14 E
W
W
LN
1.6 12 LN
GP
GP
G
1.4 10 G
DKL
DB
1.2 8
1 6
0.8 4
0.4 0.6 0.8 1 0.4 0.6 0.8 1
(a) (b)
1.5 10
8
1
Mean DKL
Mean DB
4
0.5
2
0 0
(c) (d)
Figure 6-12. 802.11gbn models evaluation metrics a) DB for various throughput values, b) DKL for various
108
(a) Throughput: 0.3495 802.11gbn (b) Throughput: 0.56793
1 1
Emperical Emperical
0.8 E 0.8 E
W W
0.6 LN 0.6 LN
CDF
CDF
GP GP
0.4 G 0.4 G
0.2 0.2
0 0
-2 0 2 4 -2 0 2 4
10 10 10 10 10 10 10 10
1 1
Emperical Emperical
0.8 E 0.8 E
W W
0.6 LN 0.6 LN
CDF
CDF
GP GP
0.4 G 0.4 G
0.2 0.2
0 0
-2 0 2 4 -2 0 2 4
10 10 10 10 10 10 10 10
Figure 6-13. 802.11gbn models vs. empirical distribution.
As previously discussed, the models presented were found to diverge from empirical
distributions for low idle-time durations (i.e., those below DCF minimum contention
window durations). Presented models converged to their corresponding idle-time
empirical distributions at durations higher than minimum contention window durations.
This work offers more emphasis on larger idle-time durations longer than that of DCF
minimum contention window. Such idle-time durations form a more significant part of
the distribution for two main reasons:
1. Idle-time durations correspond to the majority of studied distributions, meaning
that durations for the contention window are limited to a few hundred
microseconds. Nevertheless, an idle-time duration higher than the minimum
contention window can extend to several milliseconds, an order of magnitude
higher.
109
Table 10. Complete distibutions emperical modelling DKL results for 802.11 networks
Complete distribution DKL
Exponential Weibull Log-Normal Generalized Pareto Gamma
802.11b 5.47 3.88 4.13 4.37 3.78
802.11bb 1.80 1.23 1.44 1.53 1.16
802.11bbb 3.26 2.47 2.80 2.85 2.37
802.11g 5.94 5.15 5.23 5.40 5.14
802.11gg 2.31 1.59 1.70 1.61 1.68
802.11ggg 2.85 2.74 2.99 2.82 2.78
802.11n 4.85 4.66 4.81 5.17 4.61
802.11nn 2.11 1.89 2.11 1.94 1.94
802.11nnn 1.45 1.43 1.59 1.45 1.46
802.11bn 7.53 6.99 7.01 7.82 7.01
802.11gn 9.02 8.24 8.25 6.92 8.31
802.11gb 7.88 7.60 7.44 5.65 7.68
802.11bbg 9.10 8.79 8.68 6.12 8.86
802.11ggb 8.73 8.45 8.32 5.86 8.54
802.11bbn 8.99 8.30 8.16 7.94 8.40
802.11nnb 9.01 7.73 7.67 7.59 7.81
802.11nng 6.47 5.86 5.63 6.52 6.01
802.11ggn 8.62 7.62 7.69 7.75 7.82
802.11gbn 8.44 7.79 7.67 8.55 7.85
2. Larger time durations might be practically used by other coexisting technologies
to perform transmissions while minimizing collision risk with 802.11
transmission. Notably, this solution is generally not feasible for lower 802.11 idle-
time durations. For example, an 802.15.4 ZigBee header is 25 bytes long.
110
Considering ZigBee data rate of 250kbps, the header would only require 800s
for transmission, which would not be feasible within a DCF time frame.
Figure 6-14 to Figure 6-17 show idle time models compared with empirical distributions
for 802.11nnn, bn, gn, and gbn combinations at medium throughput levels. Theses graphs
represent previously discussed empirical distributions and their corresponding models for
time durations higher than 0.8 milliseconds with linear idle-time duration axis scale. The
figures show that there always exists a model that provides satisfactory representation for
empirical idle-time distributions that correspond to various 802.11 combinations.
802.11nnn, Throughput: 0.5767

1
0.999
CDF
0.998 Emperical
E
W
0.997 LN
GP
G
0.996
500 1000 1500 2000 2500 3000
Idle time (us)
Figure 6-14. Above minimum contention window802.11nnn models vs. empirical distribution
802.11bn, Throughput: 0.54784

1
0.95
0.9
CDF
0.85 Emperical
E
0.8 W
LN
0.75
GP
G
0.7
0 2000 4000 6000 8000 10000
Idle time (us)
Figure 6-15. Above minimum contention window 802.11bn models vs. empirical distribution
111
802.11gn, Throughput: 0.57143
1
0.998
0.996
CDF
Emperical
0.994 E
W
LN
0.992
GP
G
0.99
1000 1500 2000 2500
Idle time (us)
Figure 6-16. Above minimum contention window 802.11gn models vs. empirical distribution
802.11gbn Throughput: 0.56793

1
0.995
0.99
CDF
Emperical
0.985 E
W
0.98 LN
GP
G
0.975
0 1000 2000 3000 4000 5000
Idle time (us)
Figure 6-17. Above minimum contention window 802.11gbn models vs. empirical distribution
and metrics were calculated and presented for idle-time durations longer than that
of the minimum contention window. Such representation provides a more thorough
analysis of the investigated models. In turn, this analysis supports the notion of choosing
the most suitable model for each 802.11 combination.
Conclusion
A comprehensive empirical modelling of 802.11b, g and n homogeneous and
heterogeneous networks idle time distributions has been presented in this chapter.
Experimental data was collected for a complete set of 802.11 networks combinations at
112
a wide range of throughput values. Two metrics, Kullback-Leibler divergence and
Bhattacharyya distance, were employed to evaluate different models and how well they
describe the actual idle time distributions.
Results demonstrated that 802.11b and g homogeneous networks are best described using
Weibull distribution whereas 802.11n networks can be best modeled with Log-normal
models. Heterogeneous 802.11 networks combinations on the other hand were best
modeled using Exponential distribution, Log-normal distribution or Generalized Pareto
distribution depending on the combination. Nevertheless, it was observed that a
heterogeneous distribution would generally follow a Generalized Pareto distribution
whenever the medium includes transmission from one or more 802.11n pairs.
113
Table 11 presents mean for all studied 802.11 combination distributions above
minimum contention window (i.e., above 0.8ms). This table also provides mathematical
expressions describing best model for each of the studied 802.11 combinations at a given
saturation.
Notably, homogeneous 802.11b and 802.11g for one-, two-, and three-pair combinations
were best described using Weibull distribution. 802.11n homogeneous combinations are
best modeled using Log-Normal distribution. 802.11n follows enhanced MAC-layer
standards implementing frame aggregation, which explains the 802.11n variation from
both 802.11b and g. With regard to heterogeneous 802.11 networks, all three-pair
combinations containing 802.11n transmitter are best modeled using Generalized Pareto
distribution. Two- and threepair combinations containing only 802.11b and g
transmitters follow an Exponential distribution. Adding a second 802.11b pair to the
802.11bg network resulted in a sharper rise in CDF. Idle-time durations are further
fragmented, and their number is reduced due to a long duration of the 802.11b frame,
resulting from the standards lower data rate. Distribution parameters obtained in this
work for different homogeneous and heterogeneous networks combinations can be found
in Appendix I.
Conclusion
A comprehensive empirical modelling of 802.11b, g and n homogeneous and
heterogeneous networks idle time distributions has been presented in this chapter.
Experimental data was collected for a complete set of 802.11 networks combinations at
114
a wide range of throughput values. Two metrics, Kullback-Leibler divergence and
Bhattacharyya distance, were employed to evaluate different models and how well they
describe the actual idle time distributions.
Results demonstrated that 802.11b and g homogeneous networks are best described using
Weibull distribution whereas 802.11n networks can be best modeled with Log-normal
models. Heterogeneous 802.11 networks combinations on the other hand were best
modeled using Exponential distribution, Log-normal distribution or Generalized Pareto
distribution depending on the combination. Nevertheless, it was observed that a
heterogeneous distribution would generally follow a Generalized Pareto distribution
whenever the medium includes transmission from one or more 802.11n pairs.
115
Table 11. Above minimum contention window emperical modelling for 802.11 networks
Above minimum contention window distribution DKL
Combination Log- Generalized Best Saturation distribution
Exponential Weibull Gamma
normal Pareto
0.58
802.11b 1.12 0.59 0.58 0.62 0.61 () = 1 (70.11
)
0.63
802.11bb 0.39 0.14 0.15 0.18 0.16 () = 1 (38.47
)
0.65
802.11bbb 0.18 0.04 0.06 0.10 0.06 () = 1 (45.76
)
0.79
802.11g 0.65 0.46 0.46 0.49 0.47 () = 1 (58.81
)
0.96
802.11gg 0.24 0.12 0.13 0.14 0.12 () = 1 (42.77
)
0.97
802.11ggg 0.03 0.01 0.01 0.01 0.01 () = 1 (43.36
)
1 1 ln 3.58
802.11n 0.26 0.17 0.16 0.18 0.19 () = + erf ( )
2 2 1.25. 2
1 1 ln 3.3
802.11nn 0.08 0.05 0.05 0.05 0.06 () = + erf ( )
2 2 1.05. 2
1 1 ln 3.22
802.11nnn 0.01 0.00 0.00 0.00 0.00 () = + erf ( )
2 2 1.07. 2
1 1 ln 6.11
802.11bn 2.37 1.80 1.66 1.71 1.92 () = + erf ( )
2 2 2.1. 2
1
802.11gn 1.45 0.77 0.69 0.21 0.93 ( 0.5) 0.57
() = 1 (1 0.57 )
68.19

