Performance Analysis Of 802.11b Networks
Except where reference is made to the work of others, the work described in this thesis is
my own or was done in collaboration with my advisory committee. This thesis does not
include proprietary or classifled information.
Ravi Chandra Paruchuri
Certiflcate of Approval:
Foster Dai
Associate Professor
Department of Electrical and Computer
Engineering
Prathima Agrawal, Chair
Samuel Ginn Distinguished Professor
Department of Electrical and Computer
Engineering
Thaddeus A. Roppel
Associate Professor
Department of Electrical and Computer
Engineering
Stephen L. McFarland
Acting Dean, Graduate School
Performance Analysis Of 802.11b Networks
Ravi Chandra Paruchuri
A Thesis
Submitted to
the Graduate Faculty of
Auburn University
in Partial Fulflllment of the
Requirements for the
Degree of
Master of Science
Auburn, Alabama
May 11, 2006
Performance Analysis Of 802.11b Networks
Ravi Chandra Paruchuri
Permission is granted to Auburn University to make copies of this thesis at its discretion,
upon the request of individuals or institutions and at their expense. The author reserves
all publication rights.
Signature of Author
Date of Graduation
iii
Vita
Ravi Chandra Paruchuri, son of Ram Mohan Rao Paruchuri and Vijaya Paruchuri
was born on 27th October 1981, in Andhra Pradesh, India. He received his Bachelor
of Technology degree in Electronics and Communications Engineering in 2002 from VNR
Vignana Jyothi Institute of Engineering and Technology, a?liated to Jawaharlal Nehru
Technological University.
In January 2003, he joined Auburn University with an intent to pursue his Masters
degree in Electrical and Computer Engineering department. Initially he worked under Dr.
Foster Dai and Dr. Richard C. Jaeger in the fleld of wireless IC design. In January 2004, he
joined as Graduate Research assistant under Dr. Prathima Agrawal in the fleld of Wireless
Networks.
His research interests include wireless networks (802.11, 802.15 and 802.16) and high
speed ethernet networks.
iv
Thesis Abstract
Performance Analysis Of 802.11b Networks
Ravi Chandra Paruchuri
Master of Science, May 11, 2006
(B.Tech., Jawaharlal Nehru Technological University, 2002)
81 Typed Pages
Directed by Dr. Prathima Agrawal
Wireless local area networks (WLANs), especially those incorporating the 802.11b stan-
dard, have experienced rapid evolution and unprecedented widespread deployment during
the past few years. Increasing research led to the advent of new technologies and standards
in WLANs enabling them to achieve higher data rates (e.g., from 2 Mbps to 11Mbps) and
wider coverage. Since 802.11b networks operate in the unlicensed ISM (Industrial, Scien-
tiflc and Medical) band of the frequency spectrum, they experience interference from other
devices operating in the same band. Therefore, it is important to understand the per-
formance of 802.11b networks, in terms of throughput and quality (packet error rate), for
both TCP and UDP data transmissions over these networks, under interference. This thesis
presentsadetailedexperimentalstudyoftheimpactofselfinterference(other802.11baccess
points and terminals), Bluetooth interference and microwave interference from household
appliances on 802.11b networks. A mathematical model for predicting the throughput of
802.11b networks in the presence of self interference is developed. Such a model is extremely
useful in planning WLAN network deployments in indoor environments and in proactive
performance management.
v
Acknowledgments
It is my pleasure to thank all the people without whom I could not have completed the
undertaken task.
I am extremely grateful to my adviser, Dr. Prathima Agrawal, for her support and
guidance throughout the course of my graduate study at Auburn University. She has been
a constant source of inspiration for me to pursue my research ideas. I express my sincere
thanks to my previous advisors Dr. Foster Dai and Dr. Richard C. Jaeger and my committee
member Dr. Roppel for their professional guidance.
I will always cherish the opportunity to work with my research mates and I am grateful
for the various discussions, support and above all the friendship we shared. A special thanks
to Santosh Pandey for his extensive help and advice on my thesis work. I am extremely
thankful to all my friends at Auburn for being the surrogate family during my Masters.
Finally, I am greatly indebted to my parents, sister and my grand father whose love
and encouragement provided me the strength to achieve my goals and I humbly dedicate
this work to them.
vi
Style manual or journal used Journal of Approximation Theory (together with the style
known as \aums"). Bibliograpy follows van Leunen?s A Handbook for Scholars.
Computer software used The document preparation package TEX (speciflcally LATEX)
together with the departmental style-flle aums.sty.
vii
Table of Contents
List of Figures x
List of Tables xii
1 Introduction 1
1.1 Objective and Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Overview of Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.3 Thesis Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Prior Work and Interference-Free Measurements 4
2.1 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 TCP and UDP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.3 Measurement setup and tools used . . . . . . . . . . . . . . . . . . . . . . . 6
2.3.1 Experiment setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.3.2 Received Signal Strength Indicator (RSSI) . . . . . . . . . . . . . . . 7
2.3.3 Iperf . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.4 Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.4.1 Signal strength variation with distance . . . . . . . . . . . . . . . . . 8
2.4.2 Throughput variation with RSSI . . . . . . . . . . . . . . . . . . . . 8
2.4.3 Efiect of RTS/CTS signals on throughput . . . . . . . . . . . . . . . 9
2.4.4 Efiect of antenna orientation on signal strength . . . . . . . . . . . . 10
2.4.5 Fragmentation threshold on throughput . . . . . . . . . . . . . . . . 11
2.4.6 Impact of Transmission power of wireless card . . . . . . . . . . . . 12
2.4.7 MAC retransmission values . . . . . . . . . . . . . . . . . . . . . . . 12
2.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3 Interference Study of 802.11b Networks 15
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.2 Previous work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.3 Anechoic chamber . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.4 Self Interference with other 802.11b networks . . . . . . . . . . . . . . . . . 18
3.4.1 IEEE 802.11b . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.4.2 802.11b self interference . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.4.3 Co-channel and adjacent channel interference . . . . . . . . . . . . . 21
3.4.4 Impact of interfering access point (AP2) data rate . . . . . . . . . . 27
3.5 Bluetooth interference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.5.1 Bluetooth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.5.2 Interference with v1.2 devices . . . . . . . . . . . . . . . . . . . . . . 29
viii
3.5.3 Interference with v1.1 devices . . . . . . . . . . . . . . . . . . . . . . 30
3.5.4 Bluetooth interference with 802.11b and BT embedded in the same
device . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.6 Bluetooth interference - Comparison of mathematical and experimental results 32
3.6.1 Test bed and Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
4 Throughput Prediction Model in the Presence of Self Interference
for Typical Office Environments 37
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.2 Prior Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.3 Throughput prediction model - Inputs . . . . . . . . . . . . . . . . . . . . . 40
4.4 Efiect of interference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.4.1 Calculation of Practical ? . . . . . . . . . . . . . . . . . . . . . . . . 44
4.4.2 Packet Error Rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.4.3 Variation of throughput and PER with distance ?dm? . . . . . . . . . 46
4.4.4 Variation of throughput and PER with distance d1 . . . . . . . . . . 48
4.4.5 Variation of throughput and PER with distance ?d2? . . . . . . . . . 50
4.4.6 Variation of transmission rate with number of stations . . . . . . . . 51
4.5 Throughput prediction model in the presence of one interferer (two stations) 52
4.6 Throughput prediction model in the presence of ?n? stations . . . . . . . . . 53
4.7 Throughput prediction model for ?n? stations on difierent channels . . . . . 54
4.8 Experimental veriflcation of throughput prediction model . . . . . . . . . . 57
4.9 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5 Conclusion 61
5.1 Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
Bibliography 63
List of Definitions and Abbreviations 68
ix
List of Figures
2.1 Test setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Signal strength variation with distance (Result from [51]) . . . . . . . . . . 9
2.3 Throughput variation with signal strength . . . . . . . . . . . . . . . . . . . 10
2.4 Efiect of RTS and CTS on Throughput . . . . . . . . . . . . . . . . . . . . 11
2.5 Change in RSSI values because of change in the orientation. . . . . . . . . . 12
2.6 Efiect of Fragmentation threshold on Throughput . . . . . . . . . . . . . . . 13
2.7 Throughput variation with transmission power at difierent RSSI values . . . 14
2.8 Impact of MAC retransmission values on TCP throughput . . . . . . . . . . 14
3.1 802.11b Channel Assignment . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 Experimental Test bed . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.3 Throughput variation with and without interference . . . . . . . . . . . . . 21
3.4 Experimental test bed for evaluation of co-channel and adjacent channel in-
terference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.5 Behavior of co-channel and adjacent channel interference at difierent dis-
tances from main AP . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.6 Signal Strength difierence of access points and distance of N2 from AP1 at
difierent locations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.7 Co-channel and adjacent channel interference (when AP2 is on Channel 6) . 25
3.8 Throughput variation with the change in the distance (or SNR) between AP2
and N3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.9 Efiect of varying data rates of AP2 on the throughput of the main link . . . 28
3.10 Experimental setup for Bluetooth interference on 802.11b . . . . . . . . . . 29
x
3.11 Throughput of WLAN vs distance between Bluetooth nodes . . . . . . . . . 31
3.12 Test bed when BT and 802.11b are embedded in the same device . . . . . . 32
3.13 Test Bed to compare throughput values with and without interference . . . 33
3.14 Theoretical and practical throughput comparison at 4m away from the trans-
mitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.15 Throughput comparison with and without interference at 10m and 14m away
from the transmitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.16 Percentage decrease in throughput comparison . . . . . . . . . . . . . . . . 36
4.1 Variation of ? with n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
4.2 Experimental setup for studying the variation of PER and throughput with
distances dm, d1 and d2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
4.3 PER variation with dm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
4.4 Throughput variation with dm . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.5 PER variation with d1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.6 Throughput variation with d1 . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.7 Throughput comparison when the number of interferers are 4 . . . . . . . . 55
4.8 Piecewise Spectrum Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4.9 Throughput comparison for one interferer on all the overlapping channels at
d1 = 10m . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.10 Throughput comparison for one interferer on all the overlapping channels at
d1 = 6 m . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.11 Model validation for difierent d values . . . . . . . . . . . . . . . . . . . . . 59
4.12 Model validation when two interferers are on the overlapping channels and
the main station is on channel 11 . . . . . . . . . . . . . . . . . . . . . . . . 59
4.13 Model validation when interfering stations are on difierent channels and the
main station is on channel 11 . . . . . . . . . . . . . . . . . . . . . . . . . . 60
xi
List of Tables
3.1 Efiect of anechoic chamber in mitigating unwanted interference . . . . . . . 18
3.2 Bluetooth Interference on WLAN for various test cases . . . . . . . . . . . . 32
4.1 Receiver sensitivity values of Linksys WPC 11 wireless card . . . . . . . . . 42
4.2 Parameter values for difierent n . . . . . . . . . . . . . . . . . . . . . . . . . 51
4.3 Individual transmission rate variation with n . . . . . . . . . . . . . . . . . 52
xii
Chapter 1
Introduction
Phenomenal growth and rapid advancements in the fleld of wireless communications
enable us to transmit not only data but also voice and video through a variety of wireless
communication systems. These systems can be wireless cellular systems, wireless local area
networks (WLANs), wireless personal area networks (WPANs) and satellite communication
systems [7]. Among them, wireless local area networks, especially 802.11b, have experienced
rapid evolution and unprecedented widespread deployment during the past few years. It is
economical to send data through WLANs as they work in unlicensed band where the radio
spectrum does not incur any expense to the user or to the service provider. Hence, 802.11b
networks are becoming increasingly popular, both indoor and outdoor at universities, o?ces
and other public areas. Increasing research in this fleld led to the advent of new technologies
and standards in WLANs enabling them to providehigher data rates and increased coverage.
