Smart Antenna Systems and Wireless Lans The International Series in Engineering and Computer Science

SMART ANTENNA SYSTEMS AND WIRELESS LANS
THE KLUWER INTERNATIONAL SERIES IN ENGINEERING AND COMPUTER SCIENCE
SMART ANTENNA SYSTEMS
AND WIRELESS LANS
GARRET T. OKAMOTO
Assistant Professor and W. J. Nicholson Fellow Department of Electrical Engineering Santa Clara University Santa Clara, CA 95053
USA
KLUWER ACADEMIC PUBLISHERS

NEW YORK, BOSTON, DORDRECHT, LONDON, MOSCOW
eBook ISBN: Print ISBN:
0-306-47323-2 0-792-38335-4
2002 Kluwer Academic Publishers New York, Boston, Dordrecht, London, Moscow
All rights reserved
No part of this eBook may be reproduced or transmitted in any form or by any means, electronic, mechanical, recording, or otherwise, without written consent from the Publisher
Created in the United States of America
Visit Kluwer Online at: and Kluwer's eBookstore at:
http://www.kluweronline.com http://www.ebooks.kluweronline.com
This book is dedicated to my parents, Matsuo and Betty Okamoto
This page intentionally left blank
Table of Contents
Preface
Chapter 1. Introduction
xi
1
1.1 Smart Antenna Systems . . . . . . . . . . . . . . . . . . . . . 1.2 Rapid Growth of Wireless LAN Systems . . . . . . . . . . . . 1.3 IEEE 802.11 and the Smart Wireless LAN System . . . . . .
Chapter 2. Background for Smart Antenna Systems
1 2 4
7
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . .
2.2 Spatial Signature Definition . . . . . . . . . . . . . . . . . . . 7 2.3 Direction-of-Arrival Estimation . . . . . . . . . . . . . . . . . 11 2.4 Beamforming Techniques for Downlink . . . . . . . . . . . . . 11
Chapter 3. IEEE 802.11 Wireless LAN Standard 17
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 3.2 Background on the IEEE 802.11 Wireless LAN Standard . . . 17
3.3 MAC 3.3.1 3.3.2 3.3.3
3.3.4 3.3.5 3.3.6 3.3.7 3.3.8 3.3.9
Layer of the 802.11 System . . . . . . . . . . . . . . . . The Challenges of Wireless Networking . . . . . . . . . Protocol Design Objectives for IEEE 802.11 . . . . . . The 802.11 Distributed Coordination Function (DCF) and Point Coordination Function (PCF) . . . . . . . . The 802.11 Power Saving Mechanism . . . . . . . . . . The 802.11 Network Synchronization . . . . . . . . . . The 802.11 Network Operation . . . . . . . . . . . . . . Operation of the 802.11 D C F . . . . . . . . . . . . . . . Operation of the 802.11 PCF . . . . . . . . . . . . . . . Smart Antenna Systems and the 802.11 MAC Protocol
19 19 28 31 34 37 39 42 43 44
3.4 PHY Layer of the 802.11 System . . . . . . . . . . . . . . . . 46
3.5 Examples of Current Wireless LAN Products . . . . . . . . .
50
3.5.1 The Lucent WaveLAN DSSS System . . . . . . . . . . 50 3.5.2 The Proxim RangeLAN FHSS System . . . . . . . . . . 52
3.5.3 The RadioLAN BackboneLINK Narrow Band System . 54 3.5.4 The JVC Office LAN IR System . . . . . . . . . . . . . 56
Chapter 4. Smart Wireless LAN System Design 59
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . .
59
4.2 SWL Media Access Control Design . . . . . . . . . . . . . . .
60
4.3 SWL Physical Layer Design . . . . . . . . . . . . . . . . . . . 63
4.4 SWL Dynamic Slot Assignment Algorithms . . . . . . . . . . 68 4.5 SWL Timing Synchronization Algorithms . . . . . . . . . . . 70
Chapter 5. Experimental Setup for Smart Antenna Systems
79
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.2 900 MHz Testbed . . . . . . . . . . . . . . . . . . . . . . . . . 79 5.2.1 Antenna Array . . . . . . . . . . . . . . . . . . . . . . . 82 5.2.2 RF/IF Unit . . . . . . . . . . . . . . . . . . . . . . . . 83
5.2.3 Distribution Boxes . . . . . . . . . . . . . . . . . . . .
83
5.2.4 MUX/DEMUX Boxes . . . . . . . . . . . . . . . . . . . 85

5.2.5 High Speed I/O boards and Sparc 10 Workstation . . . 85
5.2.6 Stability of the Smart Antenna Testbed . . . . . . . . 87 5.3 1.8 GHz Base Station . . . . . . . . . . . . . . . . . . . . . . . 87 5.3.1 Transmit/Receive Board . . . . . . . . . . . . . . . . . 89 5.3.2 SHARC Board . . . . . . . . . . . . . . . . . . . . . . 92 5.3.3 Vocoders and Telephone Handsets . . . . . . . . . . . . 93 5.3.4 Backplane . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.3.5 Antenna Array . . . . . . . . . . . . . . . . . . . . . . . 96
5.4 1.8 GHz Handset . . . . . . . . . . . . . . . . . . . . . . . . . 96

viii
5.4.1 Transmit/Receive Channel . . . . . . . . . . . . . . . . 96
5.4.2 DSP Chip . . . . . . . . . . . . . . . . . . . . . . . . .
98
5.4.3 Handset Antenna . . . . . . . . . . . . . . . . . . . . . 98 5.5 Experimental Environment . . . . . . . . . . . . . . . . . . . . 98 5.6 Low Earth Orbit Satellite Emulation . . . . . . . . . . . . . . 99
Chapter 6. Experimental Results for Smart Antenna Systems 105
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

6.2 Calculation of Spatial Signature Variations . . . . . . . . . . . 106
6.3 Spatial Signature Stability for Stationary Users . . . . . . . . 107
6.4 Spatial Signature Variation with Frequency Shift . . . . . . . 113
6.5 Spatial Signature Variation with Displacement . . . . . . . . . 113

6.6 Spatial Signature Correlation for Closely Spaced Transmitters 119
6.7 Spatial Signature Variation Due to Small Displacement in Different Multipath Environments . . . . . . . . . . . . . . . . . 120
6.8 Multipath Angle Spread . . . . . . . . . . . . . . . . . . . . . 125 6.9 Feasibility of Downlink Beamforming . . . . . . . . . . . . . . 130
6.10 Fading Reduction and Diversity Gain . . . . . . . . . . . . . . 133

6.11 Beamforming Algorithms for SIR Improvement . . . . . . . . 141 6.12 Low Earth Orbit Satellite Emulation Results . . . . . . . . . 147 6.13 Summary of Experimental Results . . . . . . . . . . . . . . . 153
Chapter 7.
Computer Simulation Results
157
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157 7.2 SIR Performance for Beamforming Algorithms . . . . . . . . . 157
7.3 Dynamic Slot Allocation for Worst-Case SINR . . . . . . . . . 161 7.4 Dynamic Slot Allocation for Capacity Increase . . . . . . . . . 166 7.5 Timing Synchronization Results . . . . . . . . . . . . . . . . . 171
7.6 Physical Layer Simulation Results . . . . . . . . . . . . . . . . 179 7.7 Media Access Control Layer Simulation Results . . . . . . . . 185
ix
Chapter 8. Conclusion Bibliography Index
193 197 205
Preface
This book concerns two major topics, smart antenna systems and wireless local-area-networks (LANs). For smart antenna systems, it discusses the mechanics behind a smart antenna system, the setup of a smart antenna experimental testbed, and experimental and computer simulation results of various issues relating to smart antenna systems. For wireless LAN systems, it discusses the IEEE 802.11 worldwide wireless LAN standard, the operation of a wireless LAN system, and some of the technical considerations that must be overcome by a wireless LAN system designer. These two topics are combined in the discussion of the Smart Wireless LAN (SWL) system, which was designed to achieve the benefits which smart antenna systems can provide for wireless LAN systems while still remaining compatible with the 802.11 wireless LAN standard. The design of SWL calls for the replacement of the conventional wireless LAN base station (which are called access points in the 802.11 documentation) with an SWL base station, while leaving the individual terminal operation as unchanged as possible.
The material in this book is directly relevant to the design of smart antenna systems in wireless communications, including other applications besides wireless LANs such as terrestrial cellular systems, satellite cellular systems, wireless data distribution systems, and any advanced system that requires studies on the effects of advanced topics such as channel propagation and beamforming. This book details how smart antenna systems can be used to increase the capacity, decrease the effects of fading, increase battery life, and increase the range (the last two benefits due to the antenna diversity gain) of any wireless communications system. Of course, this book is especially relevant to the design of any wireless LAN system that wishes to utilize those benefits from smart antennas.
This book is organized as follows. Chapter 2 gives an overview of smart antenna and Space-Division-Multiple-Access (SDMA) concepts. The necessary background of a smart antenna system, including the definition of the spatial signature, and direction-of-arrival (DOA) estimation techniques are given. The various beamforming algorithms first studied for the SWL system are also defined and discussed. Chapter 3 presents the IEEE 802.11 wireless LAN standard, providing a brief background of the standard as well as descriptions of its MAC and PHY layers. Examples of leading 802.11 products are also given and the strategies their manufacturers take are also discussed. Chapter 4 gives a detailed description of the SWL protocol design, from the media access control to the
physical layer design. The dynamic slot assignment and timing synchro-
nization algorithms are discussed in greater detail due to their complex nature and specialization for the SWL system. Chapter 5 presents the
experimental setup for two smart antenna testbeds built at The Univer-
sity of Texas at Austin, operating at 900 MHz and 1.8 GHz. The key individual components of a smart antenna testbed are shown and discussed, and the indoor and outdoor experimental environments are also described. Chapter 6 presents the experimental measurements of the spatial signature variation with time, to study the stability of the spatial signature for a stationary transmitter, and displacement, including the scenario in which there are closely spaced transmitters. The spatial signature variation with frequency shift is also discussed, which shows that Frequency-Division-Duplex (FDD) systems are not ideally suited for smart antenna systems. The experiments which show the effectiveness of smart antenna systems in reducing fading and providing antenna diversity gain are also presented. The effectiveness of various beamforming algorithms in increasing the capacity of a system is also examined. The multiple angle spread, feasibility of downlink beamforming, and Low Earth Orbit (LEO) satellite emulation results are also presented and discussed. Chapter 7 presents the simulation results for the SWL system. The physical layer simulation results include the results for the SIR performance for the beamforming algorithms, the dynamic slot allocation
xii
results, the timing synchronization results, and the overall physical layer simulation results. The media access control layer simulation results include comparisons of the network utilization and voice delay failures for the SWL and IEEE 802.11 networks. Chapter 8 concludes the book.
This book is intended for designers and would-be designers of wireless communications systems. To limit the scope of this book to manageable proportions, the number of topics covered and the depth of coverage on some of those topics had to be limited. In the case of the computer simulation results for the MAC and PHY layers of the SWL system, for example, only representative results were given which allow for the reader to make their own conclusions as general as possible. For example, simulation results were given which as closely as possible related to the scenarios used in the experiments, to demonstrate how well the simulation results agreed with the experimental results, which both
validated the results and allowed for conclusions to be drawn from their
differences. Thus, simulation results which were too specific for the SWL or any specific smart antenna system were limited, as their usefulness to a reader desiring a different application would be limited. As another example, the discussion of the IEEE 802.11 standard is restricted to just providing the major details which influence the design of a wireless LAN system, since giving a complete description of the standard would require too much space and further information on the standard is readily available to anyone who wishes to learn more. The information provided about the standard should be sufficient for anyone seeking to acquire a general idea of how those systems work and are being implemented today, but this book is concerned about a potential future application for 802.11 and other wireless LAN systems.
The emphasis on topics important to designers results in a more detailed treatment of some topics than is traditional in academic textbooks, such as the experimental setup of a smart antenna testbed. In fact, the setup of two smart antenna testbeds (900 MHz, 1.8 GHz) are provided because information such as that is not generally available and
xiii
not duplicated in other books to my knowledge. Another example of a non-traditional topic is the wide range of experimental results provided from those testbeds, to demonstrate the feasibility of using a smart antenna system for wireless communications applications, with the SWL system designed with those results in mind. Many examples are given which relates the design of the SWL system to the knowledge gained from the experiments. This book also covers advanced topics such as dynamic slot assignment, timing synchronization, and several problems which smart antenna systems face that dont exist in conventional wireless communications systems. This book is based on my dissertation at The University of Texas
at Austin. However, the material included here is broader in scope and

has more depth because I added chapters in areas of a more general interest (such as the IEEE 802.11 wireless LAN standard chapter and the experimental setup chapter) and I was fortunate to receive the contributions of a number of my former colleagues at The University of Texas at Austin. I was also able to include in this book numerous experimental results that I collaborated on but didnt include in my dissertation because I was not the lead investigator on those experiments. This book is suitable as a first- or second-year graduate textbook,
for a class covering advanced wireless communications or signal processing topics, or for any wireless communications professional in industry
interested in these topics. Smart antenna systems are a fairly advanced
topic in wireless communications, but this book was written so that it

should be able to be understood by even those who dont have too much of a background in these advanced signal processing topics. I attempted to make the book more attractive to both audiences through the inclusion of many practical examples and a practical relation between design and the experimental and simulation results. This book is not intended to be the definitive work on either of its major topics, smart antenna systems and wireless LANs. Rather, it should provide the reader with a solid base of understanding on both
xiv
topics, as well as an understanding of the SWL system which I designed. Many of the subjects in this book would require entire books of their own to be fully explored, but this book should provide enough information
for a reader to further explore areas in which more advanced study is desired. I would like to express my deepest appreciation to Professor Guanghan Xu, my thesis advisor, for his guidance throughout the course
of my doctoral research. I would also like to express my appreciation to
Shiann-Shiun Jeng for his invaluable contribution to our collaborative

efforts, much of which is presented in the experimental results chapter. He also wrote about the 900 MHz testbed for the experimental setup chapter and contributed to the smart antenna background chapter.
I would like to thank Lars K. Hansen for writing about the experimental setup of the 1.8 GHz testbed, Adnan Kavak for writing about his LEO experiments, and John Stine for writing about the IEEE
802.11 MAC description. I would also like to thank my colleagues SangYoub Kim, Weidong Yang, Michael Montgomery, Murat Torlak, Hui Liu, Liang Dong, Roberto Vargas, Jr., Wei-Lin Yang, Ching-Fong Su, Alberto Arredondo, Kapil Dandekar, and Tae-Jin Lee for their help and valuable discussions. I owe a debt of gratitude to the members of my Ph.D. committee
for all of their suggestions which improved my dissertation and consequently this book, Guanghan Xu, Mario Gonzalez, Gustavo de Veciana,
Edward Powers, and Wolfhard Vogel. Special thanks go to my commit-
tee member (and longtime supervisor at the Jet Propulsion Laboratory) James Lesh, who provided the most comprehensive proofreading of my
dissertation that I received. While many of these colleagues and others have pointed out many errors and omissions, any remaining errors are of course my full responsibility.
Finally, I would like to thank my parents, sister Nan, and brotherin-law David for their help. Without their continual support and encourxv
agement over the years, none of my accomplishments would have been
possible. I am truly blessed to have them in my life.

I hope the result is a readable and useful book, and always appreciate comments and suggestions from the readers. Garret T. Okamoto Santa Clara, California
xvi
Chapter 1
Introduction
The use of wireless communications systems has exploded throughout the world. The exponential growth of wireless communications systems and the limited bandwidth available for those systems has created problems which all wireless providers are working to solve. One potential solution to the bandwidth limitation is the use of smart antenna systems, a concept initially developed by the military but now a field that has attracted growing interest for commercial wireless communications systems. This book will explain how smart antenna systems work, provide detailed information on two smart antenna experimental testbeds, and examine extensive experimental and computer simulation results, with a particular interest in adapting smart antenna systems for wireless LANs. Wireless LANs are particularly suited for smart antenna systems because most of the users are stationary (or at least not moving very fast) during operation, the indoor environment results in widely uncorrelated spatial signatures for even closely spaced users, and the limited bandwidth allocated makes the capacity and throughput advantages of smart antenna systems particularly critical.
1.1
Smart Antenna Systems
In wireless communications environments [1, 2, 3, 4, 5, 6, 7], a smart antenna system provides the following major benefits. First, the effect of multipath fading in wireless communications environments can be significantly reduced [8, 9]. Since the reliability and quality of a wire-
less communications system can strongly depend on the depth and rate
Introduction
of fading, this reduction of the variation of the signal (i.e., fading) greatly
enhances system performance. Second, handsets of a smart antenna system have longer battery life because the power required to transmit to
the base station is lower than that for a conventional system. This is because the antenna array at the base station achieves antenna diversity gain [10, 11], which reduces the required signal level of the handset. For example, if we achieve a 9 dB antenna diversity gain with an 8-element base station antenna array, then the required transmission power of the
handset is 9 dB less than that needed for a conventional system. Third, base stations using antenna arrays have a larger range than conventional base stations. This increased range is also due to the diversity gain of
a smart antenna system. Consequently, there is a tradeoff between the
advantages of longer battery life and increased range for a smart antenna system, with the optimal design decision depending on the requirements of the communications environment. The advantages of longer battery life and larger range result in a reduced cost of infrastructure installa-
tion for smart antenna systems. Finally, the smart antenna system can
be used to increase the system capacity [12], which is limited by the Signal-to-Interference Ratio (SIR). A smart antenna system can significantly improve the SIR of a wireless communications system, and thus significantly increase the capacity of the system [13, 14].
1.2
Rapid Growth of Wireless LAN Systems
The use of wireless local-area-network (LAN) systems has been increasing rapidly as prices have come down and a common standard has emerged. This standard is the IEEE 802.11 wireless LAN standard [15, 16, 17], which has been adopted by virtually all manufacturers
in North America (many of whom have an 802.11-compliant product line

in addition to a product line based on some other standard such as the OpenAir Standard) and has also been adopted in Europe and Japan. Wireless LAN systems have many advantages over wired systems, with users generally citing four major reasons for their use of wireless
LANs. The first, and most obvious, advantage of a wireless system is

mobility, as portable or laptop terminals can easily move around in a
Rapid Growth of Wireless LAN Systems
wireless system [18]. Most of the early wireless LAN users were those that placed a premium on mobility. Mobility is especially prized in inventory control in store and warehouse environments, point-of-sale terminals, rental car check-in, hospitals, universities, etc.
The second major advantage of a wireless LAN system is adaptability, as many environments are expensive or impractical for the cabling required in a wired system. Adaptability is critical for environments which change rapidly and require continual reconfigurations of the network (manufacturing floors, trading floors on stock exchanges, conventions and trade shows, etc.) or environments in which drilling through the walls and ceilings is undesirable or impossible (historic buildings, buildings with marble walls and floors, buildings with bad asbestos problems, etc.) [18]. Classic examples of environments that are impractical for cable are the National Gallery of Art in Washington, D.C. (which uses Lucents WaveLAN) and the space shuttle [19] (which uses Proxims RangeLAN2). The third significant advantage of a wireless LAN system is that there are no wiring costs for the terminals. This is significant because studies have shown that terminals can move 1.5-3 times per year [20], costing $200-$3000 per move not including the loss in productivity by the user disconnected from the network. In a four year study, Motorola found that equipment additions and relocations affected more than 40% of users annually [21]. A study also found that 70% of LAN problems were due to faulty wiring [21]. Examples of large organizations that switched to wireless LAN systems because they found that it would save them money over time are Centura Bank and Banco Popular Espanol.
Finally, wireless LAN systems have a huge advantage in terms of installation speed, where an 18-node wireless LAN system can be set up in as quick as an hour while a wired system could take months. A classic example of the critical nature of speedy installation is in disaster recovery scenarios [22], like how Autoliv (a Swedish company that is one of Europes largest provides of auto safety equipment) installed Lucents
WaveLAN to enable them to quickly recover from a fire that caused
Introduction
extensive damage to over 65,000 square feet of production and storage facilities.
These advantages have resulted in the tremendous growth, especially over the past 6 years, of wireless LAN systems worldwide. In 1989, there were 8000 wireless LAN nodes in the world, which grew to 28,000 at the start of 1992 [23]. Due to this rapid growth, and recognizing that because of the lack of standards no two systems from different manufacturers were compatible with each other, the IEEE 802.11 committee was formed [24]. Their charter was to create the wireless version of IEEE 802.3, commonly known as Ethernet. There was widespread industry support for the creation of a standard, and because of a variety of reasons including the positive response by corporate buyers to the concept of a common standard and a drop in price of existing products, there were 180,000 wireless LAN units in use by the end of 1992 [25]. This rapid growth has continued, with the past 6 years seeing a compound annual increase of wireless LAN systems of over 50% [19, 23, 25]. The wireless LAN worldwide market was estimated at $100 million per year in 1995 [26], and at $200 million per year in 1998 [27]. This pace is projected to continue into the future, with wireless LAN sales projected to top $1 billion per year by the year 2000 [19, 28].
1.3
IEEE 802.11 and the Smart Wireless LAN System
The current 802.11 standard has many limitations in throughput, capacity, delays, power consumption, and multimedia capability that limits the flexibility and usefulness of the systems. Because the bandwidth allocated in the standard is limited and only one user is allowed to transmit at a time in a frequency band [24], the throughput for each user is limited and users can experience long delays. While the recently selected proposal for the fast version of 802.11 may eliminate many of these problems, others may not be able to be fully solved with conventional wireless LAN systems even when running at the proposed 11 Mbps (megabits per second) maximum data rate. In addition, the standard was designed to accommodate only users with conventional data traffic and no provisions were made for multimedia or real-time applications.
IEEE 802.11 and the Smart Wireless LAN System
Not only is the throughput too low for applications such as voice when the system gets loaded, but the standard does not include the priority designations that are required for real-time applications.
To alleviate these limitations, this book proposes the Smart Wireless LAN (SWL) system [14, 29, 30, 31, 32, 33, 34, 35], which achieves throughput multiplication over the 802.11 protocol by exploiting the rich spatial diversity existing among spatially separated terminals. Spatial diversity is demonstrated by the amplitude and phase pattern of the data vectors received by an antenna array. Each transmitter location has a unique pattern, also called a spatial signature. Once the spatial signatures are acquired, different co-channel signals can be separated based on their unique spatial signatures through beamforming. Therefore, we can increase the number of virtual time slots by enabling multiple terminals to transmit in the same time slots without significantly interfering with each other. In addition to the increase in throughput, this new protocol has additional features such as simple implementation, adaptability to multimedia traffic with diverse bandwidth requests, network security, guaranteed fairness in bandwidth sharing, and ease of adaptation to the 802.11 wireless LAN standard. Also, this protocol adds priority level designations to the packets to allow for delay-sensitive communication links for multimedia applications such as voice or video. A combination of throughput performance evaluation, computer simulations, and RF experimental studies demonstrate the feasibility and benefits of the SWL system. SWL is an adaptation of the Space-Division-Multiple-Access (SDMA) technique (also referred to as the smart antenna concept) for wireless LAN systems and was designed to be compatible with the IEEE 802.11 protocol, so the individual 802.11 terminals would not have to be changed significantly. The only new requirements of this protocol are an array of antennas installed at a base station and advanced signal processing software to process the received data.
Chapter 2
Background for Smart Antenna Systems
2.1
Introduction
Smart Antenna systems provide numerous major benefits in wire-
less communications environments. These include reducing multipath fading, increasing system capacity, extending battery life of the terminals and the range of a base station, with the last two benefits due to the diversity gain achieved by the antenna array. To understand how smart antenna systems work, the mechanics behind them must be studied. While a full explanation of smart antenna systems and their applications would require many books to provide a complete explanation, this chapter provides the mathematical background explaining the fundamentals behind smart antenna systems.
2.2
Spatial Signature Definition
A diagram of a six-element uniform linear antenna array with two impinging wavefronts is shown in Figure 2.1. The antenna elements are spaced a uniform distance apart, typically 1/2 of the transmission wavelength. Figure 2.1 and the following discussion pertains to the uplink case of multiple (g) wireless terminals transmitting to a base station.
To develop the equations for our model, first let us consider the signal received at one of the antenna elements, r ( t ) , from a single source
This chapter was written in collaboration with S.-S. Jeng of Chung Yuan Christian University, Taiwan.
1
s(t),
where s(t) is the complex envelope of the transmitted baseband signal,
is the carrier center frequency, and r(t) will serve as our reference signal. The complex envelope s(t) can be represented as
where
is the sequence of complex data symbols, p(t) is a real pulse,
and is the pulse duration. Let us assume for now that we only have a direct path signal and no multipath components. The received signal x(t) at a neighboring antenna element with respect to the reference signal r(t) in the presence of noise n(t) is then
where
is the propagation delay between the reference sig-
nal and the antenna element,
is the received signal without noise,
Spatial Signature Definition
is the angle of arrival of the received signal, and c is the speed of light. For most wireless communications systems, the signal has a maximum bandwidth of 1 MHz or less, which is much smaller than its carrier frequency of around 900 MHz or 1.8 GHz. For the SWL and IEEE 802.11 wireless LAN systems, the signal has a maximum bandwidth of about 22 MHz, which is much smaller than its carrier frequency of 2.4 GHz. Experimental results have also shown that the delay spread of the significant multipath signals is well under the chip interval (approximately 100 ns) of the SWL and 802.11 systems. Additionally, if a multipath signal is delayed more than a chip interval, that signal is reduced by the Barker despreading operation so that it appears like noise to the system. Therefore, the narrowband signal model applies here, and Substituting the narrowband signal envelope yields
A snapshot vector model of the received signal due to one incident source can then be represented as
where are the propagation delays between the reference and the first to antenna elements, where M is the number of elements in the antenna array. For a uniform linear array,
where Note that at least one of the values for will be zero because the reference element has no propagation delay relative to itself. If now we assume that the signal has total signal components, consisting of a direct path signal and multipath components, our received signal is
10
where the first component is assumed to be the direct path component, contains the complex phase and amplitude information of the direct
path or multipath component, and is the angle of arrival of the component. Note that the multipath components will have different delays. Since wireless LAN signals are transmitted in a short range
local communications environment, the time delays of their significant
multipath components, are typically smaller than the chip duration of the signal, which is roughly 100 ns [36, 37]. Hence, is typically true for the multipath components in addition to the direct
path component (which was shown earlier). Our received signal can now be expressed as
multipath
where a is the spatial signature associated with the emitter defined as
If we now assume that we have time, our received signal is
terminals transmitting at the same
where is the spatial signature associated with terminal k and is the complex envelope of the transmitted baseband signal from terminal
k. Written in vector form, the received signal is
where
and
denotes the array re-
sponse matrix with columns associated with the spatial signature of each transmitted signal.
Direction-of-Arrival Estimation
11
2.3
Direction-of-Arrival Estimation
As explained in Equation (2.10),
the spatial signature is a linear combination of the different array response vectors for the direct path and the multipath signals. It obviously depends on the directions-of-arrival (DOAs) of the direct path and multipath components. Therefore, the effect of multipath DOAs on spatial signatures (SS, which abbreviation will also be used to denote its singular version, spatial signature) was studied experimentally in [38]. The high resolution direction finding algorithm, ESPRIT [39], was used to
find direct path and multipath DOAs in our experiments and computer
simulations.
Since multipath signals are sources coherent with the direct path, the signal eigenvectors will fail to span the signal subspace, and the resulting loss of rank in the signal subspace causes the ESPRIT algorithm to fail. In order to restore the dimensionality of the signal subspace, the forward and backward spatial smoothing scheme [40, 41] was used to decorrelate the coherence among multipath signals. The penalty for using the extended approach is that it can only estimate up to (2/3)M DOAs of multipath and direct path components, where M is the number
of antenna elements.
2.4
Beamforming Techniques for Downlink
The objective of the downlink is to selectively transmit signal power so that all signals with the same information can be constructively combined at a certain users receiver whereas undesired interference and noise can be nulled or suppressed. In a Time-Division-Duplex (TDD) system, the uplink and downlink schemes share the same carrier. Winters [42] proposed that the antenna array transmit the same pattern, i.e., the
12
Backgroundfor Smart Antenna Systems
vector
back to the mobile users, where
Hermitian
transpose. In this case, the channel receives the coherent sum of the signals transmitted by different antennas, while it receives the incoherent sum of the other co-channel signals
DOA-based beamforming [43], as shown in Figure 2.2, computes the downlink weight vectors based on the array response vectors, while
13
SS-based beamforming, as shown in Figure 2.3, uses the uplink spatial signatures. Four beamforming techniques among these methods were evaluated via computer simulations for the SWL system in terms of their SIR performance. They were:
1. The dominant DOA beamforming algorithm. This approach first captures the uplink spatial signatures and then finds the DOAs of the received signals using subspace based techniques such as MUSIC [44, 45] and ESPRIT. The amplitudes associated with the DOAs are also estimated. The DOA with the maximum amplitude, is selected and its array response vector is chosen as the downlink weight vector.
2. The pseudoinverse DOA algorithm. This approach is similar to the dominant DOA technique except that we take the pseudoinverse
of the array response vectors of all the DOAs instead of just that
for the DOA of the desired user. This method places nulls in all DOAs except for our desired user, which should minimize interference. To illustrate this method, suppose that the mobile unit has one direct path signal and a multipath signal with DOAs and respectively. Thus, its downlink spatial signature is . To simplify presentation, we assume that there are only two independent sources and . If the weight vectors for these two signals are and then the signal received by the first user is
is designed such that and and then . Hence, even if we transmit two co-channel signals, mobile user 1 only receives its desired signal . An intuitive explanation of this result is that we design a
If
14

