<%BANNER%>

An Experimental Study of Interference Between 802.11 and 802.15 Networks

Permanent Link: http://ufdc.ufl.edu/UFE0024455/00001

Material Information

Title: An Experimental Study of Interference Between 802.11 and 802.15 Networks
Physical Description: 1 online resource (45 p.)
Language: english
Creator: Sen, Somak
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2009

Subjects

Subjects / Keywords: bluetooth, interference, wifi
Computer and Information Science and Engineering -- Dissertations, Academic -- UF
Genre: Computer Engineering thesis, M.S.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Recent advances in the study of computer networks have established 802.11 and 802.15 networks to be the unparalleled magnates of wireless network technologies. Bluetooth is an up and coming inexpensive wireless technology designed to function within a short-range and holds great promise to be the replacement for wired communication between all portable devices and detachable components of the desktop computer. However, the performance of Bluetooth can be greatly impaired due to interference with 802.11 networks which operates in the same unlicensed wireless spectrum. Hence it becomes a matter of great concern to ascertain the extent to which the two technologies can coexist without causing interference to one another. This study aims to explain how the interference occurs in terms of the underlying transmission protocols employed by 802.11 and 802.15 technologies and experimentally attempts to determine the amount of interference between the two technologies. Our results indicate that the best results can be achieved by selecting a Bluetooth packet size of 3 which guarantees good system performance even when the network is heavily congested. Additionally, limiting the piconet size to about 20 25 piconets guarantees low packet loss and high system throughput thus causing minimum interference in the wireless medium.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Somak Sen.
Thesis: Thesis (M.S.)--University of Florida, 2009.
Local: Adviser: Liu, Chien-Lian.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2009-11-30

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2009
System ID: UFE0024455:00001

Permanent Link: http://ufdc.ufl.edu/UFE0024455/00001

Material Information

Title: An Experimental Study of Interference Between 802.11 and 802.15 Networks
Physical Description: 1 online resource (45 p.)
Language: english
Creator: Sen, Somak
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2009

Subjects

Subjects / Keywords: bluetooth, interference, wifi
Computer and Information Science and Engineering -- Dissertations, Academic -- UF
Genre: Computer Engineering thesis, M.S.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Recent advances in the study of computer networks have established 802.11 and 802.15 networks to be the unparalleled magnates of wireless network technologies. Bluetooth is an up and coming inexpensive wireless technology designed to function within a short-range and holds great promise to be the replacement for wired communication between all portable devices and detachable components of the desktop computer. However, the performance of Bluetooth can be greatly impaired due to interference with 802.11 networks which operates in the same unlicensed wireless spectrum. Hence it becomes a matter of great concern to ascertain the extent to which the two technologies can coexist without causing interference to one another. This study aims to explain how the interference occurs in terms of the underlying transmission protocols employed by 802.11 and 802.15 technologies and experimentally attempts to determine the amount of interference between the two technologies. Our results indicate that the best results can be achieved by selecting a Bluetooth packet size of 3 which guarantees good system performance even when the network is heavily congested. Additionally, limiting the piconet size to about 20 25 piconets guarantees low packet loss and high system throughput thus causing minimum interference in the wireless medium.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Somak Sen.
Thesis: Thesis (M.S.)--University of Florida, 2009.
Local: Adviser: Liu, Chien-Lian.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2009-11-30

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2009
System ID: UFE0024455:00001


This item has the following downloads:


Full Text

PAGE 1

1 AN EXPERIMENTAL STUDY OF INTERFERENCE BETWEEN 802.11 AND 802.15 NETWORKS By SOMAK SEN A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLOR IDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF SCIENCE UNIVERSITY OF FLORIDA 2009

PAGE 2

2 2009 Somak Sen

PAGE 3

3 To my loving parents

PAGE 4

4 ACKNOWLEDGMENTS I would lik e to express my gratitude to my chief supervisor and academic advisor Prof. Jonathan C. Liu (Associate Professor at Universi ty of Florida) for his constant guidance and encouragement towards the completion of this thesis. I am also indebted to the other members of my Supervisory Research Committee notably Prof. Alin Dobra (Assistant Professor at University of Florida) and Prof. Tamer Kahveci (Assistant Professor at University of Florida) who have imparted a great deal of knowledge and helped build my research interests through the cour ses they taught me as part of the curriculum. Also, a special word of thanks goes to the graduate advisors and administrative staff members of the CISE Department of the University of Florida who have painstakingly helped me with all necessary paperwork in connection with my thesis.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS...............................................................................................................4LIST OF TABLES................................................................................................................. ..........6LIST OF FIGURES.........................................................................................................................7ABSTRACT.....................................................................................................................................8CHAPTER 1 INTRODUCTION....................................................................................................................9Basic Overview.........................................................................................................................9Proposed Results............................................................................................................... ......11Related Study..........................................................................................................................132 INTERFERENCE IN THE WIRELESS MEDIUM............................................................... 17Bluetooth Frequency Hopping Technique..............................................................................17Experimental Modeling of In terference in the ISM Band...................................................... 183 EXPERIMENTAL ANALYSIS............................................................................................. 22The Network Simulator.......................................................................................................... 22Experimental Setup............................................................................................................. ....23Trace File Analysis............................................................................................................ .....25Analysis of Packet Loss Probability................................................................................25Analysis of System Throughput......................................................................................264 RESULTS AND OB SE RVATIONS...................................................................................... 28Overview....................................................................................................................... ..........28Probability of Packet Loss..................................................................................................... .28System Throughput.................................................................................................................355 CONCLUSION..................................................................................................................... ..42LIST OF REFERENCES...............................................................................................................43BIOGRAPHICAL SKETCH.........................................................................................................45

PAGE 6

6 LIST OF TABLES Table page 3-1 Case analysis of interference be tween 802.11 and 802.15 for Tx and Rx modes ............. 26

PAGE 7

7 LIST OF FIGURES Figure page 1-1 Spatial representation of coll ision between 802.11b and 802.15 packets.......................... 111-2 Connectivity model for Bluetooth network....................................................................... 141-3 Average throughput per pi conet vs. piconet load.............................................................. 141-4 Packet success probability vs. distance..............................................................................152-1 Model environmental setup for the current study.............................................................. 194-1 Packet loss probability with 20 Wi-Fi nodes in the direct case......................................... 294-2 Packet loss probability with 50 Wi-Fi nodes in the direct case......................................... 304-3 Packet loss probability with 80 Wi-Fi nodes in the direct case......................................... 314-4 Packet loss probability with 20 Wi-Fi nodes in the indirect case...................................... 324-5 Packet loss probability with 50 Wi-Fi nodes in the indirect case...................................... 334-6 Packet loss probability with 80 Wi-Fi nodes in the indirect case...................................... 334-7 System throughput with 20 Wi -Fi nodes in the direct case............................................... 354-8 System throughput with 50 Wi -Fi nodes in the direct case............................................... 374-9 System throughput with 80 Wi -Fi nodes in the direct case............................................... 384-10 System throughput with 20 Wi-F i nodes in the indirect case............................................ 384-11 System throughput with 50 Wi-F i nodes in the indirect case............................................ 394-12 System throughput with 80 Wi-F i nodes in the indirect case............................................ 40