Since 802.11b networks work in an unlicensed frequency spectrum, it experiences in-
terference from other devices working in the same band. Therefore, it is very important
to study the performance, in terms of capacity and quality, of both Transmission Control
Protocol (TCP) and User Datagram Protocol (UDP) over these networks, with and without
interference. Since TCP is the most widely used data networking protocol, its performance
is analyzed in this thesis.
1
1.1 Objective and Motivation
The objective and motivation of the research presented in this thesis is to
1. Evaluate and capture the efiects of parameters like signal strength, propagation
distance, packet size, Request to send/ Clear to send (RTS/CTS), transmission power and
medium access control (MAC) retransmissions on the network performance.
2. Analyze the efiect of interference of other 802.11b networks (self interference) in
detail on 802.11b networks.
3. Evaluate the performance of 802.11b networks under Bluetooth interference for
difierent network scenarios.
4. Develop an analytical model for predicting the throughput of 802.11b networks in
the presence of self interference. Co-channel and adjacent channel interference was modeled
to estimate the TCP throughput on all overlapping channels.
1.2 Overview of Results
The main contributions of the thesis are summarized as follows :
1. A mathematical model for predicting the throughput of 802.11b networks under
self interference was developed. The model uses empirical data and piece wise spectral
analysis. Most of the models consider the 802.11b stations to be on the same channel but
in the present model the interfering stations can be on any of the overlapping channels. It
is extremely di?cult to model the behavior of a wireless link in the presence of interfering
802.11b stations. This is due to the time varying nature of the wireless channel. The
model takes the distance between interfering wireless transmitter and receiver and the
variation of transmission rate with the number of contending stations into consideration.
Such dependencies are also not accounted for in other models [42-45].
2
2. Interesting results has been obtained for the 802.11b networks in the presence of self
interference. It was noticed that when two 802.11b stations are close to each other and are
on adjacent channels, one of the stations captures the channel and hence gives much higher
throughput than the other.
3. Interference of Bluetooth devices on 802.11b networks have been studied for various
network scenarios. The measured results of 802.11b throughput under Bluetooth interfer-
ence have been compared with the theoretical results.
1.3 Thesis Structure
Chapter 2 discusses the existing literature in signal strength and throughput mea-
surements of 802.11b WLAN networks without interference. Measured results obtained by
varying the parameters like distance, transmission power, fragmentation threshold etc., are
shown.
Chapter 3 focuses on throughput measurements of 802.11b networks in the presence of
interference. Such interference may be due to other 802.11b stations or Bluetooth stations
or household appliances like a microwave or a cordless phone. The results obtained are
presented in an intuitive fashion.
Chapter 4 discusses the throughput prediction model which estimates the throughput
in the presence of self interference on overlapping channels. Results validating the model
were also presented.
Chapter 5 summarizes the main contributions and directions for future work.
3
Chapter 2
Prior Work and Interference-Free Measurements
In this chapter, a brief review of the existing literature on the performance of 802.11b
wireless networks (without interference) is presented. Measurements showing the variation
of the throughput of WLAN networks as a function of signal strength, distance between
transmitter and receiver, power level of the transmitter , antenna orientation, request to
send / clear to send (RTS/CTS) etc., are described
2.1 Literature Review
The performance and signal strength variation (with distance) of 802.11a and 802.11b
networks are compared in [9]. It is concluded that the two networks have the same coverage
but 802.11a exhibits higher data rates. A difierent AP-card pair for both 802.11a and ?b?
networks is taken and the throughput variation with distance among difierent hardware
is observed. It was shown experimentally that the throughput depends on vendor speciflc
equipment. Similar variation has also been observed by our measurements.
The efiects of path loss and building loss measurements due to residences for outdoor
networks are described in [10]. Efiects of various shadowing objects like trees and houses
and receiver positions are measured and quantifled.
Zahur et.al.,in [11] measure the performance of 802.11b networks. The efiect of various
parameters including distance, power, and RTS/CTS on throughput is measured with varied
packet sizes using simulation. Hidden node problem is also considered to analyze the efiect
of RTS/CTS. In [12], net throughput is assessed by modeling the physical, medium access
4
control (MAC) and TCP overhead. The throughput drop because of collisions and slot
times is calculated. The modeled results match well with the measured throughput values
at all bit rates for 802.11b networks. By modeling, a net throughput of 80 percent at 6
Mbps and 55 percent at 54 Mbps was predicted for 802.11a LAN, which was still in a draft
stage then.
Performance of wireless LAN is evaluated with respect to signal to noise ratio (SNR),
flle size, number of simultaneous users and the direction of flle transfer in [13]. The following
are the observations made ; the throughput per user is quite low when there is more than
one user in a LAN, but the overall or net throughput increases giving an impression that the
lower layers of the 802.11b protocol stack reserves some resources for future users. It is also
observed that fairness is preserved by the underlying mechanism of 802.11b protocol. The
results show that the number of simultaneous users has a high impact on the throughput
compared to that due to SNR while flle size has little or no impact. Similarity of our
experimental results to these observations will be discussed in a later chapter.
2.2 TCP and UDP
TCP is a connection oriented reliable protocol. Reliability is ensured by acknowledge-
ments from the receiver and retransmissions of the lost packets at the sender. Congestion
control and  ow control mechanisms in TCP make sure that the fast sender does not swamp
too many packets on to the slow receiver. Whenever a packet is lost, TCP reduces its con-
gestion window size (a sender can only send a maximum bytes indicated by the window
size) to reduce congestion [49]. TCP is traditionally designed for wired networks wherein
the packet losses are mostly due to congestion. But in wireless networks (in 802.11 (a,b
and g) networks) the packet loss due to collisions and corruption (because of poor signal
strength) is very high. In such cases, TCP reduces its congestion window when a packet
5
Figure 2.1: Test setup
loss is reported and eventually the throughput becomes zero. The 802.11 MAC employs
MAC level retransmissions to hide these errors from TCP layer [6].
UDP, on the other hand, is a connection less unreliable protocol because there will be
no acknowledgement from the receiver of a correct packet reception and hence no retrans-
missions. Please note that these retransmissions (transport layer) are difierent from MAC
layer retransmissions.
2.3 Measurement setup and tools used
2.3.1 Experiment setup
Oursetupforevaluatingtheefiectofparameterslikesignalstrength, distance, RTS/CTS
etc., on throughput is shown in flgure 2.1. The client (or transmitter) N2 is associated with
the access point (AP) through a wireless connection while the server (or receiver) is con-
nected to AP through a wired connection.
Client Speciflcations -
6
Operating System: Linux
Processor:1.6 GHz Intel Pentium M
Wireless cards: Linksys and Cisco AP 200
Server Speciflcations -
Operating system : Windows XP
Processor: 1.6 GHz Intel Pentium M
Wireless cards: Inbuilt Dell Wireless card
Access Point - Orinoco AP 2000
Tools Used -
Iperf [15] for calculating TCP bandwidth. A brief introduction to Iperf is given is next
section.
RSSI value is measured by the Network Interface Card (NIC) and ?Wavemon? for signal
strength. ?Wavemon? is an open source tool used for signal strength measurement.
Measurements are taken for 60 seconds at each data point to get an average value.
2.3.2 Received Signal Strength Indicator (RSSI)
The IEEE 802.11 standard deflnes a mechanism by which RF energy is to be measured
by the circuitry on a wireless network interface card (NIC). "RSSI, abbreviated for Received
Signal Strength Indicator, is an arbitrary integer value, intended for use, internally, by the
microcode on the adapter and by the device driver" [49]. RSSI value can be treated as
replacement for signal to noise ratio. RSSI value of 256 correspond to 100 db SNR (or 0
db signal value) and RSSI value of 156 correspond to 0 db SNR (or -100db signal value).
The terms RSSI, signal to noise ratio and signal strength are all similar and hence are
interchangeably used in this thesis.
7
2.3.3 Iperf
Iperf is a free software copyrighted by the University of Illinois [15]. It measures the
TCP and UDP bandwidth, packet error rate, jitter and delay of a wireless connection. It
is a sophisticated version of NETPERF [16]. In Iperf, there will only be a unidirectional
tra?c (like FTP) from the client to the server [15]. Hence, this reports the maximum
throughput (bandwidth) of the link. In this thesis, throughput is the maximum throughput
or bandwidth reported.
2.4 Measurements
This section gives a quick overview of how TCP throughput of a wireless LAN varies
with parameters like signal strength, RTS/CTS, etc., TCP packets are transmitted by N2
in all the experiments unless specifled.
2.4.1 Signal strength variation with distance
Figure 2.2 [obtained from [51]] shows the variation of signal strength with distance.
We can see that the signal strength decreases exponentially as the distance increases. This
is because the path loss is a logarithmic function of the distance.
2.4.2 Throughput variation with RSSI
Throughput is measured as the number of information bits transferred in unit time.
The throughput variation with RSSI values (in dBm) is shown in flgure 2.3 (Result obtained
from [51]). Throughput increases exponentially with signal strengths and reaches saturation
condition (max throughput) even at low signal strengths values. Throughput measured
using Iperf is averaged over a time of 60 seconds. Hence, we consider average maximum
throughput (or bandwidth) achieved and not the instantaneous value.
8
Figure 2.2: Signal strength variation with distance (Result from [51])
2.4.3 Efiect of RTS/CTS signals on throughput
The 802.11 distributed coordination function (DCF) employs an optional feature called
RTS/CTS, a hand shaking mechanism, to avoid collisions due to the hidden node problem.
This mechanism reserves the channel for packet transmissions as less bandwidth is wasted
in the event of collisions. In a collision free environment, the additional over head attached
because of the RTS/CTS decreases the overall throughput and can be seen in the flgure 2.4.
Hidden node problem is brie y explained here. Consider three stations A, B and C
wherein station A can hear station B and station B can hear station C but stations A and
C cannot hear each other. In such a case, CSMA/CA will not work (between A and C)
and when stations A and C simultaneously transmit a frame to B, the frames will collide
resulting in a packet loss [4].
9
Figure 2.3: Throughput variation with signal strength
2.4.4 Efiect of antenna orientation on signal strength
The orientation of the receiver(Rx)/transmitter(Tx) is important in measuring the
signal strengths and throughput. The wireless card of the transmitter (AP) iskeptfacing the
receiver (laptop). The orientation of the receiver is changed and RSSI values are measured.