weight vector for such that the transmission pattern of the antenna array has nulls in all the DOAs except . Similarly, is designed such that the pattern has its nulls in all the estimated DOAs except . Thus, user 2 only receives its desired signal, i.e., . This strategy is illuminated in Figure 2.2.
3. The complex conjugate spatial signature algorithm. This approach is the same technique as proposed by Winters. We capture the uplink spatial signatures and use it to generate the downlink weight vectors by taking the complex conjugate of the uplink spatial signatures. The main objective of this approach is to maximize the signal power or the signal-to-noise ratio (SNR). This method does not try to null out directional interferences.
4. The pseudoinverse spatial signature algorithm. This approach is similar to the complex conjugate spatial signature technique except that we generate the downlink weight vectors by taking the pseudoinverse of the uplink spatial signatures. The pseudoinverse weight vector, is based on the uplink spatial signatures of the signal sources such that . The weight vector adjusts the relative phase and amplitude of the direct path and multipath components so that the signals are exactly canceled out at the location of the signal-of-interest (SOI). For the same example as in the pseudoinverse DOA case, we design the weight vector such that and . Hence, the signal received at the first user is
However, the array pattern corresponding to placing nulls at all the DOAs associated with . Although requires is that
is not simply and all it
It is not necessarily true that is orthogonal to both and a as shown in Figure 2.3. An intuitive understanding
15
of Equation (2.17) is that we design to control the phase and amplitude of the two paths such that they exactly cancel out at the location of mobile user 1.
Chapter 3
IEEE 802.11 Wireless LAN Standard
3.1
Introduction
The IEEE 802.11 wireless LAN standard was created to enable worldwide multi-vendor interoperability, with the standardization of components resulting in a drop in prices (with the new 802.11-based products being offered already less expensive than past proprietary offerings [46]) and product diversity in the mobile-computing marketplace. This in turn should fuel the already rapid increase in usage of wireless LAN systems worldwide. IEEE 802.11 is the wireless version of IEEE 802.3, commonly known as Ethernet, though 802.11 is more complicated than Ethernet or Token Ring because of the unreliable and unpredictable radio medium.
3.2
Background on the IEEE 802.11 Wireless LAN Standard
The IEEE 802.11 wireless LAN standard was developed by wireless LAN experts representing 42 different companies [47], chaired by Vic Hayes of Lucent since the project was begun in 1990. The standard went through six draft versions, and the standard received approval on June 26, 1997. The proposal for the fast version of 802.11 was selected on July 9, 1998, with final approval of this new standard possibly coming in 1999. Of course, published reports said that the current 802.11 standard would be delivered by the committee by as early as 1993 [48], so caution
This chapter was written in collaboration with J. Stine of the University of Texas at Austin.
1
18
should be used before counting on the new standard to be completed by any specific date.
It should be pointed out that this standard just defines two lay-
ers (MAC, PHY) of the 7-layer open systems interconnect model defined by the International Standards Organization. The 802.11 standard has a limited scope and doesnt define all necessary interfaces. For example, the standard doesnt define protocols that allow one access point to hand-off an active connection to another access point from another
vendor when the client node roams between cells [49]. Hence, customers that need roaming capabilities would have to buy all of their access points from a single manufacturer unless a solution was found. To combat this
problem, Lucent, Aironet, and Digital Ocean collaborated to develop
the inter-access point protocol (IAPP) specification [50]. The IAPP will extend the 802.11 standard by covering higher-level layers such as the logical link control that facilitates inter-access point communications (allowing for access points from different manufacturers to communicate with each other), creating a comprehensive roaming protocol. IBM has also stated its support for the IAPP, which extends the usefulness of the 802.11 standard by allowing for true multi-vendor interoperability among 802.11 systems. Of course, this interoperability wont allow for direct sequence spread spectrum (DSSS) systems to work with frequency hopping spread spectrum (FHSS) systems, a problem that exists due to the incompatible natures of those two types of spread spectrum systems, even if both the DSSS and FHSS systems are 802.11-compliant. The OpenAir Standard created in 1996 under the leadership of the Wireless LAN Interoperability Forum is the only major commercial alternative to 802.11, with the major selling point of the OpenAir Standard being the interoperability of products from numerous manufacturers. More than 30 manufacturers are building interoperable products for the OpenAir Standard. This standard is based on FHSS systems, so this enables FHSS manufacturers to be able to provide multi-vendor interoperability with an existing and widespread alternative to the 802.11 standard. Hence, FHSS manufacturers have the option of utilizing either the IAPP or OpenAir Standards to be able to provide interoperable products for wireless LAN systems.
MAC Layer of the 802.11 System
19
3.3
The developers of the IEEE 802.11 medium access control (MAC) protocol went to great effort to provide a protocol that addresses not only
the physical problems of the wireless medium but also the requirements that different types of users place on service and architecture. The result
is a MAC protocol that is relevant for a myriad of wireless network applications. The 802.11 MAC is very flexible supporting multiple qualities of service (QOS) and multiple architectural implementations. The objective of this section is to go beyond merely explaining how the protocol
works. We first try to illustrate the challenges involved in implementing a medium access control protocol for wireless networking by providing
a brief history of the evolution of wireless MAC protocols. Next, we attempt to introduce the complications involved with supporting different types of users by presenting the objectives the IEEE 802.11 committee
set for itself prior to developing the protocol. We then describe the pro-
tocol. This discussion of the IEEE 802.11 MAC protocol concludes by identifying the challenges that are specific to the use of smart antenna systems and proposes modifications to the MAC that support smart antenna applications.
3.3.1 The Challenges of Wireless Networking
Wireless networks present many challenges to medium access control protocols that are not present in the protocols for wired networks. These challenges result from both the deficiencies of the wireless medium
and the performance requirements of the potential applications. We now
present a brief evolution of wireless MAC protocols to emphasize the

challenges and to identify the proposed solutions.
There are two deficiencies in the wireless medium that do not

exist for their wired counterparts. The first is referred to as the hidden
node problem. This occurs when two nodes in the same network cannot monitor each other. One node may interrupt an ongoing transmission of another simply because it is out of range of the transmitting nodes
signal. The second deficiency is the inability to detect a collision while
20
transmitting. When two wireless nodes transmit at the same time they can only detect whether or not their transmissions have been received by whether they receive an acknowledgment (ACK). Then, there is no way for these nodes to differentiate whether the cause of not receiving an acknowledgment was due to a collision, bit errors, or the recipient was out of range. Nevertheless, the only way to detect and correct these problems is by using acknowledgments. Acknowledgments at the MAC level are essential to ensuring reliable communications in wireless networks.
The first random access protocols for packet radio used two channels, one for transmitting packets and the second for transmitting acknowledgments. This arrangement ensured that each transmission received acknowledgments in a timely manner. Stations would not have to contend to send acknowledgments since acknowledgments would be generated only after the longer transmissions had successfully contended. The obvious disadvantage of this approach is the increased complexity of operating on two channels simultaneously and the inefficient use of bandwidth on the acknowledgment channel. Nevertheless, it was superior to the alternative of requiring the acknowledgments to contend like any other packet. In this latter arrangement a successful transmission may be tagged as bad simply because the receiving node is unsuccessful in contending for the channel before the packet transmission times out. We now look at several protocols based on this two channel model.
In the first and most basic MAC protocol, Aloha, nodes simply attempt to transmit their packets as they arrive. If a transmission is not acknowledged, the node reschedules the transmission by some random backoff. Errors occur whenever a file arrives at a node while another node is transmitting. This protocol is adequate for lightly loaded networks but under optimum conditions its throughput is limited to only 18% of the channel capacity of the packet transmission channel. Five techniques were subsequently proposed to improve the performance of the distributed contention process of Aloha: slotting, carrier sense multiple access (CSMA), collision avoidance (CA), four-way handshaking, and contention resolution through splitting.
In slotting, transmissions are restricted to defined time slots.
21
Slotting improves performance since colliding transmissions perfectly overlap. The first attempt to improve Aloha was through slotting. The assumption in slotted Aloha is that all packet transmissions are the same
length. Therefore, the packet transmission channel is divided into time
slots that are each suited for sending a single packet. In slotted Aloha, nodes delay transmission of arriving packets until the start of the next time slot. At the cost of adding a timing function this simple change of slotting the packet transmission channel doubles the optimum performance. Slotted Aloha can achieve an optimum channel utilization of 36%. In CSMA, nodes first sense the channel before transmitting and then defer transmission of the call if the channel is in use. CSMA improves throughput by preventing collisions. Multiple variants of CSMA were suggested by [51], which differ by how they delay the transmission. In the simplest, non-persistent CSMA, packets that arrive when the channel is busy are rescheduled by a random distribution. The disadvantage of this approach are the delays the transmissions will suffer under loaded conditions. A second technique, referred to as p-persistent CSMA, combines CSMA and slotting. P-persistent CSMA attempts to reduce the transmission delays by first attempting to transmit packets arriving during busy periods before rescheduling their transmission. In this protocol, if a packet arrives while the channel is busy, the node defers transmission until the channel is sensed idle and then transmits with
probability p. This process continues for consecutive time slots until a
node finally gains access. If a node finds the channel busy after the start of this process or a collision occurs it reschedules the transmission by a random delay distribution. The time slot for deferring transmissions does not need to be as long as a packet transmission. It only needs to
be as long as it takes for all nodes to sense whether another node began
transmitting in the preceding time slot. The value of p can be constant or dynamically chosen. The smaller the value the less likely it is that
pending transmissions at different stations will collide, but small values
also mean transmissions will be delayed longer on average. Therefore it

is desirable to have a large value when the load is light. In a dynamic
22
approach, the value of p may start at a large value and then decrease when collisions are more prevalent.
Figure 3.1 illustrates the performance of the protocols discussed thus far for a given packet arrival pattern. We assume these arrivals occur at different nodes within the network so access is only dependent on carrier access. In Aloha, only the first packet is transmitted successfully. The others are all attempted but result in collision. In slotted Aloha, three transmissions are successful with only packets 4 and 5 colliding. Packets 4 and 5 collide since they arrived in the same transmission slot. In non-persistent CSMA, there are no collisions since after sensing the channel, nodes will reschedule packets that arrive when the channel is busy. In the illustration, packets 3 and 5 arrive when the channel is busy and are then rescheduled. Finally, in 1-persistent CSMA, the version of p-persistent CSMA where nodes attempt to send packets when the channel is next sensed idle with probability 1, all packets are transmitted successfully since no two packets arrive during the same transmission.
The CSMA protocols are not quite ideal at avoiding collisions since propagation delays can mask on-going transmissions. Figure 3.2 illustrates the effect of propagation delay on the performance of the 1-
persistent CSMA protocol. In this figure, three nodes are illustrated. Nodes 1 and 2 are in close proximity to each other so transmissions between these two nodes suffer minimal propagation delay. Meanwhile, transmissions from 1 and 2 are delayed a seconds before arrival at node
3. The figure illustrates two examples of collisions that result from prop-
23
agation delay. In the first, both nodes 2 and 3 receive packets while node 1 is transmitting. Since node 2 senses the channel idle before node 3 it begins to transmit its packet first. Node 3 senses when node 1 finishes its transmission and begins transmitting its queued packet before it can sense node 2 transmitting and a collision results. The second collision occurs when a packet arrives at node 3 in the first a seconds of node 1s transmission. Since this is before node 3 can sense node 1s transmission node 3 begins to transmit the packet and a collision occurs.
The optimum performance of the CSMA protocols is highly correlated with the duration of the propagation delay. For example, [51] finds that when the propagation delay is 10% the duration of a packet transmission that the optimum throughput of the non-persistent CSMA protocol is slightly greater than 50% of capacity; however, if the delay is 0.1% the duration of a packet transmission, the optimum throughput increases to better than 90%.
Avoiding collisions at the start of transmissions is the motivation behind the next technique of improving performance, collision avoidance (CA). In CA nodes cannot transmit until after the channel has been silent for a predesignated time. CA improves performance by preventing the collisions that occur when new packets arrive at the beginning of another nodes transmission as illustrated in the second collision of Figure 3.2. The period of silence is chosen to allow for worst case propagation delays
24
and processing delays (the time it takes a node to identify the channel as idle and then to start transmitting).
The methods above do not totally prevent collisions and when collisions occur they waste at minimum the throughput required to transmit a single packet. Four-way handshaking attempts to reduce the wasted throughput by using smaller request-to-send (RTS) packets to gain access to the medium prior to sending the larger data packets. The sequence of events that result in a successful packet transmission starts with the RTS. If the destination station receives the RTS frame it responds with a clear-to-send (CTS) frame. If the source station receives the CTS for its RTS frame it knows it has access to the channel and transmits the data packet. The destination station then completes the four-way handshake if it receives the packet without error by sending an ACK to the source. The CTS frame is the indication to the transmitting station that it has access to the medium and can transmit a data packet. If it is not received on account of a collision, only the shorter time required to transmit the RTS packet is wasted. Four-way handshaking assumes the time between the RTS and the data packet transmission (between the CTS and data packet transmission when a single channel is used for both data and ACK transmissions) is controlled such that it is not sensed by another station as an idle period. A method to accomplish this is discussed later because four-way handshaking is used by the 802.11 MAC. protocol.
When traffic is heavy, collisions are more likely. Rather than repeatedly suffering collisions some protocols attempt to resolve the collisions. These tree algorithms use either the time of arrival of the packet at the contending node, the nodes ID, or some other number to separate the contentions of the nodes into successive contention windows. The protocol works as follows. When a collision occurs all contending stations not involved in the collision cease contending and allow the contention resolution protocol to work. The set of stations that collide are then split into subsets based on their ID (or some other number). If the value of their ID is in the lower half of the possible IDs they attempt access in the next contention time slot. If there is a collision again the
25
process repeats itself. If the station successfully contends it is removed from the waiting set and the algorithm follows the adjacent branch in the tree. If the channel is idle, the resolution protocol then follows the adjacent branch of the tree. The algorithm continues along each branch until there is a successful transmission or the contention time slot is idle. Figure 3.3 illustrates the contention resolution protocol. In the first contention slot three nodes, 2, 7, and 8, collide in an 8 station network. After splitting, node 2 is in the first subset and nodes 7 and 8 are in the second subset. In the next contention slot, since node 2 is in the first subset, it attempts to contend and succeeds. Since it succeeds the algorithm then follows the adjacent branch. In the next contention slot both nodes 7 and 8 attempt to transmit and then collide again. Splitting into smaller subsets 7 and 8 are again in the same subset. In the next contention time slot no stations contend so the algorithm then follows the adjacent branch. Nodes 7 and 8 collide again in the next contention time slot and they are finally split into separate subsets. Some improvements to the algorithm can remove steps. For example in Figure 3.3
26
step 5 could be skipped since after step 4 there is enough information to resolve that stations 7 and 8 were contending for the channel.
The performance of the CSMA protocols rely on the full connectivity of the network (no hidden nodes). Hidden nodes can severely decrease performance. In the worst case, when most nodes cannot monitor each other, the CSMA protocols become Aloha. Tobagi and Kleinrock proposed in [52] the use of part of the bandwidth of the data channel for a busy tone in a protocol referred to as Busy Tone Multiple Access (BTMA). All nodes that monitor a transmission on the data channel transmit a sine wave. In this arrangement the busy tone is sensed rather
than the data transmissions. BTMA has the major disadvantage in that transmitting a busy tone consumes power. It is therefore unsuitable for a protocol for mobile nodes which often rely on battery power.
The seemingly unsolvable problem of hidden nodes on the performance of the CSMA protocols has driven many to propose protocols that attempt to resemble Time-Division-Multiple-Access (TDMA). In TDMA, nodes do not contend with each other for access since each node has a guaranteed period of time in which it can transmit without risk of collision. TDMA has the potential not only to achieve near full capacity but also to provide QOS. It is the medium access method of choice for constant bit rate (CBR) traffic. Pure TDMA is not well suited for data networks. Although full use of the channel capacity can be achieved, it would only occur when the load exceeds capacity. At lower loads, many time slots would go unused even though there is traffic to send at other nodes. At very small loads, packets would suffer unnecessary delays waiting for the sending nodes time slot in the cycle. The average delay would be half the duration of the TDMA cycle. A compromise is to combine random access with a small set of timed transmission slots. These protocols consists of repetitive cycles of random access periods followed by TDMA periods. In this modification the network can still achieve a statistical multiplexing gain. During the random access periods, the nodes contend for and reserve bandwidth in the controlled access period. Once a node has successfully contended, it can use its time slot in the controlled access period to transmit packets.
27
There are both distributed and centralized control mechanisms
for gaining access to a packet transmission slot. Examples of the distributed access methods are the Aloha and CSMA protocols discussed earlier. An example of a centralized access method is polling. In comparison, the random access methods are probably best in light loads and the
polling method is best in heavy loads. In the polling method a polling node sequentially polls the nodes of the network until a node with traffic is found. An advantage of the node polling strategy is collisions can be avoided completely since no station can transmit until directed to do so by the polling station. When traffic is light, polling can be inefficient as compared to the distributed access methods since a node with traffic
will have to wait for a series of polls with null responses before it can gain access. On the other hand, it is more efficient than the distributed
access methods if traffic is heavy and evenly distributed throughout the
network since polling avoids the frequent collisions of multiple nodes
contending simultaneously for access. So the choice between the two methods depends on the network size and load.
Control of the data transmission can also be either distributed or centralized. A study of a distributed control method where all nodes in a network maintain a list of the nodes currently authorized to transmit is provided in [53]. This list is sorted in the order the nodes are authorized to transmit. In the centralized protocols a single controlling node either
directs the other nodes of the network when to transmit or grants access
to specific repetitive transmission slots.
This brief description of the evolution of access procedures for
wireless data networks reveal that there are many challenges in wireless
access and many alternative MAC approaches to deal with them. As will be seen later, the IEEE 802.11 MAC protocol incorporates many of the methods described above and that depending on the parameters selected
for the protocol supports either a distributed or centrally controlled network.
28
3.3.2

Protocol Design Objectives for IEEE 802.11
The choice of a MAC protocol depends on the network application. Identifying which combination of MAC control mechanisms is most appropriate is dependent not only on the data traffic patterns but also the environment in which the data network will be used. To understand the diversity of requirements that would be placed on a single protocol for wireless networking one must only consider when wireless networks would be used. Naturally, they will be used when it is impractical or too expensive to use a wired network. Examples include networks with mobile nodes, networks with nodes too remote to wire, networks that will be set up for only a short period of time, and networks that are placed in a buildings that are ill equipped for wiring. Each of these offers their own unique challenges to the protocol. For example, consider the networks which include mobile nodes. Mobile nodes introduce three problems. First, fading may counteract the standard collision avoidance schemes. A fade may be sensed as a silent period. Second, since mobile nodes often rely on battery power the protocol needs to support some technique to reduce power consumption. Finally, the protocol needs to support movement of nodes amongst networks. Rather than attempt to identify all the possible requirements of a wireless MAC protocol we will simply explain the requirements the IEEE 802.11 study group established for themselves. This ambitious list of requirements includes 20 items which are now listed and explained [54]. 1. Throughput. Since the available spectrum that was to be available for the wireless LAN was a limited resource, the MAC protocol needed to provide the best throughput possible.
2. Delay. Since the wireless LAN was to support real-time services
such as voice and video, the MAC protocol not only needed efficient access delay characteristics but needed to guarantee a delay bound for sustained real-time service.
3. Transparent to Different Physical Layers. The MAC protocol
had to be designed to support the multiple physical transmission layers that were to be incorporated into the IEEE 802.11 standard. These physical layers have different designs and propagation characteristics.
29
The MAC protocol needed to define a generic set of access mechanisms and then let a MAC-PHY interface layer interpret the commands and the responses between the MAC and physical layers.
4. Ability to Serve Data, Voice, and Video.
5. Fairness of Access. The protocol needed to be designed to resolve

differences in signal strength of nodes in the network preventing nodes
with strong signals from preempting those with relatively weaker signals.
The strength of signals in a wireless network can vary significantly on
account of fading conditions.