PAGE 8

8 Abstract of Thesis Presen ted to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Science AN EXPERMENTAL STUDY OF INTERFERENCE BETWEEN 802.11 AND 802.15 NETWORKS By Somak Sen May 2009 Chair: Jonathan C. L. Liu Major: Computer Engineering Recent advances in the study of computer networks have established 802.11 and 802.15 networks to be the unparalleled magnates of wire less network technologies. Bluetooth is an up and coming inexpensive wireless technology designed to function wi thin a short-range and holds great promise to be the replacement for wired communication between all portable devices and detachable components of the desktop computer. However, the performance of Bluetooth can be greatly impaired due to interf erence with 802.11 networks which ope rates in the same unlicensed wireless spectrum. Hence it becomes a matter of gr eat concern to ascertain the extent to which the two technologies can coexist without causing interference to one another. This study aims to explain how the interf erence occurs in terms of the underlying transmission protocols employed by 802.11 a nd 802.15 technologies and experimentally attempts to determine the am ount of interference between the two technologies. Our results indicate that the best results can be achieved by selecting a Bluetooth packet size of 3 which guarantees good system performance even when the network is heavily congested. Additionally, limiting the piconet size to about 20 piconets guarantees low packet loss and high system throughput thus causing minimum interference in the wi reless medium.

PAGE 9

9 CHAPTER 1 INTRODUCTION Basic Overview Over the past few years, rese archers have spent a considerab le am ount of effort trying to ensure reliable data transfer over wireless networ ks. The reason for such concern is attributed to the basic postulate that wireless networks are prone to errors in the form of packet loss over the wireless channel. While this lo ss is trivial in case of voice service and not worth addressing, it can give rise to severe impairment in the case of transmitting multimedia, especially data in video format. Although a great many research papers have dealt with this cr itical issue of lossless multimedia transfer over a wireless channel, very few papers have attempted to formulate an actual measurement of interf erence between 802.11 and 802.15 netw orks and provide a unique solution to reduce the interference. While a lot of research papers have been published on the improvement of performance of wireless networks, and 802.11 has been explored in great detail which accounts for its popularity and recurrent citations in various research journals, the comp aratively new 802.15 counterpart is still in a state of inception. Being an emerging technology, most of the research has centered on expounding its basic functionalities rather than widening the research scope to determine its coexistence between other compe ting technologies. While it has been widely rumored that Bluetooth might eventually ga in popularity over 802.11 networks all technological innovations achieved over the years would be seriously jeopa rdized should the tec hnologies interfere and cancel each other in terms of co-e xistence. Preventing such an adverse effect by experimentally determining the level of interference is the compelling force behind this interference study. Both 802.11 and 802.15 networks have enjoyed widespread deployment during the last few years and are the most popular forms of the unlicensed wireless communication. While

PAGE 10

10 802.11 is intended for communication over longer distances, it requir es expensive hardware with larger power consumption. They use the same frequency range; however, the modulation techniques employed is different in each case. Bl uetooth is generally used to replace cables in small-scale applications, while Wi-Fi is used as a cable replacement in Local Area Networks. Bluetooth technology is ad-hoc, in that the connecti on is established spontaneously without requiring configuration to setup and teardown connections between peers. Although Wi-Fi requires a more elaborate setup procedure, it can provide full-scale network functionalities operable over larger distances with greater security. It is evident that both 802.11 and 802.15 are suited for unique dedicated purposes and it becomes imperative for both of them to co-exist with little or no interference between them. While the causes and consequences of in terference between 802.11 and 802.15 has been addressed at length in the subseque nt chapters, one reading this thesis would feel much at ease to appreciate the experimental anal ysis if one could have a prelim inary idea of what we are trying to imply when we claim that interference between 802.11 and 802.15 networks is inevitable and frequent in the wireless medium. The following diagram, taken from the paper [1 4], provides a simplistic pictorial depiction of how the collision between 802.11 and 802.15 packets occur and how obviously vulnerable this makes the wireless system given the vast di fference in size of 802.11 and 802.15 packets. While a Bluetooth packet might escape co llision quite frequently owing to its small size, a Wi-Fi packet suffers from this terrible disa dvantage of being conspicuously enormous in comparison with a Bluetooth packet and hence prone to suffer collisi on with other packets in the system Also, since Bluetooth allows us to choose from packet sizes of 1, 3 and 5, the size of the Bluetooth packet would play an important role in the analysis of packet collision. However, choosing a Bluetooth

PAGE 11

11 packet size of 1 does not guarantee the best perf ormance owing to other factors which we shall address in this study. Figure 1-1. Spatial representation of collision between 802.11b and 802.15 packets Figure 1-1 is presented from a spatial perspective meaning we are concerned with the way the packets are oriented in a 3-di mentional space with litt le or no regard to time and frequency. It is worthwhile to note that given the same time a nd frequency, we might be able to avert collision by performing a simple rotation such that the square area of the 802.11 packet faces the plane formed by the time-frequency axes. While spat ial rotation is an obvious solution to the interference problem, the situation can get increasi ngly complicated when we consider a plethora of other factors, not to menti on errors encountered in wireless experiments, which play an important role in determining the amount of interference. Proposed Results In our experim ental approach, we address the degree of interfere nce by constructing a system having a fixed number (20, 50 or 80) of Wi-Fi nodes. The number of Bluetooth nodes is varied from 0 to 100 for each of the three fi xed values of Wi-Fi node s limiting the distance

PAGE 12

12 between the nodes to either 1 meter or 5 meters. By using these values we study the change in the degree of interference with respect to not only the number of interfering nodes but also the inter-node distance and the size of Bluetooth packets. In order for us to be able to model a worstcase setup, we deliberately modify the Bluetoot h protocol by disallowing the process of frequency hopping so that the frequencies of the Wi-Fi and Bluetooth transmissions are coincident with each other. From the results of our experiment, we establ ish that the best system performance can be obtained by choosing a Bluetoot h packet size of 3 which guarantees not only substantial throughput when the system is up and running but also allows the system to continue sending packets in a congested state. We further analy ze the impact of the number of piconets in the system on the probability of p acket loss and overall system thr oughput. Our results indicate that the packet loss probability rises exponentially with the increase of the number of piconets in the system and the throughput rises in a polynomial ma nner. This simultaneous increase in packet loss and system throughput is explained using the fact that the greater the number of packets successfully transmitted, the higher are the chan ces of it getting dropped in the process of transmission, which is a tradeoff that all wirele ss media suffer from. In addition to the optimal size of 3 for the Bluetooth packets, we establish an optimum value for the number of piconets in the system that guarantees best performance. Our analysis reveals that in order to achieve best overall system performance in terms of low probability of packet loss and high system throughput, the number of piconets should li e between 20 and 25. The system becomes unsuitable for practical purposes when the number of piconets reaches 80 at which point the packet loss probability is more than 90% which implies that 90% of the packets sent into the wireless medium are either damaged or lost during the transmission. Although an optimal packet