Figure 2.5 illustrates the efiect.
The explanation of the legend in the flgure is as follows -
0 - Both the wireless cards (of Tx and Rx) are facing each other
90 - Wireless cards with a 90 degree angular variation between each other
180 - Wireless cards with a 180 degree angular variation between each other
270 - Wireless cards with a 270 degree angular variation.
In future chapters, we can see that the antenna orientation plays a major role in
determining the throughput especially under self interference.
10
Figure 2.4: Efiect of RTS and CTS on Throughput
2.4.5 Fragmentation threshold on throughput
802.11b wireless stations can use the optional feature of fragmentation to divide a large
data frame into smaller fragments, which are then sent independently to the destination.
Fragmentation threshold can be set for a wireless card, which then fragments all frames more
than the set value. If the fragmentation value is set to a lower value, it adds additional
overhead for each packet and hence the throughput decreases [11]. But setting it to a low
value can be very useful in environments where the losses are high. In our experimental
setup, the signal conditions were quite good. Hence we could only see a degradation in
throughput as shown in flgure 2.6
11
Figure 2.5: Change in RSSI values because of change in the orientation.
2.4.6 Impact of Transmission power of wireless card
The transmission power of a wireless card also plays a very crucial role in the perfor-
mance of a wireless LAN. Figure 2.7 shows the variation of throughput at difierent trans-
mission powers and at difierent signal strengths. We can see that at higher RSSI values the
transmission power has negligible efiect on throughput. However as RSSI value decreases
the efiect becomes more evident. The throughput almost reaches zero for lower power values
even at decent signal strengths. The values shown in the legend are the transmission power
levels of the wireless card while the values shown on the x-axis indicate the signal strength
of the AP perceived by the wireless station.
2.4.7 MAC retransmission values
Wireless medium is prone to high error rate when compared to the wired medium.
Hence, 802.11 MAC employs MAC level retransmissions to hide these errors from TCP
layer. This is because the TCP layer takes stringent action thinking that the loss is because
of congestion. The common MAC protocol employed is a ?Stop and Go? protocol, where
12
Figure 2.6: Efiect of Fragmentation threshold on Throughput
in the next MAC frame is sent only when it receives the acknowledgement (ACK) for the
previous frame [21]. Please note that this ACK is difierent from the TCP ACKs which work
at a difierent layer (transport layer). Figure 2.8 shows the performance of TCP throughout
at difierent MAC retransmission values.
2.5 Summary
In this chapter we have discussed the the impact of parameters like signal strength ,
distance fragmentation threshold, MAC retransmission values, RTS/CTS and transmission
power level on TCP throughput in detail.
13
Figure 2.7: Throughput variation with transmission power at difierent RSSI values
Figure 2.8: Impact of MAC retransmission values on TCP throughput
14
Chapter 3
Interference Study of 802.11b Networks
Wireless LANs are commonly deployed today in homes, o?ces and in public spaces
such as airports and shopping malls. Hence, many devices operate in the same frequency
band and in close proximity to the 802.11b networks such as Bluetooth, other 802.11b and
802.11g networks, microwave ovens, cordless phones etc., to name a few. In the present work
we study the performance degradation, in terms of throughput and packet error rate, of
802.11b networks in the presence of interference. Experiments are carried out in an anechoic
chamber so as to reduce the efiects of multi path, fading and other radio interferences.
An 802.11b network is setup in an anechoic chamber, so as to capture and understand
the true efiect of the interference introduced without the presence of other radio related
interference. The results show that there is a signiflcant degradation of performance if
two 802.11b networks work in the same or adjacent channels. Interesting phenomenon is
observed while analyzing the self interference. Performance in the presence of Bluetooth
technologies, both v1.1 and v1.2 (difierences between the versions are described later in the
chapter) was also investigated under various scenarios.
3.1 Introduction
802.11b WLANs operate in the 2.4 GHz ISM band. Many technologies such as Blue-
tooth, 802.11g and devices like microwave ovens and cordless phones work in the same
band. In many instances, these devices have to work in the presence of each other. In
such cases, there is bound to be interference and hence performance degradation in terms
15
of higher packet error rates, higher delays and lower throughput. Due to the large increase
in the number of wireless users, more 802.11b access points have to be deployed to cater to
the increasing demand. In this case, e?cient assignment of channels is di?cult and hence
802.11b networks may be operating in overlapping channels that are closely spaced.
Bluetooth devices also operate in the same frequency band as 802.11b networks. Blue-
tooth is considered to be a low power communication system and is identifled as IEEE
802.15, a standard for personal area networks (PANs). IEEE 802.15 WPANs are designed
for short range, low data rate communication (typically a few meters) unlike 802.11b net-
works that are designed to operate at a maximum range of about 300 meters. Bluetooth
(BT) devices often work in close proximity to WLAN networks. Performance of 802.11b
networks in the presence of above mentioned interference sources is experimentally mea-
sured.
3.2 Previous work
Self interference occurs when several 802.11b networks work in close proximity to each
other on interfering channels. Self Interference of 802.11b networks was not studied in detail
until now. There are a few papers that discuss the interference among 802.11b devices [25],
[26]. Since 802.11b access points have longer range (approximately 300 meters) and were
provided with 11 channels in the standard, two access points working together on the same
channel or on adjacent channels is considered to be a rare possibility. But, because of the
reasons mentioned in section 3.1, the co-existence of access points on interfering channels
is a common phenomenon. In [25] and [26], adjacent channel interference was studied, but
the emphasis was on frequency planning and analysis of channel assignment.
Performance analysis under interference between 802.11b and Bluetooth is not new;
many publications in the past have assessed the interference efiects in detail using either
simulations or experimental test beds considering various scenarios [27] - [33]. Mathematical
16
and analytical models have also been proposed to analyze the efiect [34]. But most of
the studies were conducted in indoor or outdoor environments where unwanted external
interference sources may exist which may in uence the measurements. Our experiments
were conducted in an anechoic chamber, so as to fllter out the efiect of unwanted interference
sources. The anechoic chamber experiments permitted the introduction of controlled and
intentional interference.
3.3 Anechoic chamber
Computer simulations do not really give us true picture of the scenarios considered.
Accurate models to address the behavior of various wireless network protocols under realistic
radio environments are very di?cult and complex. Approximate models, however, do not
accurately predict the network performance. So it is imperative to have a fully functional
test bed to capture and obtain the true performance of a wireless network [23].
A functional test bed too has its limitations in analyzing the wireless protocols accu-
rately. Repeatability and control over the unwanted parameters are the biggest challenges
of the test beds created. Experiments conducted in realistic environments are afiected by
interference from devices operating at the same frequency. Interference caused by re ections
and multi path also have considerable efiect on the accuracy of the results. Unpredictability
and time varying nature of the wireless channel often render experimental results to be not
reproducible [23].
So, we have conducted our experiments in the anechoic chamber, which minimizes the
in uence of external interference sources. Anechoic chambers enable us to perform exper-
iments in an interference-free and re ection-free environment. The apparent disadvantage
of using anechoic chamber is the physical space limitation. Hence power variability in the
devices becomes an important parameter to emulate real-life situations [23]. The results of
measurements taken in anechoic chamber can be treated as a benchmark for ideal behavior
17
of scenarios considered. Table 3.1 emphasizes the need for considering anechoic chambers.
In case I, the transmitter is located in an o?ce room and in case II, the transmitter is
located in an anechoic chamber. In both cases, a microwave oven is operated in an adjacent
room about 10 ft away from the transmitter. The receiver is about 20 ft from transmitter.
Throughput degradation is high in o?ce environments while it is negligible in an anechoic
chamber. This shows that the anechoic chamber successfully prevents external interference.
O?ce Environment Anechoic Chamber
(Case I) (Case II)
Throughput With Interference (Mbps) 2.63 4.9
Throughput Without Interference (Mbps) 5.06 5.06
Table 3.1: Efiect of anechoic chamber in mitigating unwanted interference
3.4 Self Interference with other 802.11b networks
3.4.1 IEEE 802.11b
IEEE 802.11b wireless local area networks are the most popular wireless systems that
have experienced rapid evolution and widespread deployment. They can use either a Fre-
quency Hopping Spread Spectrum (FHSS) or a Direct Sequence Spread Spectrum (DSSS)
technique, the latter being the most widely used. The 802.11b standard deflnes a total
of 14 frequency channels, of them; channels 1 through 11 are used in the United States.
Figure 3.1 shows the frequency channel assignment in 802.11b networks [22]. The 802.11b
signal usually occupies about 30 MHz of frequency. Since the frequency difierence between
two adjacent channels is 5MHz, a signal destined to a channel will also occupy (or overlaps
with) adjacent channels. Due to this, channel 1, 6 and 11, are the only non-overlapping
channels among the available 11 channels. WLANs are generally deployed and are operated
by assigning these channels to physically adjacent access points.
18
Figure 3.1: 802.11b Channel Assignment
802.11b standard specifles MAC layer, which co-ordinates the communication over
the wireless medium. This uses CSMA/CA (Carrier Sense Multiple Access with Collision
Avoidance). In CSMA/CA, collision happens when two or more stations send their packets
in the same slot. The performance of wireless links in the presence of self interference from
802.11b networks is analyzed in the next section.
3.4.2 802.11b self interference
Experimental Test Bed
The test bed architecture consists of three laptops and two access points-
N1, a Pentium M 1.6 GHz Windows XP machine acting as a server
N2, a Pentium M 1.6 GHz RedHat linux 9.0 laptop equipped with 802.11b interface
acting as a client
N3, a Pentium Windows 2000 laptop equipped with 802.11b interface acting as an
interfering node
AP1, a Dlink -2100 802.11b/g access point
AP2, a Dlink -514 router acting as an interfering router
19
Figure 3.2: Experimental Test bed
The dimensions of the anechoic chamber are about 36 ft x 18 ft. All laptops run Iperf,
a bandwidth estimation tool [15]. The power level of the DLink-2100 access point can be
varied from 15 dbm to 3 dbm. Signal to noise ratio (SNR) is further decreased by adjusting
the detachable antenna of the access point. The orientation of the antenna is also carefully
adjusted to obtain the desirable SNR. The power level of AP2 can also be varied from
17dbm to 10dbm. Both access points will be transmitting at their peak power levels unless
specifled.
Experimental scenario and results
Figure 3.2 shows the experimental test bed to measure the self-interference of WLAN
802.11b networks. The distance between AP1 and N2 is about 20ft. The channels of
both the access points are set to 11. The interfering access point, AP2 is set to transmit
at its maximum possible rate. Note that N2 is directly connected to AP2 and wirelessly
connected to AP1. N2 will be transmitting packets to N1 and N3 from both the available
interfaces (WLAN and Ethernet respectively). Main link (or measuring link) is AP1-N2
and interfering link is AP2-N3.