6. Battery Power Consumption. One of the primary reasons for
using wireless networks is to support mobile nodes. Mobile nodes often rely on battery power. This power is consumed at the transceivers both
in transmitting and receiving. There is great potential to conserve power
if a mobile node can doze when it has no traffic to transmit or receive. Power consumed in receiving can be as much as 50% of that consumed by
transmitting [55]. Since nodes spend substantially more time receiving than transmitting the most promising strategy for power conservation is to reduce monitoring time.
7. Maximum Number of Nodes. Potential applications identified in market studies suggested that potential users would require the network to support a large number of nodes. The MAC should be designed such that the number of nodes does not affect performance. 8. Robustness in Collocated Networks. It is likely that multiple
wireless networks will operate in the same vicinity. This introduces two severe problems, interference and security. Nodes from adjacent networks may appear as hidden nodes during periods of distributed operation.
Transmissions of nodes in adjacent networks may collide by direction if multiple centralized control mechanisms are operating concurrently.
Perhaps worst of all, nodes in one network may start to function as

nodes in the adjacent network.
9. Ability to Support Hand-off/Roaming Between Service Ar-
eas. Cellular phone services have existed for some time. The model it provides of fixed cells with mobile phones moving amongst them is a
30
model that potential service providers also want for wireless LANs. In cellular phone systems, however, hand-off between cells is controlled by a central switching office while in wireless LANs hand-off would be coordinated in a distributed fashion since there is no central control point.
10. Establish Peer-to-peer Connectivity without a Priori Knowledge. One of the major applications of a wireless LAN is the forming of temporary ad h o c networks. The concept is to allow a group of nodes to be formed into a network for the exchange of data without support of any network infrastructure. Therefore, the wireless nodes must be able to form a network without any understanding of what the overall network will look like. 11. Unauthorized Network Access Impact on Throughput. The objective is to prevent an unauthorized node from interfering with the operation of the network. Such an event may occur amongst adjacent nets when a node of one net requests service in the adjacent net. After the transmission, the adjacent net should be able to handle the interference, but, ideally, the MAC would be designed to prevent this from occurring at all. 12. Ability to Support Broadcast (Multicast).
13. Critical Delays Limit Large Area Coverage. The MAC must be robust to synchronization errors caused by varying propagation delays. Propagation delays vary by area of coverage. Using the rough assumption of 1 nanosecond delay for a foot of propagation, a 1000 square foot coverage area would have to accommodate a 1 microsecond propagation delay.
14. Insensitive to Capture Effects. Capture effects cause some nodes to be preferred over others. These effects can improve throughput but at the cost of decreased fairness. The MAC should be designed to be insensitive to these effects. 15. Support Priority Traffic. The MAC should support traffic with different priorities.
16. Ability to Support Non-reciprocal Traffic. Traffic flow to and from nodes is not reciprocal, that is each node will receive much more
31
traffic than it will transmit. This should not hinder operation of the MAC.
17. Preservation of Packet Order. Packets for real-time services
need to be transmitted in the proper order.

18. Ability to Work in a Wide Range of Systems. This is a restatement of 7 and 13 and emphasizes that the MAC should be robust to both area of coverage and number of nodes served.
19. Limit the Complexity of the Physical Layer. The MAC design should not push complexity into the physical layer as a result of designing for multiple physical layers.
20. Ability to Market and Complexity. Ultimately the MAC would have to be marketed. Simplicity usually results in higher quality products that are easier to market.
3.3.3 The 802.11 Distributed Coordination Function (DCF) and Point Coordination Function (PCF)
The term used by the 802.11 standard [56] for the group of stations that function together as a network on a single wireless channel is a basic
service set (BSS). A BSS may or may not be part of a larger network. If
it is self contained it is referred to as an independent BSS (IBSS). If it is part of a larger network and one node is identified as the access point (AP) to that network it is referred to as an infrastructure BSS. The generic MAC protocol for wireless networking that the 802.11 standard defines has two coordination functions, a Distributed Coordination Function (DCF) and a Point Coordination Function (PCF). The difference between the two is that in the DCF stations compete for access to the channel in a specified contention process while in the PCF a
single station called the point coordinator (PC) manages station access.
802.11-compliant networks may employ just the DCF or a combination of the PCF and the DCF. When supporting data only communications the DCF can be used exclusively but to achieve multiple QOS requires that both control functions be used. The standard is very explicit on the operation of the DCF. By contrast, it defines the mechanism by which
32
the PC controls traffic but leaves traffic management up to the implementer. When operating both functions, the DCF and PCF alternate. The PCF always starts on a defined period but the actual duration of the PCF and DCF may vary. The period of time the DCF is operating is called the contention period (CP) implying the stations in the network can contend for access and the period of time the PCF is operating is called the contention free period (CFP) implying the stations cannot contend for access.
The DCF defines the procedures by which all stations serviced by a single radio channel can gain access to that channel to send data. It is distributed in that the control is completely self contained at the station
attempting to send traffic. The DCF uses a carrier sense multiple access - collision avoidance (CSMA-CA) scheme where stations contend and gain access through a handshake process after a random exponential backoff. A station contending for access listens for silence and then waits a random backoff time before attempting to gain control of the
medium. If no other station requests access before the backoff expires, then the contending station may attempt to gain access through one of two modes. The first is the four-way handshake. The transmitting station first sends an RTS frame. The receiving station upon hearing this request responds with a CTS frame. And finally, if there have been
no collisions, the sending station sends its data. In the second, if the data to be transmitted is sufficiently small, the station will attempt to send the data immediately. In both cases, if the packet is successfully received, the receiving station responds with an acknowledgment.
A standard problem of wireless networks is that of the hidden node. This occurs when two stations of the same network cannot monitor each other. One may not hear the others transmission and then attempt to send data concurrently with the result of interference at both of the receiving stations. The key collision avoidance feature that helps prevent the hidden node problem is a network allocation vector (NAV) that is included in each RTS, CTS, data, and ACK frame. This vector predicts the length of time until the stations complete their current transaction. Since both the sending and receiving stations transmit the NAV, all
33
stations within listening range of either of these two stations will know how long the medium will remain busy. Each station in the net maintains a separate NAV timer which it updates upon hearing a transmission.
This NAV timer serves as a mechanism to indicate to the station that

the channel is busy even when it cannot monitor the transmissions of
one of the parties in the current data exchange. A station may only
contend for access after its NAV timer has expired and the channel has remained idle for an appropriate amount of time. New stations entering the network and stations that are waking-up from power save mode must
first successfully receive a NAV from another station before they are
eligible to contend for access. The PCF is a centralized control mechanism where a single station, the PC1, controls all the traffic on the network. The CSMA/CA and NAV mechanisms of the DCF remain active during the PCF, but the PC is able to gain control of the medium by transmitting prior to
any ordinary station in the network. This is accomplished using different interframe spaces which will be described in greater detail later. At the start of the CFP, the PC transmits a NAV that extends until the
projected end of the CFP. So all stations who hear the NAV will not attempt to contend during the CFP. Once the PC has control, it then uses polling to manage transmissions of other stations or it transmits data itself. The polls either send data, acknowledge receipt of data, direct stations to send queued traffic, or are a combination of these. A single poll may include an acknowledgment to a previous transmission, data for
the polled station and a direction to the polled station to send data it
has queued. Polled stations may send data to any station in the network
and that station in-turn responds with an ACK. The CFP will last until
the PC broadcasts an end of CFP message. At this time, monitoring stations will reset their internal NAV.
1
The PC, or point coordinator, is also the access point (which is also the base station) for the network.
34
3.3.4

The 802.11 Power Saving Mechanism
The 802.11 MAC protocol allows nodes in a network to enter the
power save (PS) mode when they have no packets to send. While in
the power save mode nodes alternate between awake and doze states.
During the periods the node is in the doze state, power is saved as a
result of not monitoring and processing transmissions. The PS nodes only stay in the awake state for a very short time unless told to remain
awake. The purpose of alternating between awake and doze states is to allow other nodes to send packets to the PS nodes. When the PS node wakes up, it listens for a traffic indication map (TIM). A TIM is a bitmap indicating which stations have packets queued at other stations. If the PS node monitors a TIM that announces another node has packets queued
for it then it will remain awake to receive those packets. The protocol defines three different procedures for broadcasting TIMs and receiving data when the power save feature is used. They differ depending on which coordination function is used and whether the network is an IBSS.
In infrastructure networks a node must first inform the PC that

it will enter the PS mode. This can be accomplished by setting Power Management bits within the Frame Control field of a transmitted frame.
When the PC receives this announcement it will buffer packets for the PS nodes until the designated time to transmit them. The method used to coordinate the transmission of this buffered traffic to the PS nodes differs by whether the network is using the DCF or the PCF. In both cases the
PS node wakes-up just prior to a scheduled beacon transmission to receive the beacon. The beacons of these networks are transmitted by the PC and consists of a timestamp and a TIM. Certain beacons are referred
to as a Delivery TIMs (DTIM). They occur at a regular interval of beacons as announced by the PC. The PC transmits all broadcast/multicast traffic following DTIMs. When the DCF is operating, the PS nodes wake up according to their ListenInterval. If they monitor a TIM indicating that packets have been buffered for them, they remain awake and contend to send a PS-Poll frame. When the PC receives the PS-Poll frame
it then contends to send the buffered packets to the PS node. PS nodes that want to receive multicast frames may also wake up at DTIMs. When
35
a PC has multicast traffic it waits until the next DTIM to send it. The PC announces its intent to send a multicast frame in the DTIM. It then transmits all multicast frames it has buffered continuing after subsequent beacons if multicast traffic is still buffered. Operation differs during the PCF. PS-nodes wake-up at every DTIM and then stay awake if multicast traffic will be sent or if they are identified in the TIM. They then stay awake until they receive traffic from the PC and are notified that no more traffic for them is buffered or they are no longer identified as having buffered traffic in a subsequent TIM. In IBSS networks, the procedures differ. In establishing a network the first station determines the beacon interval and whether the PS mode will be allowed by establishing an Ad Hoc TIM (ATIM) window. The ATIM window starts at the Target Beacon Transmission Time (TBTT) and lasts for the duration specified which is some fraction of the beacon interval. The ATIM window is reserved for the transmission of beacons and ATIM management frames. The ATIM management frames are used to announce traffic for stations in the PS mode. In such a network, a station enters the PS mode by setting the Power Management bits within the Frame Control field of its next and all subsequent transmitted frames. Other nodes that have traffic for the PS node then begin buffering that traffic until they can schedule its transmission. The PS IBSS network works as follows. Just prior to the TBTT all PS nodes wake-up. All wait for the successful transmission of a beacon. Once a beacon has been transmitted, all stations with buffered traffic for a PS node or multicast traffic contend to send an ATIM using the procedures described later. ATIMs may be directed to a single node or be multicast. If directed to a single node that node acknowledges receipt of the ATIM signaling the sending station that it can contend to send the traffic following the ATIM window. If the ATIM is multicast, the sending station will not receive an acknowledgment but will still contend to send the traffic following the ATIM window. A PS node that receives either a directed ATIM or a multicast ATIM remains awake for the remainder of the beacon interval. Meanwhile, after the ATIM window, all stations that sent either an acknowledge directed ATIM or a multicast ATIM during the ATIM window and all nodes attempting to send traffic to
36
non-PS nodes contend for channel access. If one of these stations fails to gain access prior to the next TBTT then the process starts over again.
Figures 3.4-3.6 illustrate the 3 power save procedures. Figure 3.4 illustrates the infrastructure PCF. The PS node wakes up each DTIM. It stays awake after the first DTIM to receive a multicast frame from the PC. It stays awake after the second DTIM as well to receive directed frames from the PC. Note that it stays awake for several transmissions past the next beacon until it has received all of the packets the PC has
37
buffered for it. Figure 3.5 illustrates the infrastructure DCF. The PS node wakes up on its own listen interval which in this illustration is every two beacons. In this manner it also monitors the DTIMs and can receive multicast packets. When it awakens the first time, the DTIM indicates a multicast transmission will follow so the PS node stays awake to receive that packet. After receiving it and since no additional packets are to be multicast the node returns to the doze state. It awakens a second time two beacons later. This time it is identified in the TIM of the beacon. The PS node then contends and sends a PS-Poll transmission to the PC. The PC receives the PS-Poll and subsequently sends the packet to the PS node. The PS node acknowledges the packet and returns to the doze state since there are no additional packets reserved for it. At the third ListenInterval the PS node wakes-up only to return to the doze state after the DTIM since it was not identified in the TIM. Figure 3.6 illustrates the IBSS DCF. Three power save nodes are shown. All wake-up prior to the TBTT and in the case of the first interval remain awake since they received a multicast ATIM. Note also that PS node 3 has traffic buffered for PS node 1 and sends a directed ATIM to it. The ATIM is subsequently acknowledged by PS node 1. Following the ATIM window both the multicast and the directed frames are sent. In the second ATIM window the only ATIM transmitted was a directed ATIM to PS node 2. All the PS nodes return to the doze state after the ATIM window except PS node 2 which waits for the frame announced in the ATIM window. Finally, in the third ATIM window a directed ATIM is sent to PS node 3 which stays awake past the ATIM window while the other PS nodes return to the doze state.
3.3.5 The 802.11 Network Synchronization
Ad hoc networks consists of nodes, perhaps notebook computers, that are networked to exchange data with each other as opposed to a network infrastructure. These networks are created in a extemporaneous manner. The first node instantiates the network and establishes some key parameters such as the beacon interval and the ATIM window. After the network is established no particular station has the role of syn-
38
chronization, instead this becomes a collective function. At the TBTTs all nodes suspend contending to send packets and contend to send the beacon. Once a beacon is successfully transmitted all other nodes cease their attempt to send the beacon and compare the Timing Synchronization Function (TSF) timer of the transmitted beacon with their own. If it is later than their own they then adopt it as their own. In this manner all stations keep synchronized. After the beacon is transmitted all stations follow the procedures of the DCF both in the ATIM window and for the remainder of the beacon interval.
The access point of an infrastructure network is responsible for
timing. It establishes the Beacon Period and DTIM Period and then transmits all beacons and DTIMs. In networks where the PCF is used these parameters help define the transitions between contention and contention free periods. 802.11 has four variables that determine how the transitions between the CP and CFP occur. They are BeaconPeriod, aDTIMPeriod, aCFPPeriod, and aMaxCFPDuration. Table 3.1 defines each of these variables. These variables are the primary design variables that affect the operation of the network.
Figure 3.7 illustrates the relationship of these variables. The CFP always starts at the beginning of the aCFPPeriod and the CP always ends at the end. The transition from the CFP to the CP may
occur at any time after a minimum CFP up to aCFPMaxDuration. The selection of aCFPMaxDuration must allow a minimum duration CP.
39
3.3.6
The 802.11 Network Operation
Interframe Spaces
The key to the operation of both the DCF and the PCF is the use of different interframe spaces. Interframe spaces are defined periods of required silence prior to transmission. Each type of transmission must wait for a specific interframe space. The duration of the interframe space determines the priority that the transmission has within the protocol. For example, stations that want to contend for access must wait a longer interframe space than a station that needs to acknowledge the receipt of a message. Similarly, the PC seizes control by using a shorter interframe space than that used by a contending station but longer than the interframe space for an acknowledgment. The 802.11 MAC defines four interframe spaces. The first and shortest in duration is the Short Interframe Space (SIFS). The SIFS is the period of no transmission between the receipt and response to a frame. It is the space that exist between the CTS and RTS frames, data frames and the subsequent ACK, and the polling frames and responses during the operation of the point coordination function. The second, and next shortest in duration is the PCF Interframe Space (PIFS). The PIFS is used by the point coordinator to send beacons and to seize and maintain control of the medium during the contention free period. The third, and next longest in duration is the DCF Interframe Space (DIFS). This is the required duration of silence before contenders begin to count down
40
their backoff timers. The fourth and longest is the Extended Interframe Space (EIFS). It is used by a station waiting to contend instead of a DIFS when the traffic that it last detected was not a valid MAC frame.
The duration of the first three of these interframe spaces is dependent on the physical layer. They are determined by two values, aSIFSTime and aSlotTime. The aSIFSTime is the time it takes from the end of receipt of a message for a station to process the message and then respond with the first symbol of the preamble. The aSlotTime is the time it takes a station to recognize a channel is busy or idle plus the time it takes to process a frame, prepare a response, transmit it, and
for it to propagate to the receiving station. The duration of the SIFS is aSIFSTime. The PIFS is one aSlotTime longer than the SIFS and the DIFS is two aSlotTimes longer the SIFS. The EIFS is equivalent to the
time it takes to transmit 8 consecutive ACK frames plus the duration of one SIFS and one DIFS.
Transmitting Data
In a network operating both the DCF and the PCF, data traffic may be sent in either the CP or the CFP. The choice of how to send the data is made by the sending station. It either contends to send data immediately to the receiving station during the CP or it contends and requests that the PC mediate the transmission of the traffic during the CFP. When data is sent in the CP it may either be fragmented or sent as a single transmission. Fragmentation occurs when the message is larger than a threshold called aFragmentationThreshold. This threshold is used to minimize the overall transmission time of the file when retransmission on account of bit errors is considered. A second threshold exists for sending data during the CP called aRTSThreshold. This threshold determines when a file is sufficiently small such that it is more advantageous to send data without the RTS/CTS handshake. Its size is dependent on the probability of a collision on the wireless link. Table 3.2 lists the different thresholds and the actions taken for each.
Each station maintains a counter variable which counts the number of times a station attempts to contend for a given message. If the
41
station continues to contend without success at reaching the destination
it may take one of two actions, drop the transmission or request that
the PC manage it. The reason for not having success may be one or
more of the following: the destination is no longer in the network; the destination cannot hear the transmitter; or the destination is in the PS mode. Going to the PC assumes the latter of these three are the cause of
the contention failure. If this is the case the PC will awaken the destination prior to orchestrating the transmission. If it is not the case, the PC
knows which stations are on the channel and can assist the transmitting station in making the decision to drop the transmission. (This is not explicitly part of the 802.11 standard.)
Transmitting Voice and Other Real Time Traffic

Voice transmissions require the network to provide a guaranteed
constant bit rate. This is only possible during the CFP where the PC can manage the bandwidth. So when a station wants to transmit a voice
message it contends sending a message to the PC requesting that the PC
control the traffic. If there is sufficient bandwidth, the PC will acknowledge the request otherwise the call is blocked. The specific process by which the PC manages the bandwidth during the CFP is not defined by 802.11. A military application of the 802.11 MAC which supports voice
communications is described in [57]. A modification to the 802.11 MAC that allows voice communications during the DCF is proposed in [58].
42
3.3.7

Operation of the 802.11 DCF
Figure 3.8 illustrates the contention process and the different types of traffic in a network with a PCF. The process starts when traffic arrives at a station and needs to be transmitted. That station then calculates a backoff using the equation
The variable i in this equation is the number of times the station
has previously contended to send this data. So the first time a station attempts to send the traffic, the backoff is between 1 and 7 aSlotTimes, the second time between 1 and 15, the third time between 1 and 31 and so on until the maximum of between 1 and 255 is reached. In Figure 3.8 there are 4 stations contending for backoff. The backoff timer of the first is 8 aSlotTimes and it has a data packet that is smaller than aRTSThreshold to send. The backoff timer of the second is 2 aSlotTimes and it has a packet larger than aRTSThreshold but smaller than aFragmentationThreshold. The backoff timer of the third is 12 aSlotTimes
43
and it has a data packet larger than a aFragmentationThreshold. The fourth station has a backoff of 5 aSlotTimes and it is seeking access from the PC for either real-time traffic or a file that is too large to be
efficiently transferred during the CP. Also shown in the illustration is
a generic receiving station and the point coordinator. Note that after
each contention period all backoff timers are decreased by the smallest
backoff time. Observe the use of the interframe spaces. Since CTS and
ACK frames are transmitted only after aSIFSTime they have priority over beacons and contention backoff. In turn, beacons are transmitted
after aPIFSTime and therefore have priority over contention backoff.
3.3.8
Operation of the 802.11 PCF
The operation of the PCF is not fully described by the 802.11 standard. Besides providing the mechanisms that allow the PC to seize control of the channel and defining the different types of polling frames, the standard leaves the rest up to the implementer. Figure 3.9 illustrates an implementation of the point coordination function where it is
assumed the PC may control five types of traffic. Traffic from the one station to a second. Traffic from a station to the PC. Traffic from the
44
PC to a station. Broadcast from a station. And finally, a broadcast from the PC. Figure 3.9 illustrates these five types of traffic. The following transmissions occur sequentially in Figure 3.9. Station 1 is directed to transfer data to another station, not the point coordinator. Station 2 is directed to send data to the point coordinator. Station 3 receives data from the PC. Note that the PCs transmission includes the ACK to Station 2s transmission as well as both the Poll and Data for Station 3. Next, all stations receive a broadcast from the PC. This broadcast occurs in one fragment. The PC does not wait for an acknowledgment. Immediately after a SIFS the PC attempts to transmit data to a station but in this case the station does not hear the transmission and sends no ACK. Note that the PC uses the PIFS to regain control of the medium. In the final set of transmissions the PC directs Station 3 to send a broadcast transmission. Note again that the PC regains control of the network after waiting only a PIFS.
3.3.9 Smart Antenna Systems and the 802.11 MAC Protocol