PAGE 13

13 size of 3 for Bluetooth packets guarantees a non-negligible throughput in this worst-case scenario, the system is practically stalled due to high level of conge stion thus rendering it unusable for practical wireless applications. Related Study The problem of reducing interference has b een approached from various perspectives by different groups of researchers. This section is dedicated to enumerating the contributions of various researchers that pert ain to the study of Bluetooth networks and addressing the vulnerabilities it faces when existing in tandem with other wireless media. In order to address the coexistence problem, we need to analyze the threats that Bluetooth devices are faced with when attempting to operate in the unlicensed 2.4 GHz Industrial, Scientific and Medical (ISM) band shared by Bluetooth and Wi-Fi devices. The vulnerability issues that Bluetooth might face in order to se rve as a large-scale cab le replacement technology have been addressed in depth by Ve rgetis et al in [18]. Research ers have approached this issue from varied perspectives which fall into either one of the following categories 1. Study involving collision of packets in the wireless me dium based on coincidence according to time, frequency and space. 2. Specific algorithms which do not directly at tempt to study the interference but help mitigate the amount of interference ca used as a concomitant side-effect. Similar studies involving the coexistence issu es faced by Bluetooth technology have been conducted in [3], [12], [15] a nd [17]. One of the papers which has maximum relevance to our study and which provides us with the basic ma thematical relations to help formulate our experimentation is the one proposed by Cordeiro et al, in [5] wh ich experimentally models the concept of inter-piconet interference and establishes several important mathematical relationships aimed at facilita ting analysis at the Medium Ac cess Control (MAC) level. The modeling takes into account the st rength of the received signal after suffering attenuation during

PAGE 14

14 propagation from sender to receiver and determines whether this strength is greater than a fixed threshold value in order to be able to cause in terference at the receiver. The basic connectivity model used in this paper is: Figure 1-2. Connectivity model for Bluetooth network In this study we use a similar approach as [5] to model our own experimental environment without confining ourselves solely to Bluetooth piconets. We include Wi-Fi nodes in our analysis to build a congested network for studying the worstcase interference. The graphs obtained in [5] for average throughput and packet success probab ility are shown in Figure 1-3 and Figure 1-4. Figure 1-3. Average throughput pe r piconet vs. piconet load

PAGE 15

15 Figure 1-4. Packet succe ss probability vs. distance Examples of other notable algorithms propos ed to mitigate the amount of interference between 802.11 and 802.15 networks in clude but are not limited to the MAC scheme proposed in [7] that allow for the statistical sharing of th e total available bandwidth across channels, the Interference-aware. Bluetooth Segmentation (iBL UES) algorithm in [4] to select Bluetooth packet types based on the packet success proba bility, the coexistence mechanism based on Time Division Multiplexing (TDM) proposed in [16], the study involving carrier-t o-interference ratio in [11] performed using realistic parameters, the packet fragmentation technique in [13] aimed at reducing the probability of packet collision, the Bluetooth Interference-aware Scheduling (BIAS) algorithm in [8] for effective channel allo cation based on frequency hopping, the Adaptive Automatic Repeat Request (ARQ) timeout techni que in [1], the adaptive frequency hopping and scheduling technique in [10], the dynamic slot assignment and piconet partitioning mechanisms in [2], the handoff algorithms proposed in c onnection with dynamic spreading to support multimedia traffic in [20], the problem of es tablishing connected topology in terms of algorithmic complexity in [8] and orthogonality fact or as in [6] are examples of such specific algorithms which result in reduction of interf erence. The optimization technique for uplink

PAGE 16

16 scheduling in [19] is a similar study of interf erence between Wi-Fi and cellular third-generation (3G) networks. Having studied the related research papers dealing with coexiste nce issues between 802.11 and 802.15 and novel algorithms aimed at improveme nt of system throughput we attempt to delve deeper into the intri cate underlying mechanisms to unde rstand how the interference is caused. Consequently, we attempt to create our own experimental set up to allow Wi-Fi and Bluetooth devices to run within the same envi ronment and subsequently determine a measured estimate of interference based on the test results.

PAGE 17

17 CHAPTER 2 INTERFERENCE IN THE WIRELESS MEDIUM Bluetooth Frequency Hopping Technique The 802.11 Medium Access Control (MAC) protoc ol in Wi-Fi networks uses CSMA/CA (Carrier Sense Multiple Access with Collision Avoidance) to transmit data packets. Before transmitting a packet, a station senses the channel to ensure that it is not busy and refrains from transmitting should it find the channel busy. A set of dedicated packets entitled Request to Send (RTS) and Clear to Send (CTS) are exchanged in order to signal th e transmitting stations that the channel is free and that they might carry on the transmission. This collision avoidance protocol is used particularly because collision detecti on techniques cannot be implemented in wireless medium since a station cannot listen to the channel and transmit data at the same time. Bluetooth, on the other hand, uses a completely different approach to transmit packets. It partitions the frequency range of 2.402 GHz to 2.480 GHz into 79 channels and employs an intelligent technique of Frequency Hopping to hop between channels on each packet transmission, i.e. opting to transmit each packet on a different randomly-selected channel, thus reducing the probability of packet collision. It makes a total of 1600 hops/second which restricts the slot time to approximately 625 microseconds. The advantage of having such an approach is the fact that frequency hopping allows for transmission errors to be quickly and effectivel y detected and remedied than would have been the case if the packets were transmitted at a constant frequency. This is further facilitated by the fact that Bluetooth packets are smaller in size and can be transmitted faster than 802.11 packets. Given, the high bit-error rates of wireless networks, collisions between packets are inevitable. However, a collision (or, to be mo re precise, an interfer ence) between an 802.11 and an 802.15 packet would result in the entire 802.1 1 transmission to be corrupted while, for the

PAGE 18

18 802.15 packet, only that particular packet needs to be retransmitted. This clearly is a critical drawback. Also, since the 802.11 pa ckets are larger in size, th ey would be susceptible to collision than 802.15 packets which cannot be avoided. In anticipation of the interf erence in the wireless channel between Wi-Fi and Bluetooth devices, we intend to obtain a mathematical estimat e that provides a measure of this interference. Experimental Modeling of Interference in the ISM Band The Industrial, Scientific and Medical (ISM) Band is a general-purpos e publicly-accessible wireless channel used ubiquitously by industrial, scien tific and m edical organizations to transmit wireless data. Bluetoot h and Wi-Fi devices operating in th is ISM band are prone to face a considerable amount of interference from othe r unlicensed devices bei ng operated on the same channel probably using the same frequency. As e xplained earlier, the amount of data impairment in case of Wi-Fi will be substantially high in co mparison to that of the Bluetooth devices. Our concern is to use simulation tec hniques to perform experimentation in this free-license wireless channel. Having enumerated the different approaches an d varied opinions expressed by researchers regarding the vulnerabilities of coexistence in the ISM band and regarding the fact that it is impossible to select any one of the proposed solutions as an unanimously accepted and universally applicable techni que, the task remains to model our own ISM environment and design a simulation based on such a model. In keeping with all the design issues discussed so far, we want to model an experimental setup that would enable us to determine the degree of interference in the wireless medium. Furthermore, this setup should be compatible w ith the experimental environment that we intend to use in order to perform the analysis. Figur e 2-1 shows a feasible experimental modeling scheme for our interference problem.