20
Figure 3.3: Throughput variation with and without interference
Theresultsobtainedinthiscaserepresenttheworstcaseofthethroughputdegradation,
as both the access points are set to same channel and the interfering access point is set to
transmit at its peak data rate (about 5 Mbps). Figure 3.3 shows the experimental results at
difierent SNR values of the main link. We can see that the throughput dropped from around
5 Mbps to around 1.47 Mbps even at good SNR (SNR of AP1 is higher than AP2). The
SNR from AP2 is about 45 db. The throughput starts to degrade once the SNR crosses
a threshold value and we would see a steep decrease in throughput. Throughput in the
graphs shown is the throughput of the main link.
3.4.3 Co-channel and adjacent channel interference
Co-channel interference occurs when two access points working close to each other
are on the same channel (e.g., Ch11- Ch11 on both access points) and adjacent channel
interference occurs when they are on interfering channels (e.g., Ch11-Ch10, Ch11-Ch9,
Ch11- Ch8).
21
Figure 3.4: Experimental test bed for evaluation of co-channel and adjacent channel inter-
ference
Experimental test bed
The experimental test bed to see the efiect of co-channel and adjacent channel inter-
ference on 802.11b network is shown in flgure 3.4. The distance between AP1 and AP2 is
20 ft while the distance between AP2 and N3 is 4 ft. Measurements are taken while moving
N2 from positions 1 through 5. Positions 1 through 5 are at a distance of 5 ft, 7 ft, 11 ft, 15
ft and 17 ft. respectively from AP1. AP2 will be on channel 11 throughout the experiment
while the channels of AP1 are changed from 11 to 1. The wireless card of N3 is associated
with AP2 and that of N2 is associated with AP1. Let SS1 and SS2 be the signal to noise
ratios of AP1 and AP2 observed by N2. No RTS/CTS was used throughout the experiment.
SNR of the interfering link is about 60 db.
Results
Figure 3.5 shows the throughput variation of the main link at difierent channels of
AP1, the other access point being at Channel 11. The difierence in the signal strengths
measured at each position is also listed in the flgure. From the results, Channels 6 through
22
Figure 3.5: Behavior of co-channel and adjacent channel interference at difierent distances
from main AP
1 were unafiected by interference while a decrease in throughput was observed on channels
11 through 7. Measured data shows that at positions 1, 2 and 3, throughput increases as the
channel distance increases which is quite expected except on channel 8. Individual channel
throughputs decrease as N2 moves closer to AP2. This is due to the fact that as the signal
level of AP2 starts dominating the wireless link, the number of packets lost due to collisions
increases resulting in the decrease in throughput. Figure 3.6 gives the location along with
the signal to noise ratios at difierent positions.
Anomaly observed
Careful observation of the results in flgure 3.5 leads to a very interesting phenomenon
when N2 is close to AP2. At position 4, throughput on channel 10 is observed to be less
23
Figure 3.6: Signal Strength difierence of access points and distance of N2 from AP1 at
difierent locations
than channel that on 11. A similar observation was seen on channel 9 at position 5 where
the throughput on channel 9 is far below than that on channel 11. We observed a decent
throughput of 1.33 Mbps at position 5 while the same on channels 10 and 9 are 0.310 Mbps
and 0.61 Mbps, respectively. In short, lower throughput was observed on adjacent channels
than on co-channel when node (N2) is close to the interfering access point. Percentage of
throughput degradation was also more on adjacent channels than on co-channel. This kind
of behavior is also true when AP1 is transmitting instead of N2.
Similar result was observed on overlapping channels of AP1, when AP2 was set to
channel 6, where in the main link on channels 5 and 7 at position 3 record lower throughputs
than on channel 6. Results are shown in flgure 3.7.
The throughput on overlapping channels, under interference, not only depends on the
signal strength (or distance) of the interfering access point but also on the signal level of
the link between the interfering access point (AP2) and the receiver (N3). As the SNR of
24
Figure 3.7: Co-channel and adjacent channel interference (when AP2 is on Channel 6)
the link between AP2 and N3 is decreased, an increase in throughput is observed on the
overlapping channels of main link (AP1 and N2).
A decrease in SNR on interfering link is achieved by either changing the distance
between AP2 and N3 or by decreasing the power level on AP2. In the present scenario,
since physical distance is a limitation due to the size of anechoic chamber, the transmitting
power of AP2 is set to 12.5 percent of the maximum available power. The distance between
AP2 and N3 was set at 8 ft. Figure 3.8 gives the comparison of the throughput change.
Let SNR1 and SNR2 be the signal strengths observed by N3 before and after the change in
transmission power. The measurements are taken at position 3.
We can see from flgure 3.8 that after the transmission power of interfering access point
AP2 is reduced, there is an increase in the throughput on all channels. A signiflcant increase
25
Figure 3.8: Throughput variation with the change in the distance (or SNR) between AP2
and N3
in throughput on channel 10 before and after the adjustment of power level (or SNR) re-
emphasizes the point that the link with better SNR captures the channel when the links
are on difierent channels. After the adjustment, the main link on Channel 10 has a better
SNR than the interfering link and hence there is a signiflcant increase in the measured
throughput. The interfering link sufiers in this case. When both links are close to each
other, one link capturing the channel is also observed in [25], but the paper does not describe
the adjacent channel link behavior.
Discussion
Ideally, we would like to have both the interfering access points on orthogonal channels
which is not always possible. The other alternative is to assign the access points on over-
lapping channels. But from flgures 3.5 and 3.7 it can be concluded that it is better to have
26
both the APs on the same channel rather on adjacent channels at lower SNR values. This is
because when N1 nears AP2, the throughput degradation is more on adjacent channels. The
threshold level of performance is drawn depending on the application in use. Depending on
the acceptable baseline for performance, the appropriate channel is chosen. The percentage
of throughput drop was also observed to be higher on adjacent channels when compared to
that on co-channel.
3.4.4 Impact of interfering access point (AP2) data rate
Test bed and results
The location of AP2 being so close to AP1 (about 20 ft away) is not a realistic situation.
AP2 would be a good 50-100 ft away in real-life situations. Since physical space is a
limitation in our case, the data rate at which AP2 is transmitting is varied to emulate larger
distances (or lower signal strengths). This is done considering the fact that at difierent signal
strengths, the data rate would be difierent and the access point wont be transmitting at
high data rate always. So, instead of varying the physical location of AP2, we changed
its data rate to emulate difierent distances between AP2 and N3. Figure 3.9 shows the
throughput variation of the main link at difierent transmitting data rates of interfering
AP2. The measurements are taken at position 1.
From flgure 3.9 we can see that the throughput level would become unacceptable when
the data rate of the interfering access point is more than 1Mbps.
3.5 Bluetooth interference
3.5.1 Bluetooth
Bluetooth devices use frequency hopping where signal is hopped over 79, 1 -MHz chan-
nels at 1,600 hops per second. So, at each channel the packet transmission time will be
27
Figure 3.9: Efiect of varying data rates of AP2 on the throughput of the main link
625 microseconds. The data rate is around 728 Kbps. Recent advances in Bluetooth tech-
nologies provide good range (up to 100m) for Bluetooth devices by increasing the power.
Bluetooth uses a master-slave concept for communication. A group of Bluetooth devices
communicating in master-slave fashion is called a piconet. The master chooses the fre-
quency hopping sequence of the piconet. They use difierent links - Synchronous Connection
Oriented (SCO) and Asynchronous Connection Less (ACL) for communication. SCO is
used for voice while ACL is used for data [27]. Since a channel in a WLAN DSSS system
occupies about 22 MHz of frequency, the probability that a Bluetooth device transmitting
simultaneously in the same channel is roughly 28 percent (22/79). Because of this inter-
ference, there would be performance degradation in both WLAN and Bluetooth systems.
The severity of degradation depends upon the power level of the devices operating and the
proximity of Bluetooth devices to 802.11b devices [27].
28
Figure 3.10: Experimental setup for Bluetooth interference on 802.11b
Bluetooth Version 1.1 and 1.2
The difierences between Bluetooth v1.1 and v1.2 technology are enhanced data rate,
addition of new proflles and services and better communication and co-existence with other
devices in the ISM band. One of the features of v1.2 that is most striking is Adaptive
frequency hopping (AFH). AFH improves the performance by identifying the busy channels
and excluding them. This is done by collecting the statistics on metrics such as packet
error-rate, bit-error rate or RSSI [35], [43].
3.5.2 Interference with v1.2 devices
Experimental test bed and Results
Our test bed consists of Anycom Bluetooth USB-240 adapters which are compliant
with Bluetooth v1.2 technology with a range of 330 ft. They are placed on devices N3 and
N1 and are made to communicate in an adhoc fashion as shown in flgure 3.10. The rest of
the set-up is same as in flgure 3.3.
29
DuetoAFH,wehardlyseeadifierenceinthethroughput withandwithoutinterference.
AFH technique usually takes some time to adapt and performance degradation was seen
only for the flrst few seconds (10-20 seconds) after the transmission starts. So, the Bluetooth
v1.2 efiectively prevents throughput degradation.
3.5.3 Interference with v1.1 devices
Experimental test bed and results
Test bed for one test scenario is same as that in flgure 3.10. N1 and N3 are equipped
with Dlink DBT-120 Bluetooth v1.1 USB adapters. The power level of Bluetooth USB
devices are 0 dbm and the range is about 33 ft peer to peer. Here, N3 is the transmitter
and N1 is the receiver. Let d be the distance between N1 and N3. N3 is moved from N1
towards N2. Figure 3.11 shows the variation in throughput as a function of distance d. As
d increases, the throughput decreases, because of increase in the distance between N3 and
N1 and also because of N3 moving towards a stronger transmitter N2. Signiflcant impact
was felt only when Bluetooth devices are close to each other (approx. 5 ft). SNR mentioned
in the flgure is the SNR of the wireless link. Please note that in the flgure, No Int. is the
throughput without BT interference.
3.5.4 Bluetooth interference with 802.11b and BT embedded in the same de-
vice
Test Bed
This experiment was conducted with both 802.11b network interface card and Blue-
tooth adapter collocated in the same device. Figure 3.12 shows the experimental test bed
of such a setup. The distance between N2 and N3 is 4 ft throughout the experiment. The
30
Figure 3.11: Throughput of WLAN vs distance between Bluetooth nodes
output power level of the wireless card is 15 db. Four scenarios are considered in this test
bed -
Case 1: N2 transmitting on both the 802.11b and Bluetooth link
Case 2: N2 transmitting on 802.11b link and receiving on Bluetooth link
Case 3: N2 receiving on both the links
Case 4: N2 receiving on 802.11b link and transmitting on Bluetooth link
This work is an extension of the one in [27] where emphasis was to study the efiect of
WLAN interference on Bluetooth. Table 3.2 summarizes the measurement details.