802.11gb 0.48 0.51 0.53 NA 0.50 () = 1 328.85

802.11bbg 1.29 1.34 1.35 NA 1.33 () = 1 431.15

802.11ggb 0.60 0.62 0.64 NA 0.61 () = 1 334.41
1
802.11bbn 3.97 3.30 3.02 1.82 3.52 ( 0.5) 0.67
() = 1 (1 0.67 )
74.85
1
802.11nnb 4.85 3.45 3.19 1.94 3.69 ( 0.5) 0.16
() = 1 (1 0.16 )
149.82
()
1
802.11nng 1.18 0.75 0.55 0.46 1.00 ( 0.5) 1.81
= 1 (1 + 1.81 )
26.57
()
1
802.11ggn 1.62 0.95 0.73 0.57 1.23 ( 0.5) 0.96
= 1 (1 + 0.96 )
26.93
1
802.11gbn 2.18 1.49 1.32 0.97 1.70 ( 0.5) 0.36
() = 1 (1 0.36 )
108.03
116
Chapter 7: Applications and Case Studies
Previous chapters have detailed temporal characterization of 802.11 networks in the ISM
band. This chapter presents two applications of such characterizations: 1) 802.15.4
ZigBee adaptive packet size, and 2) Wireless standards identification through machine
learning [76]. Algorithm and measurement methods used to obtain spectrum occupancy;
802.11 temporal distributions; and the empirical modelling of 802.11 idle time
distributions are discussed.
The first application aims at enhancing 802.15.4 ZigBee performance in the presence of
802.11 interference by adaptively changing 802.15.4 ZigBee packet size to maximize the
probability of coexistence, and thus duration, based on the mean value of 802.11g idle
time distribution. Achieved results indicated reduced 802.15.4 ZigBee PER. The second
application is to blindly differentiate among the various 802.11 standards and identify the
number of transmitters with transmission range without demodulating 802.11 frames
transmitted over the medium. This process was achieved constructing 802.11
distributions using energy detection. Features extracted from these distributions were
used by machine-learning algorithms to identify wireless networks and terminals.
Notably, both experimental and simulation setups presented in Chapter 3 are employed
in the work presented in this chapter to achieve the two aforementioned distributions.
More details on the developed methodology to construct the applications and the obtained
results are provided in the following sections.
117
Application 1: 802.15.4 ZigBee adaptive packet size
This section presents a case study highlighting usability of channel temporal traffic
pattern occupancy awareness for enhancing performance of wireless technologies
coexisting with 802.11 networks. 802.15.4 ZigBee, referred to as ZigBee from this point
forward, was used as an example of ISM band technology coexisting with 802.11g
network in this application. For the purpose of this proof of concept, ZigBee simulation
and 802.11g simulation were used. A validation of ZigBee simulation is first discussed,
and then methodology and results of this application are presented in the following
subsections.
802.15.4 ZigBee simulation validation
A set of experiments was designed and executed to compare ZigBee PER results obtained
from experimental test to those obtained from simulation.
Experimental setup included two 802.11g terminals (i.e., a transmitter and a receiver)
placed equidistant from the ZigBee device under test (DUT), see Figure 7-1. Transmit
power for 802.11g terminals was set to 16 dBm, and data was exchanged at varying DC
from low values to saturation. ZigBee RSS was maintained at -70 dBm (1 dB) for all
tests to ensure that ZigBee receiver was situated at the edge of its cell with zero percent
PER [7]. Test runs were performed in a 6.6m x 4m x 3m anechoic chamber to avoid
unintended interference. For each test run, 1000 ZigBee packets were sent.
To obtain ZigBee PER through simulation setup, 802.11g activity times were generated
for one pair 802.11g network. ZigBee simulation was then run, wherein a packet was
transmitted using the 802.15.4 CSMA/CA scheme. Both 802.11g and 802.15.4 ZigBee
118
simulations are described in greater detail in Chapter 3. ZigBee PER was collected for
increasing 802.11g DC values.
Wi-Fi
Tx
1m
Zigbee 5.9 m
DUT 1m
Wi-Fi
Rx
Figure 7-1. 802.15.4 ZigBee and 802.11g coexistence experimental setup
A comparison of PER results for both the experimental setup and simulation is provided
in Figure 7-2. Clearly, simulation PER results are comparable with experimental PER
results. PER difference was higher for larger DC values; difference in PER did not exceed
7% for all DC values.
100
Experimental
90 Simulation
80
70
60
PER (%)
50
40
30
20
10
0
0 10 20 30 40 50
DC (%)
Figure 7-2. 802.15.4 ZigBee PER simulation vs. experimental results
119
Methodology and results
This subsection offers the methodology for implementing adaptive ZigBee packet size,
as well as obtained results from this investigation. ZigBee packet size was adaptively
changed based on 802.11g idle time distribution using simulation setup. Mean 802.11g
idle time value was calculated for a distribution region higher than the minimum
contention window (i.e., third idle time distribution region, which was described in
Chapter 4). ZigBee packet size was set to the immediate integer number of Bytes with
time duration smaller than calculated 802.11g idle time distribution mean. Simulation
implemented a worst-case interference scenario (i.e., any overlap between ZigBee and
802.11 traffic renders a ZigBee packet corrupted).
802.15.4 ZigBee standard limits maximum ZigBee packet size to 127 Bytes, including a
25 Byte header. Adaptive packet size was constrained by the standards boundaries in this
work: an upper bound of 127 Bytes and lower bound of 26 Bytes. ZigBee adaptive packet
size PER and throughput were tracked for increasing 802.11g DC. Results were compared
to ZigBee fixed packet size PER and throughput. Fixed packet sizes, including headers,
were 125, 100, 75, and 30 bytes. Table 12 shows calculated adaptive ZigBee packet size
for various throughput/DC values.
Table 12. ZigBee adaptive packet sizes

802.11G DUTY CYCLE (%) 802.11G THROUGHPUT (MBPS) ZIGBEE PACKET SIZE (BYTES)
2.32 1 127
7.09 3 127
11.8 5 80
16.6 7 58
21.33 9 47
26.2 11 39
30.87 13 34
35.59 15 30
39.97 17 27
44.87 19 26
120
Figure 7-3 and Figure 7-4 show ZigBee PER and throughput, respectively, for both
adaptive (in black) and fixed packet sizes. Large fixed packet sizes exhibited a high
throughput for low 802.11g DC and poor performance at high 802.11g DC. In contrast,
small fixed packet sizes demonstrated poor performance for low 802.11 DC and improved
performance for high DC. Results indicate a tradeoff between PER and packet payload-
to-header ratio. Large packet sizes perform better at low 802.11g duty cycle due to a high
ratio of packet payload-to-header, whereas small packet sizes perform better at high
802.11g DC given that they suffer fewer collisions.
The adaptive packet size method introduced in this work demonstrates a throughput
comparable to the highest fixed packet size performance, regardless of 802.11g DC value,
as shown in Figure 7-3. The method revealed throughput improvement of up to 66% over
fixed packet size of 30 bytes at low duty cycle of 2.3%. Moreover, the method showed
significant improvement over fixed packet size of 125 bytes at medium and high
utilization of channel DC.
Adaptive ZigBee packet size also showed significant improvement in PER, as shown in
Figure 7-4. For certain cases adaptive ZigBee packet size resulted in PER drop of up to
25% when compared to the upper end fixed packet size of 125 Bytes.
121
0.7
Zigbee packet size: 125B
0.6 Zigbee packet size: 100B
0.5
Zigbee throughput
Zigbee packet size: Adaptive
0.4
0.3
0.2
0.1
0
0 5 10 15 20 25 30 35 40 45
802.11g duty cycle (%)
Figure 7-3. ZigBee throughput for fixed and adaptive packet sizes
100
90
80
70
60
PER (%)
50 Zigbee packet size: 125B