Networks that use smart antennas are also infrastructure networks which use the PCF. A single station with a smart antenna serves as the PC. The spatial multiplexing gain only occurs during the PCF when the PC can coordinate communications amongst stations whose spatial signatures are known. Therefore, in order to capitalize on the spatial multiplexing gain, most traffic must be sent during the CFP. Stations contend as normal during the CP but avoid sending packets directly to other stations opting to request the PC mediate their transmissions. The contention process is not spatially multiplexed and is very important. The PC not only receives the requests transmitted but also determines the spatial signatures of the stations sending them. Traffic management during the PCF would be much different for a Smart LAN. In order to achieve the most from the multiplexing gain, transmissions need to be synchronized. The network would require a standard packet transmission size. A fixed transmission size ensures that all spatial channels are fully used during uplink and downlink. The protocol would have to establish procedures to ensure ACKs
MAC Layerof the 802.11 System
45
following downlink transmissions are synchronized from the end of the last transmission rather than the end of each stations received message. Finally, new procedures would be required for probing polls searching for buffered traffic. In the standard PCF, the PC polls stations authorizing those nodes access without knowing a priori of buffered traffic. In the smart LAN, the PC may still poll but to avoid the risk of only one of several stations polled transmitting packets the protocol would establish a distinction between polls directing transmission and those probing for buffered traffic. Nodes which receive probing polls would simply acknowledge the probe with an appropriate indication of whether it has buffered traffic. The PC would perform the traffic management function of determining when the nodes actually send the much larger data packets.
Due to the rich diversity of spatial signatures, Smart LANs cannot easily support mobility during ongoing transmissions without advanced adaptive or blind signal processing algorithms since even slight movements would change the spatial signatures. Spatial signatures are only stable when a node is stationary, as shown by the experimental results presented in Chapter 6. This does not preclude movement of nodes so long as moving nodes do not transmit during a spatially multiplexed time period and provisions are made to update the signatures after each change in location. For uplink transmissions, the PC can use the signa-
tures acquired when the nodes either contended to gain access or gained access by responding to a successful probing poll. Downlink transmissions, however, are performed using the most recent update which may be invalid if no traffic had been sent by the destination node after its last move. To insure that the update is valid can be a joint responsibility of the PC and the mobile node. Once the node has finished moving it is appropriate for it to contact the PC for the purpose of providing the PC
its new spatial signature. Meanwhile, the PC also needs a method that
allows it to probe for and acquire a stations spatial signature. The PC may need to validate a spatial signature when transmissions to a particular node are not successful. This could occur if the node completed a move during the CFP. The PC can obtain spatial signatures by sending separate probing polls to the missing node using an omni-directional
46
transmission. The nodes response provides the spatial signature. An alternate solution uses adaptive or blind algorithms to allow for the rapid acquisition or prediction of the spatial signatures of a mobile user. Many theoretical, computer simulation, and experimental studies are currently underway to find the best solution to the real-time mobility problem for smart antenna systems. For example, the 1.8 GHz smart antenna testbed at The University of Texas at Austin described in Chapter 5 allows for two users to conduct real-time simultaneous voice communications with a smart antenna base station even when both users are moving (walking or driving in cars).
3.4
PHY Layer of the 802.11 System
The 802.11 standard actually allows three very different physical layer options for manufacturers. The members of the 802.11 working group chose allow three physical layer options to allow systems designers to choose a technology that matched the price, performance, and operations profile of a particular application, similar to the choices between l0Base-T, 10Base-2, and 100Base-T in Ethernet [50]. The infrared (IR) option is mainly for implementations within a room [50], as the diffuse IR selected for 802.11 has a very limited range and has difficulty passing through walls. However, IR has the advantages of being interference free and secure. The two radio frequency (RF) options are in the unlicensed industrial, scientific, and medical (ISM) 2.4 GHz frequency band. The actual bandwidth allocated is from 2.400-2.4835 GHz, except in Japan (2.471-2.497 GHz), Spain (2.445-2.475 GHz), and France (2.4465-2.4835 GHz). In the ISM band, manufacturers can choose between DSSS and FHSS options. The other two ISM bands (902 MHz and 5.8 GHz) werent selected because the 902 MHz band is already used extensively by other wireless technologies (which would serve as interference sources), has lim-
ited allocated bandwidth, and is not available worldwide for networking purposes, while the 5.8 GHz band in less desirable for wireless LANs because of the higher cost of systems at that frequency band and because higher frequencies have less propagation, so their signals are less capable of penetrating through walls. The 802.11 committee chose two RF
47
physical layer options to provide a choice between the implementation complexity and costs necessary to achieve performance and reliability [49]. Transmit power is limited to 100 mW, with power control specifications required for any system capable of transmitting above the 100 mW level to insure that the level of the emission is reduced to a level at or below 100 mW. Systems must support a data transfer rate of 1 Mbps
(megabits per second) to be 802.11-complaint, with an optional data rate of 2 Mbps also specified. Typical ranges of the transmitters are around
500 feet in office situations and 1000 feet in open space environments.
FHSS systems jump from narrowband to narrowband within a wideband, transmitting one or more packets at each carrier frequency. The dwell time, which must be no longer than 400 milliseconds and is typically 20 milliseconds [50], is the fixed length of time the transmitter sends data on a given frequency before it switches to another frequency for another fixed length of time. The jumping pattern is specified in the standard and is tracked by both transmitter and receiver. The idea behind FHSS systems is that there may be significant noise that corrupts the signal at any frequency band; however, when the system hops around to another frequency band, it may be able to achieve error-free transmissions. If interference occurs at one frequency band, then the data is retransmitted on a future hop on another frequency band. There are 79 such frequency bands (with approximately 1 MHz of bandwidth each) available in the standard except in Japan (23 frequency bands), Spain (27), and France (35). The standard contains three distinct sets hopping patterns, each of which is composed of 26 sequences (4 sequences for Japan, 9 for Spain, 11 for France). The patterns within a set were chosen to exhibit desirable properties; for example, the consecutive frequencies in a given pattern are spectrally separated by at least 6 MHz (5 MHz for Japan) to avoid a narrowband interferer. FHSS systems use 2 level Gaussian Frequency Shift Key (GFSK) modulation for the 1 Mbps data rate, and 4 level GFSK modulation for the 2 Mbps data rate, with a nominal bandwidth bit-period (BT) of 0.5.
DSSS systems modulate the original data signal by a spreading code which is known to both transmitter and receiver, similar to CodeDivision-Multiple-Access (CDMA) systems. However, unlike multicode
48
CDMA systems, only one spreading code is used, an 11-chip Barker code. The factor by which the bandwidth of the signal is expanded is the pro-
cessing gain, which is 10.4 dB for the 11-chip code. The idea behind DSSS systems is that spreading the signal over a wide frequency band allows the system to transmit successfully even if there is a large interference or noise source at a certain frequency, in addition to benefiting from the processing gain already mentioned. Hence, FHSS systems experience narrowband interference only for a fraction of the time (with potentially
significant losses during that time) and DSSS systems experience a fraction of the narrowband interference all of the time (with potential losses due to the decrease in system performance due to the constant presence of the interference). There are 11 DSSS center frequencies defined in 802.11 for North America (13 for Europe, 2 for Spain, 4 for Frnace, and
1 for Japan), but only 3 of those may be used without overlap (only 1 in Spain, France, and Japan). Note that it is possible that Spain may certify 802.11-compliant products created for the French frequency set, as they did with the Lucent WaveLAN-1 (not 802.11-compliant) product, so Spanish 802.11 products may have 4 center frequencies available instead of 2. DSSS systems use Differential Binary Phase Shift Key (DBPSK) modulation for the 1 Mbps data rate, and Differential Quadrature Phase
Shift Key (DQPSK) modulation for the 2 Mbps data rate.

IR systems use near-visible light in the 850-950 nm range, similar to the spectral range for consumer devices like IR remote controls and data communications equipment like IrDA (Infrared Data Association) devices. However, the use of diffuse infrared means that both reflected and line-of-sight (LOS) IR energy is used for communications. Hence, the receiver and transmitter dont need a clear LOS, but the penalty for this type of IR system is that the range is limited to only about 10-20 meters. The IR signals do not pass through walls and are significantly attenuated when passing through most exterior windows, which can allow the network to be contained within a single physical room. This may be a desirable characteristic for users, as it allows for different IR systems to be used in adjacent rooms separated by only a wall without interference, and makes for a secure network by eliminating the possibility of eavesdropping. The IR systems specified are also limited because
49
they will only work in indoor environments (since IR interfaces cannot be exposed to direct sunlight), unlike the DSSS and FHSS systems. IR
systems use 16-Pulse Position Modulation (PPM) for the 1 Mbps data rate, and 4-PPM for the 2 Mbps data rate.
The Physical Layer Convergence Protocol frame format and operation specified in the 802.11 standard is described in the Physical Layer Design section of the next chapter on the Smart Wireless LAN system design. Other 802.11 operations such as the scrambling and descrambling of the DSSS bits are also described in that section. The information is
presented in that chapter because the methods in which the SWL system implements those 802.11 requirements are discussed there. Some of the
MAC operations specified in the 802.11 standard are discussed in the Media Access Control Design section of the SWL system design chapter for the same reason.
The proposal which was selected by the IEEE 802.11 committee to create a new high-speed wireless LAN system was developed by
Lucent Technologies and Harris Semiconductor. It will enable wireless LANs to transmit data at 11 Mbps, resulting in a greater than five-fold increase in data rate and boosting the throughput beyond the 10 Mbps
Ethernet standard. The new standard will facilitate compatibility between high-speed wireless LAN equipment from multiple suppliers and
is backwards-compatible with other (DSSS) 802.11-compliant products being sold today. It operates in the same 2.4 GHz ISM frequency band as the current standard, and uses the same channelization scheme defined in the current 802.11 standard for DSSS transmission at the 1 and
2 Mbps data rates. Since it allows for only the DSSS physical layer, FHSS and IR products will be excluded from this 11 Mbps data rate
and will not be compatible with products created for the new standard because of the fundamental differences between the three physical layer options. To achieve its top two data rates, the standard will use a special coding scheme called Complementary Code Keying (CCK). It will
support data rates of 11 Mbps and 5.5 Mbps, as well as the 2 Mbps and
1 Mbps data rates from the current 802.11 standard. CCK is extremely resistant to echoes and has extremely tight delay spread specifications,
50
100 nanoseconds at 11 Mbps and 250 nanoseconds at 5.5 Mbps. Due to its excellent delay spread characteristics, it can work in a variety of hostile RF environments. The eight-chip complementary code spreading sequence of CCK will enable it to achieve the high transmission speeds without sacrificing the security or robustness of the system. Smart antenna systems should be able to utilize their advantages for capacity, fading reduction, diversity gain, etc., with this standard; however, the Smart Wireless LAN system described in the next chapter will need to be modified to incorporate this new standard with the current 802.11 standard.
3.5
Examples of Current Wireless LAN Products
It would take an entire book (which would probably be outdated even before it could be published) to cover all of the wireless LAN products currently available. Hence, this book will discuss four wireless LAN
products, selected because their manufacturers are among the leaders in this field and they cover the major options for wireless LAN systems today.
3.5.1 The Lucent WaveLAN DSSS System
Lucent 1 , according to [59] the biggest manufacturer in the wireless LAN equipment market with about 40 percent market share, is the
leading proponent of DSSS wireless LAN systems. Their WaveLAN products, shown in Figure 3.10, operate in the 2.4 GHz ISM band, use DSSS with BPSK/QPSK modulation, and are fully compatible with 802.11. Their system has a maximum throughput of 2 Mbps, can use multiple channels at that 2 Mbps rate depending on the regulatory agency (11 channels with the FCC, 13 channels with the ETS, 4 channels in France, 1 channel in Japan, and 2 or 4 channels in Spain as explained earlier, with up to 3 non-overlapping channels with the FCC and ETS bands)
1
The photos and some of the information in this section were provided by Lucent
Technologies, www.wavelan.com
51
and a cell radius of 1400 feet in open environments (1200 feet at the 2 Mbps data rate) and 650 feet in semi-open environments (550 feet at the 2 Mbps data rate), with range extender antennas available to increase the range of the system by 50% in an open (LOS) environment, 15% in an semi-open (blocked) environment. Hence, if those range extender antennas were use by both sides in the communication link, the range would therefore be extended by 100% in open environments and 30% in semi-open environments.
Until late June 1998, their WaveLAN products werent compatible with 802.11, so having such a big player in the wireless LAN market adapt their product to 802.11 was significant for the standard. Their access points (WavePOINT-II), shown in Figure 3.11, have two PCMCIA card slots which allows for the user to use one slot for their 802.11compliant card and another for their older WaveLAN system, which provides an easy migration path for their installed user base to 802.11. Later, users can then use one of those card slots for the fast version of 802.11, providing a clear migration path for their current users to move to the current and future standards with the same base station. It should be noted that Lucents compliance with the 802.11 standard
is no surprise, as Vic Hayes from Lucent chaired the 802.11 committee
52
and Lucent took a key role in drafting the standard. Having the major wireless LAN manufacturers such as Lucent adopt the standard should create economies of scale that will lead to lower costs of 802.11-compliant components and thus to lower wireless LAN system costs. This, in turn, should aid the proliferation of wireless LAN usage worldwide.
3.5.2
The Proxim RangeLAN FHSS System
Proxim1 has been the driving force behind the OpenAir standard, the wireless LAN industrys most widely adopted wireless LAN technology available since 1994 [60]. Their RangeLAN2 system for the OpenAir standard has been a 2.4 GHz, FHSS, 1.6 Mbps, 1000 foot (in open space) cell radius (500 foot cell radius in typical office environments), 15 channel, alternative to 802.11 systems. However, they now also support the 802.11 standard with their RangeLAN802 system, which is a 2 Mbps, FHSS system. The RangeLAN802 PC cards, shown in Figure 3.12, may be used in PCs at a radius of 1000 feet in open space and 500 feet in
1 The photos and some of the information in this section were provided by Proxim Inc., www.proxim.com
53
typical office environments from their access point, shown in Figure 3.13. Proxim also has an extended range option, called their XR technology, which can increase the coverage area of their access points by nearly 30% indoors and 100% outdoors.
54
Proxim recommends their RangeLAN2 system for applications which require high mobility, extended battery life, interoperability with a wide range of third party devices, will require support for 10 or more clients per access point, or will require a wide range of adapters and access points for any network situation. They recommend their RangeLAN802 system for applications which require a lot of large file transfers, use devices that have access to power or where batteries are readily available, do not require interoperability with a wide range of third party devices, and have typically less than 10 clients per access point. They also recommend their RangeLAN802 system for clients who feel more comfortable with (and possibly require) the IEEE stamp of approval on their system. Hence, Proxim is keeping their options open in the standard wars, creating product lines for both their OpenAir standard and the 802.11 standard.
3.5.3 The RadioLAN BackboneLINK Narrow Band System

RadioLAN1 provides an interesting alternative to RF 802.11 systems, which is why it was selected to be the only RF system discussed here which is not 802.11-compliant. Their products operate at the 5.8 GHz unlicensed ISM band in a low-power (50 mW peak) narrow band system, quite different than the spread spectrum systems mandated in 802.11 and used by Lucent and Proxim. This narrow band single frequency system is similar to a frequency hopping system that never changes its frequency. Their product was the first stand-alone access unit that bridged wireless traffic onto an existing l0Base-T network at full 10 Mbps Ethernet speed. They are limited to using just one channel, but they can support up to 128 active wireless stations per access point on that channel.
The RadioLAN PC CardLINK T M , shown in Figure 3.14, or ISA TM CardLINK may be used by PCs at a radius of 300 feet in an open environment and 120 feet in a semi-open environment (with the shorter
1 The photos and some of the information in this section were provided by RadioLAN Inc., www.radiolan.com
55
56
range due to the lower power and higher data rate of their system) from their BackboneLINK TM access point, shown in Figure 3.15. The BackboneLINK access point was designed to be flexible enough to be connected to any 10Base-T device via a RJ-45 connector, compact so that it is can be placed in almost any location, and fast enough to perform like a wired 10Base-T system. The new 11 Mbps 802.11 standard could provide a big problem for manufacturers such as RadioLAN which are not 802.11-complaint, as it would seem pose a significant threat to their established niche.
3.5.4
The JVC Office LAN IR System
The JVC1 Office LAN is an IEEE 802.3-compliant Ethernet compatible IR system, with a maximum throughput of 10 Mbps. Their basic PC IR interface, called the Node, can be used up to 33 feet away from their access points (16.5 feet between Nodes in an ad hoc network configuration). Their extended range PC IR interface, shown in Figure 3.16, is called the Station and can be used up to 66 feet away from their access points. Their base station options are the wall-mounted W1
The photos and some of the information in this section were provided by JVC,
www.jvc.net
57
Satellite, with a 150 angle of communications, and the ceiling-mounted S-Satellite, shown in Figure 3.17, with a 360 angle of communications,
A major selling point of this system is that it is compatible with Ethernet (802.3) and allows for the use of IR beams to replace cable connections in a LAN while retaining the full 10 Mbps Ethernet speed. However, laptop and desktop computers without a built-in l0Base-T network connector must use their PCMCIA network adapter, the MNode card, which has only a 23 foot transmission range (which is still much better than their previous option, the PC-Card, which had just a 3.3 foot range). Even without this limitation, the 66 foot maximum range, lack of penetration through walls (because it is an IR system) and price of the system may limit the general purpose usefulness of the system, so like all IR systems it is the best choice for users with specific types of applications. Examples of these types of users are those who desire the 10 Mbps Ethernet compatibility, enhanced security of an IR system, or who would most benefit from the immunity to RF interference that an IR system provides.
Chapter 4
Smart Wireless LAN System Design
4.1
Introduction
The Smart Wireless LAN system was created to adapt the smart antenna concept for the IEEE 802.11 standard. The 802.11 specifications were not designed with smart antenna systems in mind, so special
care had to be taken in order to utilize the unique advantages of smart antenna systems. This chapter will describe the MAC and PHY layers of
the SWL system and demonstrate how the system operates. Remember, this system was created so that 802.11-compliant terminal operations would not need to be changed (which significantly increased the design difficulty but also increased the potential usefulness of the system), so only the SWL base station (also called the access point or the point coordinator) operation was allowed to significantly deviate from conventional
802.11 systems. Also, keep in mind that the SWL system was designed to be compatible with the DSSS specifications in the 802.11 standard and is
not compatible with any FHSS system, even if it is 802.11-complaint. It was decided not to have the SWL design implement a FHSS system for now because the frequency hopping was deemed less desirable for smart antenna systems after the experimental results showed that the spatial signatures would vary when the frequency band changed, hence requiring more frequent updating of the spatial signatures or the utilization of
blind or adaptive techniques. The second generation of the SWL system

will allow for operation while a terminal is moving, which will likely require blind or adaptive techniques. Hence, the second generation of the SWL system might support FHSS systems as well as DSSS systems, to
60
demonstrate the feasibility of utilizing smart antenna systems for FHSS systems.
4.2
SWL Media Access Control Design
The SWL media access control (MAC) design extends the smart antenna concept and the efficient scheduling scheme for wireline networks [61, 62, 63, 64] to create a new packet reservation protocol to significantly increase throughput. The SWL algorithm divides each TimeSpace frame into two periods, the contention period and the contention free (data) period. When a node wishes to join a network, it issues a request for authentication (RFA) during the contention period. The RFA includes the type of transmission, the amount of time slots per frame requested, and the data length (for data transfers) or timing requirements (for voice, video). If the RFA is successful, the base station responds by sending out a block of random text to the potential users node. The user encrypts the text string using the users password as a cipher key. If the network gets the proper encrypted string of data back, the node is authorized for network access and will be allocated a time slot (or slots, if more than one slot per frame is requested) when one is available in the data period for its transmission.
The RFA can only succeed during the contention period if its request satisfies the 802.11 Distributed Coordination Function (DCF) criterion. In the DCF, if more than one request is sent in a contention period slot, then all requests in that slot are assumed to fail. Note that this criterion is conservative because, due to the capture effect [65], it may be possible for a particular contention request to succeed even though the number of requests during that time slot exceeds the maximum allowable number. This is because the loss due to fading and propagation over radio channels fluctuates the received power of transmitted signals, so a base station still might correctly receive a packet with the maximum received power, even if two or more packets happen to collide. The length of the contention period is adaptively selected, subject to maximum and minimum parameters set by the network. Each slot that has
a contention failure increases the length of the contention period by one,
SWL Media Access Control Design
61
each slot that has no contention requests reduces the contention period by one, and the length is unchanged if the contention is successful. The contention period is also reduced in size when the network is operating in an overloaded state, as described later. In addition, if all the time slots in the contention period are used in a frame, then the contention period is increased in size.
An example of a Time-Space SWL frame is shown in Figure 4.1. There are a total of N time slots in each frame, with the number of time slots in the contention and data periods (Nc and Nr, respectively) adaptively adjusted depending on the input traffic conditions as explained above. There are Ns (four in this example and in the simulations used for comparison purposes between SWL and 802.11) rows in the contention free period because the use of a smart antenna system allows for Ns users to transmit at once. There is only one row in the 802.11 frame simulation because only one user is allowed to transmit at a time in a frequency band. Note that the SWL simulation assumes an eight-element uniform linear antenna array at the base station to separate the four transmissions in the same frequency band at the same time. The shaded slots represent terminals (1-7) transmitting the high-priority voice data. The other numbered slots represent terminals (8-19) that are transmitting e-mail messages or performing file transfers. The slots with no numbers
62
in them are time slots that have not been assigned to any terminal. Terminals 4, 6, and 17 have successful RFA requests to join the network and will probably use those time slots in the next frame. In fact, those terminals may even push the network into a low-overload situation, as discussed later.
The global scheduler operation can be summarized as follows. First, all terminals are given the same priority in the contention period and only after they succeed in joining the network are their priority designations considered. This is done for fairness to all terminals and because it is impractical to assume that we know the type of data being transmitted from a terminal before that terminal joins the network. Second, all high priority data requests are satisfied before any low priority data request is considered. If the high priority data requests cannot be satisfied in one frame, the requests are served in a round-robin fashion. Third, if there are remaining data slots, they are shared in a round-robin fashion by the low priority requests. Fourth, when the network is operating in high-overload (the high priority data requests cannot be satisfied in one frame) or low-overload (the low priority data requests cannot be
satisfied in one frame) modes, the network increases the length of the
data period, which reduces the length of the contention period. This has the beneficial effects of both increasing the number of time slots available in the data period where they are needed by the overloaded network and reducing the probability that new data requests will succeed in the contention period and further overload the network. Finally, one segment of the data period is reserved for the base station, which uses it to broadcast to all users and potential users the timing parameters for the next frame, the length of the contention period, and other necessary parameters for the next data period. Other segments can be reserved by the base station to handle administrative or bookkeeping tasks.
This protocol mirrors the 802.11 protocol as closely as possible and mobile terminal hardware would not need to be changed, though there would have to be a software upgrade for terminals who wanted their data to be classified as high priority. However, this new protocol has the following major features: 1) multiple uplinks and downlinks can
SWL Physical Layer Design
63
be carried out in the same time slot and same frequency band; 2) the transmission of active terminals with different bandwidth is organized by using a global scheduler to achieve fair delay performance; and 3) priority designations allow the network to meet guaranteed maximum delay requirements needed for delay-sensitive multimedia applications
such as voice links.
4.3
A simplified block diagram of the computer simulation of the physical layer for the downlink for the SWL system is shown in Figure 4.2. This structure was created to be 802.11-compliant while allowing for the use of a smart antenna system and was simulated using MATLAB. Data is randomly generated for the multiple simultaneous users, and is placed in the Physical Layer Convergence Protocol (PLCP) frame format explained later. The DOAs for each direct path and multipath signal are randomly generated, with some test scenarios requiring that the DOAs differ by at least a specified minimum angle separation. The amplitude and phase of each received signal is randomly generated. The spatial signatures of each terminal can then be computed since we know the DO A, amplitude, and phase of the signal. Encryption is an optional feature that is performed using the Wired Equivalent Privacy (WEP) algorithm used in 802.11. The WEP algorithm is a 40-bit RC4 encryption algorithm designed specifically for LAN communications and is basically a bitwise XOR with a pseudorandom key sequence created by a users secret key. The data is modulated with DBPSK or DQPSK to achieve the 1 and 2 Mbps data rates specified in 802.11. Each terminal uses the same code to spread their data, an 11-chip Barker code. The channel model is randomly generated each time the simulation is run and can vary significantly from simulation to simulation. Additive noise is scaled to achieve the desired SNR between the desired signal and the gaussian noise. When all of the signals are combined, the other simultaneous terminals serve as interference sources for the desired terminal. Beamforming is performed when the base station receives the uplink signal in order to mitigate the effects
64
of the interference, or when the base station transmits a downlink signal to the multiple terminals simultaneously, in order to mitigate the interference that each terminal will receive. The signal is then despread, demodulated, deinterleaved, decrypted, and Viterbi decoded in the standard manner. The resulting signal is compared to the original signal to
find the bit error rate (BER) of the system. Note that we do not have
to resolve the 180 degree phase ambiguity (a result of the despreading operation) because of the use of differential modulation. The PLCP frame structure is shown in Figure 4.3, which is specified in the 802.11 standard and utilized by SWL. The MAC (Media Access Control) Protocol Data Unit (MPDU) is the field that contains the data, which may be any length (including 0 bytes) up to the maximum length specified by the base station in the periodic management frames it transmits to all terminals. Each frame transmitted by the base station or terminals has 128
65
synchronization bits, a pattern of alternating zeros and ones. The user searches through this data to synchronize its despread timing with the received signal. This search is necessary because the exact timing for the start of the Barker code is unknown by the terminal, so the terminal searches a sliding 10-bit (110-chip) window to find the maximum despread signal, which gives the synchronization timing. This operation is described in the 802.11 standard and is extremely useful for smart antenna implementation.
This search for synchronization works because the beamforming routine reduces the interference signals below the desired signal. Also, because the Barker code is uncorrelated with any unsynchronized part of itself, the terminal will find a low signal until the chips are aligned correctly and the timing is synchronized (where the SWL simulation then finds a significantly higher peak). After synchronization is achieved, the terminal then searches for the known Start Frame Delimiter (SFD) pattern (3FA0h in the 802.11 DSSS specification, where h denotes hexadecimal notation), which indicates that the PLCP Header and MPDU (the data) will immediately follow. The synchronization and SFD fields comprise the 144 preamble bits for the PLCP.
The PLCP header consists of the Signal, Service, Length, and
66
CRC fields. The Signal field denotes the modulation type and data rate being used by the transmitted data. The modulation types for the DSSS version of 802.11 (the only version that SWL is currently compatible with) allow for two selections, DBPSK and DQPSK. The data rate is given in 100 kbps (kilobits per second) increments, with 1 Mbps the maximum for DBPSK and 2 Mbps the maximum for DQPSK. The PLCP preamble and header bits are always transmitted with DBPSK, then the MPDU data bits are transmitted with the modulation type specified in the Signal field. The Service field is reserved for future use, with 00h being required for 802.11 compliance. The Length field is an unsigned 16 bit integer which indicates the number of microseconds required to transmit the MPDU. The CRC (cyclic redundancy check) field protects the Signal, Service, and Length fields with a CCITT CRC-16 FCS (frame check sequence). This FCS is the ones complement of the remainder generated by the modulo 2 division of the protected PLCP fields by the polynomial The protected bits are processed in transmit order. All FCS calculations are made prior to data scrambling.
All bits transmitted with DSSS are scrambled and descrambled with the polynomial
The feedthrough configuration of the scrambler and descrambler is self synchronizing which requires no prior knowledge of the transmitter initialization of the scrambler for receive processing. The scrambler may be initialized to any state except all ones when transmitting. Figures
67
4.4 and 4.5 show typical implementations of the data scrambler and descrambler, respectively.
Other features of 802.11 also aid the implementation of a smart antenna system. Timing considerations, critical for smart antenna systems, are aided by 802.11 specifications such as the regularly broadcast frame by the base station which synchronizes the clocks of all terminals
in its local area. Other examples of features in 802.11 that aid smart antenna systems are power control, the synchronization bits in the header of each frame, the frame structure with contention and contention free periods, and terminal identification fields in each header.
When multiple users wish to uplink to the base station at the same time, the base station can separate the signals because it already knows the spatial signatures of the individual users. Using that knowledge, the base station performs a simple pseudoinverse spatial signature beamforming operation to obtain the matrix that will be multiplied with the received waveform to obtain the multiple data vectors containing the signals of the multiple simultaneously transmitting terminals. In pseudoinverse spatial signature beamforming, the weight vector w can be found by
where X is the received signal. The least squares solution for this beamforming algorithm, is designed so that the weight vector for and In effect, the weighting routine steers the beam in the direction associated with the array
68
response vector of the source and places nulls in the other source directions to minimize their interference. Other beamforming methods being studied include dominant DOA, complex conjugate spatial signature, and pseudoinverse DOA, which have better performance in some scenarios than the pseudoinverse spatial signature method [38]. Also being tested is a new method for designing downlink weighting vectors
based on the filter bank beamforming concept [66] which simulations found to be superior in all scenarios to the pseudoinverse technique, but which requires more computational complexity.
The assumption that we already know the spatial signatures of each terminal is a reasonable one because each terminal is required to transmit multiple frames to the base station during the contention period in order to join a network. Since only one user is allowed to transmit at a time during this period (adhering to the Distributed Coordination Function in 802.11), we can easily find that users spatial signature. Experimental studies discussed in Chapter 6 found that the spatial signature remains fairly constant over time for a stationary terminal. If the transmission quality from a terminal appears to be degrading, the base station can send a request to that terminal to transmit again during the contention period to update its spatial signature.
4.4
SWL Dynamic Slot Assignment Algorithms
A First Fit technique similar to the First Fit SDMA/TDMA technique in [67, 68] is used by SWL for dynamic slot assignment of users based on their spatial signatures. Beginning from the first time slot, the base station tries to assign a new user to the first time slot that is below its capacity The resulting signal-to-interference-plusnoise ratio (SINR), is computed for that slot, and the result is compared with where is the minimum SINR allowed by the base station for a user to be added to a time slot. If adding the spatial signature of user to the time slot results in the base station keeps trying open time slots (those below their capacity) until the test is passed, possibly placing the user in a time slot with no other users.
SWL Dynamic Slot Assignment Algorithms We define the signal covariance matrix as
69
where * denotes complex conjugate, denotes matrix transpose, I is the identity matrix, and is the mean noise power per element. The local mean SINR after beamforming is
where w is the N 1 complex weight vector found by the pseudoinverse