Results and Discussion
Measurements for all test cases are taken at three difierent signal strengths. In case 1,
at high SNR, the drop in throughput is not high as WLAN is transmitting. Since the output
power level of the wireless card is much higher than that of Bluetooth, WLAN dominates
the data transfer in cases 1 and 2. At lower SNR, since the Bluetooth is transmitting in case
31
Figure 3.12: Test bed when BT and 802.11b are embedded in the same device
1, it has lower throughput than case 2. In case 3, a decrease in the throughput is observed
compared to earlier cases as the WLAN card is in the receiving mode. The impact of the
presence of Bluetooth interference on WLAN is very strongly felt in cases 3 and 4 especially
at lower signal strengths.
SNR=60 db SNR=48db SNR=21db
Case 1 4.71 4.7 3.2
Case 2 4.7 4.7 4.65
Case 3 4.31 3.77 0.565
Case 4 4.13 3.4 0.677
Table 3.2: Bluetooth Interference on WLAN for various test cases
3.6 Bluetooth interference - Comparison of mathematical and experimental
results
Several papers [46], [47], [30] [48] and [60] discuss the mathematical analysis of the
Bluetooth interference on 802.11b networks. In this section, we compare the mathematical
32
Figure 3.13: Test Bed to compare throughput values with and without interference
results obtained in [47] to that of our test bed results and present the same in an intuitive
fashion.
3.6.1 Test bed and Results
Test bed for the present scenario is shown in flgure 3.13. Laptops with Bluetooth
devices are about 2 ft away from the main receiver, N2, and about 3 ft away from each
other. The positions 1, 2 and 3 are at distances of about 4m, 10m and 14m away from AP1.
FTP (flle transfer protocol) flle transfer takes place between N3 and N4 (Bluetooth nodes)
and hence the BT load factor is 100 percent.
Case 1 : N2 at position 1
Figure 3.14 shows the throughput change with and without the Bluetooth interference
at 4m away from the 802.11b transmitter. We can see that there is no throughput degra-
dation. This is due to the fact that signal to interference ratio (SIR) is greater than 10
db.
Case 2 : N2 at position 2 and 3
33
Figure 3.14: Theoretical and practical throughput comparison at 4m away from the trans-
mitter
From flgure 3.15, we can observe the decrease in the throughput because of the efiect
of interference at all packet sizes at positions 2 and 3, respectively. Understandably, the
throughput degradation in both the cases is not the same even if the SIR is less than 10 db.
This highlights the fact that even though involved in collision, not all Bluetooth packets
involved in collision are lost. But most of the papers on mathematical analysis, do not take
this aspect (variation of throughput with SIR) into account and assume that all packets
involved in collision are lost. Hence the predicted throughput will be the same for both the
distances. One interesting observation in flgure 3.15 is that the throughput for 750 byte
packet size is greater than 1500 byte packet size. Similar observation was also made in
[47]. This is because smaller 802.11b packets take less time to transmit and hence have low
probability to collide with Bluetooth packets. But, if the packet size decreases further, the
over head increases (even though the probability of collision decreases) and hence we see a
decrease in throughput.
34
Figure 3.15: Throughput comparison with and without interference at 10m and 14m away
from the transmitter
It would be more meaningful to compare the change in throughput because of inter-
ference than the absolute throughput. Figure 3.16 shows the comparison of percentage
decrease in throughput due to interference for the test bed results and the results obtained
in [47].
The reason for comparing the theoretical results to the practical values is to highlight
the fact that most of the models neglect the signal to interference ratio (or the ratio of
distance between main transmitter and receiver to the interfering transmitter and the main
receiver) factor in determining the packet error rate and throughput and assume that the
packet under collision is lost. Even in the model which consider this factor the through-
put values predicted by the model are not compared to the practically observed ones to
determine the error.
35
Figure 3.16: Percentage decrease in throughput comparison
3.7 Summary
In this chapter we have studied the interference efiects of WLAN and Bluetooth on
802.11b networks. Though the emphasis was on performance measurement under interfer-
ence, the results are very useful in interpreting how they can be used for proactive network
management. In the presence of self interference, where 802.11b networks have to operate
in proximity, the performance when the devices are on adjacent channel was worse than
that when they are on co-channel at lower SNR.
Since the power level of Bluetooth devices are 0 dbm we can only see interference if the
Bluetooth devices are close to each other. But the performance degradation is signiflcant
when the 802.11b device is in receiving mode instead of in transmitting mode, if they are
embedded in the same device. Simultaneous usage of both technologies should be avoided
for better performance. Safe distance must be maintained for e?cient resource usage.
36
Chapter 4
Throughput Prediction Model in the Presence of Self Interference for
Typical Office Environments
This chapter describes a mathematical model for predicting the throughput of the
802.11b wireless LAN in the presence of self interference. The throughput prediction model
is one of the key contributions of this thesis.
4.1 Introduction
Prediction models aid in estimating the performance, throughput in our case, or any
other parameter of interest. Knowing the performance before hand via modeling, can be
very useful for several reasons:
1) Reduces a lot of efiort in actually setting up a test bed and measuring and
2) Helps in proper network planning and performance management
Modeling the behavior of throughput for difierent scenarios with an appropriate set of
mathematical equations forms the basis of our prediction model. The present model can be
viewed as a combination of empirical and analytical parts since it makes use of experimental
measurements in the process of building a model.
4.2 Prior Work
Mathematical and analytical models have been proposed for the performance of 802.11b
in the presence of Bluetooth interference. Many publications [53] - [60] have done an exten-
sive analyses of the performance under such conditions. Few papers have actually compared
37
the results (throughput or PER) obtained by the models to the measured ones. But very
few papers have concentrated on the performance variation under self interference. We
will discuss here a few empirical/analytical throughput prediction models that are already
developed.
In [50], the performance of public wireless LANs in the presence of multiple users
is experimentally evaluated through extensive measurements. A throughput prediction
model is built based on the measurements. Single user and multi-user environments are
considered. An AP is placed inside the building and measurements are taken outside at
difierent signal strength locations - 11 for single user measurements and 3 for multi-user
measurements at each of the 3 restaurant sites. The piece-wise linear model and exponential
model are validated for throughput. The results show that the predicted throughput for a
new environment is within the confldence intervals of the measured one. The authors also
developed an empirical model which estimates the throughput with ?N? number of stations
in a WLAN.
In [49], a measurement based approach is used to predict the throughput of rate
adaptive 802.11a WLANs. Throughput and packet error rate (PER) variation with sig-
nal strengths and distance, the dependence of average RSSI with physical data rate rate
is measured for difierent environments (Indoor line of sight (LOS), outdoor line of sight
and indoor non line of sight (NLOS)). The physical layer simulation results and MAC layer
analyses along with the measurements are used to predict the throughput. Reasonable as-
sumptions are made depending upon the observations from measurements - 1) wireless card
is adapting data rates so as to keep the PER constant and 2) The average RSSI is strongly
related to the average physical data rate value. Results show that the predicted throughput
closely matches the measured values. The measurements exhibited strong correlation with
the environment being used.
38
While [50] is entirely empirical, [49] is a combination of simulation and analytical
modeling. But the common key aspect of both research approaches is the conversion of
signal to noise ratio into either PER or throughput. Piece-wise or exponential model is
used in [50] where as simulations are used in [49]. The discussion here is to emphasize the
fact that it is necessary to either use an empirical mathematical model or simulations to
map the SNR to PER or throughput.
In his highly acclaimed work [8], Bianchi proposed an analytical model which predicts
the throughput of 802.11b DCF (distributed coordination function) networks. The perfor-
mance of networks largely depends on parameters like number of stations and contention
window size.
From [8], the probability that a transmission occurring in the channel is successful is
the probability that exactly one station transmits given that at least one station transmits.
This is given by,
Ps = n?(1??)
n?1
1?(1??)n (4.1)
where ? is the probability that a station transmits in that slot time. Therefore, the
probability that the transmitted packet is not successful i.e., it encountered a collision is,
Pc = 1?Ps (4.2)
Using Markov chain model, the probability with which a station transmits in a slot
time, ?,is calculated. To achieve maximum throughput, the optimum ? value is given by
the following equation.
? ? 1npT?
c =2
(4.3)
39
Where T?c = Tc/slottime. Tc is the period of time sensed busy by other non colliding
stations and n is the number of stations. The above shows that ? is inversely proportional
to n. But, Bianchi?s model does not relate throughput and PER variation with the change
in SIR.
The models discussed in [50], [49]and [8] do not predict the throughput when the
interfering stations are on other channels. Such a model is important because, in enterprise
environments, the stations which act as interference sources will be on difierent channels.
The complexity of predicting the throughput, when the stations are on difierent channels
and at difierent distances from each other, is so high that some reasonable assumptions are
to be made for multi-user, multi-channel environments.
4.3 Throughput prediction model - Inputs
The inputs to the throughput prediction model are
1) the distance between the transmitter and receiver of the measuring/main link i.e.,
measuring link distance (say dm)
2) distance between interfering transmitter (either an AP or a station) and main re-
ceiver (say d1)
3) distance between interfering transmitter and receiver (say d2)
4) channel distance between the main WLAN and interfering WLAN (channel distance
is deflned as the difierence between the operating channels of the two WLANs)
We assume that the interfering stations are at the same distance from the main receiver.
We also assume that every station had a packet to send at all times.
The model is based on DCF (distributed coordination function) access method of the
802.11b MAC. We flrst present the model which estimates the throughput when the in-
terferer is on the same channel and extend the model further to ?n? stations all being on
40
difierent channels. Main/measuring WLAN here mean the wireless transmitter- receiver
(or client-server) pair for which the throughput is being measured.
Since indoor environment is considered for the model, SNR is calculated from the
distance ?dm? according to the path loss equations given in [54]
PL = 32:45 + 20log(f:dm)::::: if dm < 8m
= 58:3 + 33log(dm=8):::::: if dm ? 8m
where f is the frequency in GHz (2.4 GHz in our case) and dm is the distance between
the main transmitter and receiver pair.
A brief derivation of the above path loss model is given below.
The basic log-distance path loss model for distance dm is given by the equation,
PL(dm) = PL(do) + 10nlog(dm=do)
where, do is the reference distance and n is the path loss exponent
The theoretical free space path loss, PL(do) is given by
PL(do) = 20log(4?do=?)
substituting ? = c=f in the above equation we get
PL(dm) = 20log(4?do=?)+20log(f)+20log(dm=do)
Line of sight is assumed for the flrst 8 meters and non-line of sight for distances greater
than 8 meters. Hence, the path loss exponent in the above equation is chosen accordingly
[54],
n = 2 and do = 1m for dm < 8m
n = 3:3 and do = 8m for dm ? 8m
Substituting the above parameters and by rearranging we obtain the path loss equation
deflned.
The received power, PR, in dbmW is calculated as [55] -
PR = PT ?PL where PT is the transmitted power
41
SNR is obtained according to the equation
SNR = PR ?SR (4.4)
where, SR is the receiver sensitivity in dbmW. Receiver sensitivity (a threshold value)
indicates how low a signal can get before it cannot be detected. Receiver sensitivity is a
very important feature of the receiver and it varies with the data rate employed. Typical
receiver sensitivity values for a Linksys WPC11 wireless card is given in table below [56].
This model of the wireless card is used as the main transmitter for our measurements.