PAGE 19

19 Figure 2-1. Model environmenta l setup for the current study The system consists of two piconets which do not overlap and a ubiquitous wireless Local Area Network whose range covers a ll the devices in the system. The system contains two devices which communicate with each other via Wi-Fi network and each of which is, in turn, part of one non-overlapping piconet. Each piconet contains two Bluetooth devices. Additionally, the system contains an Access Point (AP) which c onnects to an Ethernet wired network. We are concerned with the in terference in the Wi-Fi device contained in the left-most piconet (highlighted using a rectangular frame in Figure 2-1) which we shall hereafter refer to as the reference node. This device attempts to comm unicate with the other Wi-Fi device located in the other piconet and, concurren tly exchanges Bluetooth packets with one of its Bluetooth neighbors in its own piconet. Cl early, interference is inevitable, should the Bluetooth frequency (as selected randomly by the frequency hopping technique) coincide with the frequency on which the Wi-Fi packets are being transferred. A set of metrics which might be consid ered for the experimental setup are:

PAGE 20

20 The spatial distance between the reference nod e and the Bluetooth node s in the vicinity. The spatial orientation of the Bl uetooth nodes within the piconet. .The size of the transmitted Bluetooth packets. The number of Bluetooth nodes expressed in terms of the number of piconets formed. The frequencies on which the packets are being transmitted. In order to determine the worst-case interference, we might assume that fre quency hopping does not take place and the Bluetooth packets are transmitted on a constant frequency. The total duration we should system the system to run before convincing ourselves that the output is acceptable with out loss of generality. The experimental modeling of the above interf erence scheme calls for the selection of an experimental setup that takes the above menti oned factors into consideration. Since we are interested in documenting the results for the worst-case scenario, we might deliberately set constant values to some of these metrics in an attempt to aggravate the performance and drive the system to follow a worst-case behavior. Also, we need to be cognizant of the pros and cons of using a software simulation in contrast with an actual physical setup and determine whether feasible values for the above me trics can be assigned without a dditional overhead in our chosen scheme of experimentation. While a physical setup might guara ntee the most realistic results, flexibility of change for the values of parameters might be severely re strained. As an example, a worst-case scenario might be modeled by having a piconet size of 100 where each piconet has a full-capacity of 8 Bluetooth devices. This would re quire us to have 800 Bluetooth devices working in parallel which is almost impossible to set up in a co llege laboratory. Furthermore, having limited physical apparatus would not create enough interfere nce for us to actually arrive at a general conclusion. To illustrate this point, we might have a situation where we might need to introduce

PAGE 21

21 additional Bluetooth devices in multiples of 20 to achieve a measurable change in system performance. In our case, we have attempted to measure the interference using a software simulator rather than a physical setup. In order for us to analyze the performance in a worst-case scenario, we have assumed that the Blue tooth devices transmit packets on a constant frequency value rather than hopping over frequencies. Also, the sp atial distance between the Wi-Fi and Bluetooth devices is assumed to be a constant value. This experimental study uses tw o fixed values for this distance, namely a unit distance of 1 meter which is very neglig ible to cause signal attenuation and a distance of 5 meters which has a possibili ty of drastically modi fying the results as a consequence of signal attenuati on caused during propagation from sender to receiver. It is important to note that Corderio et al. performed a similar spatial analysis in [5] to measure system throughput in the face of in terference in a system consisting solely of Bluetooth devices. This study takes into account the probability that the strength of the propagated signal, after suffering attenuation, is strong e nough to cause interference at the receiving device and the subsequent throughput is expresse d as a function of this probab ility. The reader is strongly encouraged to refer to this paper to get a prelimin ary idea of the spatial approach to interference and the associated mathematical modeling for the consideration of signal a ttenuation as a primal factor in the computation of inte rference. In our analysis, we ha ve identified two basic cases: one of which takes into account the signal attenuation and the other which does not. Using these two cases we have attempted to measure system perfo rmance in terms of the number of interfering Bluetooth and Wi-Fi devices, the size of the transmitted Bluetooth packets. The only assumption made in order to simplify the analysis is that the Bluetooth packets are transmitted on the same frequency without hopping across different frequency values.

PAGE 22

22 CHAPTER 3 EXPERIMENTAL ANALYSIS The Network Simulator In the research community, sim ulation tools do exist for Wi-Fi and Bluetooth individually. However, they do not co-exist in the same simulation framework. During this research investigation, we believe Wi-Fi and Bluetooth a pplications should be run via an appropriate simulator with wireless network support. The Network Simulator (NS-2) developed at the University of Southern California was selected as the simulation platform. However, original work has been performed to create the Bluetooth module for extending the missing capability of NS-2. By integrating our own Bl uetooth modules, experiments were possible to determine the degree of interference between 802.11 and 802.15 networ ks. NS-2 is a discrete event simulator with support for simulating Transmission Control Protocol (TCP), User Datagram Protocol (UDP), Routing and Multicast protocols over wired and wireless networks. This simulator is still in a stage of development and errors are periodically being re ported and eliminated by the developers. This simulator is an open-source project wr itten as a collection of C++ programs which was downloaded free of charge from the project website at: http://downloads.sourceforge.net/nsnam/ns-allinone-2.29.3.tar.gz In order for the Bluetooth program s to run, th e following library was built into the NS-2 installation: http://www.cs.uc.edu/~cdmc/ucbt/src/ucbt-0.9.9.2a.tgz This add-on im plements the protocol stack for Bluetooth and allows th e user to use this built-in protocol stack to run Bluetooth appli cations in collaborati on with other wireless

PAGE 23

23 standards. The whole suite was set up on a C ygwin environment running over a Windows Vista personal laptop computer. Scripts designed to simulate the network se tup were written in Tool Control Language (TCL) scripts and fed to the NS-2 simulator to get an output modeled to reflect the dynamic network behavior over a period of time. Experimental Setup The experim ental environment that we assume is based on the office settings in major corporations. In this environmen t, there are many Wi-Fi and Blue tooth devices co-existing in a confined space. To simplify the complex interaction, we simply assume that these devices are stationary at this moment. However, their positions should be randomly placed within a limited distance. Various experiment configurations were set up using TCL scripts and the generated output trace files were analyzed. As explai ned in the introduction, the outcome of the interferences will appear as the probability of packet loss and system throughput impact. Therefore, via the output traces, our aim is to a rrive at an estimate of the interference between 802.11 and 802.15 networks in terms of the probabil ity of packet loss an d measured throughput. There are many factors to be taken into acc ount when designing th e experiments. We believe the following factors are the critical ones: 1. The possibility of co-existence betw een 802.11 and 802.15 networks in the same environment needs to be ensured. The Network Simulator allows the user to define nodes which implement the Wi-Fi and Bluetooth prot ocols. This can be achieved by defining a node type to be either Mac/BNEP fo r Bluetooth or Mac/802_11 for Wi-Fi. 2. A packet-wise trace of the system needs to be generated by the simulation which can then be analyzed to identify the behavior of each packet in the system. Network Simulator uses a print function that can generate such detail ed trace files in a tabular format where each column represents a different parameter for the experiment. Additional details of this format may be found at http://nsnam.isi.edu/nsnam /index.php/NS-2_Trace_Form ats This output can either be read manually by the pr ogrammer or fed to the associated Network Animator (NAM) program to visualize the system behavior.