spatial signature beamforming method discussed earlier. A First-Come First-Served (FCFS) technique for dynamic slot assignment has also been studied for comparison purposes. In this method,
the number of users in a time slot is set at just as that number was set during the first few years of development of the SWL system when it was called the Spread Spectrum with Space-Division-Multiple-Access system (SS-SDMA) [31, 35], and was not then 802.11 compatible. The
first users are assigned to the first time slot with the second users assigned to the second time slot and so on, with no SINR test used for assignment. This is different than the Random algorithm in [67, 68], where the new user is assigned to the first available time slot and then has all of the users SINR in that time slot tested. If the SINR passes
the test, then that Random algorithm assigns the user to that time slot; otherwise, it is assigned to a new time slot with no other users. That Random method was proven to perform extremely inefficiently, which
is why a different algorithm was selected for SWL for comparison purposes. The authors in [67, 68] also at times call their Random method
their FCFS method, which is why the different operation between the
FCFS technique used in this research and their Random technique were
explained. The First Fit and FCFS algorithms were simulated in both sorted and unsorted formats. Sorting the spatial signatures in order of ascending power is beneficial because it helps prevent those with higher power
70
from overpowering the other users. This is because the spatial signatures tend to be grouped together as the users are placed in the network when the First Fit algorithm is used, assuming that most of the users pass their initial SINR test. For the FCFS algorithm, the users are always grouped in order, so the sorting can have an even greater benefit from this. Another benefit from sorting is that because lower power signals are also competing with a relatively large amount of noise, it is best to allocate them at the start of the frame when there are more unallocated terminals to test. This provides more time slot options for the lower power signals to choose from, and the lower power signals benefit more from having a greater amount of options than the higher power signals do. The FCFS algorithm does not benefit from this advantage because it does not test the spatial signatures before allocating them to a time slot.
There are many major differences between this research and the research in [67, 68]. First, their research was strictly attempting to maximize the number of users in a time slot, instead of being concerned with the SINR performance the users experienced in the network. Second, they use optimal SINR beamforming, which requires much more computational complexity than the pseudoinverse spatial signature beamforming used by SWL. Third, their simulations do not take advantage of the antenna or spreading gains that SWL achieves. Fourth, they are concerned with fairly rapidly changing channel environments, on the order of milliseconds, while our experimental results show that for stationary terminals the environment remains essentially stable over long periods of time. However, the excellent work in [67, 68] provides an invaluable comparison tool between the various algorithms under selected scenarios.
4.5
SWL Timing Synchronization Algorithms
Spread spectrum techniques similar to those used in CDMA are used in 802.11 with the unfortunate (for smart antenna systems) requirement that each terminal use the same spreading code. This requirement probably did not seem to be a major problem to the 802.11 committee because they assumed that the system would only have to support one
71
terminal transmitting or receiving data at a time in a frequency band. Unfortunately, this requirement means that a straightforward implementation of the spreading algorithm in the SWL system (with each terminal using the same code) would result in a loss of the 10 dB of spreading gain that a single user achieves. Consequently, a chip delay is inserted by the SWL system between each of the resulting signal vectors before downlink, as shown in Figure 4.6. Each row represents the spreading code used by a terminal and the actual starting point of the Barker code is shown by a solid vertical line. Because of the ideal property of the Barker code (minimal correlation between the various parts of the code), this allows the coding gain from spreading to be achieved despite all terminals using the same code. In a similar manner, the uplink coding gain is achieved by adjusting the timing of the start of each users uplink transmission, taking into account the propagation delay found earlier. Due to the beamforming effectiveness as shown later and in [11, 14, 69], each user receives one significant and distinct signal to synchronize to and can easily despread its signal in the standard manner. The results obtained through MATLAB are compared with experimental results from our smart antenna testbeds at the J.J. Pickle Research Campus of The University of Texas at Austin which are described in Chapter 5, as well as other published results.
When the downlink transmission containing the frames for the multiple users is received by a user, it is able to distinguish its signal from the signals directed to the other users by using the PLCP preamble and taking advantage of the effectiveness of the beamforming algorithm.
72
The PLCP preamble, shown earlier in Figure 4.3, contains the SYNC
and SFD fields. The SYNC field consists of an alternating 0-1 pattern, and the user searches through this data to synchronize its despread timing with the receive signal. This is necessary because the exact timing for the start of the Barker code is unknown by the terminal, so the terminal searches a sliding 10-bit (110-chip) window to find the maximum despread signal, which gives the synchronization timing. This search works because the beamforming routine reduces the interference signals below the desired signal. Also, because the Barker code is uncorrelated with any unsynchronized part of itself, the terminal will find a low signal until the chips are aligned correctly and the timing is synchronized (where the
SWL simulation then finds a significantly higher peak, approximately 10
dB higher, just as predicted by the theory). After synchronization is achieved, the terminal then searches for the known SFD pattern, which indicates that the PLCP Header and MPDU (the data) will immediately follow.
The round trip propagation time is found for each terminal when they communicate with the base station during the contention period. As stated earlier, terminals use this period to join the network, change their priority level or bandwidth allocation, and to update their spatial signatures periodically or at the request of the base station. During a
contention time slot, multiple messages are sent between the base station and terminal, with no other terminal allowed to transmit until their communications are completed. During this time, the base station can
measure the round trip time for messages between it and the terminal,
which should remain constant unless the terminal moves. This is another
reason behind the assumption that a terminal remains stationary during operation for many of the simulations for the first generation of the SWL system. The first step in the SWL timing synchronization algorithm is the
computation of the round trip propagation times for all the users that
the dynamic slot allocation algorithm has assigned to the current time slot. The user with the smallest round trip time is used as a reference, and the relative round trip chip offsets, of the other terminals in the
73
time slot are computed relative to that reference, modulo 11. Because the spreading code repeats itself every 11 chips, mod 11 arithmetic is required for the computation of the relative offsets. For example, two terminals with round trip times offset by exactly 11 chips would in effect transmit with no chip offset between one another. Unless a timing adjustment is added to one of the terminals, and ignoring the effects of
beamforming, the signals from the two terminals would merge together and be indistinguishable from each other by the base station, eliminating all benefit gained by the spreading of the signal and creating a large source of interference. Beamforming can reduce the loss of spreading gain, but only a perfect null created for each interference source can preserve the entire spreading gain when there is no chip offset. Thus, the goal for the timing synchronization algorithm is to find the necessary chip delay adjustments, which result in all the overall chip offset times, being unique. If we can insure that and for where the overall chip offset times are computed by
where represents the user number, then the spreading gain for each user in the time slot will be preserved.
There are 11 possible chip delay adjustments, ranging from 0 10, as shown in Figure 4.7. No chip adjustment can be used twice, because the chip adjustment is the relative offset from the reference users timing (again, using modulo 11 arithmetic) used by the base station
74
for the downlink information or for the message informing the multiple terminals that they have been chosen to be allowed to send an uplink signal in that time slot. If for then the base station would be simultaneously transmitting to terminals i and j with identical start times for the Barker code, thus eliminating much or all of the spreading gain. This problem is shown in Figure 4.8, where the first two users have In this case, the spreading gain will be partially or entirely lost for users 1 and 2.
The terminal with the smallest round trip time is designated terminal 0 and has its round trip chip offset (relative to terminal 0, itself), set to 0. The chip delay adjustment for that terminal, b is also set to 0. Thus, the overall trip offset for user 0,
A flowchart demonstrating the operation of the timing synchronization algorithm is shown in Figure 4.9. The symbols and terms used in the flowchart are defined in this section. The dotted line shows the termination of the Quick algorithm. The dashed line demonstrates how the sort routine may be skipped by the algorithms.
The rest of the terminals may be sorted by the algorithm in order of ascending round trip chip offset delays. The chip delays were sorted in the original algorithm because it made verification of the algorithm by hand and a step-by-step debugging of the MATLAB program easier. However, the simulation results showed that the sorting had just a minor benefit because the algorithm searches for any possible valid timing offset. Thus, the terminals besides terminal 0 in this algorithm are either numbered on a FCFS basis or in ascending order of round trip chip offset
75
delays (where ties are broken on a FCFS basis). In either case, terminal 0 is selected as the terminal with the smallest round trip time, as stated earlier.
The initial set of valid chip offset adjustments for the remaining terminals, D, contains 1 10, since 0 is already taken. After each iteration in which a chip adjustment is found for a terminal, that chip delay adjustment is removed from D. For example, if terminal 1 is assigned a chip adjustment of 3, then D would then contain 1, 2,410. If terminal 2 was then assigned a chip delay adjustment of 8, then D would contain 1,2,47,9,10.
The set of valid overall chip offsets, T, is handled in the same manner as D. Initially, T contains 110, since 0 is already taken by the terminal 0. Each time is computed for user i, that value of is checked with the set of remaining valid times, T. If then the chip offset adjustment used to obtain that is accepted, and the values of are removed from D and T, respectively. The minimum element in D is selected for and is tested to
76
check if it will result in a valid overall chip offset. If it does, then that offset is selected for the terminal, that value for is removed from D, and the algorithm begins the search for the next terminal. If the resulted in an invalid chip offset (in other words, ), then the next smallest value in D is selected and the overall chip offset is checked for validity. If none of the elements in D result in a valid value, then that terminal cannot be placed in the time slot and the algorithm moves on to check the next terminal.
It is quite possible, and even fairly common in the simulations, that a terminal may not be able to find a valid overall chip from the available chip adjustments in D but some of the as yet untested terminals may be able to find valid chip adjustments. This is due to the different round trip delays for the terminals because they are at varying distances from the base station, which causes different relative round trip chip delays. A simulation was created for comparison purposes which stops after it finds a terminal for which it cannot find a valid round trip delay from the valid chip delay adjustments in D. This modified algorithm was labeled the Quick algorithm, and the algorithm described earlier was labeled the Standard algorithm. The Quick algorithm simulation was able to execute faster than the Standard algorithm, but experienced a noticeable drop in performance. This proved the point that later terminals may be able to find a valid chip adjustment even though an earlier terminal cannot. The simulations stop either after they attempt to place every terminal in the available time slots, or after they find a terminal who cannot find a valid overall chip offset time.
It can easily be shown that the Standard algorithm can always guarantee to be able to find chip offsets to preserve the spreading gain for at least six terminals per time slot. Hence, due to the 11 chip Barker code used in SWL and 802.11, the minimum number of terminals which will be able to be placed in a time slot while retaining the spreading gain is 6, while the maximum number is 11. It is trivial to show that the maximum is 11 because each terminals transmission must begin on a different chip and there are only 11 chips available. A simple explanation to show that at least 6 terminals can have their timing synchronized
SWL Timing Synchronization Algorithms is below. Definitions: = Designation of user i in the time slot. = Round trip chip offset for user i relative to user 0 (mod 11). = Chip delay adjustment added by the base station for user i. = Total chip timing delay for user i (mod 11).
77
Initialization: = 0 Delay of user 0 relative to itself is 0. = 0 User 0 is given no chip delay adjustment, so Optional sorting of users via their round trip delays. Goal: Assign any 6 randomly selected users to a time slot while preserving the
spreading gain. To preserve the spreading gain, we must have:
Required for downlink spreading gain. Required for uplink spreading gain.
Given:
Assume 5 of the 11 possible delays have been assigned. There are 11 possible chip adjustments. Assume that the 5 total chip delays are unique.
The relative round trip delay for the next user is known. Proof: (mod 11) (mod 11) (mod 11) (mod 11) (mod 11) (mod 11) There are six possible timing totals for user 5. The timing totals are unique because the chip offsets are
unique.
We have 6 different values for to choose from, and only 5 values which have been reserved by the previously assigned users. Hence, there
78
is at least one timing result which is different, with a corresponding chip adjustment which is different from what was assigned before. Hence, we can always assign at least 6 users to a time slot and satisfy the requirements to preserve the spreading gain for uplink and downlink.
Chapter 5
Experimental Setup for Smart Antenna Systems
5.1
Introduction
The experimental testbeds described in this chapter were developed at the Electrical Engineering Research Laboratory (EERL) located in the J. J. Pickle Research Campus of The University of Texas at Austin, under the leadership and supervision of Prof. Guanghan Xu. Three smart antenna testbeds were completed, with the 1.5 GHz testbed (which consisted of a receive-only base station) completed in 1993, the 900 MHz testbed (transmit and receive to enable uplink and downlink beamforming) completed in 1994, and the 1.8 GHz tested (capable of real-time uplink and downlink beamforming) completed in 1997. The setup for both the 900 MHz and 1.8 GHz testbeds will be described in this chapter. The indoor and outdoor experimental environments described in this chapter applies to the experiments conducted with all three testbeds unless otherwise noted.
5.2
900 MHz Testbed
A variety of experiments were carried out using smart antenna testbeds designed as in Figure 5.1. The 900 MHz testbed, shown in Figures 5.2 and 5.3, is comprised of the following subsystems: (1) One
This chapter was written in collaboration with L. K. Hansen of the University of Texas at San Antonio, S.-S. Jeng of Chung Yuan Christian University, Taiwan, and A. Kavak and G. Xu of the University of Texas at Austin.
1
80
900 MHz Testbed
81
82
8-element patch antenna array and four 1-element dipole antennas. The 8-element patch antenna array, arranged in a linear fashion with separation of about one half wavelength, is the base station. The dipole antennas are used by the four mobile units. (2) Twelve RF and IF up/down converters and switches operating in the RF band at around 900 MHz and IF band at around 144 MHz. (3) Two distribution boxes providing synthesized sources for RF and IF local oscillator signals. (4) Twelve A/Ds and 24 D/As. (5) Four digital multiplexing (MUX) and demultiplexing (DEMUX) boards. Each MUX/DEMUX board is connected to one of two high speed I/O boards installed in the s-bus slots of a Spare 10 workstation. (6) Two bi-directional high speed I/O boards installed in a Spare 10 workstation.
5.2.1
Antenna Array
An 8-element patch antenna array is installed on a wood supporting structure and arranged as a uniform linear array with separation between antenna elements of about a half wavelength (16 cm). The center frequency of each patch antenna is around 900 MHz. The return loss of each antenna is about 10 dB. The bandwidth is about 10 MHz. The
900 MHz Testbed
83
feeding port of the patch antenna is connected to a RF switch which can operate from DC - 2.5 GHz. Figure 5.4 displays our 900 MHz antenna
array. Our 1.8 GHz 8-element patch antenna array (also arranged as a uniform linear antenna array), shown in Figure 5.3, is very similar to the 900 MHz antenna array except that it is much smaller because the patch antenna elements are placed a half wavelength apart and the size of the antenna elements is smaller at the higher frequency.
5.2.2 RF/IF Unit There are 12 RF/IF downconverter and upconverter units operating around Each RF downconverter and upconverter is built on the same printed circuit board (PCB) with low-cost integrated circuits (ICs). The gains for both the RF downconverter and upconverter
are about 20 dB and the noise figure of the downconverter is about 4 dB.
The IF frequency is around 144 MHz and the gains for both the IF down-
converter and upconverter are adjustable from 20 dB to 50 dB. Among the 12 RF/IF downconverters and upconverters, 8 of them are dedicated to the base station smart antenna system while 4 of them are connected to mobile antennas with several hundred feet of RG-8 cables. The RF I/O port is connected to the RF switch. The IF I/O port is connected to D/A and A/D converters in the MUX/DEMUX box. Figures 5.5 and 5.6 displays one of our RF/IF boxes for four channels.
5.2.3 Distribution Boxes
Two distribution boxes were built to generate flexible RF and IF local oscillating signals. They can provide 16 coherent RF local oscillating signals with an adjustable range from 600 MHz to 2.5 GHz and 32 IF local oscillating signals centered around 144 MHz. Since quadrature detection is done in software, the IF local oscillating signals for the upconverters and downconverters are slightly different. The difference can be tunable with a resolution of 2 Hz. Figure 5.7 displays one of our distribution boxes.
84
900 MHz Testbed
85
5.2.4
MUX/DEMUX Boxes
There are 12 sets of 8-bit A/D converters, 24 sets of 8-bit D/A
converters and four digital MUX/DEMUX boards in the MUX/DEMUX box. The sampling rate of the A/D converters can be set up to 5 MHz. The sampling rate of the D/A converters can be set up to 2.5 MHz. Each A/D or D/A pair is connected to the baseband I/O port in the RF/IF unit. Each MUX/DEMUX board is connected to one of two high speed S-bus I/O boards installed in a Spare 10 workstation. Figures 5.8 and 5.9 shows our multiplexer/demultiplexer box. 5.2.5 High Speed I/O boards and Sparc 10 Workstation Each of the two I/O boards can read or write from or to the Sparc 10 memory at a speed of 40 MBps (Megabytes per second) independently. In the current setup, one I/O reads or writes at a speed of 20 MBps while another works at a rate of 10 MBps. The console of our testbed is a Sparc 10 Model 51 workstation with an 8 GB hard disk and 96 MB of RAM.
86
1.8 GHz Base Station

5.2.6 Stability of the Smart Antenna Testbed
87
To measure the long-term stability of the smart antenna testbed, we connect a calibration source through a one-to-eight power divider to each channel of the smart antenna system. We measured the spatial signatures over 8 hours by taking one snapshot every 5 minutes. The results are shown in Figures 5.10 and 5.11. The smart antenna testbed is very stable with time and temperature. The spatial signature variation due to hardware effects was determined to be less than 1 % relative angle change and relative amplitude change over the duration of an 8-hour measurement. A system calibration was performed before and after each propagation experiment.
5.3
The 1.8 GHz testbed consists of two basesites (BS) and multiple mobile handsets (HS). The experimental results described in Chapter 6 do not utilize all the capabilities of the testbed. However, in this chapter
88
we will describe the entire 1.8 GHz testbed setup so that the reader can fully appreciate the capabilities of this system.
The testbed supports two identical basesites. These programmable basesites were designed with flexibility in mind, in order to support multiple research efforts. In fact, in addition to the wireless LAN work, they have supported CDMA, TDMA, and SDMA experiments.
Each basesite consists of a PC computer with EZ-ICE In-Circuit Emulator, 4 transmit/receive (T/R) boards (with two T/R channels per board), a DSP board with two Analog Devices SHARC chips, and asso-
ciated vocoder chips and telephone handsets. The EZ-ICE Emulator is used to boot, monitor, and generally control the operations of the two SHARC chips. Each SHARC chip is connected to 4 T/R channels. The two SHARC chips are connected to each other via a link port, allowing for communication and synchronization between processors. Figure 5.12
is a photograph showing the larger of the two basesites with its controlling PC and the two telephone handsets. (Although identical electrically, one basesite is packaged in a smaller chassis than the other.)
89
5.3.1
Transmit/Receive Board
Inside of a basesite, connected to the backplane, are four T/R boards. Each T/R board has two T/R channels. Figure 5.13 is a photograph of a T/R board. The copper foil provides RF shielding to the analog circuits on the board.
A block diagram of one T/R channel is pictured in Figure 5.14. Each T/R board has two such channels. The blocks in the diagram will be discussed in the following sections.
Carrier, RF, and IF Frequencies

The T/R channels transmit and receive on a nominal carrier frequency of 1.88 GHz. Received signals are mixed down with a 1.81 GHz RF frequency to generate a 70 MHz IF signal. We refer to these frequencies as nominal since the internal RF frequency is generated by a frequency synthesizer; this frequency is tunable over a tens of megahertz range. Similarly, transmitted signals are mixed up from 70 MHz to the 1.88 GHz carrier frequency via the 1.81 GHz RF frequency.
90
The receive T/R channels mix down the IF signals with an I-Q pair of 70 MHz IF frequencies to generate the in-phase (I) and quadrature (Q) baseband signals. Similarly, the transmit T/R channels mix up the I-Q baseband signals to a 70 MHz IF. In order to form this I-Q pair of 70 MHz frequencies, the T/R frequency synthesizer actually supplies a 140 MHz signal which is frequency-divided to 70 MHz, and then one leg
is phase-shifted by 90 degrees.
Sampling Rates
On the basesite T/R channels, incoming baseband signals are lowpass filtered and then sampled at 3.072 Msamples/sec per channel. This extremely high sample rate was selected in order to support CDMA experiments. Experiments not requiring this high sample rate simply
decimate the incoming samples.
Incoming samples are digitized to 10 bits, latched, inserted into a 16-bit word, and then put out onto the system bus. A separate Field Programmable Gate Array (FPGA), not shown in the drawing, supplies
91
92
all of the timing signals necessary for latching and sampling. In addition,
it supplies the logic signals necessary to transfer data to and from the T/R board on the system bus.
Outgoing 10-bit data is inserted into 16-bit words and latched

into the transmit latches at a rate of 384 ksamples/sec per channel. The data is then interpolated up by a factor of 4 by Zilog DSP chips.
The resulting 1.536 Msamples/sec per channel data is digital-to-analog converted and then I-Q modulated up to the 70 MHz IF. It is then up-
converted to 1.88 GHz and transmitted.