Zigbee packet size: Adaptive
10
0
0 5 10 15 20 25 30 35 40 45
Duty cycle (%)
Figure 7-4. ZigBee PER for fixed and adaptive packet sizes
Application 2: Wireless technology identification through machine learning
This section presents identification of 802.11 standards and number of transmitters as an
application of 802.11 temporal characterization. Identification was performed blindly
using 802.11 idle time and activity distributions obtained via energy detection. Features
characteristic of different standards and number of transmitters are first extracted from
802.11 idle time and activity distributions. Features are then input into a trained machine-
122
learning algorithm that is responsible for classifying a particular observation into
corresponding standard and number of transmitters. Wireless technology identification
enables situational awareness to improve coexistence and reduce interference among the
devices. Details on the methodology and obtained results for this application are provided
in the following subsection.
Methodology
This subsection details the methodology employed to perform wireless technology
identification. Presented work is based on an energy detection technique with the
following three stages:
Stage 1, Data collection and processing:
Simple energy detection scheme is implemented to measure RSS. Collected
measurements are compared with a pre-set threshold to determine time periods while the
channel is active (i.e., transmission is detected) or inactive (i.e., channel is idle). Temporal
characteristics consisting of activity and idle time distributions are then derived by
constructing histograms of active time periods and idle time periods, respectively. Time
bin widths used in the histograms were: 1) 1.25s for idle time periods; and 2) 2.5s for
activity periods. Time bin duration was short enough to accurately capture MAC layer
temporal characteristics of the protocols under study (i.e., 802.11b/g/n). The lowest
expected idle time period corresponds to the 802.11g/n MAC SIFS of 10s, while the
lowest expected activity period corresponds to 802.11b/g MAC acknowledgment (ACK)
packet length of 30s.
123
Stage 2, Feature extraction and data set construction:
Distinguishing features are extracted from activity and idle time distributions in Stage 2.
A dataset consisting of features extracted from multiple scans is constructed to train a
specific classifier, and then to test its accuracy for identifying the wireless technology in
use. The set of features extracted from each particular energy scan is referred to as an
observation. A set of multiple observations for different wireless technologies is referred
to as a dataset.
Since each wireless standard/technology has a unique MAC and PHY protocol
implementation, a distinguishable temporal traffic characteristic exist. Temporal traffic
patterns in this work are described via activity and idle time distributions, which were
constructed from energy scanning data, as discussed in great detail in Chapter 4.
Specific features able to differentiate wireless technologies are then extracted from
aforementioned distributions. A combination of features were used to make an
identification decision on wireless technologies utilizing the spectrum. A set of features
was extracted from each distribution for a specific test run representing a single
observation. Examples of such features include frequency of idle time durations at specific
distribution regions; distribution mean and its standard deviation; and mean and standard
deviation of a specific region within the distribution. Observations from multiple test runs
are then grouped into a single dataset for further analysis. To clarify, a distribution region
represents a specific time span in idle time or activity distribution. This time span contains
information characteristic of a particular wireless technology (e.g., activity time region
near 111s representing 802.11b ACK). Frequency of 802.11b ACKs, along with other
124
characterizing features, can be employed to construct the dataset that can be used for
wireless technology identification.
A total of 596 tests for 802.11 networks were used to evaluate the proposed method in this
work. The experimental setup for these tests is discussed in Chapter 3. 802.11 time
distributions and specific features chosen to identify different 802.11 networks are
discussed in the following sections. However, these features are not discussed in this
section to purposefully maintain a generic methodology. Identification of 802.11 networks
is intended merely as an example of the methods functionality and presented as such. That
said, the methodology presented in this work is applicable to different wireless
technologies.
Stage3, Wireless technology identification:
Wireless technology identification is performed in Stage 3 using the scanned frequency
channel. Figure 7-5 shows the overall structure of Stage 3. The dataset constructed in
Stage 2 will train a classifier and assess its performance. This dataset is divided into two
equal subsetsone for training purposes (i.e., training set) and the other to test
identification accuracy rate (i.e., testing subset). Observations in testing set act as new
scans for which the wireless device must perform technology identification. Note that
Increasing the number of features is not associated with proportional increase in
classification accuracy. On the contrary, additional features might result in poor
identification accuracy because features from different classes that correspond to different
technologies might overlap. As such, the classifier training process is iteratively repeated
using different feature sets to select those that provide best performance.
125
Dataset
Testing Divide Training

set dataset set
Build
classifiers
Test
accuracy
Repeat
process using
different
features set
Figure 7-5. Wireless technology identification stage
Two types of classification techniques, namely nave Bayes and K-nearest neighbor, were
investigated in this research to build the classifier. Testing two machine-learning
techniques aids in examining the classifier effect on the identification accuracy versus the
extracted feature effect. Though machine-learning is not the focus of this work, a brief
description is provided to aid in understanding utilized classification methods:
Nave Bayes classification method:
Bayesian Networks or Belief Networks (BN) are hybrid of graph theory and probability
theory. The aim of learning with BN is to determine structures and conditional
probabilistic tables that distinguish between nodes via input training data [24, 25]. In a
Bayesian network implementation, parent nodes (i.e., a class) represent a particular
802.11 standard or number of transmitters while other nodes (i.e., continuous) represent
features used to distinguish between classes. When using a Bayesian network for
126
classification purposes, probability of features that are conditional on parents are
estimated using a training set. These probability functions are then used later to make a
classification decision using a testing set [26]. This work utilizes a Nave Bayesian
classification method, which assumes conditional independence of a feature that is
conditional on its parents [27]. Thus, probability of a particular 802.11 observation O =
(1 , 2 , , ) conditional on class c is given by the following equation:

(1 , 2 , , |) = ( |) (27)
=1
The probability of a class c given observation O can be obtained by:
(|) = (|1 , 2 , , )
(1 , 2 , , |)()
=
(1 , 2 , , ) (28)
k=1 p(fk |c)p(c)

n
=
p(f1 ,f2 ,,fn )
Based on the value of adjacent class probabilities ratio, a classifier will make a decision
on observation O. Notably, nave Bayesian assumption of conditional independence of
parents is inaccurate in most real life applications. However, this condition achieves a
surprisingly superior performance. The reason being that it utilizes zero-one error
function, which penalizes according to classification accuracy rather than accuracy of
probability estimation. This means that given probability estimation is poor and
classification is accurate, correct functionality is assumed [28].
127
Nave Bayes is a simple and widely used algorithm in machine learning based on a strong
assumption of independence between features and classification. Despite the unrealistic
assumption of independence, Nave Bayes is considered plausibly efficient [77].
Supervised learning is used for training the algorithm in order to predict a class based on
accumulated information evidence. Simply put, the algorithm provides a classification
decision that attempts to select the most probable outcome. Applications of Nave Bayes
are diverse and span from image processing [78] and text classification [79] to networking
[80].
K-nearest neighbor classification method:
k-Nearest Neighbors algorithm (KNN) [81] is a non-parametric classification method
wherein an algorithm polls neighbors of a data point, and then votes to assign a given class.
Voting is carried out by weighting neighbor contributions. Neighbors that are closer to
the point of interest will have higher participation than those farther away from the point.
An obvious drawback of this method is the high dependence on the structure of the dataset.
KNN has been used to detect anomalies in wireless sensor networks in [82], to perform
medical data mining in [83], and to implement biometric recognition in [84].
Results
This section details experimental wireless technology identification results obtained for
802.11 homogeneous and heterogeneous networks. This works attempts to blindly identify
number of 802.11 transmitters and the transmitters wireless technologies using statistical
information obtained from 802.11 time distributions. Statistical information for a
particular test run are used as input features for a machine-learning algorithm in an effort
128
to classify the number of wireless technologies and the number of transmitters used for the
corresponding run.
Choosing a greater number of features does not necessarily guarantee improved
identification performance, as aforementioned. Features corresponding to different
wireless technologies might be strongly overlapping. Thus, adding such features to the
training/testing datasets will confuse the classifier, as well as reduce the probability of
accurate classification.
An example of overlapping features for homogeneous 802.11 technologies is
depicted in Figure 7-6. Features are extracted from idle time distribution and are
identified as total idle time and mean of idle time region between 20s and 180s.
Both exhibit strong overlap between 802.11 observations corresponding to different
technologies/standards.
Notably, when total idle time was replaced by mean of idle time region between 20s
and 305s, as depicted in Figure 7-7, separation between observations corresponding to
different technologies was enhanced considerably. This feature was extracted from idle
time distribution, as well. Consequently, including features such as total idle time would
degrade classifier performance.
129
5000
802.11b: one, two and three pairs
802.11g: one, two and three pairs
4000 802.11n: one, two and three pairs
Total idle time mean (us)