PAGE 24

24 3. The system should be able to produce an output for any number of nodes. NS-2 is a robust simulator that can handle up to any number of nodes and generate th e trace files for the entire run. 4. The system should be able to generate output for different Bluetooth packet sizes. It is a matter of prime importance to determine wh ether the throughput can be increased by varying the size of the gene rated Bluetooth packets. 5. The system should also allow the programmer to modify the spatial configuration of the system, i.e. the distance between the nodes. Based on our experimental modeling, a set of TCL scripts were devised to simulate the coexistence environment between Wi -Fi and Bluetooth. What we feel interested in most is the accumulated impact due to the scalability of Wi-F i and Bluetooth networks. In the near future, we envision that hundreds of Bl uetooth devices can co-exist w ith hundreds of Wi-Fi devices in the company settings. To the best of our know ledge, the potential impact due to these hundreds of devices is still unknown in re search literatures. Wh at makes it more interesting is that the Bluetooth protocol allows multiple packets to be se nt in a contiguous burst. It is unclear how the burst of Bluetooth packets will affect the overall interference. Thus, we have designed the experiments by varying the number of Bluetooth nodes and also specifying the size of Bluet ooth packets to be 1, 3 or 5. A ccordingly, the number of Wi-Fi nodes was fixed to be 20 in the base case then varied up to 80, at whic h point the packet loss probability approaches 100% and the system is st alled. These experiments were repeated twice, once under the assumption that the Bluetooth piconets are located within unit distance from the Wi-Fi nodes such that the signal attenuation is negligible which causes the maximum interference; and again by moving the Bluetooth nodes at a distance of 5 meters from the Wi-Fi nodes. In either case, the system was allowed to run for a specified duration of time sufficient enough to allow for the nodes to perform an enquiry scan before formation of the piconets, achieve a stable state and start transmission. Al so the system is required to allow to get

PAGE 25

25 congested by simultaneous transmission and rece ption by a huge of cluster of nodes so that packet loss is unavoidable. Finally, considering the large size of the trace files, only the columns required by our analysis is retained from the trace file and subseq uently analyzed to determine the probability of packet loss and overall throughput. Furthermore, simple programs were written in Java programming language to scan the extracted co lumns of the trace f ile and calculate the probability of packet loss and the overall system throughput. Trace File Analysis The trace file generated by the NS-2 pr ogram is processed by a text editor and subsequently analyzed to get the probability of packet loss and system throughput. Prior to calculating the packet loss probability and the throughput, the required columns are filtered out using the Unix grep utility and certain log information such as those documenting the InquiryScan and Standby states of the Bluetooth nodes are removed manually. Analysis of Packet Loss Probability The generated trace file contains a colum n which specifies whether the packet was successfully sent during a transm it operation (Tx) or receive ope ration (Rx). Any other generated code for this column is considered an inco rrect transmission and the corresponding packet is considered lost. Our approach analyzes the trace file to count the total number of times a packet is not transmitted correctly and calculate a ratio between the number of correct and incorrect transmissions The above analysis for the probability of packet loss is performed for each packet transmitted between a set of nodes in the system. The packet ID for the packet and the node IDs of the sender and receiver nodes are used to determine the number of packets that are successfully transmitted between the selected node pairs

PAGE 26

26 Our analysis of the packet loss probability draws upon a similar experimental technique employed by Golmie et al in [9] to determine the potential threats of interference under the assumption that both Wi-Fi and Bluetooth sy stems use the same frequency. Table 3-1 summarizes the four cases that were identified in their experiment according to whether the devices are in a Transmit Mode (Tx) or Receive Mode (Rx). Table 3-1. Case analysis of interferen ce between 802.11 and 802.15 for Tx and Rx modes Rx(BT) Tx(BT) Rx(Wi-Fi) Interference from Wi-Fi to BT and from BT to Wi-Fi. Wi-Fi received signal depends on distance from Access Point (AP). BT received signal depends on distance from other BT device and adaptive gain control. Severe interference from BT to Wi-Fi Tx(Wi-Fi) Severe interference from Wi-Fi to BT Interference to remote devices. Also may cause signal distortion at the Power Amplifier (PA) As evident from Table 3-1, the maximum amount of interference occurs from one of the devices is in the process of receiving a transm ission while the other is in the process of transmitting a packet. The least impairment occurs when both are transmitting in which case the victim is a third remote device which might be receiving transmissions from neighboring devices. In case when both devices are operating in the Receive Mode, we need to consider the spatial distances in order to determine the interference. Analysis of System Throughput In order to m easure the throughput, we need to obtain an estimate of the total size of data transferred in a unit amount of time. This is achi eved by analyzing the trace file and filtering out the relevant columns which, in this case, are th e timestamp values, the pa cket ID, the node IDs

PAGE 27

27 for the sender and receiver nodes and the packet size in bits. The total number of bits sent in a specified time interval is determined to obtai n a system throughput in kbps. This process is repeated for varying sizes of the Bluetooth pack ets to obtain the throughput based on packet size and the number of Bluetooth nodes. In order to perform the calculations for pack et loss and throughput, simple programs were written in Java to read the columns form the trac e files, extract the numerical values therein and use simple arithmetical formulation to compute the values required for graphical representation. The graphs were obtained by feeding the numerical values into a Microsoft Excel Worksheet and using the built-in Scatter Graph utility to obtain the graphical results.

PAGE 28

28 CHAPTER 4 RESULTS AND OBSERVATIONS Overview As explained in Chapter 3, the sim ulations were run for the following two cases and the corresponding values were plotted. DIRECT CASE. In this case, the Bluetooth nodes were spatially oriented in a manner such that the Bluetooth nodes are lo cated at a unit distance from the Wi-Fi nodes. This is the case which is guaranteed to cause minimu m amount of signal attenuation due to propagation and hence maximum interference. INDIRECT CASE. In this case, the Bluetooth nodes were spatially oriented in a manner such that the Bluetooth nodes are locat ed at a distance of 5 meters from the Wi-Fi nodes. Since the signals need to travel a distance of 5 meters in order to propagate between neighboring Wi-Fi and Bluetooth nodes, the amount of in terference should be conceptually less as compared to the direct case. The graphical results for the probability of packet loss and the system throughput are provided in order to analyze the system be havior under different node configurations. Probability of Packet Loss In both the direct and indirect cases, the experim ents were repeated three times for Bluetooth packets of sizes 1, 3 and 5 and the pack et loss probabilities for each of the three runs were plotted in the same graph. In case of packet loss probability, the maximum packet loss occurred in case of Bluetooth p ackets of size 5 and minimum packet loss was observed in case of packet size of 1. Figure 4-1, Fi gure 4-2 and Figure 4-3 provid e the packet loss probabilities incurred in the direct case wh ile Figure 4-4, Figure 4-5 and Fi gure 4-6 correspond to the results obtained in the indirect case. We attempt to de termine the conditions in which the packet loss probability is optimum and also the worst-case scenario when this probability reaches an exceptionally high value thus clogging the syst em. Once we have the results, we want to generalize the findings by analyz ing similarities between the graphs and determining the overlapping piconet range that gu arantees an optimum amount of packet loss in all the cases.