5.3.2 SHARC Board
The SHARC board contains two Analog Devices SHARC 20160 floating-point DSP chips plus supporting interface chips. Figure 5.15 is a photograph of the SHARC board. The two SHARC chips are clearly
visible as the large square chips inscribed with metallic circles.

Computational Units
93
A SHARC chip uses the highly-parallel Harvard architecture. There are separate data paths for program and data memory allowing for dual-fetches of data and coefficients. The floating-point multiplier and ALU are also in parallel. Each instruction executes at a 30 MHz rate. The highly-parallel architecture allows for fast DSP algorithm execution.
Especially important is the fact the the SHARC performs computations in 32-bit floating point. This allows the SHARC to support highly complex operations such as matrix inversions and SVDs without the headaches of fixed-point arithmetic.
DMA Capability
The SHARC chips support direct memory access (DMA) capability for almost all of their I/O ports. In particular, this capability allows
the programmer to set up a DMA to send/receive data to/from the T/R boards, and then proceed on to other computations. The DMA transfer happens in the background over dual-ported memory so that there is no loss of executions cycles when DMA is going on.
I/O Ports
The SHARC has several 4-bit link ports. As configured in the basesites, the link ports are used to communicate with the controlling PC and with the sister SHARC chip.
The SHARC also has two serial ports. One serial port is configured to interface with the control logic of the T/R boards. The second serial port transmits and receives data from the vocoders chips.
5.3.3 Vocoders and Telephone Handsets
Each basesite has two vocoder chips. The vocoders use Motorola DSP56166/7 fixed-point DSP chips plus support circuitry. Speech is compressed to a rate of 8600 bits/second. The vocoders work internally with 20 ms segments of speech data. However, the data is transferred
94
from the chips every 10 ms. At the receiving end, the receiving vocoder then appends two 10 ms frames together to assemble the 20 ms internal
frame.
The vocoders interface to the basesite telephone handsets. Figure 5.16 is a photograph of the two telephone handsets. Although they
appear to have buttons, etc., these telephone handsets have been gutted and contain only a microphone and a speaker. All intelligence is
contained in the programmable basesite hardware.
5.3.4
Backplane
The four T/R boards and single SHARC board draw power from
and communicate via a backplane. Figure 5.17 is a photograph of the
backplane.
Figure 5.18 is a photograph of the backplane, T/R, and
SHARC boards.
95
96
5.3.5

Antenna Array
A basesite can handle up to eight antennas, with each SHARC receiving the date from four of the elements. Thus it is possible to run just a single SHARC with four antennas. However, the wireless LAN experiments were conducted using both SHARCs and the full eightelement array.
5.4
1.8 GHz Handset
Although not utilized in the wireless LAN experiments, the testbed also supports multiple portable handsets. Each handset consists of a RF board with a single T/R channel, a Motorola DSP56166 chip for modem functions, a Motorola DSP56166 chip for vocoder functions, an EPROM, RAM for the speech data, and a telephone handset with peripheral support hardware. The handsets are programmed via an emulator connected to a PC computer. Figure 5.19 is a photograph of a handset, emulator board, and supporting PC. Figure 5.20 is a close-up of a handset attached to its emulator board.
Software for the handsets is developed with the handset tethered to its emulator and supporting PC. However, when the software development is completed, the handsets can be programmed with bootstrap code and their executable code can be blown into the EPROM. This allows for experiments with mobile handsets.
5.4.1 Transmit/Receive Channel
Each handset has only a single T/R channel. It differs from the
basesite T/R channels in the following respects:
1. The FPGA in the handset generates fast interrupts to the DSP chip, rather than the supporting DMA.
2. The A/Ds on the handset are 8 bit, not 10 bit.
3. The handset T/R board does not have an on-board DSP chip to support interpolation of the transmit data. Therefore, rather than
1.8 GHz Handset
97
98

sending TX data at a rate of 384 ksamples/second, the handset must send TX data at a rate of 1.536 Msamples/second.
4. The handset T/R channel samples the received baseband data at a rate of 768 ksamples/second. 5.4.2 DSP Chip
The handset DSP chip is a Motorola DSP56166 16-bit fixed-point chip. It also uses the highly parallel Harvard architecture. The 56166 runs at an internal clock rate of 60 MHz.
Fast Interrupts
The 56166 does not have DMA capabilities. Instead it uses fast
interrupts to send and receive data from the T/R channel. A fast interrupt takes only two instruction cycles to complete. However, it is an interrupt and does consume CPU cycles, as opposed to the DMA transfers on the basesite SHARC chip.
Vocoder Chip
The handset vocoder chip is also a Motorola DSP56166. It runs basically the same code as the vocoder chips at the basesite. The hand-
set DSP chip communicates with the handset vocoder chip via a serial interface. 5.4.3 Handset Antenna
Each handset uses a single antenna. Figure 5.21 is a photograph of a handset antenna element. Each element is about 3 inches long.
5.5
Experimental Environment
We conducted a series of spatial signature variation measurements inside and outside of the EERL . Photographs of the indoor and outdoor
Low Earth Orbit Satellite Emulation
99
measurement sites are shown in Figures 5.22-5.26. The indoor environment is a rectangular shaped one-story building with several office rooms partitioned with lumber and concrete block construction and a large laboratory also containing a shielded room with a metal door. The building is equipped with various microwave and electronic instruments, personal computers and peripherals, an assortment of cables, wooden and metal desks, file cabinets, a van for mobile communications measurements, and bookshelves. The outdoor environment is a paved area surrounded by several buildings and metal chain-link fences.
5.6
The LEO environment, shown in Figure 5.27, was emulated by our smart antenna array and transmitting tower unit shown in Figure 5.28. The tower can be raised up to a height of 20 meters and the 900 MHz smart antenna testbed used for this experiment was described earlier in this chapter.
The distance between the tower and the antenna array was ap-
100
101
102
103
104
Experimental Setup for Smart Ant enna Systems
proximately 20 meters. The circularly polarized spiral antenna at the top of the tower was moved upwards in equal steps of 10 cm. The -20 dBm unmodulated sine-wave carrier around 900 MHz was generated by the signal generator placed inside the van. The signal was fed to the antenna and transmitted continuously while the tower was being elevated. The initial height of the tower was 6 meters and was increased to 9 meters, which covered the elevation angle range 12-21.
Chapter 6 Experimental Results for Smart Antenna Systems
6.1
Introduction
A spatial signature is the response vector of an antenna array to a mobile unit at a certain location. The mobile users at different locations exhibit different spatial signatures. The exploitation of spatial diversity (or the difference between the spatial signatures) is the basic idea of the
so-called SDMA or smart antenna scheme, which can be used to signifi-
cantly increase the channel capacity and quality of a wireless communications system. Although many existing and new array signal processing
techniques [70, 71, 72] are being proposed for the SDMA scheme, most of the performance evaluation to date relies on computer simulations or theoretical analysis based on ideal assumptions. Spatial signature variations due to the channel propagation effect have not yet been taken into account much, even though the channel propagation effect is critically important to a design engineer of a smart antenna system. For example, the variation of the spatial signature due to motion may help her determine the appropriate update rate of the spatial signature. The purpose of this chapter is to describe various propagation experiments carried out in typical scenarios, to derive the spatial signature and direction-of-arrival variations and to demonstrate the rich spatial diversity and potential benefits of using an antenna array for wireless communications. All of
This chapter was written in collaboration with S.-S. Jeng of Chung Yuan Christian University, Taiwan, and A. Kavak and G. Xu of the University of Texas at Austin.
1
106
Experimental Results for Smart Antenna Systems
these experimental results were obtained under the supervision of Prof. Guanghan Xu of The University of Texas at Austin.
6.2
Calculation of Spatial Signature Variations
If there is only one source present, as assumed in Equation (2.10), the data vector received by the antenna array can be written as
where is the source, is the spatial signature of the source, and n( . ) is background noise. Suppose that n( . ) is spatially white noise, i.e., the correlation of the noise between two antenna elements is zero. Then the spatial signature can be easily obtained via a singular value decomposition (SVD) [43, 45, 73] of the data matrix or an eigenvalue decomposition of the sample covariance matrix, i.e., , where denotes a complex conjugate and transpose operation.
Since the spatial signature is a vector instead of a scalar, it is more difficult to characterize its variation. Here, we quantify its variation by measuring both the relative amplitude and angle changes. The relative angle change of two spatial signatures and is defined as:
The relative amplitude change of two spatial signatures defined as:
and
is
where
denotes the norm of vector b.
The relative angle change defined in Equation (6.2) helps determine the update rate of the weight vectors for downlink beamforming. For example, in a Time-Division-Duplex (TDD) system where the uplink and downlink share the same carrier, we can design and keep a weight
Spatial Signature Stability for Stationary Users
107
vector of a smart antenna system based on the spatial signature received at the time slot such that for the downlink. At the time slot, the signal received by the mobile user will be s(t), where and are normalized vectors. If the update rate is fast enough, so that or the relative angle change the mobile user will
receive the maximum signal power. However, if the update rate is slow,
so that or the relative angle change the mobile user will not receive any signal power, causing deep fading. A physical interpretation of the relative angle change is that signals with the same angle (0 degree relative angle difference to the base station) have a 0% relative angle change while signals which are orthogonal (90 degree relative angle difference to the base station) with each other have a 100%
relative angle change. The relative amplitude change defined in Equation (6.3) gives
the relative signal power that can be received from an incident field
with spatial signature
at the
time slot relative to the signal power
received with spatial signature at the time slot. This definition is very similar to the standard definition of multipath fading seen by a
single antenna.
6.3 Spatial Signature Stability for Stationary Users
As explained in Chapter 2, the spatial signature represents the response of an array antenna to an emitter at a certain location in a given environment. In a wireless communications system, the users are located in different locations and therefore have different spatial signatures at the base station antenna array. Exploiting the differences of the spatial signatures, we can selectively receive and transmit multiple cochannel signals without creating significant interference among the users.
To achieve a certain required level of isolation between two not widely separated sources in a typical scenario, one needs to know the stability of the spatial signature.
For this experiment, we placed a fixed emitter first indoors and then outdoors and measured the spatial signatures over 8 hours, taking
108
one snapshot every 5 minutes with our 900 MHz smart antenna testbed. The results are shown in Figures 6.1 and 6.2, which illustrate that the relative amplitude change of the spatial signature was within and that the relative angle change of the spatial signature was less than 14%. Thus, the spatial signatures do not vary significantly over a long period of time as long as the mobile unit is stationary. It is also apparent that the spatial signatures measured outdoors vary more than those measured indoors. Some of the spikes in the outdoors case may have been caused by moving objects, such as vegetation in the wind or passing vehicles. The outdoor environment cannot be maintained as stationary as the indoor environment for such a long time. It should be noted that the outdoor experimental environment was a research campus, which may have less traffic density than other environments. Furthermore, the indoor environment was a laboratory with 4 adjoining offices which was occupied by only five people at the time of the measurement. Other indoor environments would probably be busier. If the environment has greater traffic density, then we may expect more variation in the spa-
109
tial signature. The results for the indoor environment only are shown in Figures 6.3 and 6.4 because of their importance to wireless LAN systems and because they were somewhat obscured by the scale needed by
the outdoor results. These figures illustrate that in the indoor case the
relative amplitude change of the spatial signature was within

and that the relative angle change of the spatial signature was less than
4%. Figures 6.1-6.4 show that the spatial signatures for both indoor and outdoor environments do not vary significantly over a long period of time as long as the mobile unit is stationary, which reduces the rate which
the spatial signature measurements need to be updated. The stability of the spatial signatures for our 1.8 GHz testbed was also examined. For this scenario, we again placed a fixed emitter in
the same indoor environment and measured the spatial signatures over a period slightly longer than 8.5 hours, taking 100 snapshots over that period with our 1.8 GHz smart antenna testbed. The results are shown
in Figures 6.5 and 6.6, which are fairly consistent with our results from the 900 MHz testbed. It appears that our 1.8 GHz testbed is less stable
110
111
112
than the 900 MHz testbed, which is consistent with other measurements
comparing the two testbeds. The 1.8 GHz testbed had just slightly more relative amplitude change than the 900 MHz testbed for the stationary transmitter, dB as compared to dB. The relative angle change between measurements was also slightly higher for the 1.8 GHz testbed
than the 900 MHz testbed, 6% to 4%, for this stationary transmitter
indoor spatial signature stability test. However, the results for the 1.8 GHz testbed still show that the spatial signature is largely stable over a
long period of time for a stationary transmitter. The for the implementation of a smart antenna system.
dB relative
amplitude change and less than 6% relative angle change is not significant
We also studied the stability of spatial signatures for stationary transmitters when there were people moving around in the environment,
as would be the case in a realistic environment. Figure 6.7 shows the
Spatial Signature Variation with Frequency Shift
113
spatial signature changes within 1 minute for a stationary transmitter when there were 6 people walking around the antenna array and the transmitter using the 900 MHz testbed. From the figure, it is easy to see that the spatial signature does not vary significantly within 1 minute, and definitely does not vary significantly between 10 second intervals. The spatial signature changes for the more ideal case of a stationary transmitter without the people walking around the antenna array and the transmitter produced similar but slightly better results, as expected.
6.4
Spatial Signature Variation with Frequency Shift
In current cellular telephony near 800 MHz, downlink and uplink frequencies have about a 45 MHz separation [1]. It should therefore be interesting to determine the variation of the spatial signature across such a frequency difference. The result illuminates the feasibility (or lack
thereof) to expand the capacity of the already existing infrastructure of reusing uplink spatial signatures for spatially selective transmission in
the downlink. In the experiment, we fixed the location of the mobile transmitter and varied the carrier frequency from 874 MHz to 924 MHz in 5 MHz steps. This measurement was made outdoors at a site with two dominant DOAs. Figures 6.8 and 6.9 depict the variation of the spatial signature with carrier frequency. Obviously, the spatial signature variation is quite significant even for a small percentage change (< 5%) of the carrier frequency. Although the amplitude variation may be due to the narrow bandwidth of the patch antennas, the angle variation clearly shows the high sensitivity of the spatial signature to the carrier frequency. This will be explained in detail in Section 6.7. The results implies that the uplink spatial signature can not be used directly for downlink beamforming in a Frequency-Division-Duplex (FDD) system.
6.5
Spatial Signature Variation with Displacement
Another issue we studied was how the spatial signatures differ for transmitters at different locations. If the spatial diversity between
two terminals is too small and they are assigned to the same time slot,
114
115
then the base station may not be able to separate them. We studied the variation of spatial signatures at different locations by randomly placing a transmitter at 16 different locations to transmit to our 900 MHz testbed. A location for comparison purposes was chosen at random for each of the 16 locations and their relative amplitude and angle changes were calculated, with the results shown in Figures 6.10-6.11. From Figures 6.10-6.11 we can see that the spatial signatures vary significantly between different locations. The relative amplitude change between two locations may be up to 25 dB. Furthermore, the relative angle change between two locations may be up to 98%. The mean and standard deviation of the relative amplitude change is approximately 0 dB and 6.55 dB, respectively. The mean and standard deviation of the relative angle change is 82.98% and 17.42%, respectively.
We also studied the variation of spatial signatures by using our 1.8 GHz testbed. We picked various locations in our indoor environment shown earlier and then moved the transmitter slightly to record the spatial signature at 21 different closely spaced positions at that location.
116
Each of the 21 closely spaced positions were compared to the previous position and their relative amplitude and angle changes were calculated, with the results shown in Figures 6.12-6.13 for a typical location. The results from a second location are shown for comparison purposes in Figures 6.14-6.15. We see from both locations that the small displacement between positions results in only a small change in amplitude, less than
1 dB. However, even with this small displacement between positions, the
relative angle change is significant. The major differences between these results and the previous displacement results are that these results are for our 1.8 GHz testbed instead of the 900 MHz testbed and the displaced locations were selected to be closely spaced instead of at random locations in our indoor
environment. From the experimental results from both testbeds, we can
conclude again that the spatial signature varies greatly between different locations.
Therefore, due to a large amount of short-delay multipath in an

indoor environment, transmitters at different locations usually possess
117
118
SS Correlation for Closely Spaced Transmitters
119
significant spatial diversity. In the unlikely event that two terminals have a high spatial signature correlation, the base station can keep a record of the pair and take care not to assign them to the same time slot.
6.6
Spatial Signature Correlation for Closely Spaced Transmitters
Another study was conducted to examine the spatial signature correlation for closely spaced transmitters. If the spatial diversity between two terminals is too small and they are assigned to the same time slot, then the base station may not be able to separate them. In order to ascertain the feasibility of the smart antenna concept in wireless communications environments, a uniform linear antenna array of five elements with separation 10 cm and carrier 1.5 GHz was built and experimental studies were conducted of some typical scenarios. This is the third testbed whose results are included in this chapter, though this testbed was actually the first one constructed. The description of the smart antenna testbed with Figure 5.1 and the indoor environment of Figures 5.22-5.23 still apply to this testbed. The antenna array was placed near the ceiling of the laboratory at a height of around 3 m. The 50 transmitters, whose positions are shown in Figure 6.16, transmitted QPSK signals and were placed at heights of 1 m (locations 1-25) and 1.3 m (locations 26-50).
The correlations between closely spaced nodes with a range from 10 cm to 50 cm, placed at various points in our laboratory are shown in Table 6.1. From the table, is easily seen that most closely spaced
120
pairs have significantly different spatial signatures. Therefore, due to a large amount of short-delay multipath in an indoor environment, a little
spatial separation usually means significant spatial diversity.
6.7
Spatial Signature Variation Due to Small Displacement in Different Multipath Environments
It is well known that the fading rate of a mobile terminal is determined by its speed [2, 3]. One would also expect that the rate of spatial signature variation is proportional to terminal speed. This rate determines the beamforming update rate for an antenna array system. In our experiments, we chose three cases each for indoor and outdoor scenarios and our 900 MHz testbed to study the rate of spatial signature variation with small displacement: (a) one strong DOA, (b) two strong DOAs, (c) many DOAs. In each case, we calculated the spatial sig-
SS Variation from Small Displacement in Multipath Environs
121
nature variations by moving the transmitter along a line with a step size of 3 cm (equivalent to about ). The total displacement was 30 cm (equivalent to about ). The relative angle and amplitude changes between two adjacent spatial signatures for the indoors and outdoors cases are shown in Figures 6.17 and 6.18 and in Figures 6.19 and 6.20, respectively. The mean and standard deviations of the relative angle and amplitude change for those three cases were also calculated and are listed in Table 6.2. It should be noted that the results exhibited in Figures 6.17-6.20 and Table 6.2 are typical cases, representing the range of spatial signature variations of all 23 sets of measurements taken.
From Figures 6.17 and 6.19, we can see that the relative angle change in case (a), i.e., with one strong DOA, is much smaller than that in cases (b) and (c). This can be explained as follows: the spatial signature in case (a) is virtually a scalar multiple of a , while those in cases (b) and (c) are a linear combination of i.e., as given by Equation (2.14). Although small displacement does not cause too
122
SS Variation from Small Displacement in Multipath Environs
123
124
much change of the amplitude of it does change the path length of each multipath component, and therefore changes the phase of Consequently, the spatial signature may also change significantly with small displacement of the mobile terminal. This also explains why the spatial signature is sensitive to the carrier frequency. The 45 MHz difference in the uplink and downlink carriers only accounts for 5 % change
in the range of 800 900 MHz. Such a small change does not cause
any significant change to given by Equation (2.14). However, the
5 % change can lead to significant variation of the phase of each multipath component and eventually the spatial signature. It is interesting to note that the spatial signature varies more significantly in case (b)
than in case (c). One possible explanation is that the cancelation effect
is stronger with two DOAs since the spatial signature may change from
We seem to find a type of averag-
ing effect in our scenarios with many multipath DOAs, which was an
expected and desirable result.
Figures 6.18 and 6.20 confirm that the amplitude changes in all three cases are small, less than 4 dB. The mean and standard deviations of the relative amplitude changes are all less than 1.03 dB. These results demonstrate that a smart antenna system with an update rate faster than the rate of change of the propagation channel can, in most cases, mitigate multipath fading by exploiting spatial diversity. Figure 6.21
further illustrates this point with a typical sample taken from the entire
set of 23 measurements, comparing worst case amplitude variations for a single antenna to those for an antenna array with 2, 4, and 8-elements. It
is evident from Figure 6.21 that the averaging effect among the antenna elements clearly results in an improved worst case amplitude variation as
the number of antenna elements are increased. Again, this was a result
predicted from the theory and is intuitively understandable. The one

exception to this is the case when there are many multipath components with similar DOAs, which will be discussed in the next section.
Multipath Angle Spread
125
6.8
The angle spread of multipath signals is generally defined as the maximum difference of DOAs among significant multipath components.
The angle spread of multipath signals is critical to the multipath fading characteristics of an antenna array as well as the feasibility of using direction finding techniques for selective uplinks and downlinks. If the angle spread is small, i.e., , then by (2.10)
In this case, the spatial signature is basically a scalar multiple of the array manifold This situation may be illustrated with the following two hypothetical scenarios: (1) There is only one direct path without any other significant multipath. (2) There are many multipath components with similar DOAs (localized scatterers). The fading characteristics of these two scenarios are completely different. In the first case, there is virtually no fading even with a single antenna, while in the latter case,
126
fading can be quite severe even with an antenna array because the spatial diversity cannot help much.
To quantify these fading characteristics, we chose two canonical cases and measured relative amplitude changes. In the first case, the transmitter was placed in an open field at a distance from the base station of about 400 meters. Since there was no scatterer close to the transmitter, only one dominant DOA was detected (the specular reflection has the same DOA). In the second case, the transmitter was set up in the parking lot close to a research building. The parking lot was nearly filled with cars, all acting as scatterers. Since the distance between the transmitter and the scatterers was small compared to the distance between the transmitter and the base station, we could view this as the case with many multipath components with similar DOAs. The relative amplitude changes are given in Figures 6.22 and 6.23. The relative amplitude changes shown in Figure 6.22 for case (1) measured with either a single antenna or the antenna array were all less than 3 dB. There was no severe fading point even with a single antenna; on the other hand,
the maximum relative amplitude change shown in Figure 6.23 for case
(2) measured with the antenna array was up to 6 dB. Therefore, fading was very severe even with an antenna array. In either case, DOA-based selective uplinks and downlinks may be quite effective, however, since direction diversity is the only diversity left to exploit. If the angle spread is large, the above conclusions are the opposite, i.e., direction finding based techniques are not that effective and the spatial diversity of multiple antennas can be exploited to combat multipath fading. To find the DOA distribution in different environments, the following three cases were chosen for measurement of their multipath angle spread: (1) The base station and the mobile transmitter were place inside the laboratory space in the EERL , as shown in Figure 5.22 for the indoor measurement. (2) The base station was placed outside the EERL, and the mobile transmitter was set up in an open field. There was nothing between the base station and the transmitter blocking the direct path of the outdoors line-of-sight (LOS) measurement. (3) The
127
128
base station was located outside the EERL, and the mobile transmitter
was placed in a location where the LOS was blocked by a building. In each scenario, we moved the mobile transmitter to several randomly selected positions. At each position, the number and intensity of DOAs were detected by the ESPRIT algorithm and forward/backward spatial
smoothing techniques. Limited by the number of array elements and the direction finding algorithm, we can reliably estimate at most 4 major DOAs. In the experiments, we obtained and analyzed 102 sets of DOA
data but only exhibit 40 typical cases measured at 16 positions for the
indoors measurement, 16 positions for the outdoors LOS measurement
and 8 positions for the outdoors blocked LOS measurement. At each position, we measured the DOA angle spread with signal level thresholds of
3 dB, 6 dB, and 9 dB. For this reason, we defined DOA spread as the maximum angle of arrival difference between the DOA of the direct
path and any other DOAs whose intensity was higher than the specified threshold.
The mean and standard deviation of the angle spread results for all three cases and thresholds are summarized in Table 6.3. From these results, we can see that the angle spread of outdoors LOS case was
much smaller than that of the indoors and outdoors blocked cases. In the outdoors LOS environment, the angle spread of the multipath may
be small since most of the significant multipath components are from
129
local scatterers within 100-1000 wavelengths, which is small in comparison with the distance between the base station and the mobile terminal. This is probably not the case in an indoors environment. Since the path lengths of direct path and multipath components are not so different in the indoors environment, we expect to see more significant multipath components and wider angle spread. For the outdoors blocked environment, since there is no direct signal path, multipath components can be from scatterers in any direction. The angle spread of the significant multipath components is quite arbitrary in the blocked case. Therefore the mean and standard deviation of angle spread in the blocked case are larger than those of the other two cases examined, as expected.
The RMS angle spread, which takes into account the relative power of each multipath component, was calculated as:
where
where P is the total number of positions measured in each case, is the measured DOA and represents the phase and amplitude difference between the multipath and the direct path at the position. The results are summarized in Table 6.4. We can see that the RMS angle spread statistics shown in Table 6.4 are consistent with those shown in Table 6.3.
130
The experimental results suggest that DOA-based techniques may not be too effective in indoors and outdoors blocked scenarios. We also have shown that multipath fading can be overcome by exploiting
spatial diversity. However, if the angle spread is the result of some well defined specular components arriving at the smart antenna, with a reasonably well-designed downlink beamforming strategy, some significant
downlink improvement might still be achieved.
6.9
Feasibility of Downlink Beamforming