Data rate (Mbps) Sensitivity dBm
11 -82
5.5 -85
2 -89
1 -91
Table 4.1: Receiver sensitivity values of Linksys WPC 11 wireless card
In order to determine the throughput, a model which maps the SNR to throughput is
required. An empirical exponential throughput prediction model, proposed in [52], is used
and is expressed as,
T = Tmax(1?e?Ae?(SNR?SNRo)) (4.5)
Where, Tmax is the maximum throughput that can be achieved using a wireless card.
SNRo is the signal where the throughput becomes zero. Ae is the rate at which the
throughput decreases with respect to the change in SNR. Tmax, SNRo and Ae are constants.
The constants are calculated empirically for Iperf tool in [50] by taking extensive mea-
surements at various data points and using a MATLAB curve fltting algorithms (for e.g.,
nlinflt and polyflt). The constants are reported as,
42
For Orinoco card,
Tmax = 5 Mbps, Ae= 0.069 and SNRo= 6.9
For Cisco card,
Tmax = 5.3 Mbps, Ae= 0.07 and SNRo= 5.4
Equation 4.5 is a single-user (and no-interference) throughput prediction model which
estimates the throughput from SNR. Please note that we also haven?t considered the inter-
ference efiect till this point. Ae and SNRo for both cisco and orinoco cards have almost
same values and hence we can assume them to be the values mentioned above for most of
the wireless cards.
4.4 Efiect of interference
Let us flrst consider the scenario when both the main and interfering stations are on the
same channel. We would later extend this when the interfering stations are on overlapping
channels. Please refer to flgure 3.1 for channel allocation in 802.11b standard.
In the presence of interference, when both the main and interfering stations are on
the same channel, (time) sharing of the channel takes place. So if there are two stations
competing for the channel, the throughput achieved is approximately half of the single user
throughput. This is with an assumption that all transmitting stations are at approximately
the same signal strength from the associated receiver(when dm ? d1). Similarly, when there
are ?n? number of stations involved, the throughput gets divided among the ?n? stations.
Though 802.11 stations have CSMA/CA mechanism, when two stations sense the chan-
nel at the same time, they transmit the packets in the same slot time. This would result
in collisions and hence packet loss. Assuming perfect channel sensing by the stations, the
probability that the transmitted packet is not successfully received (due to collision), Pc,
when ?n? stations are competing for the channel is given by equation 4.2
43
Pc = 1? n?(1??)
n?1
1?(1??)n (4.6)
where ? is deflned by the equation 4.3
When the interfering stations are close to each other then the packet error rate is
same as Pc. Pc can also be interpreted as maximum packet error possible. In other words,
when all the packets failed to reach the receiver due to collisions, the PER ? Pc. The
PER is calculated using UDP transmission since the Iperf tool reports PER only for UDP
transmission. Hence, the reported error rate might be slightly more than the actual PER
which is neglected.
4.4.1 Calculation of Practical ?
Earlier, it is assumed that all stations have perfect channel sensing. But, in practice it
is proved that perfect channel sensing is not possible. Because of this, ? deflned should be
modifled for practical purposes.
For this, we flrst measure the maximum PER in the presence of ?n? stations where ?n?
is 2 , 3 and 4. Now, we calculate the empirical ? values from the PER using equation 4.6.
The theoretical values of ? are obtained using equation 4.3. The theoretical and empirical
values of ? and their variation with the number of stations is shown in flgure 4.1. The flgure
in turn re ects the signiflcant difierence between the theoretical and practical PER.
From the empirical values, we derive a new equation for ? expressed as
?prac = (k?n+c)?? (4.7)
where k and c are constants and n is the number of stations. The constants show the
variation ?prac/? with n. A linear curve fltting algorithm called ?polyflt? is used to calculate
44
Figure 4.1: Variation of ? with n
the constants in MATLAB [58]. k and c values are calculated to be 0.6111 and 1.5377
respectively. From the above equation for ?prac we can calculate PER values for any ?n?.
4.4.2 Packet Error Rate
It was observed from our experiments, that the PER values are high and almost con-
stant for the flrst few meters. As d1 increases further, the PER almost varies linearly with
d1 according to the flgure 4.5(explanation of such variation is explained in section 4.4.4.
Therefore,
PER = Pc:::when the interfering transmitter is close to main Rx
(when d1 < 3m)
PER = Pc(a?d1+b+1)::::when the interfering station is away from the Rx
(when d1 ? 3m) (4.8)
where, d1 is the distance between the the interfering transmitter and main receiver and
?a? and ?b? are constants determining the variation of PER with d1.
45
Figure 4.2: Experimental setup for studying the variation of PER and throughput with
distances dm, d1 and d2
Throughput varies exponentially with SNR or SIR [50], [57]. It was shown in [48] that
the throughput-distance relationship (variation of throughput with d) in a typical o?ce
building follows a linear relationship. Hence, from flrst order approximation we can say
that the throughput varies linearly with d1. The relationship between throughput and
PER is also linear [49]. Hence, the relation between PER and d1 follows a linear rule.
Calculation of ?a? and ?b? values is explained in the next section.
The test bed for the mathematical model is shown in flgure 4.2. One of the inputs (dm,
d1 and d2) is changed keeping the other two constant and its efiect on PER and throughput
is studied. The aim is to analyze the variation of the inputs individually on PER and
throughput and flt in an appropriate mathematical equation to capture the variation.
4.4.3 Variation of throughput and PER with distance ?dm?
In this section, dm is changed keeping the other distances constant. d1 and d2 ? 5ft.
46
Figure 4.3: PER variation with dm
PER and throughput measurements are taken at each data point and are shown in
flgures 4.3 and 4.4 respectively. Since physical space is a limitation in our experiments,
RSSI values are considered instead of dm. From flgure 4.3 we can say that as dm increases
(as RSSI decreases), a constant packet error rate difierence is observed with and without
interference. This shows that the PER, because of the interfering transmitter on the main
wireless link at all dm values is constant. So in other words, the throughput of the main
wireless link is determined primarily by the distance ?dm?. Hence the throughput would
follow equation 4.5. Since two stations are contending for the channel, the channel is shared.
Therefore, the throughput equation will be,
Tdm = [T2 ?(1?PER)] (4.9)
where T is deflned by equation 4.5 and PER is deflned by equation 4.6
47
Figure 4.4: Throughput variation with dm
4.4.4 Variation of throughput and PER with distance d1
The test bed to see the efiect of d1 on the throughput and PER of the main link
is shown in flgure 4.2. Note that the measurements are taken considering one antenna
orientation for the entire experiment.
At each of the flve data points the PER and throughput are measured for ?1? interferer.
Similar measurements are also taken for ?2? and ?3? interferers. The results of which are
shown in flgure 4.5 and 4.6. The measurements are taken multiple times and the value
indicated in the graph is the average value. even though the variation is not very linear we
assume a linear relationship in order to flt an appropriate equation.
The measured PER values and the distances d1 are run through a MATLAB function
called polyflt which flts the data to a linear function and gives the slope ?a? and intercept
?b? of the curve [50]. Parameters ?a? and ?b? signify the change in the PER with d1. The
48
Figure 4.5: PER variation with d1
values of ?a? and ?b? for one interferer are 0.0158 and 0.0585, respectively. Similarly ?a? and
?b? parameters are calculated for n = 3 and n =4 (for 2 and 3 interferers, respectively).
From the measurements, we have observed that the percentage of throughput increase
, as d1 increases, is greater than the percentage of PER decrease. To accommodate this
we incorporate two other parameters fi and fl. The parameters fi and fl are calculated in
similar fashion using polyflt function. These parameters are useful to closely match the
theoretical values to the experimental values. These parameter values also change with d1.
While?a? and?b?showsthechangein PER,fi andfl showsfurtherchangeinthroughput.
The values of fi and fl for one interferer are 0.0145 and 0.0098 respectively.
All the parameters (a, b, fi and fl) are made as a function of the number of interferers
(n-1) and are shown in table 4.2.
The net values of the parameters are obtained by taking the average value of each of
the parameter.
49
Figure 4.6: Throughput variation with d1
Hence, the throughput in the presence of one interferer (after incorporating the efiect
of d1 and d2) now would be,
Tdm;d1 = [T2 ?[fi?d1+fl +1]?(1?PER)] (4.10)
where PER is deflned by equation 4.8
4.4.5 Variation of throughput and PER with distance ?d2?
The distance between the interfering transmitter and interfering receiver is changed and
its impact on the performance of the main link is analyzed in this section. It is extremely
di?cult to concretely conclude anything from such an analysis because of the complications
involved in the test setup. But it can be inferred that as the interfering wireless link
50
n a b fi fl
2 -0.0634 0.1135 0.0145 0.0098
3 -0.0362 0.0836 0.0199 -0.1144
4 -0.0251 0.0580 0.0168 -0.1217
Table 4.2: Parameter values for difierent n
throughput decreases, because of the low SNR, the available throughput is being used by the
main wireless link. Please note that the vice versa was also true i.e., when the main receiver
moves away from main transmitter the lost throughput is used by the interfering link. The
throughput of the interfering WLAN, without interference, will also follow equation 4.5.
Hence, the throughput in the presence of one interferer after incorporating the efiect
of d2 would be,
Tdm;d1;d2 = [T2 ?[fi?d1+fl +1]?(1?PER)+ Tmaxi ?Ti2 ] (4.11)
4.4.6 Variation of transmission rate with number of stations
The individual transmission rate of each of the transmitter will decrease as the number
of stations increases due to channel sharing. But the combined transmission rate of all the
stations will increase with ?n? and is shown in Table 4.3. Similar observation in throughput
is noticed in [13]. Hence the variation in the transmission rate because of the change in n
has to be accounted.
It was also observed that when compared to other cards this particular card (Linksys
WPC 11) with higher transmission rate gives more packet error rate and hence the transmis-
sion rate and PER are card - dependent. We have taken measurements, built and verifled
the model based on the measurements taken on this card. It was observed that some
of the wireless cards have better transmission rate than others because of manufacturing
variations.
51
number of stations Individual Transmission rate
of Linksys WPC 11 wireless card(Mbps)
1 6.03
2 3.4
3 2.6
4 2.2
5 1.9
Table 4.3: Individual transmission rate variation with n
Please note that to measure the PER and transmission rate we use UDP packets, with
MAC retransmissions equal to zero.
The throughput equation now would be as follows -
T(dm;d1;d2;txrate) = [T2 ?[fi?d1+fl +1]?(1?PER)?
(change in transmission rate because of n)+
Tmaxi ?Ti
2 (4.12)
4.5 Throughput prediction model in the presence of one interferer (two sta-
tions)
The factors afiecting the throughput of the main link have been analyzed in the previous
section. The net throughput in the presence of one interferer is expressed as
Tinterference(2) = [T2 ?[fi?d1+fl +1]?(1?PER)?