PAGE 29

29 Figure 4-1. Packet loss probability with 20 Wi-Fi nodes in the direct case To our surprise, even when no Bluetooth node is present, the packet loss for Wi-Fi nodes occurs with a probability of 28%. We have spent a significant amount of time to trace the causes. Apart from the vulnerabilities of the wireless me dium (roughly about 10% e rror rate), we realize that the packets transmitted by the Bluetooth nodes can be significantly large even before the piconet is formed. It is because the nodes need to perform an inquiry scan prior to the formation of the piconets. These signaling packets do interfe re with the Wi-Fi transmissions and cause a significant amount of packet loss. Since a piconet can only accommodate 8 devices, more piconets will be formed when the nodes are increased. With more pi conets co-existing with the Wi -Fi nodes, the interference is expected to increase. Thus, the probability of lost pack ets rises rapidly from 28% to 40% with a corresponding increase of 12 picone ts. Furthermore, we observe that a packet loss of 60% occurs when the number of piconets is close to 20 for a ll Bluetooth packet sizes. This probability of loss rises exponentially and reaches approximately 75% and 80% when the number of piconets increases to 40 and 60 respectivel y for all the three packet sizes. The system degrades to a substantial degree when the number of piconets reaches 80 at which point the packet loss

PAGE 30

30 probability is close to 90%, which is inadmi ssible in practical purposes. As the number of piconets increases to more than 80, the packet loss probability rises expo nentially till it reaches 100% and the system is practically halted becau se every packet in the system is dropped. One noteworthy fact in the Figure 4-1 is that the system behavior in case of a Bluetooth packet of size 1 differs drastically from the one s of size 3 and 5. The packet loss for a Bluetooth packet of size 1 is much less than that for si zes 3 and 5. When the number of piconets is 20, the packet loss probability is approximately 50% for p acket size of 1 while it is 60% for sizes 3 and 5. When the number of piconets increases to 40, 60 and 80, the respective packet loss probabilities in case of packet size of 1 are approximately 58%, 67% and 75%. These respective probabilities in the cases of packet size 3 and 5 are approximately 80%, 85% and 90%. It is interesting to note that for a Bluetooth packet of size 1, the system does not stall when the number of piconets reaches 100, and packets con tinue to be transmitted at a loss rate of 80%. However, we find that the packet loss probabi lity is around 97% for the other two packet sizes which renders the system practically non-functional for practical purposes. Figure 4-2. Packet loss probability with 50 Wi-Fi nodes in the direct case

PAGE 31

31 When the number of Wi-Fi nodes is increased to 50, the packet loss probability increases slightly than in Figure 4-1 and the initial inte rference rises from 28% to 33%. The packet loss probabilities in case of Bluetooth packet size of 1 are close to 50% 62%, 70%, 77% and 81% when the number of piconets reaches 20, 40, 60, 80 and 100 respectiv ely. The corresponding probabilities for the other two packet sizes are approximately 62%, 78%, 88%, 90% and 98%. The optimal piconet size is somewhere between 15 and 20 which agrees with the results in our previous case. Figure 4-3. Packet loss probability with 80 Wi-Fi nodes in the direct case Finally, when the number of Wi-Fi nodes reac hes 80, the initial inte rference becomes 35% which proves that the interference is caused mainly by the pres ence of Bluetooth nodes which do not use the collision avoidance scheme used by the 802.11 protocol. This in itial interference is attributed to the exchange of inquiry scan p ackets between Bluetooth devices vying to form a piconet with the immediate master in the vicini ty. In case of Bluetooth packets of size 1, the respective packet loss probabil ities for piconet sizes of 20, 40, 60, 80 and 100 are approximately 51%, 63%, 77%, 80% and 85%. The corresponding pr obabilities for the othe r two packet sizes

PAGE 32

32 are 65%, 80%, 88%, 93% and 99%. In accordance with our earlier results, the lowest packet loss is caused when Bluetooth p ackets of size 1 are chosen. Figure 4-4. Packet loss probability with 20 Wi-Fi nodes in the indirect case In the indirect case, contrary to our anticipation, we find that the initial interference in the presence of 20 Wi-Fi nodes rises to 43% from 28% in the direct case. Also, the curves rise in an exponential manner quite similar to the direct case causing a packet loss probability of 90% when the number of piconets reaches 70. Along th e lines of our interp retation of the earlier results the packet loss for Bluetooth sizes of 3 and 5 are significantly greater than that for packets of size 1. The respective probability values fo r the piconet sizes of 20, 40, 60, 80 and 100 are 55%, 68%, 75%, 82% and 88% for packets of si ze 1 and 62%, 78%, 90%, 95% and 99% for packets of size 3 and 5. The optimal choice for th e number of piconets is about 15 in which case the corresponding packet loss is about 50% for all Bluetooth packet sizes. A possible reason for the rise in the amount of initi al interference would be that in the indirect case the packets need to travel a di stance of 5 meters from the source to reach the destination and would thus have a greater chan ce of colliding which any other packet that it might encounter along its path. Also, while we would have expected this probability to be

PAGE 33

33 considerably high as compared to the direct case, the signal attenuati on caused while traveling the distance of 5 meters mitigat e the expected interference. Figure 4-5. Packet loss probability with 50 Wi-Fi nodes in the indirect case In case of 50 Wi-Fi nodes, the initial interference is 46%. The respective probabilities for the piconet sizes of 20, 40, 60, 80 and 100 are 60%, 75%, 81%, 87% and 92% for packets of size 1 and 69%, 83%, 91%, 98% and 100% for packet s of size 3 and 5.The optimal number of piconets is a little less than 15, implying that the system performance degrades slightly. Figure 4-6. Packet loss probability with 80 Wi-Fi nodes in the indirect case

PAGE 34

34 In the final case, when we have 80 Wi-Fi nodes, the initial interferen ce caused solely due to the interference betw een Wi-Fi nodes and Bluetooth inquiry scan packets is 57% which is considerably high. The respective probability va lues for the piconet sizes of 20, 40, 60, 80 and 100 are 71%, 80%, 84%, 93% and 96% for packets of size 1 and 80%, 90%, 96%, 99% and 100% for packets of size 3 and 5. Packet loss with probability close to 10 0% occurs when the number of piconets increases beyond 90 for a ll Bluetooth packet sizes. Under the present configuration, we can almost rule out the possib ility of achieving impressive system performance with Bluetooth packet sizes of 3 and 5. It is ev ident that the packet loss reaches 80% when the number of piconets rises to barely 20. This mean s that using packet sizes of 3 or 5 is totally impractical in a public environment where the number of piconets is guaran teed to exceed to way larger than 20. As evident from the above results, there is a considerable probability of loss even when there are no piconets formed in the system. With increase in the number of piconets, the probability of packet loss incr eases exponentially with maximum packet loss for large Bluetooth packets and minimum for lower packet sizes. When the number of Bluetooth devices is suffi ciently large, the system is clogged because of the swamping of packets all over the wireless medium and packet loss is inevitable for every packet transmitted. This accounts for the packet loss probability reaching 100% after as the number of piconets increase over time. An optimum value of 50% for the probabil ity of packet loss might be guaranteed by limiting the number of Bluetooth piconets to lie somewhere within 10 to 20 to obtain best performance. The system becomes practically unus able when the packet loss probability exceeds 80% with the number of piconets reaching 50.