In Time-Division-Duplex (TDD) systems, such as DECT and CT-
2, the carrier frequencies for uplinks and downlinks are the same. In this
case, the uplink spatial signatures obtained from blind-type algorithms can be directly used for downlink beamforming. To verify this claim experimentally, we formed downlink weighting vectors based on the uplink spatial signatures of two mobile users and transmitted two user-specific tones back to the mobiles, applying their corresponding weighting vectors. Figures 6.24 and 6.25 show the spectra of the signals received by the two mobile users with two different angular separations, 20 and 3 degrees, respectively. We can see that the co-channel suppression achieved by spatial signatures downlink beamforming is quite significant (> 13 dB) even when these two sources are very close.
In Frequency-Division-Duplex (FDD) systems, such as GSM, IS54, and IS-95, the uplink and downlink frequencies are significantly different (45 MHz apart). In this case, we must construct the beamforming weights according to the DOAs of all the sources. We also implemented a simple weight design method, i.e., taking the pseudoinverse of a matrix with the steering vectors corresponding to all the direct path and multipath DOAs. The results are shown in Figures 6.26-6.27. The result for 20 separation shown in Figure 6.26 is comparable to that in Figure 6.24, while the result for 3 separation shown in Figure 6.27 is much worse than that in Figure 6.25, since the DOA-based beamforming has a fundamental resolution limitation. This result is another example of the difficulties that smart antenna systems have with FDD systems.
Feasibility of Downlink Beamforming
131
132
Fading Reduction and Diversity Gain
133
6.10
The benefits of antenna diversity to mitigate multipath fading were studied extensively in our experiments. The reason this is important is because multipath fading is a serious problem in wireless communications systems and can be the limiting factor in a system. Two typical path scenarios, LOS and blocked, were chosen for the measurement of diversity gain and fading. In each scenario, the mobile transmitter was moved to eleven consecutive positions with 3 cm increments.
To study the performance of antenna systems with 1, 2, 4, and 8-elements, the following three cases were chosen for measurement of diversity gain and fading: (1) The base station was placed outside the EERL , and the mobile transmitter was set up in an open field. There was no obstacle between the base station and the transmitter blocking the direct path of the outdoor LOS measurement. We did not expect any significant multipath components in this scenario as there were no flat objects near the mobile transmitter to generate significant specular reflections. (2) The base station was placed outside the EERL, and the mobile transmitter was set up in front of a nearby research building. Again, there was no obstacle between the base station and the transmitter blocking the direct path of the outdoor LOS measurement. We expected one significant multipath component, however, because of the building behind the mobile transmitter. (3) The base station was located outside the EERL, and the mobile transmitter was placed in a location where the LOS was blocked by a building. Since there was no direct path signal, we expected to have many significant multipath components for this scenario. In these three cases, we collected data at eleven neighboring positions by moving the mobile transmitter in small 3 cm steps along a
straight line. We then measured the received power for systems with 1,
2, 4, and 8-antenna elements and compared their system gains and variations. For the uplink, we captured the SS, cophased and combined the individual signals, and compared each systems gain. The SS was computed at each uplink transmission point. For the downlink, we used the SS from the first position to calculate the weight vector for beamformed
134
transmission. As we move the user to each of the 10 other points, we would naturally expect that the users received power should decrease from its initial value. This is because, as shown in Section 6.7, the spatial signature of the mobile unit changes as it is moved; hence, our estimated weight vector should perform poorer as the mobile unit is moved away from the first position.
It should be noted that the cases described below and in Section 6.11 were chosen from and are representative of a total of 23 sets of measurements. Finding exhaustive statistics for typical environments would require many more measurements, but the results presented here give insight into the magnitude of diversity gain, fading reduction and beamforming performance of an antenna array.
To compare system gains and variations in each case, the diversity gain for the uplink is calculated as
where is the power received by the antenna array with the first i antenna elements when the mobile terminal is at the position. For example, is the power received by the antenna array using just the first antenna element when the mobile terminal at the first position. The diversity gain for the downlink is similar to that for the uplink except that the power is received by the mobile terminal. The peak-to-peak fading for the uplink is calculated as
where and are the maximum and minimum power received by the antenna array using the first k antenna elements when the mobile
terminal was moved within (11 positions to select from), respectively.
The peak-to-peak fading for the the downlink is similar to that for the uplink except that the power is received by the mobile terminal. The mean diversity gain and peak-to-peak fading of an antenna array moved within for the uplink and downlink in the three cases are shown in
Tables 6.5 and 6.6, respectively.
135
In case 1, we received a single dominant DOA component with no significant multipath signals because we had a LOS scenario with the mobile unit in an open area. In this ideal case, we did not expect that fading would be a problem because the SS should not change much. This can be seen from the spatial signature formula
with all the Thus, because , the array manifold vector does not vary much for small displacements. Results for the uplink and downlink scenarios are plotted in Figures 6.28 and 6.29, respectively.
136
As expected, the system experienced insignificant fading in all antenna cases, with only a small fading reduction (0.7 dB peak-to-peak) for the 8-element system over even the 1-element system. It should be noted that the gain in received power by using an antenna array is evident in this ideal scenario, with Figures 6.28 and 6.29 showing that the 8element system obtained about a 8 dB gain improvement (close to the 9 dB theoretical maximum) over the 1-element system. In this case, the SS does not need to be updated frequently, because the SS does not change much when we only have a single dominant DOA component.
In case 2, we received two significant components, one along the direct path (because we had a direct LOS) and one dominant multipath (from the building behind the mobile transmitter). Our results for the uplink and downlink scenarios are shown in Figures 6.30 and 6.31, respectively. We found that we could not mitigate multipath fading effectively in any of the antenna element cases, with even the 8-element case experiencing significant fading. However, the amount of fading decreased slightly as the number of antenna elements was increased. For
137
the uplink, the peak-to-peak fading was found to be 9.0 dB for the 1element system, 7.6 dB for the 2-element system, 7.2 dB for the 4-element system, and 6.3 dB for the 8-element system. For the downlink, the
peak-to-peak fading was found to be 9.0 dB, 7.6 dB, 7.6 dB, and 6.7
dB, for the 1, 2, 4, and 8-element systems, respectively. The reason