3000
2000
1000
0
65 70 75 80 85 90 95 100 105 110
Mean of region between 20us to 180us (us)
Figure 7-6. Overlapping idle time distribution features
180
Mean region between 20us to 305us (us)
802.11b: one, two and three pairs

802.11g: one, two and three pairs
160
802.11n: one, two and three pairs
140
120
100
80
60
65 70 75 80 85 90 95 100 105 110
Mean of region between 20us to 180us (us)
Figure 7-7. Idle time distribution features with improved technology separation
A set of five features extracted from activity distributions was used in the identification
process, see Table 13. Note that frequency of occurrence refers to the number of
occurrences that a particular fragment was repeated during unit time.
130
Table 13. Features used for wireless standards identification
Feature Distribution type
Frequency of occurrences at 111s Activity
Frequency of occurrences between 150s and 200s Activity
Frequency of occurrences between 1100s and 1300s. Activity
A total of 13 features were used to perform number of transmitters identification, see Table
14.
Table 14. Features used for number of transmitters identification

Feature Distribution type
Highest frequency of occurrence Idle time
2nd highest frequency of occurrence Idle time
Time duration of 2nd highest frequency of occurrence Idle time
3rd highest frequency of occurrence Idle time
Time duration of 3rd highest frequency of occurrence Idle time
Mean of region extending from 20us to 180us Idle time
Mean of region extending from 180us to end of distribution Idle time
Mean of region extending from 20us to 305us Idle time
Mean of region extending from 305us to end of distribution Idle time
Standard deviation of region extending from 20us to 180us Idle time
Standard deviation of region two from 180us to end of distribution Idle time
Most contributing idle time fragment duration to the total idle time Idle time
Number of fragments at 36us (representing number of 802.11n block Activity
acknowledgments
Homogeneous 802.11 networks identification
Technology identification for 802.11 homogeneous networks was performed using a
dataset of 127 observations at variable throughput divided into training and testing sets.
Frequency channel was scanned for duration of one minute per observation. One-, two-,
or three-pair networks exchanged traffic at different throughput levels, ranging from low
throughput to saturation. At low throughput, recorded channel utilization can be as low
as 12%, whereas channel utilization can reach up to 98% for 802.11n saturation.
131
Figure 7-8 depicts identification accuracy obtained from both KNN and Nave Bayes
classifiers. An accuracy rate of 96.9% was achieved for homogeneous networks. Figure
7-9 shows identified observations distribution for nave Bayes classifier. Only two out of
63 802.11g observations in the testing set failed correct identification.
Figure 7-8. Homogeneous networks identification accuracy
Figure 7-9. Distribution of identified observations for homogeneous networks
Number of transmitters identification for homogeneous networks via energy detection
was investigated and presented in this work, as well. An observation was classified as
132
one, two, or three transmitters. Maximum accuracy of 85.71% was achieved when
directly implementing number of transmitters identification to a certain observation. See
Figure 7-10.
Figure 7-10. Number of transmitters identification accuracy
A two-stage classifier can be implemented wherein a wireless technology/standard is
identified at first, and then transmitter number is identified. As a means to enhance
accuracy, number of transmitter identification was investigated in two stages. The reason
for this approach is the satisfactory accuracy rate obtained from homogeneous standards
identification presented earlier. Per-standard accuracy rates for two stages of transmitter
identification is depicted in Figure 7-11.
Figure 7-11. Per-standard, two stages, number of transmitters identification accuracy
133
Heterogeneous 802.11 networks identification
This section provides identification results for a combined dataset comprised of both
homogeneous and heterogeneous 802.11 networks. Heterogeneous 802.11 networks may
consist of two- or three-pair networks operating on different 802.11 standards. A total of
469 observations were collected and utilized to construct the dataset. Observations were
collected at varying throughput values for the following combinations:
- One-pair: 802.11b, g and n.
- Two-pair: 802.11bb, gg, nn, bg, bn and gn.
- Three-pair: 802.11bbb, ggg, nnn, bbg, bbn, ggb, ggn, nnb, nng and bgn.
Figure 7-12 depicts achieved identification accuracy. An accuracy of up to 85.9% was
obtained when using a Nave Bayes classifier. Figure 7-13 shows distribution of identified
observations over various 802.11 standard combinations.
Figure 7-12. Heterogeneous networks identification accuracy
134
Figure 7-13. Distribution of identified observations for heterogeneous networks
Figure 7-13 demonstrates that overlap occurred between combinations characterized with
the same wireless technology. For example, a major overlap was observed between
802.11n one-pair and two- and three-pair heterogeneous networks with 802.11gn
combination. The overlap results from temporal parameter similarities between the two
technologies at low 802.11n throughput values. Whenever 802.11n transmitter aggregates
two sub-frames, it results in aggregated frame duration comparable to 802.11g frame size.
Three-pair 802.11bgn combination observations were identified correctly at an accuracy
rate of 100%. Observations for other combinations composed of two technologies were
confused with the 802.11bgn combination, although error rate for these cases did not
exceed 10%.
Notably, Nave Bayes classifier outperformed KNN classifier for all reported
homogeneous and heterogeneous technology identification experiments. KNN classifier
provided better results than Nave Bayes only for homogeneous two-stage number of
transmitter identification.
135
Conclusion
This chapter presented two applications of 802.11 temporal characterization aimed at
providing situational awareness to wireless technologies coexisting with 802.11 networks
and enhancing their performance. Spectrum utilization awareness provides coexisting
wireless devices with an opportunity to enhance performance.
An application based on channel-timing parameters was investigated and presented
wherein an adaptive rather than a fixed 802.15.4 ZigBee packet size was implemented.
The method demonstrated superior performance over traditional fixed packet size.
A novel method for wireless technology and number of transmitter identification using
simple energy detection techniques was researched, as well. The presented method does
not require demodulation, making its application feasible when using the majority of
currently used transceivers. Experimental results have been provided to investigate
applicability of the proposed identification method and to assess its performance. Various
802.11 homogeneous and heterogeneous wireless networks were utilized in experimental
assessment. Identification accuracy of up to 96.9% was achieved under homogeneous
network conditions and up to 85.9% under heterogeneous condition.
136
Chapter 8: Conclusion and Future Work
An extensive temporal characterization of 802.11b, g and n spectrum occupancy and
traffic patterns was presented in this dissertation. Applications that benefit from such
characterization were introduced and implemented.
Literature review was provided, including an overview of wireless technologies
employed in this work. An energy detection methodology free from frame demodulation
was provided and implemented. Presented results delivered critical information about
channel utilization in terms of DC, throughput, activity distributions, and idle time
distributions.
Comprehensive empirical modelling of 802.11b, g and n networks was also presented in
this work. Experimental data was collected for a complete set of 802.11 network
combinations at a wide range of throughput values. Results demonstrated that 802.11b
and g homogeneous networks are best-described using Weibull distribution; 802.11n
homogeneous networks are best modeled with Log-Normal models. Heterogeneous
802.11 network combinations were best-modeled using exponential distribution, Log-
normal distribution, or Generalized Pareto distribution, depending on the combination of
standards employed. A heterogeneous distribution was found to generally follow a
Generalized Pareto distribution when the medium included transmission from one or
more 802.11n pairs.
Two applications based on channel-timing parameters were also presented and discussed
in an effort to highlight usability of 802.11 temporal characterization for enhancing
coexistence in the ISM band. The first application introduced adaptive 802.15.4 ZigBee
137
packet size based on detected 802.11 temporal distributions. ZigBee packet size was
changed adaptively to reduce interference with an 802.11g network and to take advantage
of longer idle time durations. The method demonstrated superior performance over
traditional fixed packet size.
The second application presented a novel method for wireless technology identification
using simple energy detection techniques. The presented application enables
transceivers to identify wireless technology and the number of transmitters using the
spectrum. Results of identifying 802.11 homogeneous and heterogeneous wireless
networks were presented to prove the concept. Identification accuracy of up to 96.9%
was achieved for homogeneous networks; 85.9% accuracy was achieved for
heterogeneous networks.
Future work
Work presented in earlier chapters provided a comprehensive study of 802.11 temporal
behavior in the 2.4GHz ISM band. The experimental setup was designed to focus on the
core operational scenarios that would represent a wide range of spectrum temporal
behavior variations for the investigated networks. Nevertheless, outlier scenarios may
exist which could introduce changes to the investigated networks behavior. To
accommodate for such scenarios, topology of the networks being studied in the
experimental setup can be modified. Notably, when changes to the topology are
introduced, we may start observing phenomena resulting for both PHY and MAC layers
rather than only isolated MAC layer effects. Two test topologies, depicted in Figure 8-2
and Figure 8-3 are proposed as future work. Test setup depicted in Figure 8-1 aims at
investigating effects of hidden node on the spectrum behavior. It is of interest to observe
138
spectrum occupancy near the hidden node transmitter and receiver and the other
transmitters in the network. 802.11 networks operate at frequency above 2.4GHz with
limited transmit power leading to increased levels of attenuation. As such, spectrum
temporal behavior may prove to be localized and demonstrate different results at either
communication end. Figure 8-3 depicts a test setup intended for investigating over
exposed node effects. Similarly, spectrum measurements shall be conducted at both
802.11 transmitter and receiver to characterize the resulting spectrum temporal behavior.
Realizing these test setups would obviously require changes to spacing between nodes
and modification of transmit power for different transmitters to replicate the hidden node
and exposed node problems. Details of the hidden node and over exposed node problems
can be found in [85]. Investigating such test scenario is directly achievable using methods
and tools presented in this dissertation.
Figure 8-2. Proposed hidden node test setup
139
Figure 8-3. Proposed exposed node test setup
Future work also includes performing a temporal and frequency characterization of
coexisting wireless technologies in the ISM band. Wireless technology identification
study can then performed for such wireless technologies similar to the one performed for
802.11 networks. This helps in providing a holistic view and understanding of the ISM
band spectrum occupancy. Also, it serves as a step to enable implementation of extensive
cognitive functionality of coexisting devices in unlicensed bands based on temporal and
frequency characterization. Such activities help in providing a better wireless devices
performance and an efficient use of available spectrum.
Extending these efforts to the 5.9GHz band is proposed as a future work as well. U.S
Federal Communication Commission has allocated a 75MHz in the 5.9GHz band for
Directed Short-Range Communications (DSRC). Wireless Access in Vehicular
Environment (WAVE) is the base communication technology for V2X systems in the
United States. PHY and MAC layers implementation for Wave is defined in IEEE
140
802.11p [86] and IEEE 1609 [87] standards. Wi-Fi alliance is expected to share this band
as a secondary user [88]. Efficient use of the band while avoiding interference with V2X
network shall be investigated. Work presented in this dissertation provides a method for
characterizing available time and frequency in the 5.9GHz band to enable an improved
coexistence between
Also, these efforts can be extended to other unlicensed bands to investigate their spectrum
occupancy in an effort to increase efficiency of spectrum utilization and improve
performance of wireless technologies utilizing these spectrum bands.
141
References
[1] David Watkins (Strategy Analytics), Embedded WLAN (Wi-Fi) CE Devices: US