(txrate2txrate
1
)]+ Tmaxi ?Ti2 (4.13)
T and PER are deflned by the equations 4.5 and 4.8 respectively. txrate2 and txrate1
are the transmission rates when there are ?n? stations (2 in this case) and ?1? station,
52
respectively. The ratio signifles the change in the transmission rate and is necessary because
we are calculating the throughput of the main link with just one station.
fi, fl, a and b parameters (introduced earlier) are estimated as explained in the previous
section using a curve fltting algorithm called polyflt. The values of the parameters change
as the number of interferers change.
The last fraction Tmaxi?Ti2 becomes signiflcant when d2 is large. When d2 is small (as
is the case in most of the experiments conducted to verify the model) this factor can be
neglected.
Note that the number of interferers will be denoted as n?
where,
n0 = n?1
n is the number of stations contending for the channel.
4.6 Throughput prediction model in the presence of ?n? stations
When there are ?n? stations on the same channel, equation 4.13 is modifled as shown
below. We assume here that all interfering stations are at about the same distance from
the main WLAN receiver.
Tinterference(n) = [Tn[(fid1+fl)n0 +1]?(1?PER((ad1+b)n0 +1))?
(txratentxrate
1
)]+
P
i:1ton0(Tmaxi ?Ti)
n (4.14)
Where
Ti is the throughput of the interfering wireless station (without interference) deflned
by the equation 4.5. Tmaxi is the maximum throughout achieved by that station.
53
The a , b , fi and fl values for the above equation can be obtained by taking the average
of the values mentioned in table 4.2.
The fraction
P(T
maxi?Ti)
n , as explained earlier accomodates the change in the through-
put of the main wireless link when the interfering receiver is away from the interfering
transmitter. But the equation is not entirely correct as the throughout division again de-
pends upon the signal strength difierence perceived by each of the wireless station. So, the
signal strength weightage should also be included in the fraction. But this is extremely
di?cult to model. So, for this factor was not extended. The present model developed takes
most of the factors that efiect the throughput under interference into consideration.
We have experimentally measured the throughput and PER for 1 , 2 and 3 interferers
and build the model. We have to check if this model works for 4 interferers. Figure 4.7
shows the comparison of experimental values to the predicted values. The percentage error
is deflned by the following equation.
Error percentage = (Measured throughput?Predicted throughput)Measured throughput ?100 (4.15)
From the flgure, we can say that the model predicts the throughput fairly accurately
even in the presence of n stations.
4.7 Throughput prediction model for ?n? stations on difierent channels
Until now we have considered the scenario where in all the stations are on the same
channel. But in practice, this assumption may not be realistic. We have difierent stations
operating on difierent channels.
The hypothesis for the model is that the throughput can be studied by piece wise
analyses of the spectrum. We verify our hypothesis from the measured results.
54
Figure 4.7: Throughput comparison when the number of interferers are 4
Also, to model such scenario the following assumptions are made
1) the frequency spectrum can be partitioned into non-overlapping sections.
2) the PER occurring in one part of the spectrum is independent of the PER occurring
on the other part of the spectrum, if the spectrum is seen individually.
For better explanation of model behavior, we take an example where we have one
station each on channel 11, 10 and 9. Let us say that we want to flnd the throughput of the
station on channel 11. So the stations on channel 10 and 9 act as interferers. The spectrum
division in such a case is shown in flgure 4.8
Each channel is 22 MHz wide and the difierence between the center frequencies of
adjacent channels is 5 MHz [2].
We divide the spectrum into 3 pieces for analysis as shown. The flrst 5 MHz is used by
only one station and hence experiences no interference. The next 5MHz is used by stations
on channel 11 and channel 10. So two stations share this part of the channel frequency.
The next 12 MHz is used by stations on channel 11, 10 and 9. Hence three stations share
this part of the frequency on channel 11. The throughput of the station on channel 11 is
therefore deflned by the following equation.
55
Figure 4.8: Piecewise Spectrum Analysis
Ttotalinterference = 522 ?T + 522 ?Tinterference(2) + 1222 ?Tinterference(3) (4.16)
where,
T is deflned by equation 4.5
Tinterference(2) and Tinterference(3) are deflned by equation 4.16 when n = 2 and n = 3,
respectively.
The flrst factor( 522 ? T) takes care of 5/22 th part of the channel frequency which is
occupied by just one station with no interference. The second factor ( 522 ?Tinterference(2))
takes care of 2 stations sharing the next 5/22 part of the channel. The third factor (1222 ?
Tinterference(3)) takes care of the remaining channel spectrum occupied by all the three
stations.
56
Figure 4.9: Throughput comparison for one interferer on all the overlapping channels at d1
= 10m
4.8 Experimental veriflcation of throughput prediction model
Figure 4.9 shows the throughput comparison on all over lapping channels for one in-
terferer for d1 = 10 m. The results show that the predicted throughput closely matches
with experimental throughput. Therefore we can state that the throughput can be studied
by piece-wise analyses of spectrum. The error percent is calculated and indicated for each
measurement in the flgure.
Figure 4.10 shows the same for d1 = 6 m. We can also observe the anomaly we had
discussed in section 3.4.3 in the flgure when channel distance is 1.
Figure 4.11 shows the throughput comparison between the measured results and the
model at difierent dm values. The RSSI values indicated in the flgure represent difierent
dm values. Figure 4.12 shows the throughput comparison when 2 interfering stations are
on the same channels other than the main station. Both the interfering stations are on the
same channel indicated on the graph while the main station is on channel 11.
57
Figure 4.10: Throughput comparison for one interferer on all the overlapping channels at
d1 = 6 m
Figure 4.13 shows the throughput comparison for 2 interfering stations which are on
difierent channels other than the main station. The main station, as in the previous case,
is on channel 11.
We can see that as the number of stations increases the accuracy of the throughput
prediction model decreases marginally.
4.9 Summary
A throughput prediction model combining empirical and analytical modeling is devel-
oped. The model is validated through experimentation.
The experimental results indicate that the throughput under interference is very much
card dependent as no two cards give the same throughput values for a given setup. So it is
extremely di?cult to generalize the model for all card types.
58
Figure 4.11: Model validation for difierent d values
Figure 4.12: Model validation when two interferers are on the overlapping channels and the
main station is on channel 11
59
Figure 4.13: Model validation when interfering stations are on difierent channels and the
main station is on channel 11
60
Chapter 5
Conclusion
A detailed study of the interference efiects of 802.11b WLAN and Bluetooth on 802.11b
networks are performed. The beneflts of performing the experiments in the anechoic cham-
ber is discussed and demonstrated experimentally. The performance of 802.11b networks
in the presence of co-channel and adjacent channel interference is studied in detail. In the
presence of self interference, the throughput of an 802.11b network is greater when the other
802.11b station (acting as interferer) is on the same channel than on adjacent channel when
the devices are close to each other. Performance evaluation in the presence of both versions
(with and without AFH) of Bluetooth standards is conducted. An extension to the already
existing work in [27] is done. The theoretical results obtained from [47] are compared to
our experimental results.
Since performance analysis is an important ingredient of network management, such an
analysis of the wireless 802.11b networks under interference helps in drawing a base line for
acceptable performance. This helps in proactive performance management which enables
us to take necessary corrective measures for the imminent problem before hand and extract
maximum performance in cases where interference is inevitable.
A mathematical model which estimates the performance of 802.11b network under self
interference is developed. The factors which are neglected in the previous models have
been taken into consideration. The model predicts the results fairly accurately within the
error tolerance of less than 20 percent in most cases. We have observed that it is extremely
di?cult to model the behavior and predict the results under interference as difierent wireless
61
cards gave difierent results. We have also learnt that the orientation of the wireless cards
play an important role in the performance of a network under interference. So we have used
only one particular antenna orientation throughout our experiments. Piece wise analysis of
spectrum was employed when the interfering stations are on overlapping channels.
Knowing the throughput before hand using such models or tools will be extremely
useful in network planning where the throughput prediction need not be very accurate. It
saves a lot of efiort in such cases in actually setting up a test bed and taking measurements
considering various possible test scenarios into account.
5.1 Future Work
The model that we have developed has neglected certain factors like difierent antenna
orientations which have to be considered. We have also assumed that the interferers are
at about the same distance from the receiver. This may not be a realistic assumption.
802.11b networks are fast becoming obsolete with the emergence of 802.11g networks. Since
802.11g also operate in the same frequency band, our model can be used as a frame work for
developing network planning tools for 802.11g networks. Such analyses can also be extended
to WiMAX networks even though it is di?cult because of their high range (around 2-3
miles).
62
Bibliography
[1] IEEE Standard 802.11, Part 11 Wireless LAN Medium Access Control (MAC) and
Physical Layer (PHY) Speciflcations, 1999.
[2] IEEE Standard 802.11b-1999, Part 11 Wireless LAN Medium Access Control (MAC)
and Physical Layer (PHY) Speciflcations: Higher-speed PhysicalLayer in the 2.4 GHz
Band, 1999.
[3] R. Paruchuri and P. Agrawal, \Interference Study of 802.11b Networks for Proactive
Network Management", to appear in 10th IEEE/IFIP Network Operations and Man-
agement Symposium (NOMS), April 3-7, 2006.
[4] A.S. Tanenbaum , Computer networks, Prentice-Hall, 1996.
[5] T. S. Rappaport , Wireless Communications, Prentice-Hall, 1996.
[6] L. Peterson and B. Davie, Computer Networks, Morgan Kaufmann, 1996
[7] T. Sheldon, Encyclopedia of Networking and Communications, McGraw Hill, 2001.
[8] G. Bianchi , \Performance Analysis of the IEEE 802.11 Distributed Coordination Func-
tion, IEEE Journal on Selected Areas in Communications", vol. 18, no. 3, pp. 535-547,
March 2000.
[9] D. Dobkin. The correlation of data throughput with link loss for commercial WLAN
devices. High Frequency Electronics Magazine, January 2003.
[10] G.Durgin, T.Rappaport, andH.Xu,\Measurementsandmodelsforradiopathlossand
penetration loss in and around homes and trees at 5.85 GHz", IEEE Trans. Commun.,
vol. 46, pp. 1484 - 1496, Nov. 1998.
[11] Y. Zahur, M. Doctor, S. Davari, and Yang, T. A. \802.11b Performance Evaluation".
M.H. Hamza (Ed.): Proc. of the Int. Conf. on Communications, Internet, and Infor-
mation Technology (CIIT)., Scottsdale, Arizona. Nov. 17-19, 2003
[12] A. Kamerman, G. Aben, \Net throughput with IEEE 802.11 Wireless LANs," 2000
IEEE Wireless Communications and Networking Conference, vol. 2, pp. 747-752.
[13] T. Demir, C. Komar, C. Ersoy, \Measured Performance of an IEEE 802.11 Wireless
LAN", Proceedings of the Fifteenth International Symposium on Computer and Infor-
mation Sciences, pp. 246-254, Istanbul/Turkey, October 2000.
63
[14] A. Doefexi, S. Armour, B.-S. Lee, A. Nix, and D. Bull, \An evaluation of the per-
formance of IEEE 802.11a and 802.11g wireless local area networks in a corporate
o?ce environment," International Conference on Communications (ICC), Vol.2, pp.