PAGE 35

35 System Throughput Under sim ilar conditions as those for the pack et loss probabilities, the system was run for both the direct and indirect cases to measure the overall system throughput. Figure 4-7, Figure 48 and Figure 4.9 correspond to the system throughput obtained in the direct case, while Figure 410, Figure 4-11 and Figure 4-12 provide graphica l representation of the system throughput for the indirect case. As in the case of packet loss probabilities, we attempt to generalize the results to obtain a range of values for which the throughput is maximized. Figure 4-7. System throughput with 20 Wi-Fi nodes in the direct case In Figure 4-7 we note that there is a minimum amount of system throughput of 450 kbps even in the absence of any Bluetooth devi ce when the system r uns entirely using Wi-Fi devices as in case of packet lo ss probability. This is easily expl ained along the same lines as those of packet loss by the fact that the throughp ut is accounted for by the packets transmitted by the Wi-Fi nodes alone and also the initial inquiry scan packets se nt by the Bluetooth nodes prior to forming the piconet We further notice the fact that the throughput increases in a polynomial manner and attains a maximum value which is the greatest (around 3700 kbps) in case of Bluetooth packet size of 5

PAGE 36

36 and least for packet size of 1 (around 2200 kbps). This is attributed to the fact that even under the assumption that the same number of collisions occur and the same number of packets are dropped in the process, the amount of data carried by a packet of size 5 is 5 times greater than that carried by a packet of size 1; hence the total amount of data transmitted successfully by a packet of size 5 is bound to be greater than that transmitted successfully by a packet of size 1. After attaining a maximum value, the throughpu t reduces due to increased congestion in the system as explained by the concomitant incr ease in packet loss probability as shown in Figure 4-1. The maximum value is obtained for a piconet size of 20 which is in keeping with the earlier observations. As in the case of packet loss, the throughput curves for Blue tooth packet sizes of 3 and 5 are close enough to each other but drastically di fferent for the curve obtained by specifying a Bluetooth packet size of 1. The throughputs in case of packet sizes of 1 are 1500 kbps, 2250 kbps, 2220 kbps, 1900 kbps and 1500 kbps when the number of piconets rises to 20, 40, 60, 80 and 100 respectively which satisfy a polynomial relationship. The co rresponding values of throughputs for Bluetooth packets of size 3 and 5 are close to 3600 kbps, 3400 kbps, 2400 kbps, 1400 kbps and 800 kbps. One interesting thing to note is that even through a packet of size 5 results in maximum throughput it performs poorly (about 700 kbps) when the system suffers high packet losses owing to congestion. On the cont rary, a packet size of 1 retain s a throughput close to 1500 kbps even under highly congestive scenar ios which is more than twice the throughput for a packet size of 5. This is because when packets are dropped w ith a high probability, it is easier for a smallsized packet to avoid co llision and reach from source from des tination than a large packet to pass through the congested environment.

PAGE 37

37 In terms of packet sizes, an optimum choice w ould be a packet size of 3 which would not degrade the system performance to the extent as a packet size of 5 does in the face of congestion, but which would guarantee a maximum throughput wh ich is considerably larger than that obtained for packet size of 1. Figure 4-8. System throughput with 50 Wi-Fi nodes in the direct case Repeating the experiment for 50 Wi-Fi nodes we find that the overall throughput decreases owing to increase in interfer ence. The maximum achievable throughput for optimal Bluetooth packets of size 3 is about 3250 kbps which guara ntees a worst case throughput of 900 kbps when the system becomes congested. For Bluetooth packet of size 1, the value of thoroughputs corresponding to piconet sizes of 20, 40, 60, 80 and 100 are 1250 kbps, 1900 kbps, 1850 kbps, 1700 kbps and 1250 kbps. The corresponding values for Bluetooth packets of sizes 3 and 5 are approximately close to 2600 kbps, 3000 kbps, 2200 kbps, 1300 kbps and 650 kbps. The worst-case throughput in case of Bluet ooth packet size of 1 is about 1300 kbps while that for Bluetooth packet size of 5 is only 550 kbps. In keeping with our analysis so far, we find that the optimal choice for th e Bluetooth packet size is 3.

PAGE 38

38 Figure 4-9. System throughput with 80 Wi-Fi nodes in the direct case In Figure 4-9, the throughput further deteriorates and the maximum throughput becomes 3250 kbps. For the optimal packet size of 3, we find that the worst case throughput is 850 kbps while the maximum throughput is as high as 2950 kbps. The corresponding throughput values for piconet sizes of 20, 40, 60, 80 and 100 in case of Bluetooth packets of size 1 are 1100 kbps, 1800 kbps, 1850 kbps, 1500 kbps and 1200 kbps. The corresponding approximate values for the other two packet sizes are 2250 kbps, 2700 kbps, 1900 kbps, 1250 kbps and 600 kbps. Figure 4-10. System throughput with 20 Wi-Fi nodes in the indirect case

PAGE 39

39 In the indirect case, the signal attenuati on plays a significant role in performance degradation in that the maximum throughput cease s to increase significantly even when the Bluetooth packet size is increa sed from 3 to 5. Under the circumstances, the optimal choice for packet size is 3 because it yi elds the greatest worst-case throughput close to 1250 kbps although the corresponding maximum value of throughput is slightly above 2500 kbps. The corresponding throughput values for piconet sizes of 20, 40, 60, 80 and 100 in case of Bluetooth packets of size 1 are 1300 kbps, 1800 kbps, 1850 kbps, 1800 kbps and 1500 kbps. The corresponding approximate values for Bluetooh packet sizes of 3 and 5 are 2500 kbps, 2550 kbps, 2000 kbps, 1500 kbps and 1100 kbps. Figure 4-11. System throughput with 50 Wi-Fi nodes in the indirect case In case of 50 Wi-Fi nodes, the system degr ades considerably ow ing to increase in congestion. The optimal packet size of 3 results in a worst case throughput of around 1200 kbps. The maximum value of throughput for this optimal case is only 2300 kbps. The corresponding throughput values for piconet sizes of 20, 40, 60, 80 and 100 in case of Bluetooth packets of size 1 are 1200 kbps, 1700 kbps, 1650 kbps, 1500 kbps and 1300 kbps. The corresponding

PAGE 40

40 approximate values for the other two p acket sizes are 2000 kbps, 2250 kbps, 1900 kbps, 1550 kbps and 1000 kbps. An interesting observation in Figure 4-11 is the fact that the shape of the plots for Bluetooth packet sizes of 3 and 5 are somewhat flat near the maximum value. Comparing this with the previous results we observe that the sy stem reaches a temporary state of equilibrium or saturation in that the throughput lingers at the maximum valu e for a while instead of plummeting sharply as in the previous cases. Figure 4-12. System throughput with 80 Wi-Fi nodes in the indirect case In the case of 80 Wi-Fi nodes, we witness the worst case of interferen ce obtained using our experimental model. The throughput, although incr easing in a polynomial manner, has a smaller degree of variation over a range of continuously increasing values for the number of piconets thus giving rise to a shape representing a plateau. The correspondi ng throughput values for piconet sizes of 20, 40, 60, 80 and 100 in case of Bluetooth packets of size 1 are 900 kbps, 1550 kbps, 1650 kbps, 1600 kbps and 1300 kbps. The corresponding approximate values for the other two packet sizes grouped together are 1000 kbps, 1550 kbps, 1700 kbps, 1300 kbps and 700 kbps.

PAGE 41

41 From the graphical analysis of the system throughput, we notice that the optimum piconet size which guarantees maximum throughput is between 20 Also, the throughput in a congested system can be maximized by choosing an optimum Bluetooth packet size of 3. By choosing any other packet size would guarantee either a maximu m attainable throughput or a maximum worst-case throughput, but not both under the same configuration. Comparing the packet loss probability with the system throughput we observe that an increase in system throughput incu rs a high probability of packet loss. This can be explained by the fact that the greater the data transferred by the packet, the greater is the pay-off in case the packet is transmitted successfully and the greate r is the data loss in case the packet is dropped during the transmission which is an inhe rent tradeoff that cannot be averted. We can summarize our experimental result s by observing the following two facts which are established in the course of analyzing th e graphical results. 1. Best system performance may be achieved by limiting the number of Bluetooth piconets to 20. An increase in the number of piconets would degrade the system performance to the extent that it will become uns uited for practical applications when this number reaches 80. 2. An optimum choice of the Bluetooth packet si ze is 3 since it not only provides a maximum throughput value close to that obtained with a packet size of 5, but also guarantees higher throughput than that guaranteed by a packet size of 5 when the system gets congested. 3. Better throughput incurs a greater probability of packet loss which is an inevitable tradeoff that must be accepted. While the experimentation for this thesis was performed using the NS-2 simulator, an actual physical simulation would be more realistic in that it would take into consideration not only additional real-time paramete rs but also a slew of unfores een errors and imperfections encountered during runtime which a simulato r cannot guarantee in a convincing manner.