Market Forecast, 2013. [Online]. Available:
http://www.strategyanalytics.com/default.aspx?mod=reportabstractviewer&a0=8452.
[2] H. Takagi and B. H. Walke, Spectrum requirement planning in wireless
communications: Model and methodology for IMT-Advanced, vol. 2. John Wiley &
Sons, 2008.
[3] IEEE Recommended Practice for Information technology-- Local and metropolitan area
networks-- Specific requirements-- Part 15.2: Coexistence of Wireless Personal Area
Networks with Other Wireless Devices Operating in Unlicensed Frequency Bands,
IEEE Std 802.15.2-2003. pp. 1150, 2003.
[4] Y. Zhu, X. Tian, and J. Zheng, Performance analysis of the binary exponential backoff
algorithm for IEEE 802.11 based mobile ad hoc networks, 2011 IEEE Int. Conf.
Commun., 2011.
[5] A. Sikora and V. F. Groza, COEXISTENCE OF STANDARDIZED WIRELESS
SYSTEMS IN THE 2 . 4 GHZ-ISM-BAND *, vol. 4, no. 1, pp. 511, 2005.
[6] L. Angrisani, M. Bertocco, D. Fortin, and A. Sona, Experimental study of coexistence
issues between IEEE 802.11b and IEEE 802.15.4 wireless networks, IEEE Trans.
Instrum. Meas., vol. 57, no. 8, pp. 15141523, 2008.
[7] N. J. LaSorte, S. a Rajab, and H. H. Refai, Developing a reproducible non-line-of-sight
experimental setup for testing wireless medical device coexistence utilizing ZigBee.,
IEEE Trans. Biomed. Eng., vol. 59, no. 11, pp. 32219, Nov. 2012.
[8] N. LaSorte, S. Rajab, and H. Refai, Experimental Assessment of Wireless Coexistence
for 802.15. 4 in the Presence of 802.11 g/n, 2012 IEEE Int. Symp. Electromagn.
Compat., pp. 473479, 2012.
[9] W. Guo, W. M. Healy, and M. Zhou, Impacts of 2.4-GHz ISM band interference on
IEEE 802.15.4 wireless sensor network reliability in buildings, IEEE Trans. Instrum.
Meas., vol. 61, no. 9, pp. 25332544, 2012.
[10] Radio Frequency Wireless Technology in Medical Devices - Guidance for Industry and
Food and Drug Administration Staff, 2013. [Online]. Available:
http://www.fda.gov/MedicalDevices/DeviceRegulationandGuidance/GuidanceDocument
s/ucm077210.htm. [Accessed: 01-Jan-2014].
[11] S. Atapattu, C. Tellambura, and H. Jiang, Energy Detection for Spectrum Sensing in
Cognitive Radio, Springer, 2014.
[12] A. Mariani, A. Giorgetti, and M. Chiani, SNR Wall for Energy Detection with Noise
Power Estimation, 2011 IEEE Int. Conf. Commun., pp. 16, Jun. 2011.
[13] J. Avila and K. Thenmozhi, Boosted Energy Detector Based Spectrum Sensing
Methodology For Cognitive Radio, Middle-East J. Sci. Res., vol. 20, no. 9, pp. 1031
1035, 2014.
[14] K. Chabbra, G. Mahendru, and P. Banerjee, Effect of dynamic threshold & noise
uncertainty in energy detection spectrum sensing technique for cognitive radio systems,
2014 Int. Conf. Signal Process. Integr. Networks, vol. 1, no. 2, pp. 377361, Feb. 2014.
142
[15] H. M. Baradkar and S. G. Akojwar, Implementation of Energy Detection Method for
Spectrum Sensing in Cognitive Radio Based Embedded Wireless Sensor Network
Node, 2014 Int. Conf. Electron. Syst. Signal Process. Comput. Technol., pp. 490495,
Jan. 2014.
[16] M. Matinmikko, M. Mustonen, M. Hyhty, T. Rauma, H. Sarvanko, and A. Mmmel,
Distributed and directional spectrum occupancy measurements in the 2.4 GHz ISM
band, in 2010 7th International Symposium on Wireless Communication Systems
(ISWCS), 2010, pp. 976980.
[17] M. Matinmikko, M. Mustonen, M. Hyhty, T. Rauma, H. Sarvanko, and A. Mmmel,
Cooperative spectrum occupancy measurements in the 2.4 GHz ISM band, in 2010
3rd International Symposium on Applied Sciences in Biomedical and Communication
Technologies (ISABEL), 2010, vol. 00, no. 134624.
[18] D. Denkovski, M. Pavloski, V. Atanasovski, and L. Gavrilovska, Parameter setting for
2.4GHz ISM spectrum measurement, in 2010 3rd International Symposium on Applied
Sciences in Biomedical and Communication Technologies (ISABEL), 2010, pp. 37.
[19] S. Geirhofer, L. Tong, and B. M. Sadler, Dynamic Spectrum Access in the Time
Domain: Modeling and Exploiting White Space, IEEE Commun. Mag., vol. 45, no. 5,
pp. 6672, 2007.
[20] L. Stabellini, Quantifying and modeling spectrum opportunities in a real wireless
environment, in 2010 IEEE Wireless Communications and Networking Conference
(WCNC), 2010.
[21] M. Lopez-Benitez and F. Casadevall, Empirical time-dimension model of spectrum use
based on a discrete-time markov chain with deterministic and stochastic duty cycle
models, IEEE Trans. Veh. Technol., vol. 60, no. 6, pp. 25192533, 2011.
[22] M. Lpez-Bentez and F. Casadevall, Time-dimension models of spectrum usage for
the analysis, design and simulation of cognitive radio networks, IEEE Trans. Veh.
Technol, vol. 62, no. 5, pp. 20912104, 2013.
[23] X. Wang, A. Wong, and P. Ho, Extended knowledge-based reasoning approach to
spectrum sensing for cognitive radio, IEEE Trans. Mob. Comput., vol. 9, no. 4, pp.
465478, Apr. 2010.
[24] L. Tang, Y. Chen, E. L. Hines, and M.-S. Alouini, Effect of primary user traffic on
sensing-throughput tradeoff for cognitive radios, IEEE Trans. Wirel. Commun., vol. 10,
no. 4, pp. 10631068, 2011.
[25] K. Chang and B. Senadji, Spectrum sensing optimisation for dynamic primary user
signal, IEEE Trans. Commun., vol. 60, no. 12, pp. 36323640, 2012.
[26] S. Zarrin and T. Lim, Throughput-sensing tradeoff of cognitive radio networks based
on quickest sensing, 2011 IEEE Int. Conf. Commun., pp. 15, Jun. 2011.
[27] M. Guerrini, L. Rugini, and P. Banelli, Sensing-throughput tradeoff for cognitive
radios, 2013 IEEE 14th Work. Signal Process. Adv. Wirel. Commun., pp. 115119, Jun.
2013.
[28] F. Zhang, W. Wang, and Z. Zhang, A primary traffic aware opportunistic spectrum
sensing for cognitive radio networks, 2011 IEEE 22nd Int. Symp. Pers. Indoor Mob.
Radio Commun. (PIMRC), pp. 700704, 2011.
143
[29] Y. Lin and V. W. S. Wong, WSN01-1: Frame Aggregation and Optimal Frame Size
Adaptation for IEEE 802.11n WLANs, Global Telecommunications Conference, 2006.
GLOBECOM 06. IEEE. pp. 16, 2006.
[30] Y.-D. Lin, J.-H. Yeh, T.-H. Yang, C.-Y. Ku, S.-L. Tsao, and Y.-C. Lai, Efficient
dynamic frame aggregation in IEEE 802.11s mesh networks, Int. J. Commun. Syst., vol.
22, no. 10, pp. 13191338, 2009.
[31] N. Hajlaoui, I. Jabri, and M. Ben Jemaa, Analytical study of frame aggregation in error-
prone channels, 2013 9th Int. Wirel. Commun. Mob. Comput. Conf., pp. 237242, Jul.
2013.
[32] M. Kim and C. Choi, Joint rate and fragment size adaptation in IEEE 802.11 n wireless
LANs, Netw. Conf. (CCNC), 2011 IEEE, pp. 942947, 2011.
[33] G. Kim, C. Shin, and H. Park, Adaptive frame size estimation using extended Kalman
filter for high-stressed WLANs, 2012 IEEE 23rd Int. Symp. Pers. Indoor Mob. Radio
Commun. -, pp. 272277, Sep. 2012.
[34] P. Teymoori, A. Dadlani, K. Sohraby, and K. Kim, An Optimal Packet Aggregation
Scheme in Delay-Constrained IEEE 802.11n WLANs, Wireless Communications,
Networking and Mobile Computing (WiCOM), 2012 8th International Conference on.
pp. 14, 2012.
[35] J. Deng and M. Davis, An adaptive packet aggregation algorithm for wireless
networks, Wireless Communications & Signal Processing (WCSP), 2013 International
Conference on. pp. 16, 2013.
[36] D. Kim and S. An, Throughput enhancement by Dynamic Frame Aggregation in multi-
rate WLANs, Communications and Vehicular Technology in the Benelux (SCVT), 2012
IEEE 19th Symposium on. pp. 15, 2012.
[37] J. Yin, X. Wang, and D. P. Agrawal, Optimal packet size in error-prone channel for
IEEE 802.11 distributed coordination function, Wireless Communications and
Networking Conference, 2004. WCNC. 2004 IEEE, vol. 3. pp. 16541659 Vol.3, 2004.
[38] E. Rocha, D. Corujo, and R. Aguiar, Implementing and evaluating improved MAC
efficiency through payload extension in 802.11n networks, Communications
Workshops (ICC), 2013 IEEE International Conference on. pp. 982987, 2013.
[39] E. N. Ciftcioglu and O. Gurbuz, Opportunistic Scheduling with Frame Aggregation for
Next Generation Wireless LANs, Communications, 2006. ICC 06. IEEE International
Conference on, vol. 11. pp. 52285233, 2006.
[40] N. Shetty, S. Pollin, and P. Paweczak, Identifying Spectrum Usage by Unknown
Systems using Experiments in Machine Learning, Wireless Communications and
Networking Conference, 2009. WCNC 2009. IEEE. pp. 16, 2009.
[41] H. Nguyen, N. Nguyen, G. Zheng, Z. Han, and R. Zheng, Binary Blind Identification of
Wireless Transmission Technologies for Wide-Band Spectrum Monitoring, Global
Telecommunications Conference (GLOBECOM 2011), 2011 IEEE. pp. 16, 2011.
[42] R. Miller, W. Xu, P. Kamat, and W. Trappe, Service Discovery and Device
Identification in Cognitive Radio Networks, in 2007 2nd IEEE Workshop on
Networking Technologies for Software Define Radio Networks, 2007, pp. 4047.
144
[43] K. Kim, C. M. Spooner, I. Akbar, and J. H. Reed, Specific Emitter Identification for