1196-1200, 2003.
[15] IPERF, Link to product page. http://dast.nlanr.net/Projects/Iperf/
[16] NETPERF, Link to product page. http://www.netperf.org/netperf/NetperfPage.html
[17] J. Bardell. Converting signal strength percentage to dBm values, Technology white
paper,Wildpackets Inc., November 2002.
[18] B. Bing. Measured Performance of the IEEE 802.11 Wireless LAN. Proc. IEEE Con-
ference on Local Computer Networks, LCN?99, Lowell/Boston, Massachusetts, USA,
October 1999.
[19] N. P. K.Benekos, G.Kalivas, G.papadopoulus, A.Tzes, \TCP Performance Mea-
surements in IEEE 802.11b-based Wireless LANS," presented at IEEE MELECON,
Dubrovnik, Croatia, 2004.
[20] Atheros. Atheros press release. http://www.atheros.com/news/Performance.html.
[21] MAC tutorial Link to the page. http://www.hpl.hp.com/personal/
[22] A. Babbar, \WLAN and 802.11 architecture",Technical Report, July 2005.
[23] N. H. Vaidya et al., \Illinois Wireless Wind Tunnel: A Testbed for Experimental
Evaluation of Wireless Networks", Technical Report, UIUC, 2004.
[24] D. Raychaudhury, \Orbit - wireless network testbed," July 2004. WINLAB, Rutgers
University. http://www.winlab.rutgers.edu/pub/docs/focus/ORBIT.html
[25] D. Leskaroski and W.B. Mikhael, \Frequency Planning and Adjacent Channel Interfer-
ence in a DSSS Wireless Local area Network (WLAN)", Journal on Wireless Personal
Communcations Bluetooth and Other Technologies, pp. 169-180, 2001.
[26] J. Park, S. Park, P. Cho, and K. Cho,\ Analysis of spectrum channel assignment for
IEEE 802.11b wireless lan", The 5th International Symposium on Wireless Personal
Multimedia Communications, pp. 1073-1077, October 2002.
[27] D. Famolari and P. Agrawal, \Architecture and Performance of An Embedded IP
Bluetooth Personal Area Network", Proceedings of ICPWC, 2000, pp. 75-79.
[28] J. Del Prado and S. Choi, \Experimental Study on Co-existence of 802.11b with Alien
Devices," in Proc. IEEE Vehicular Technology Conference VTC01-Fall, Atlantic City,
October 2001.
[29] M. Fainberg and D. Goodman, \Analysis of the Interference Between IEEE 802.11b and
Bluetooth Systems", in Vehicular Technology Conference, October 2001, pp. 967-971.
64
[30] Jung-Hyuck Jo; N. Jayant, \Performance evaluation of multiple IEEE 802.11b WLAN
stations in the presence of Bluetooth radio interference", IEEE International Confer-
ence on Communications, ICC 2003, pp.1163 - 1168 vol.2
[31] R.J.Punnoseetal., \ExperimentalresultsforinterferencebetweenBluetoothandIEEE
802.11b DSSS systems, in Vehicular Technology Conference, Fall 2001, Pg: 67-71.
[32] J. Park, D. Kim, C. Kang, Daesik Hong, \Efiect of Bluetooth Interference on OFDM-
based WLAN, VTC 2003 Fall, Florida, USA, 6-9 Oct. 2003.
[33] A. Stranne, F. Florn, O. Edfors and B.-A. Molin, \Throughput of IEEE 802.11 FHSS
Networks in the Presence of Strongly Interfering Bluetooth Networks", in Proceedings
of the 13th IEEE International Symposium on Personal, Indoor and Mobile Radio
Communications, PIMRC 2002, volume 1, pp. 161-165, 2002.
[34] C.F. Chiasserini and R.R. Rao, \Performance of IEEE 802.11 WLANs in a Blue-
tooth Environment", IEEE Wireless Communications and Networking Conference 2000
(WCNC 2000), Chicago, IL, September 2000.
[35] Bluetooth Core Speciflcation V1.2 http:// www.bluetooth.org/spec, Nov 2003.
[36] O. Karjalainen, S. Rantala and M. Kivikoski, \ The Performance of Bluetooth System
in the Presence of WLAN Interference in an O?ce Environment", IEEE International
Conference on Communication Systems, ICCS 2002, 25-26 November, 2002.
[37] I. Vilovic and B.Zovko-Cihlar, \Performance analysis of Wireless Network Using Blue-
tooth and IEEE 802.11 Devices", 46th International symposium on Electronics in Ma-
rine, ELMAR-2004, 16-18 June 2004,Croatia.
[38] Jin-A Park et al., \Experiments on radio interference between wireless LAN and other
radio devices on a 2.4 GHz ISM band", The 8th International Conference on Commu-
nication Systems, 25-28 Nov. 2002 , pp. 628 - 631 vol.2.
[39] A.Dauferi. S.Armour, B.S. Lee, A. Nix, D. Bull, \An Evaluation of the Performance
of IEEE 802.11a and 802.llg Wireless Local Area Networks in a Corporate O?ce En-
vironment", ICC 2003, Alaska, May 2003.
[40] G. Judd and P. Steenkiste, \Repeatable and realistic wireless experimentation through
physical emulation", in HotNets-II, November 2003.
[41] J. Kurose et al., \Report of NSF Workshop on Network Research Testbeds," Nov 2002.
[42] http://www.us.anritsu.com/downloads/flles/MustHave- Wireless.pdf
[43] C. Hodgdon, \Adaptive Frequency Hopping for Reduced Interference between Blue-
tooth and Wireless LAN", Ericsson Technology Licensing, May 2003.
[44] A.S.M. De Franceschi, M. A. da Rocha, Henrique L. Weber, C. B. Westphall. \Proac-
tive Network Management Using Remote Monitoring and Artiflcial Intelligence Tech-
niques," iscc, vol. 00, p. 167, 2nd 1997.
65
[45] A.S.M. De Franceschi, L.F. Kormann, C.B. Westphall, \Performance Evaluation for
Proactive Network Management in Proceedings of the ICC96 International Conference
on Communications", vol. I, Dallas, USA, Jun., 1996. p. 22-26.
[46] J. Zyren, \Extension of Bluetooth and 802.11 Direct Sequence Interference Model,
Doc.: IEEE 802.11-98/378, Nov. 1998.
[47] J. Zyren, \Reliability of IEEE802.11 Hi Rate DSSS WLANs in a High Density Blue-
tooth Environment", Technical Paper, Bluetooth ?99, 1999.
[48] K. Pahlavan and P. Krishnamurthy, Principles of Wireless Networks, Prentice Hall:
Upper Saddle River, NJ, 2002, ISBN 0-13-093003-2.
[49] P. Gopalakrishnan, \Methods For Predicting The Throughput Characteristics Of Rate-
Adaptive Wireless Lans," in Electrical and Computer Engineering. Rutgers: The State
University of New Jersey, 2004.
[50] J. K. Chen and T.S. Rappaport, \Throughput Measurements and Models of Public
IEEE 802.11b Wireless Local Area Networks," The University of Texas at Austin,
Technical Report, WNCG-TR-2004-05-4, 2004.
[51] S.Vasudevan, \Asimulatorforanalyzing thethroughput of IEEE802.11bwireless LAN
systems," Masters Thesis in Electrical Engineering. BlacksBurg: Virginia Polytech and
State University, 2005
[52] B. E. Henty and T. S. Rappaport, \Throughput Measurements and Empirical Predic-
tion Models for IEEE 802.11b Wireless LAN (WLAN) Installations,"Dept. of Electrical
and Computer Engineering, Virginia Polytechnic Institute and State University, Tech-
nical Report MPRG 01- 08, 2001.
[53] A. Conti, D. Dardari, G. Pasolini, and O.Andrisano, \Bluetooth and IEEE802.11b
coexistence: analytical performance evaluation in fading channels," IEEE Journal on
Selected Areas in Communications, vol. 21, no. 2, pp. 259 - 269, feb. 2003.
[54] A. Kamerman, \Coexistence between Bluetooth and IEEE 802.11 CCK: Solutions to
avoid mutual interference," in IEEE P802.11 Working Group Contribution, IEEE P802.
I1-00//162r0, July 2000.
[55] N. Golmie, R. E. Van Dyck, and A. Soltanian, \Interference of Bluetooth and IEEE
802.11: Simulation Modeling and Performance Evaluation," Proceedings of the Fourth
International ACM Workshop on Modeling, Analysis and Simulation of Wireless and
Mobile Systems, July 2001, Rome, Italy.
[56] Linksys Product Speciflcations. Link to the product page www.linksys.com
[57] F. Berggren and S.-L. Kim, \Energy-E?cient Control of Rate and Power in DS-CDMA
Systems," IEEE Transactions on Wireless Communications, vol. 3, no. 3, pp. 725-733,
2004.
[58] Mathworks Inc., Link to the product page www.mathworks.com
66
[59] I. Howitt, \WLAN and WPAN coexistence in UL band," IEEE Trans. Veh. Technol.,
vol. 50, pp. 11141124, July 2001.
[60] C. F. Chiasserini and R. R. Rao, \Performance of IEEE802.11 WLANs in Bluetooth
environment," in Proc. IEEEWireless Communications and Networking Conf., vol. 1,
IL, Sept. 2000, pp. 9499.
[61] C. Chowdhury , Makela and K. Pahlavan,\Statistical information transfer in random
crossing of infostation coverage" In: Proc. 2005 Finnish Signal Processing Symposium
(FINSIG?05), Kuopio, Finland, Aug 25, 2005.
[62] V.Bhatia, W.Y. Chen, and L.T.Mereaud, \MQP Project",WPI, December 2003.
67
List of Definitions and Abbreviations
Self interference - Interference of other 802.11b stations on the 802.11b network of
interest.
Bluetooth interference - Interference of Bluetooth devices on 802.11b network.
Co-channel interference - Co-channel interference occurs when two wireless stations
working close to each other are on the same channel.
Adjacent channel interference - Adjacent channel interference occurs when two wireless
stations working close to each other are on any of the overlapping channels.
SNR/RSSI - Signal strength of the access point perceived by the wireless station.
WLAN Wireless Local Area Network
AP Access Point
RSSI Receive Signal Strength Indicator
MAC Medium Access Control
PHY Physical
LOS Line Of Sight
NLOS Non Line Of Sight
PER Packet Error Rate
SNR Signal-to-Noise Ratio
TCP Transmission Control Protocol
68
UDP User Datagram Protocol
RTS Request to send
CTS Clear to Send
FTS Fragmentation Threshold Set
CSMA-CA Carrier Sense Multiple Access-Collision Avoidance
CW Contention Window
ACK Acknowledgement
MS Mobile Station
DCF Distributed Coordination Function
PCF Point Coordination Function
DIFS DCF Inter Frame Space
SIFS Short Inter Frame Space
NIC Network Interface Card
IP Internet Protocol
Tx Transmitter
Rx Receiver
69