PAGE 42

42 CHAPTER 5 CONCLUSION While it has been envisioned by m any researcher s that interference in the public Industrial, Scientific and Medical (ISM) wireless band is inevitable and unavoidable in practical cases and subsequent algorithms have been proposed to redu ce if not eliminate this interference, an actual mathematical estimate for this interference w ould prove beneficial to speculate upper bounds on system performance prior to performing future wireless experiments and compare the actual results with the theoretical analysis. Our study dealt with an experimental outlook on modeling this interference using experimental results base d on the output files generated by the simulations done using the open-source Network Simulator (NS-2) software. Our analysis indicates that the best results can be achieved by selecting a packet size of 3 for the Bluetooth packets which guarantees good system performance even when the network is congested with too many transmitting nodes. Furthermore, care might be taken to limit the piconet size to an optimum value of 20 piconets in which case the best performance might be guaranteed in terms of low packet loss and high system throughput. Any modeling based on test data has its in herent limitations. The throughput obtained by this experimental approach might be substa ntially improved by cons idering certain other parameters and unanticipated r eal-time errors that are far-fet ched in a network simulation software such as NS-2 but which can be impl emented in a mobile-computing laboratory by using physical Bluetooth devices. However, it becomes practically cumbersome to physically create a network system with a large numbe r of piconets in order to analyze a worst-case scenario as accomplished in this study.

PAGE 43

43 LIST OF REFERENCES [1] L. Chen, R. Kapoor, K. Lee, M. Y. Sanadidi and M. Gerla, Audio Streaming over Bluetooth: An Adaptive ARQ Timeout Approach Proceedings of the 24th International Conference on Distributed Computing Systems Workshops (ICDCSW), 2004, pp. 196. [2] C. Cordeiro, S. Abhyankar and D. P. Agrawal, Scalable and QoS-Aware Dynamic Slot Assignment and Piconet Partiti oning to Enhance the Performance of Bluetooth Ad Hoc Networks, IEEE Transactions On Mobile Computing, Vol. 5, No. 10, October 2006, pp. 1313. [3] C. Cordeiro, S. Abhyankar, R. Toshiwal and D. P. Agrawal, A Novel Architecture and Coexistence Method to Provide Global A ccess to/from Bluetooth WPANs by IEEE 802.11 WLANs Performance, Computing, and Commun ications Conference, 2003. Conference Proceedings of the 2003 IEEE International, pp. 23. [4] C. Cordeiro and D. P. Agrawal, Employing Dynamic Segmentation for Effective Colocated Coexistence between Bluetooth and IEEE 802.11 WLANs Global Telecommunications Conference, 2002, pp. 195. [5] C. Cordeiro, D. P. Agrawal and D. H. Sadok, Interference Modeling and Performance of Bluetooth MAC Protocol IEEE Transactions On Wireless Communications, Vol. 2, No. 6, November 2003, pp. 1240. [6] O. Dousse, F. Baccelli and P. Thiran, Impact of Interferences on Connectivity in Ad Hoc Networks, IEEE/ACM Transactions on Networking, Vol. 13, No. 2, April 2005, pp. 425 436. [7] C. T. Ee, S. Shenker, B. Chun and W. Hong, Interference Avoidance in Wireless Multihop Networks Second Annual IEEE Communications Society Conference on Sensor and Ad Hoc Communications and Networks, 2005. [8] N. Golmie, N. Chevrollier and I. ElBakkouri, Interference Aware Bluetooth Packet Scheduling Global Telecommunications C onference, Vol. 5, 2001, pp. 2857. [9] N. Golmie and F. Mouveaux, Impact of Interference on the Bluetooth Access Control Performance: Preliminary Results IEEE P802.15Working Group for Wireless Personal Area Networks, September 2000. [10] N. Golmie, O. Rebala and N. Chevrollier, Bluetooth Adaptive Frequency Hopping and Scheduling Military Communications C onference (MILCOM), 2003, pp. 1138. [11] R. Hekmat and P. V. Mieghem, Interference in Wireless Multi-hop Ad-hoc Networks ACM Digital Library, Vol. 10, Issue 4, July 2004, pp. 389. [12] I. Howitt, IEEE 802.11 and Bluetooth Coexistence Analysis Methodology 53rd IEEE Vehicular Technology Conference, vol. 2, May 2001, pp. 1114.

PAGE 44

44 [13] I. Howitt and F. Awad, Optimizing IEEE 802.11b Packet Fragmentation in Collocated Bluetooth Interference IEEE Transactions On Communications, Vol. 53, No. 6, June 2005, pp. 936. [14] I. Howitt, J. Gutierrez and V. Mitter, Tools for Evaluating Bluetooth Coexistence with Other 2.4GHz ISM Devices IEEE P1451.5 Project, Wireless Sensor Working Group. [15] J. Lansford, A. Stephens and R. Nevo, Wi-Fi (802.11b) and Bluetooth: Enabling Coexistence IEEE Network September/October 2001, pp. 20. [16] L. Ophir, Y. Bitran and I. Sherman, Wi-Fi (IEEE802.11) and Bluetooth Coexistence: Issues and Solutions 15th IEEE International Sym posium on Personal, Indoor and Mobile Radio Communications (PIMRC), 2004, pp. 847 852. [17] G. Pasolini, Analytical Investigation on the Co existence of Bl uetooth Piconets IEEE Communications Letters, Vol. 8, No. 3, March 2004, pp. 144. [18] E. Vergetis, R. Gurin, S. Sarkar and J. Rank, Can Bluetooth Succeed as a Large-Scale Ad-Hoc Networking Technology IEEE Journal on Selected Areas In Communications, Vol. 23, No. 3, March 2005, pp. 664. [19] J. Wang and J. C. L. Liu, Optimizing Uplink Scheduling in an Integrated 3G/WLAN Network Special Issue of International Jour nal of Wireless and Mobile Computing (IJWMC) on The Integration of 3G and WLAN Networks, 2007, pp.288. [20] J. Wang, J. C. L. Liu and Y. Cen, Handoff Algorithms in Dynamic Spreading WCDMA System Supporting Multimedia Traffic IEEE Journal on Selected Areas in Communications, Vol. 21, No. 10, December 2003, pp. 1652.

PAGE 45

46 BIOGRAPHICAL SKETCH The author is a m asters student in the Department of Comp uter and Information Science and Engineering at the University of Florida, USA, where he ha s been enrolled since the fall of 2007. He has been working on this thesis toward partial fulfillment of his masters degree. The authors present research interests ar e centered on high-speed wired and wireless networks, wireless ad-hoc networks, network flow and routing and multimedia support over wireless networks.