Cognitive Radio with Application to IEEE 802.11, in IEEE GLOBECOM 2008 - 2008
IEEE Global Telecommunications Conference, 2008, pp. 15.
[44] V. Brik, S. Banerjee, M. Gruteser, and S. Oh, Wireless device identification with
radiometric signatures, in Proceedings of the 14th ACM international conference on
Mobile computing and networking - MobiCom 08, 2008, p. 116.
[45] Z. Shi, M. Liu, and L. Huang, Transient-based identification of 802.11b wireless
device, in 2011 International Conference on Wireless Communications and Signal
Processing (WCSP), 2011, pp. 15.
[46] C. Neumann, O. Heen, and S. Onno, An Empirical Study of Passive 802.11 Device
Fingerprinting, in 2012 32nd International Conference on Distributed Computing
Systems Workshops, 2012, pp. 593602.
[47] S. Hu, Y. Yao, and Z. Yang, MAC protocol identification using support vector
machines for cognitive radio networks, IEEE Wirel. Commun., vol. 21, no. 1, pp. 52
60, Feb. 2014.
[48] R. Challoo, a. Oladeinde, N. Yilmazer, S. Ozcelik, and L. Challoo, An overview and
assessment of wireless technologies and coexistence of ZigBee, bluetooth and wi-fi
devices, Procedia Comput. Sci., vol. 12, pp. 386391, 2012.
[49] P. Chatzimisios, A. C. Boucouvalas, and V. Vitsas, Effectiveness of RTS/CTS
handshake in IEEE 802.11a Wireless LANs, Electronics Letters, vol. 40, no. 14. pp.
915916, 2004.
[50] IEEE, Supplement to IEEE Standard for Information Technology- Telecommunications
and Information Exchange Between Systems- Local and Metropolitan Area Networks-
Specific Requirements- Part 11: Wireless LAN Medium Access Control (MAC) and
Physical Layer (PHY) Sp, IEEE Std 802.11b-1999. pp. i90, 2000.
[51] I. Ramachandran and S. Roy, Clear channel assessment in energyconstrained wideband
wireless networks, Wireless Communications, IEEE, vol. 14, no. 3. pp. 7078, 2007.
[52] I. Ramachandran and S. Roy, WLC46-2: On the Impact of Clear Channel Assessment
on MAC Performance, Global Telecommunications Conference, 2006. GLOBECOM
06. IEEE. pp. 15, 2006.
[53] B. Pearson, Complementary code keying made simple, Intersil, Milpitas, CA, USA,
Appl. Notes AN9850, vol. 1, 2000.
[54] B. E. Henty, A Brief Tutorial on the PHY and MAC layers of the IEEE 802.11 b
Standard, July, vol. 12, pp. 1216, 2001.
[55] IEEE, IEEE Standard for Information Technology- Telecommunications and
Information Exchange Between Systems- Local and Metropolitan Area Networks-
Specific Requirements Part Ii: Wireless LAN Medium Access Control (MAC) and
Physical Layer (PHY) Specifications, IEEE Std 802.11g-2003 (Amendment to IEEE Std
802.11, 1999 Edn. (Reaff 2003) as amended by IEEE Stds 802.11a-1999, 802.11b-1999,
802.11b-1999/Cor 1-2001, and 802.11d-2001). pp. i67, 2003.
[56] M. Debbah, Short introduction to OFDM, White Pap. Mob. Commun. Group, Inst.
Eurecom, 2004.
[57] M. Ergen, IEEE 802.11 Tutorial, Univ. Calif. Berkeley, vol. 70, 2002.
145
[58] IEEE, IEEE Standard for Information technology-- Local and metropolitan area
networks-- Specific requirements-- Part 11: Wireless LAN Medium Access Control
(MAC)and Physical Layer (PHY) Specifications Amendment 5: Enhancements for
Higher Throughput, IEEE Std 802.11n-2009 (Amendment to IEEE Std 802.11-2007 as
amended by IEEE Std 802.11k-2008, IEEE Std 802.11r-2008, IEEE Std 802.11y-2008,
and IEEE Std 802.11w-2009). pp. 1565, 2009.
[59] A. Saif and M. Othman, Frame Aggregation in Wireless Networks: Techniques and
Issues., IETE Tech. Rev., vol. 28, no. 4, p. 336, 2011.
[60] Z. Alliance, IEEE 802.15. 4, ZigBee standard, http//www. zigbee. org, 2009.
[61] IEEE, IEEE Standard for Telecommunications and information Local and metropolitan
area networks Specific requirements Part 15 . 4: Wireless Medium Access Control (
MAC ) and Physical Layer ( PHY ) Specifications for Low-Rate Wireless Personal Area
Networks (, vol. 2006, no. September. 2006.
[62] RouterBoard.com: RB433UAH. [Online]. Available:
http://routerboard.com/RB433UAH. [Accessed: 16-Oct-2015].
[63] Manual:TOC - MikroTik Wiki. [Online]. Available:
http://wiki.mikrotik.com/wiki/Manual:TOC. [Accessed: 16-Oct-2015].
[64] RF Signal Analyzers - National Instruments. [Online]. Available:
http://sine.ni.com/nips/cds/view/p/lang/en/nid/203041. [Accessed: 16-Oct-2015].
[65] N. LaSorte, D. Bloom, S. Rajab, S. Asadallahi, H. H. Refai, R. Zhang, and W. He,
Comparison of duty cycle measurement techniques of 802.11 b/g in the frequency and
time domain, 2013 IEEE Int. Instrum. Meas. Technol. Conf., 2013.
[66] M. Lopez-Benitez and F. Casadevall, Methodological aspects of spectrum occupancy
evaluation in the context of cognitive radio, Wireless Conference, 2009. EW 2009.
European. pp. 199204, 2009.
[67] A. Holt and C.-Y. Huang, 802.11 Wireless Networks: Security and Analysis, 1st ed.
Springer Publishing Company, Incorporated, 2010.
[68] S. A. Rajab, W. Balid, and H. H. Refai, Comprehensive study of spectrum occupancy
for 802.11b/g/n homogeneous networks, in 2015 IEEE International Instrumentation
and Measurement Technology Conference (I2MTC) Proceedings, 2015, pp. 17411746.
[69] W. Balid, S. A. Rajab, and H. H. Refai, Comprehensive study of spectrum utilization
for 802.11 b/g/n networks, in Wireless Communications and Mobile Computing
Conference (IWCMC), 2015 International, 2015, pp. 15261531.
[70] J. O. Rawlings, S. G. Pantula, and D. A. Dickey, Applied regression analysis: a research
tool. 1998. Springer, 1998.
[71] r2, a measure of goodness-of-fit of linear regression, 2014. [Online]. Available:
http://www.graphpad.com/guides/prism/6/curve-
fitting/index.htm?r2_ameasureofgoodness_of_fitoflinearregression.htm.
[72] D. Mercedes, M. Plata, . Gabriel, and A. Retiga, Evaluation of energy detection for
spectrum sensing based on the dynamic selection of detection-threshold, vol. 35, pp.
135143, 2012.
146
[73] C. Walck, Hand-book on STATISTICAL DISTRIBUTIONS for experimentalists,
Hand-b. Stat. Distrib. Exp., no. September, pp. 2635, 2007.
[74] W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery, Numerical Recipes
3rd Edition: The Art of Scientific Computing, 3rd ed. New York, NY, USA: Cambridge
University Press, 2007.
[75] A. Bhattacharyya, On a measure of divergence between two multinomial populations,
Sankhy Indian J. Stat., pp. 401406, 1946.
[76] S. A. Rajab, W. Balid, M. O. Al Kalaa, and H. H. Refai, Energy Detection and Machine
Learning for the Identification of Wireless MAC Technologies, in Wireless
Communications and Mobile Computing Conference (IWCMC), 2015 11th
International. IEEE, 2015.
[77] H. Zhang, The optimality of naive Bayes, in FLAIRS, 2004.
[78] S. Aksoy, K. Koperski, C. Tusk, G. Marchisio, and J. C. Tilton, Learning bayesian
classifiers for scene classification with a visual grammar, IEEE Trans. Geosci. Remote
Sens., vol. 43, no. 3, pp. 581589, Mar. 2005.
[79] S. Kim, K. Han, H. Rim, and S. H. Myaeng, Some Effective Techniques for Naive
Bayes Text Classification, IEEE Trans. Knowl. Data Eng., vol. 18, no. 11, pp. 1457
1466, Nov. 2006.
[80] C. Livadas, R. Walsh, D. Lapsley, and W. Strayer, Usilng Machine Learning
Technliques to Identify Botnet Traffic, in Proceedings. 2006 31st IEEE Conference on
Local Computer Networks, 2006, pp. 967974.
[81] T. Cover and P. Hart, Nearest neighbor pattern classification, IEEE Trans. Inf. Theory,
vol. 13, no. 1, pp. 2127, Jan. 1967.
[82] M. Xie, J. Hu, S. Han, and H.-H. Chen, Scalable hypergrid k-NN-Based online
anomaly detection in wireless sensor networks, IEEE Trans. Parallel Distrib. Syst., vol.
24, no. 8, pp. 16611670, Aug. 2013.
[83] M. Tseng, Medical data mining using BGA and RGA for weighting of features in fuzzy
k-NN classification, in 2009 International Conference on Machine Learning and
Cybernetics, 2009, no. July, pp. 30703075.
[84] E. Acar and M. S. Ozerdem, An iris recognition system by laws texture energy measure
based k-NN classifier, in 2013 21st Signal Processing and Communications
Applications Conference (SIU), 2013, pp. 14.
[85] A. Jayasuriya, S. Perreau, A. Dadej, and S. Gordon, Hidden vs exposed terminal
problem in ad hoc networks. ATNAC 2004, 2004.
[86] IEEE Std. 802.11-2007, Part 11: Wireless LAN Medium Access Control (MAC) and
Physical Layer (PHY) specifications, 2007.
[87] IEEE 1609.4-2006 WAVE Multi-Channel Operation, 2006.
[88] M. ORielly and J. Rosenworcel, Steering into the Future with More Wi-Fi by Sharing
the Upper 5 GHz Band, 2015. [Online]. Available: https://www.fcc.gov/news-
events/blog/2015/09/16/steering-future-more-wi-fi-sharing-upper-5-ghz-band.
147
APPENDIX A
This appendix provides distributions parameters that were obtained from the empirical
modelling activities presented in this work.
Table 15. Homogeneous networks idle time distributions empirical modelling distributions parameters
Combination Throughput Scale Shape Location Distribution

0.140845 1151.251 0.31634 NA
0.28169 440.6714 0.331248 NA
0.422535 288.5521 0.352516 NA
802.11b
Weibull
0.56338 216.5685 0.38959 NA
0.704225 150.3286 0.420093 NA
0.84507 110.6426 0.457851 NA
0.985915 80.31762 0.533705 NA
0.25641 454.4337 0.327704 NA
0.512821 154.6577 0.36692 NA
0.769231 91.01028 0.429571 NA
0.948718 56.419 0.532209 NA
0.512821 137.558 0.371585 NA
802.11bb
Weibull
0.641026 112.3662 0.397244 NA
0.897436 64.70541 0.486399 NA
0.948718 54.43224 0.512986 NA
0.769231 83.78105 0.435607 NA
0.935897 42.97834 0.587049 NA
0.948718 45.56229 0.578959 NA
0.395257 235.6976 0.358938 NA
0.922266 64.30312 0.552894 NA
802.11bbb
0.948617 58.17729 0.595449 NA

Weibull
0.948617 59.0059 0.579327 NA

0.943347 47.31613 0.64903 NA
0.961792 58.52603 0.57807 NA
0.988142 45.27752 0.659201 NA
0.177936 296.2906 0.401876 NA
0.355872 188.0788 0.474716 NA
802.11g
Weibull
0.533808 136.1997 0.545248 NA

0.711744 94.95738 0.613136 NA
0.907473 67.13438 0.720941 NA
148
0.20202 243.4112 0.420061 NA
0.3367 150.9766 0.459805 NA
0.484848 92.07866 0.522523 NA
0.673401 80.93936 0.582589 NA
0.750842 57.12293 0.666552 NA
802.11gg
Weibull
0.37037 175.1613 0.4992 NA
0.673401 87.96291 0.595932 NA
0.838384 61.70147 0.69289 NA
0.905724 49.57255 0.775471 NA
0.538721 119.0187 0.581204 NA
0.96633 43.87613 0.94169 NA
0.707071 90.92134 0.645902 NA
0.505051 118.8617 0.524905 NA
0.673401 87.09919 0.60096 NA
0.973064 46.49123 0.928603 NA
0.96633 46.475 0.933612 NA
0.841751 63.76732 0.718561 NA
0.96633 43.60345 0.989488 NA
0.96633 44.32544 0.975629 NA
0.962963 43.71256 0.985164 NA
802.11ggg
0.969697 44.02202 0.968904 NA
Weibull
0.942761 44.77692 0.95326 NA
0.912458 51.25997 0.807983 NA
0.962963 44.60241 0.947918 NA
0.976431 44.68183 0.951513 NA
0.986532 43.60212 0.982651 NA
0.979798 43.47336 0.97884 NA
0.976431 43.22909 0.984923 NA
1 44.22538 0.979422 NA
0.983165 43.74223 0.978745 NA
0.976431 43.30836 0.969737 NA
0.177305 2.069346 NA 4.203107
Log-normal
0.35461 1.592815 NA 3.822592

802.11n
0.531915 1.211674 NA 3.534159

0.707447 1.202277 NA 3.527166
0.884752 1.204307 NA 3.533672
0.204461 1.945643 NA 4.148856
Log-normal
0.371747 1.60833 NA 3.732821

802.11nn
0.557621 1.201955 NA 3.382773

0.743494 1.056183 NA 3.35401
0.923792 0.970632 NA 3.410213
149
0.966543 1.043055 NA 3.324704
0.390335 1.492385 NA 3.828994
0.743494 1.007904 NA 3.401058
0.921933 1.063105 NA 3.344005
0.979554 1.053032 NA 3.311238
0.576208 1.109256 NA 3.554679
0.762082 1.117833 NA 3.554234
0.901487 1.118302 NA 3.537706
0.576699 1.269116 NA 3.318556
0.743689 1.232356 NA 3.188379
0.860194 1.171053 NA 3.225668
0.881553 1.189282 NA 3.195587
0.850485 1.111687 NA 3.207965
0.741748 1.189706 NA 3.226072
0.879612 1.162332 NA 3.201583
0.885437 1.092253 NA 3.197527
0.891262 1.124227 NA 3.193137
0.897087 1.098211 NA 3.200927
0.838835 1.0693 NA 3.205014
0.794175 1.082335 NA 3.187564
0.825243 1.101089 NA 3.19606
0.928155 1.11306 NA 3.218478
0.88932 1.086084 NA 3.195579
Log-normal
802.11nnn
0.908738 1.068917 NA 3.211776

0.757282 1.122891 NA 3.210936
0.893204 1.112785 NA 3.206724
0.924272 1.118681 NA 3.208861
0.88932 1.098844 NA 3.223486
0.926214 1.210991 NA 3.211035
0.902913 1.140704 NA 3.222747
0.92233 1.097092 NA 3.186204
0.961165 1.104124 NA 3.18189
0.912621 1.060313 NA 3.20389
0.873786 1.083727 NA 3.192075
0.906796 1.058571 NA 3.187935
0.935922 1.063335 NA 3.200623
0.897087 1.095473 NA 3.188811
0.92233 1.070957 NA 3.176276
0.933981 1.103175 NA 3.185043
0.953398 1.13054 NA 3.203061
0.945631 1.12596 NA 3.198065
150
0.920388 1.126939 NA 3.193299
0.947573 1.113638 NA 3.184497
Table 16. Heterogeneous networks idle time distributions empirical modelling distributions parameters
Combination Throughput Scale Shape Location Mean Distribution

0.272682 1.10298 0 4.656476 0
0.520575 1.069559 0 4.618012 0
0.76351 1.295357 0 4.757562 0
0.979177 1.908868 0 5.704975 0
1 2.113918 0 6.109859 0
0.366882 1.333006 0 5.227343 0
Log-normal
802.11bn
0.547843 1.44634 0 5.363297 0

0.783342 1.653392 0 5.498698 0
0.884978 1.845906 0 5.751182 0
0.840357 1.939403 0 5.778878 0
0.371839 1.326488 0 5.231814 0
0.540407 1.447206 0 5.356055 0
0.701537 1.508741 0 5.405192 0
0.946951 1.913603 0 5.810937 0
0.285714 52.27314 -0.34669 0.5 0
0.47619 52.9427 -0.33988 0.5 0
0.666667 132.3694 -0.27529 0.5 0
0.853333 167.4889 -0.03088 0.5 0
0.952381 79.08634 -0.49533 0.5 0
0.380952 58.46963 -0.49444 0.5 0
0.571429 47.67757 -0.25662 0.5 0 Generalized Pareto
0.739048 37.33339 0.33333 0.5 0
0.845714 27.90672 0.814537 0.5 0
802.11gn
0.99619 70.99381 -0.57591 0.5 0

0.502857 72.43361 -0.46505 0.5 0
0.620952 73.51602 -0.17785 0.5 0
0.75619 46.95976 0.132992 0.5 0
0.857143 40.27006 0.485429 0.5 0
0.980952 73.88522 -0.50547 0.5 0
0.51619 58.16333 -0.37951 0.5 0
0.6 45.64311 -0.13294 0.5 0
0.739048 135.7636 -0.1678 0.5 0
0.851429 28.60806 0.775861 0.5 0
Exp
ent
2.1
on
80
1g
ial
0.242915 0 0 0 240.6198
b
151
0.445344 0 0 0 408.0854
0.850202 0 0 0 445.5406
1 0 0 0 197.1214
0.404858 0 0 0 331.6408
0.607287 0 0 0 341.6722
0.615385 0 0 0 172.9214
0.611336 0 0 0 336.0893
0.453441 0 0 0 329.7796
0.59919 0 0 0 169.3101
0.623482 0 0 0 329.5189
152

2016 Rajab Samer Dissertation

Uploaded by

Copyright:

Available Formats

2016 Rajab Samer Dissertation

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

2016 Rajab Samer Dissertation

Uploaded by

Copyright:

Available Formats

UNIVERSITY OF OKLAHOMA

TOWARD ENHANCED WIRELESS COEXISTENCE IN THE 2.4GHZ ISM BAND

VIA TEMPORAL CHARACTERIZATION AND EMPIRICAL MODELING OF

SUBMITTED TO THE GRADUATE FACULTY

in partial fulfillment of the requirements for the

A DISSERTATION APPROVED FOR THE

continuous support of individuals whom Id like to take this opportunity to

Acknowledgment goes to my friends whose help and support was a major

motive for me to succeed.

Michelle Farabough is acknowledged for her instrumental help in editing this

The University of Oklahoma is very well appreciated and acknowledged for

support and embracement

modelling of 802.11 temporal behavior. A detailed characterization of 802.11b/g/n

distribution and channel utilization.

networks. Spectrum occupancy measurements are essential for wireless networks

planning and deployment.

Detailed characterization of 802.11g/n homogeneous and heterogeneous network traffic

enhanced wireless coexistence.

802.11 network idle time distribution.

In this work, a wireless technology identification algorithm was assessed experimentally.

measured and used as algorithm input. Identification accuracies of up to 96.83% and

demonstrate improved ZigBee performance and significant enhancement in throughput

as a result of using adaptive packet size transmissions.

significantly in recent years, rendering wireless spectrum an expensive and scarce

Fi one of todays prevailing wireless communications technologies. New bandwidth-

limited, frequency spectrum.

Various wireless technologies are uniquely characterized according to time, frequency,

and power domains. Wireless technology characterization and identification provides a

anticipated transmission periodicity, frequency bandwidth, and transmit power, among

other parameters. Awareness about these characteristics is useful, especially in

heterogeneous frequency bands like the ISM band.

expensive resource. Although a number of spectrum occupancy measurement campaigns

2.4GHz ISM band spectrum occupancy.

Spectrum scarcity makes heterogeneous spectrum access a necessity in current wireless

frequency band [2]. As a consequence, wireless coexistence among various wireless

parameter adaption for coexisting wireless technologies, thus enhancing transmission

as a timely example of issues related to heterogeneous spectrum access. Spectrum

occupancy measurement serves as a tool to assist wireless device manufacturers to make

cognizant decisions when choosing transmission parameters, including transmission

intelligently characterize the spectrum occupancy and identify wireless technologies

utilizing the spectrum at a particular time.

the received signal strength is above a certain threshold [4].

understanding spectrum occupancy for heterogeneous Wi-Fi networks can aid in

improving coexisting network performance. Measuring frequency channel DC is one

method for assessing spectrum occupancy.

This work in part investigates time distributions modelling of 802.11 heterogeneous

networks in the 2.4GHz ISM band, including a detailed characterization of 802.11b/g/n

coexisting wireless technologies. Time distribution modelling focuses on 802.11

1) The relatively limited 80MHz spectrum, and

implement wireless technologies in their devices for a number of practical reasons,

including patient convenience. Medical device applications for wireless connectivity

range from transmitting regular telemetry data to patient life-critical applications.

Medical devices must utilize low-power consumption technologies to minimize service

interruption. Several manufacturers employ ISM band wireless technologies as often as

communication technologies currently implemented in medical devices. Devices using

wireless coexistence risk assessment of wireless medical systems must be performed

certification process [10]. The experimental wireless technology identification techniques

modified for improved performance based on channel temporal traffic patterns.

wireless technology characterization and identification techniques. For example, cognitive

Cognitive radio network requires spectrum usability assessment, particularly in licensed