for the lack of success in significantly reducing multipath fading is that
the two dominant signals have very similar DOA angles, i.e., Thus, As shown in [38], usually changes rapidly with movement; consequently, fading in this scenario can be quite severe, even with an antenna array. Hence, using an antenna array in this type of scenario is not effective to combat fading because spatial diversity does not help much.
It should be noted that the gain in received power was again improved as we increased the number of antenna elements in this scenario, as in the first case. For the uplink, the diversity gain improvement was 3.2 dB, 7.3 dB, and 11.4 dB, for the 2, 4, and 8-element systems, respectively. For the downlink, the diversity gain improvement was 3.2, 7.1,
138
139
and 11.2 dB, for the 2, 4, and 8-element systems, respectively. Just as in the first case, the SS does not need to be updated frequently because the SS does not change much with two dominant signals, which is evident from how the downlink results (which used the SS from the first position for the downlink beamforming weight vectors for all 11 positions) mirrored the uplink results.
In case 3, we received many significant multipath signals because the LOS was blocked by a building. Our results for the uplink and downlink scenarios are shown in Figures 6.32 and 6.33, respectively. Due to the lack of a LOS signal and the numerous multipath signals, each of which are extremely sensitive to movement, we found that the 1-element system had significant fading (12.5 dB peak-to-peak), as expected. This is because as we move, all of the values change, with a displacement resulting in a 360 phase change. Consequently, the 1-element
antenna is unable to combat multipath fading, with more than 10 dB
of fading within just a 30 cm displacement, which could be catastrophic in a real system. The reason that systems with more antenna elements were able to successfully combat the fading problem can be seen from Equation (6.8). The fading is caused by the fluctuating values. Since is independent for different antenna elements, the averaging effect over the antenna elements reduces fading. Hence, for the uplink, systems with multiple antenna elements performed significantly better in terms of both fading and gain, with the performance increasing as the number of antenna elements increased, as expected.
The peak-to-peak fading for the uplink was 3.8 dB, 2.4 dB, and 2.2 dB, for the 2, 4, and 8-element systems, respectively. Note that all of these fading values are significantly lower than the 12.5 dB measured in the 1-element case. The downlink results were surprising, with peak-topeak fading of 12.5 dB, 2.5 dB, 4.5 dB, and 7.3 dB, for the 1, 2, 4, and 8-element systems, respectively. In this downlink scenario, the 8-element case performed worse than the 2 and 4-element cases in terms of fading reduction. It can be seen from Figure 6.33 that the performance loss was caused by spatial signature mismatch over space, with the transmission loss increasing with displacement. Similar results were found for the
140
Beamforming Algorithms for SIR Improvement
141
downlink diversity gain measurements. Consequently, the SS needs to be updated much more frequently in this case than in the other two cases because the SS changes rapidly with displacement. We can conclude from our results that the SS needs to be updated within a (1/2) displacement for this blocked (no LOS) scenario.
6.11
The SIR of a system is critical because in many cases it is the limiting factor that determines the capacity of a system. In order to measure the SIR of various systems, we simulated a situation with two moving users. We then calculated the SIR of the two mobile users based on the four beamforming approaches discussed in Section 2.4 in three distinct but typical cases.
In case 1, we received a single dominant DOA with no significant
multipath signals because we had a LOS scenario with the mobile users in open areas. We separated the two mobile users by approximately 100 meters; however, we placed the mobile users such that they were very close to each other in the angular direction from the base station (less than 1 of angular separation). This is a particularly catastrophic scenario for the DOA techniques because it is extremely difficult for the base station to separate the two mobile users and isolate the interference source, especially since there are no significant multipath signals to distinguish one mobile user from the other. This difficulty is evident in Figures 6.34 and 6.35, the SIR results for the two mobile users, where it is clear that the complex conjugate SS and the two DOA methods could not distinguish between the two mobile users, which caused the SIR to drop to around 0 dB. The complex conjugate SS, maximum DOA, and pseudoinverse DOA techniques were unable to find a null point for each mobile user, so they were not very useful in this case. The pseudoinverse SS method performed well for the stationary case but had its SIR decrease below 10 dB when a mobile user was moved as little as 0.1 While the pseudoinverse SS method achieved the best SIR of all the systems, it performed poorly in terms of diversity gain, as shown in Figures 6.36
142
and 6.37. Hence, there is a tradeoff among the beamforming techniques between SIR and diversity gain.
In case 2, user 1 received only one significant signal, from the direct path, while user 2 received two significant signals, from the direct path and a single dominant multipath. The two mobile users were placed so that they had more than 10 degrees of angular separation. The direct path and multipath signals of the two mobile users both had an angular separation of more than 10 degrees. Consequently, the separation was large enough that all four techniques achieved a SIR level greater than 12 dB, even when the displacement of a user was up to Again, the pseudoinverse SS technique performed best among the techniques. The
SIR results for each beamforming technique are exhibited in Figures 6.38
and 6.39 for users 1 and 2, respectively. From the first two cases we can note that the angular resolution of the system is critical for the performance of the DOA techniques, with the DOA techniques failing when the angular separation is too small. This is a known limitation of the DOA techniques and was expected.
143
144
145
In case 3, we placed user 1 such that he was blocked from the base station while user 2 had a LOS with the base station. User 1 therefore
had many significant multipath signals while user 2 had a dominant
direct path signal. The SIR results for the four beamforming techniques
for users 1 and 2 are plotted in Figures 6.40 and 6.41, respectively. It
is evident from Figure 6.40 that the blocked user (user 1) experienced
poor SIR for displacement greater than . This poor performance is
reasonable because the SS changes significantly in the blocked case with
even slight movement, as we also demonstrated in case 3 of the previous section. The LOS user (user 2), on the other hand, achieved a higher SIR and was much less sensitive to movement. We can conclude from this
case that the SS estimate must be updated frequently for a blocked user,
while the estimate for a LOS user can be updated much less frequently.
146
Low Earth Orbit Satellite Emulation Results
147
6.12
In current LEO satellite systems like Iridium [74], handset units using traditional omni-directional single antenna elements receive both direct path and multipath signals when the LOS is clear. These multipath signals, which degrade the reception, are the result of either diffuse scattering or ground specular reflection, depending on the roughness of the surface. With the LEO satellite distance ranging from 500 to 2000 km, a handset using a single antenna element to transmit may not have sufficient power to directly uplink because of the large path loss, coupled with the multipath losses. Utilizing a two-element phased-array antenna [75] and an Advanced Signal Processing Module (ASPM), a smart antenna system in the handset units provides a potential solution to this problem. In the downlink case, the handset can determine the highest combined received power by adaptively suppressing the fading
and adjusting the phase. For Iridium-like TDD LEO satellite systems,
for which uplink and downlink share the same carrier frequency, the spatial signatures estimated during the downlink time interval can be used during the uplink time interval. Therefore, the channel and multipath reflection effects can be eliminated and the required transmitting power can be obtained.
The uplink effective power, , received by the satellite can be maximized by assigning relative weights to each antenna element. Let us assume that the spatial signature for the downlink at time is a = [ ] and we design the uplink weight vector at time The amplitude and phase change of the received signal caused by the multipath and channel distortion effects are included in vector a. Then the transmitted signal to the satellite, will be received as
and the power received by the satellite is
where PS is the power of the transmitted signal. Assuming that the maximum power amplifier output is P watts for each antenna element,
148
then the optimization problem to find the desired weight vector is
subject to
For example, let us assume that the satellite requires 8 watts power and power amplifiers having a maximum output of just 2 watts were connected to the antenna elements in Figure 5.27. To maximize the power received by the satellite, the angle between the vectors w and a in (6.10) should be 0 degrees. The physical constraints on the w are and Assuming and . , then will be watts. This is a significant advantage because we would have transmitted only 4 watts if we had used a single omni-directional antenna. Hence, once the spatial weight vector (i.e., w) is designed for the optimum power setting, it can be used for transmitting from the handset as long as the propagation characteristics of the environment does not change.
In real scenarios, the changes in the environment characteristics may increase the difficulty in using the smart antenna system approach with current handset units. If the characteristics of the reflecting surface change as the handset is moved, we cannot make use of the downlink spatial signatures. The variation of the propagation channel characteristics will be mostly due to the position change of the satellite. The handset can be considered almost stationary when compared to the motion of the LEO satellite. For example, the Iridium system is designed to have an orbiting period of 100 minutes at a distance of 700 kilometers above the earth. The propagation delay will be approximately 2.5 milliseconds.
Many experiments were conducted to emulate a LEO satellite scenario using the 900 MHz testbed with the transmitting tower described in Section 5.6. In the outdoor environment used for this experiment, the reflecting surface was asphalt. Both ground reflected and direct path signals were received by the vertically placed array antenna and processed by the smart antenna system. Directivity of both antennas ensured the
149
proper transmission and reception. Two different experiments were conducted. In the first experiment, all eight channels of the RF unit were connected to the eight patch elements of the antenna array. In the second experiment, two channels (channels 7 and 8) were connected to the patch elements numbered as 5 and 6 as counted from bottom to top, while the antenna was lowered to a height of 1.25 meters. The path length difference between the direct path and the multipath signals was very small (e.g., several nanoseconds), and it was difficult to identify them based on their time-diversity. However, since they had significantly different angles of arrival, we measured their DOAs using our antenna array testbed. Spatial smoothed ESPRIT was used to estimate the DOAs. From these DOA values, we made sure that we received only two strong direct and ground reflected signals. During the experiment, we observed that the relative strength of these two signals changed as the transmitter moved upwards. In both experiments, the received power variation for each channel was calculated and normalized.
150
The variation of the normalized power level as the elevation angle of the transmitter increased is shown in Figure 6.42. When the power received by channels 3 and 5 were combined by equal gain combining, the power fluctuation was limited to 3 dB. On the other hand, the fluctuation of the power levels of the individual channels 3 and 5 were around 9 dB and 7 dB, respectively.
In the second experiment we found that we obtained a smoother fluctuation of the received power when two channels were combined, as shown in Figure 6.43. Fluctuation was around 2 dB until the tower was elevated approximately by 1 meter, and remained almost constant after that point. The variation was observed to be around 7 dB for each individual channel.
The relative amplitude and angle changes of the spatial signatures of array elements 3 and 5 are shown in Figures 6.44 and 6.45, respectively. The spatial signatures for each measurement set were estimated as the elevation angle of the transmitting tower was increased. As
151
152
seen from Figure 6.44, the relative amplitude change of the spatial signature was around 0.5 dB in the elevation angle ranges 13.5-17 and 1820.5. The relative angle change shown in Figure 6.45 is more sensitive to the elevation angle change. The relative angle change was approximately 10% in the elevation angle ranges 13.5-17 and 18-20.5. Because we had strong DOAs from both the direct and ground reflected paths, the relative angle of the spatial signature vectors is affected significantly by even a slight change in the elevation angle.
In Iridium-like LEO satellite systems, the uplink and downlink time interval is less than 90 ms. It is important to evaluate the spatial signature change due to the movement of the satellite within 90 ms. Based on preliminary calculations, the elevation angle change within 90 ms is less than 0.3. Therefore, we need to evaluate the effect of the spatial signature change by calculating the correlation of the spatial signatures with an elevation angle change of 0.3. The results are shown
Summary of Experimental Results
153
in Figure 6.46, in which the elevation angle ranges from 12 21. We can see that the effect due to the satellite motion is not significant, usually smaller than a 0.1 dB variation of power the satellite receives, with the worst case being a 0.5 dB difference.
The estimated DOAs using ESPRIT [76] vs. tower height change are displayed in Figure 6.47. A spatial smoothing factor of 2 was used. These DOAs represent the elevation angle of direct and ground reflected signals. Both angles changed nearly 10. As expected in theory, the direct path angle was always smaller than that of the reflected path.
6.13
The variations of the spatial signature with time, frequency, and displacement were presented. Multipath angle spread and spatial signature and DOA-based beamforming algorithm performance were examined. The results from our smart antenna testbeds showed: (1) The
154
variation of the spatial signature is very small if the mobile unit and its surroundings are stationary. (2) The spatial signature changes significantly with the carrier frequency. (3) The results of spatial signature variation with small displacement in different multipath environments show that there exists rich spatial diversity to be exploited by an antenna array for capacity expansion and performance improvement. (4) The angle spread of the multipath is critical to the multipath fading characteristics of an antenna array as well as the feasibility of using direction finding techniques for selective uplinks and downlinks. The multipath angle spread is small in outdoors LOS environment, while wider angle spread is observed in indoors and outdoors blocked environments. (5)
The co-channel suppression achieved by applying spatial signature-based
downlink beamforming is quite significant even if two mobile users are very close to each other. DOA-based beamforming could apply to both TDD and FDD systems in environments with limited numbers of significant multipath signals such as rural areas, though the DOA-based beamforming methods have a resolution limitation.
From these experimental results we conclude that in a TDD system, spatial signature-based beamforming could achieve significant co-
channel interference suppression even when the sources are very close to each other if there are rich multipath signals. In an FDD system, the spatial signature-based beamforming will not work since the downlink spatial signature will be significantly different from the uplink spatial signature due to the frequency difference of the links. In a TDD system, spatial signature-based beamforming is very suitable to communications applications, e.g. PCS, wireless PBX/LAN and wireless static networks, where the mobile units do not move rapidly.
We also presented experimental results for evaluating the performance of smart antenna systems for diversity gain, fading reduction, and SIR performance. In terms of fading reduction, we found that all antenna systems performed well for the single DOA case, all antenna systems performed poorly for the similar DOAs case, and only the 1-element system performed poorly for the many DOAs with wide angle spread case. For diversity gain, we found that increasing the number of antenna elements
155
in the system increased the diversity gain for the uplink. In all cases, the 8-element antenna system performed best for fading reduction and diversity gain for the uplink. These same conclusions can be made for the downlink if the update rate is adequate.
For SIR performance, we found that when the users were close in angular direction, only the pseudoinverse SS technique performed well for a stationary user. However, even that technique performed poorly when the users were moved slightly. When the users had adequate angular separation, all four algorithms performed well. When one user was blocked from the base station while the other user had a clear LOS, all four algorithms performed well for the LOS user, while only the pseudoinverse SS algorithm performed well for the blocked user (when that user was kept stationary). In fact, the pseudoinverse SS method achieved the best SIR performance among all the algorithms for the stationary case; however, it sometimes performed poorly in terms of antenna diversity gain as compared to the conjugate SS and maximum DOA algorithms.
Experimental studies also showed the feasibility of employing a smart antenna system in LEO satellite handset units to mitigate the losses due to multipath fading and increase the uplink power efficiency. The results showed that multipath fading could be significantly reduced via the use of a phased antenna array instead of a single antenna. The results also showed that there was not much variation in the relative amplitude and angle of the spatial signatures of the two-element array when the elevation angle of the LEO satellite changed by 0.3 for the angle range 12 21. In TDD Iridium-like LEO satellite systems, the uplink and downlink lime interval is less than 90 ms, during which time the satellite would change its elevation angle by less than 0.3. Hence, it is feasible to utilize the downlink spatial signature for uplink, which would result in significantly more power received by the satellite. With a simple phased array antenna (a 2-element array) and signal processing software, we can significantly improve the power amplifier efficiency, resulting in a potential considerable reduction of complexity and cost.
Chapter 7
7.1
Introduction
Numerous computer simulations were created using MATLAB to aid in the design and evaluation of the SWL system. These results were also compared with the experimental results to validate the models used and to try to extend the conclusions that could be drawn. Since an 802.11-compliant smart antenna system was not available to test the SWL system design of Chapter 4, computer simulations were the only way to evaluate the system design. These results were then used to upgrade the design of the SWL experimental testbed which will be built at Santa Clara University.
7.2
SIR Performance for Beamforming Algorithms
The SIR is critical because in many cases it is the limiting factor that determines the capacity of a system [12]. The simulations studying the SIR performance of the beamforming algorithms for the SWL system made the significant assumption that all users were stationary. The stability of the spatial signature for stationary users was shown earlier by the experimental results. The standard scenario envisioned for this firstgeneration simulation for a wireless LAN terminal is for the terminal to remain stationary while in use. If a user moves or if its spatial signature changes significantly for any reason, the base station will request that the user send a message during the contention period to update its spatial signature. Experiments studying the case of an indoor moving user have been completed (using the 1.8 GHz testbed) and a model will be
158
constructed after the data is analyzed in order to create an adaptive algorithm to try to allow for real-time updating of spatial signatures when there is movement by users.
The four beamforming techniques described in Chapter 2 were simulated via MATLAB to study their performance in terms of interference rejection [14]. These techniques were the pseudoinverse SS, conjugate symmetric SS, pseudoinverse DOA and dominant DOA algorithms. Other beamforming techniques have also been studied for the SWL system but their results are not shown here due to their more advanced and specialized nature.
The SIR results, averaged among the users, are shown in Table 7.1 for the direct-path only case with 2- and 4-users in the ideal situation where the spatial signature is known exactly by the base station. It is evident from the results that all algorithms performed adequately when the DOAs are spread (greater than 10 degree separation), with the pseudoinverse methods clearly superior. However, the closely spaced case (less than 2 degree separation) had poor results for all but the pseudoinverse methods. The reason behind the poor performance is that the conjugate symmetric SS and dominant DOA beamforming techniques cannot separate terminals when their angle separation is too small. The closely spaced case mirrors the results from Case 1 in Section 6.11 (where the terminals were placed within 1 degree of each other, as in the simulation) of the experimental results, and the case where the DOAs are spread mirrors the results of Case 2 in Section 6.11. The
SIR Performance for Beamforming Algorithms
159
major discrepancy in the results is that the SIR results are better in the computer simulation than in the experiments, especially for the pseudoinverse DOA case. Also note that the direct-path only scenario results in the two pseudoinverse methods having basically the same results because, in the absence of noise and with perfect knowledge of the spatial signature, they are identical in the 2-user case and very similar in the 4-user case. This similarity is also true between the conjugate symmetric SS and dominant DOA techniques. When the performance was greater than the numerical accuracy of the MATLAB simulation, the results are indicated as > 30 dB. The reason for the extremely high performance of the pseudoinverse beamforming methods is that, for this ideal case of perfect spatial signature knowledge and direct-path only signals, the pseudoinverse methods are able to create ideal nulls for the interference sources. Since the interference signals are effectively eliminated, the SIR (which now has a signal power level around 1 divided by an interference power level around 0) approaches infinity. The performance of the first simulation was much better than that of the experimental results, a result that was immediately investigated to find the cause of the discrepancy. To check whether the assumption of exact knowledge of the spatial signature was too ideal, a new computer simulation was created that assumed that all signals and the previously obtained spatial signatures were corrupted by independent Gaussian noise before the weight vectors were computed by the beamforming algorithms. The noise level was set such that the desired SNR level would be experienced by each terminal and the previously obtained spatial signature would be corrupted by the same level of noise (with all noise sources generated independently) before the weight vectors were computed by the beamforming algorithms. The SIR results for this more realistic simulation with a 10 dB SNR level are shown in Table 7.2 for the direct-path only case. Clearly, each beamforming algorithms performance suffered from the additive noise and imperfect spatial signature knowledge for the beamforming algorithms, especially the pseudoinverse DOA technique. The simple direct-path only scenario is shown because it allowed for comparison with the experimental results and the previously shown simulation results. The SIR results obtained
160
for both the closely spaced and spread cases are consistent with the experimental results when the terminals were stationary.
The SIR results for the scenario where all users have multipath and direct-path signals with the same type of imperfect spatial signature knowledge for the beamforming algorithms and a 10 dB SNR level are in Table 7.3. The conjugate symmetric SS and dominant DOA techniques suffered the most from the addition of multipath and the pseudoinverse SS technique still performed the best, all results as expected from our experimental results.
In all cases, the pseudoinverse SS technique performed the best, as in our experiments. The pseudoinverse DOA technique performed better than the dominant DOA and conjugate symmetric SS techniques, also as in our experimental results for the stationary case. The performance of the computer models was directly affected by the noise level in
Dynamic Slot Allocation for Worst-Case SINR
161
the system and by the accuracy of knowledge about the spatial signature, as expected.
7.3
The results for the First Fit and FCFS algorithms, sorted and unsorted, are shown in Figure 7.1 when the system has a 10 dB SNR level, 16 users placed by the algorithms, and a maximum of four users were allowed to transmit in a time slot. In terms of worst-case SINR, which is the desired criterion here for comparison purposes, the sorted cases for the First Fit and FCFS algorithms are superior, followed by the unsorted First Fit algorithm, with the unsorted FCFS algorithm performing the worst. This was as expected from the theory and is consistent with the results from [67, 68], which were the major research papers located about this subject for smart antenna systems at the time when these simulations were constructed. Conventional research on wireless LAN systems such as 802.11 do not study dynamic slot allocation algorithms because the requirement that only one terminal may transmit at a time in a frequency band makes dynamic slot allocation unnecessary.
It should be noted that all of these dynamic slot allocation simulations did not include the spreading gain achieved from the use of the spread spectrum techniques. Hence, the SNR values shown are the effective SNR values for the system, and includes the 10 dB of spreading gain the system achieves. Thus, a 15 dB SNR value for one of the plots corresponds to an actual SNR (before spreading) of about 5 dB. The criterion for the First Fit algorithms for the addition of a terminal to a time slot was an SINR level of 10 dB or greater.
The results for the four algorithms studied when the system has a 5 dB SNR are shown in Figure 7.2, again for a scenario with 16 users and a maximum of four users in a time slot. Under this more difficult scenario, the sorted First Fit algorithm was superior (again, in terms of worst-case SINR), followed by the unsorted First Fit algorithm, then the sorted and unsorted FCFS algorithms. When the system experienced poorer conditions, the First Fit algorithms achieved much superior performance over the FCFS algorithms, even to the point that the unsorted
162
163
First Fit algorithm performed better than the sorted FCFS algorithm. This is a clear change from the 10 dB scenario, and is consistent with the results in [67, 68], which showed a much greater advantage for their
First Fit algorithm over their Random algorithm under poor conditions,
with all algorithms performing at comparable levels when the SNR was
at a sufficient level.
The worst case SINR results for the algorithms at various SNR
levels are shown in Figure 7.3 when there are again 16 users in the
system and a maximum of four users placed in a time slot. As was
expected from the earlier results, the First Fit algorithms were superior under low SNR conditions, while the First Fit and FCFS results became
much more comparable as the SNR was increased. Also as expected, the
sorted algorithms were superior to the unsorted algorithms, especially for the low SNR cases. The excellent worst case performance for the FCFS algorithm for high SNR shows that the manner of selection of
terminals for a particular time slot isnt very important when conditions
164
are favorable, as virtually any randomly selected spatial signature will do. The lack of a significant benefit from sorting when the SNR was high shows that the power levels of the various spatial signatures in a time slot is much less important when conditions are favorable.
One point should be noted about the First Fit algorithm results in all of the above plots. The First Fit algorithm only tested the SINR for the new user, not all users, in order to investigate whether or not merely testing the new user would be sufficient for implementation of the algorithms. This would save a significant amount of computational complexity, with a potential performance cost to the users already in the time slot. The results went below the 10 dB SINR benchmark for some of the other users when a new terminal was added and the SNR was poor, which led to the conclusion that the SINR for all users should generally be checked before a new user is admitted to a time slot via the First Fit algorithm when conditions are poor. However, when conditions are good (SNR greater than 10-15 dB), a system using the First Fit algorithm might be able to only test the new user.
A comparison between testing only the new user and testing all users in a time slot for the First Fit algorithm is shown in Figure 7.4 in terms of their worst case SINR results for typical runs of the simulations when 16 users are in the system and a maximum of four users are allowed in a time slot. As expected, the algorithm which checked every spatial signature in the time slot before accepting a terminal (termed First Fit All) performed better for low SNR than the algorithm which merely checked the new terminal before making the decision (termed First Fit New). As the SNR increased, the performance difference between the two algorithms became negligible, which was expected because the SINR for all terminals became comfortably above the 10 dB SINR limit. This was as expected from the results from Figures 7.1-7.3, which showed that all the terminals performed above the limit when the SNR was at a reasonable level for the First Fit New simulations. Also as expected, the algorithms with sorting performed better than their unsorted versions. From these results, coupled with the earlier results, we can conclude that the First Fit New algorithm may safely be substituted for the First
165
Fit All algorithm when the SNR is about 15 dB. Because this SNR does not reflect the 10 dB of spreading gain that is achieved with the Barker code, this means that we only need an actual SNR of around 5 dB to be able to safely use the First Fit New algorithm.
The comparison between testing only the new user and testing all users in a time slot for the First Fit algorithm is shown in Figure 7.5 when the system has an SNR of 5 dB, again for typical runs of the simulations when there are 16 users in the system and a maximum of four users in a time slot. As expected, the First Fit All algorithm performed better in terms of worst case SINR. The First Fit New algorithm had terminals with better performance than any of the First Fit All terminals, but also had much worse performance in the worst case. As having all users with good performance is deemed superior for the network to having some users with great performance and others with unsatisfactory performance, we can conclude that testing all terminals before adding a terminal to a time slot is necessary for low SNR conditions.
166
7.4
Dynamic Slot Allocation for Capacity Increase
Another critical measure for the performance of a dynamic slot allocation algorithm is the number of terminals it can place in its time slots while still satisfying minimum SINR requirements. The minimum SINR requirement set for these simulations is 10 dB, the same requirement used in the worst case SINR simulations. Each simulation run computed the average number of terminals placed per time slot, and the average of at least 100 simulation runs were taken to find each data point.
Figure 7.6 shows the performance of the Dynamic Slot Assignment algorithms for the low SNR cases of 3 and 5 dB. Again, this SNR includes the spreading gain (from the use of the Barker Code for the spread spectrum techniques) but not the antenna gain from the use of the antenna array. Figure 7.6 shows the average number of terminals that the algorithms were able to assign to the time slots in use. This average should generally be expected to increase as the number of terminals increase, since the greater selection of spatial signatures provides
167
more opportunities to find compatible combinations with the already allocated spatial signatures. The First Fit All algorithm was used for these low SNR scenarios, as it was shown from the previous results to be the most appropriate algorithm for dynamic slot allocation purposes for these low SNR types of scenarios.
There is a significant difference in performance by the 3 dB and 5 dB cases of Figure 7.6, approximately 1.5 terminals per time slot. This is because it becomes increasingly harder to place terminals in a time slot as the SNR drops to levels where the antenna gain barely boosts the optimum SINR above the 10 dB requirement. In difficult scenarios such as these, it becomes increasingly critical to find spatial signatures with little correlation to each other so that the beamforming routine can eliminate all possible interference.
It is also interesting that the algorithms with sorting only performed slightly better than their unsorted versions. From this and all
168
of the other scenarios studied (many of which are beyond the scope of this book), it was evident that the sorting did not significantly impact the average number of terminals per time slot. Coupled with the SINR results from the previous section, the simulations showed that sorting helps the performance experienced by the individual terminals, but does not improve the capacity of the network appreciably.
Figure 7.7 shows the performance of the Dynamic Slot Assignment algorithms for the high SNR cases of 10 and 15 dB. The 15 dB case performed better than the 10 dB case as expected, but by generally less than 0.5 terminals per time slot. The sorted versions of the algorithm did not perform significantly better than the unsorted versions, as was also the case for the low SNR scenarios. The First Fit All algorithm was used here for comparison purposes with the low SNR scenarios, even though the First Fit New algorithm would probably have been adequate, especially for the 15 dB case.
169
It is interesting that a 2 dB change (from 3 dB to 5 dB) resulted in a 1.5 terminals per time slot difference in capacity when the SNR was low, while a 5 dB change (from 10 dB to 15 dB) resulted in a 0.5 terminal per time slot difference in capacity when the SNR was high. The 5 dB change between the low SNR case of 5 dB and the moderate SNR case of 10 dB also has about a 2 terminal per time slot difference. These results reaffirm the conclusion that terminals have increasingly more difficulty in being added to a time slot as the SNR drops to low levels. In fact, when the SNR was reduced from 3 dB to 1 dB, the average number of terminals per time slot was cut in half.
It is clearly evident from Figures 7.6-7.7 that the average number of terminals per time slot is not a continuously increasing function, and there are a fair number of fluctuations and trend reversals in the results. While this may seem counter-intuitive and against the predictions, it
was actually an expected result. The reason behind this is that there are many situations when a terminal is added but cannot be fit into
any of the existing time slots. Consequently, the terminal is placed by itself in a new time slot, which significantly drops the average number of terminals per time slot in the system. This is a natural phenomenon which cannot be prevented, and it occurs quite consistently for certain numbers of terminals at certain SNR values. Thus, a system with 8 dB SNR and 25 terminals may have a much higher average than a system with 26 terminals because the terminal may require a new time slot. Hence, it is better to study the overall trend of the plots rather than any individual data points or fluctuations in the data. The effect of this lone terminal phenomenon should be reduced as the number of terminals is increased, and the results in the figures are consistent with that. The fluctuations in the average number of terminals in both low and high SNR scenarios decrease noticeably in frequency and size as the number of terminals in the system becomes large.
Other ways to present the average number of terminals per time slot were also analyzed, in an attempt to eliminate the fluctuations due to new time slots with just a single terminal being added. However, it was found that is was not feasible to eliminate this phenomenon, and in
170
the worst case doing so actually resulted in skewed or extremely incorrect conclusions to be drawn from the data. There were many reasons why
excluding terminals that are by themselves in a time slot didnt work. First, there were certain scenarios (normally created when the spatial signature variations were purposely reduced in the simulations) when
terminals were by themselves in a time slot other than the last slot. This
was normally in the second to last time slot, and was due to the scenario
when the last few terminals had spatial signatures which were fairly close to the terminal in that slot, and the SNR was poor so the interference dropped the time slots SINR below the threshold value when any of those terminals were attempted to be added to that time slot. Because
this was due to the SINR test, and not merely a time slot overload forcing
a new slot to be used, it would not be fair to exclude that time slot from
the analysis. Similarly, a terminal could be placed by itself in the final time slot because its spatial signature was incompatible with those in the
other time slots, and excluding that terminal would unfairly boost the results. Second, low SNR cases (SNR values of 3 dB or less, for example), especially with a lesser number of users, had less than two terminals per
time slot on average. In these cases, single terminals in a time slot are common and excluding single terminals from the average provides extremely skewed results. Additionally, because single terminals in time slots occurred much more often in those poor scenarios, excluding single
terminal cases could easily result in misleading conclusions being drawn from those results as compared with the results from the more favorable scenarios. Third, excluding time slots with single terminals allowed time
slots with just two terminals due to the same overcrowding problem to distort the data. There then became a significant difference between systems in which the time slots were full and a terminal had to begin a new time slot (which was then excluded) and the same case with a second terminal added to the new time slot. However, excluding time
slots with either one or two terminals in them wasnt feasible and would result in data which wouldnt allow for any reasonable conclusions to
be made, since that would necessitate eliminating many or nearly all of the time slots from many of the poor SNR scenarios. Due to these and
other reasons, it was concluded that the best course of action was to
Timing Synchronization Results
171
present all the data, including the fluctuations, and to pay attention to the overall trends and not the individual data points or fluctuations. A
good analogy can be made between this type of study and the study of
the stock market over the past 30 years.
7.5
The results for the SWL timing synchronization algorithm for 100,000 independent runs of the program are shown in Table 7.4. The algorithm that was described in Chapter 4 is the Standard S case, where S denotes that the users were sorted in order of ascending relative round trip propagation chip delays. For the standard scenario, the
best cases were able to find selections of the chip adjustments to allow for 11 users to be received with different chip offsets, which is the maximum number of users possible because we have an 11-chip code. The worst cases were still able to achieve 6 users per time slot, which may be adequate for our purposes because 8 simultaneous users is the maximum
number we can accommodate under the best possible conditions with the 8-element antenna array that SWL uses, while we can probably only
satisfy 5-6 users in the same frequency band and time slot in a real-world system. The average number of timing offsets found was slightly over 9,
much more than what we need for our system. A second algorithm was tested without the initial sorting of the
propagation chip delays and is labeled Standard U in Table 7.4. The chip delays were sorted in the original algorithm because they made verification of the algorithm by hand easier. However, the search routine in
the algorithm may make the sorting unnecessary because it searches for any possible valid timing offset. Consequently, the Standard U case in Table 7.4 has very similar results as the Standard S case, as expected.
The results for the Standard S algorithm are actually slightly better than that for the less computationally intensive Standard U algorithm,
but by less than 0.1 users, which is within the variance of the simulation.
In fact, further runs for both the Standard and Unsorted routines
found that their average performance varied by more than 0.1 users from
run to run, depending on the specific groupings selected by the random
172
propagation delay selections. However, the sorted algorithm did generally maintain a slight performance edge over the unsorted algorithm.
A less computationally intensive algorithm was tested for com-
parison purposes, and is labeled Quick in Table 7.4. The sorted version of this algorithm is called Quick S and the unsorted version of this algorithm is called Quick U . This algorithm required much less computation because it stopped after the first timing conflict was found that couldnt be quickly resolved. The idea behind this algorithm is that all the extra searching and shifting in order to attempt to allow more users to be allocated in the time slot may not be worth the effort. The diminishing returns for the searches after the first unsolvable resolution may make that search not worthwhile, especially when it is noted that the overall system will probably only be able to satisfy 5-6 users per time slot anyway. Hence, since a faster algorithm may achieve the desired number of users that are needed, we may be able to realize significant computational savings by utilizing it.
The results show that these Quick algorithms performed slightly worse than the Standard algorithms. However, its performance may be adequate for the SWL system, as it achieved at least 5 users in all cases and averaged about 7.6-7.7 users. The original plan for the SWL system was to allow only 4 users per time slot, a number that this algorithm would easily be able to handle the synchronization for. The contrast between the Quick routines and the Standard routines allows for another tradeoff between computational complexity and performance. The sorted and unsorted versions of the Quick algorithm had virtually the same results, well within the normal variance from simulation
173
run to simulation run, with the sorted version slightly outperforming their unsorted counterparts again. The unsorted version was selected for SWL because the results are comparable and the whole purpose for the Quick algorithm was to save processing time, which skipping the sort routine is consistent with. The solution desired for the SWL system is to use a combination of the Quick and Standard algorithms. After the number of users is determined for a time slot by the dynamic slot allocation routine, the unsorted Quick algorithm is run. If it succeeds in finding the necessary chip offsets to satisfy the timing synchronization for all the users assigned to the time slot, then the timing search stops. However, if not all users are able to be assigned, then the unsorted Standard algorithm picks up where the Quick algorithm left off to find the necessary timing offsets for the users. Because the Standard algorithm is simply an
extension of the Quick algorithm (the Quick algorithm was designed
specifically for this purpose), no computations are wasted and moving to the Standard algorithm is handled smoothly. If it is found that the time slot has a particularly unfortunate set of propagation offsets and cannot satisfy all the users that the dynamic slot allocation algorithm assigned, then the excess users are sent to another time slot. However, since our 8-element antenna array can only handle at most 8 users under the most ideal of conditions, and probably only will be able to satisfy 5-6 users in a time slot, and since the Standard routine can always satisfy at least 6 users, this should not be a very big problem. Also, it should be noted that the Standard algorithm succeeds in finding the chip offsets for at least 7 users over 99.9% of the time.
The previous timing synchronization results created 11 terminals with random round chip delays and studied how many of them could be assigned to a time slot. However, if the dynamic slot algorithm assigns the terminals to the time slot before the timing algorithm is run, a more meaningful measure for the SWL system would be how successful the timing algorithm is at allowing all the users in the time slot to transmit simultaneously while attaining the 10 dB of spreading gain by finding different chip adjustments, which for each terminal results in unique
174
overall chip delays, for each terminal. The timing results just presented show the potential for the timing synchronization algorithm, and
the following results will show the effectiveness of the algorithm.
The effectiveness of the various algorithms in placing as many terminals as possible when 11 terminals (with randomized propagation delays) were assigned to a time slot while preserving the spreading gain
by finding chip offsets for the downlink signals that results in the necessary timing for both the uplink and downlink signals is shown in Table 7.5. This scenario is the same as that of the results from Table 7.4,
with the breakdown for the number of users able to be accommodated

in each simulation run shown. The Standard algorithm usually found
timing offsets to allow 9 or 10 users per time slot, while the Quick algorithm usually achieved 7 or 8 users. The maximum, minimum, and breakdown of users allocated is as expected from the results in Table 7.4. For example, the 38.53% value for the Standard U algorithm under the 10 users column means that of the simulation runs for the 11 user case, 10 users (exactly 10 users, not 10 users or more) were able to be placed by the Standard U algorithm 38.53% of the time, with all 11 users being place 1.76% of the time. The sorting routine did not have any significant benefit for either algorithm, which was also expected from Table 7.4. Again, the results from Tables 7.4-7.11 are the results of 100,000 independent simulation runs.
It should also be noted that the breakdowns in some parts of the following tables do not add up to 100%. This is because the Quick algorithm terminated sometimes after only placing 5 users. The guarantee
175
that offsets can always be found for 6 users in a time slot only applies to the Standard algorithm, which searches all possible combinations for timing synchronization. The minimum number of users with timing offsets found by the various algorithms for a time slot was 5 for the Quick algorithm and 6 for the Standard algorithm, regardless of the number of users assigned to the time slot (assuming that the number of users assigned to the time slot was more than the minimum number of 5 or 6). The effectiveness of the various algorithms when they attempt to place as many terminals as possible when 10 terminals (with randomized
propagation delays) were assigned to a time slot while preserving the
spreading gain is shown in Tables 7.6 and 7.7. The Standard algorithm
was able to find timing offsets to satisfy 9 out of the 10 terminals in most situations, with 8 and 10 terminals being placed a fair amount of
times. The sorted Standard algorithm was usually able to place all 10 terminals more often, resulting in a higher mean number of terminals for
176
it. The Quick algorithm benefited even more from the sorting routine, as the sorted version was able to place 9 terminals at a significantly higher rate, leading to a higher mean number of terminals. The Standard algorithm was able to place a little more than a terminal more per time slot than the Quick algorithm. The effectiveness of the various algorithms when they attempt to place as many terminals as possible when 9 terminals (with randomized propagation delays) were assigned to a time slot while preserving the spreading gain is shown in Tables 7.8 and 7.9. As in the 10 terminal scenario, the sorted versions of the algorithms slightly but noticeably outperformed their unsorted counterparts in the 9 terminal scenario. The unsorted Standard algorithm usually accommodated all 9 users, with a comparable amount of times with 8 users, while the sorted version was able to place all 9 users at a significantly higher rate. Similarly, the unsorted Quick algorithm usually accommodated 8 users, with a large number of results with 7 users, while its sorted version had significantly
177
more 8 user results and significantly fewer of the lesser user results. Again, the Standard algorithm was able to place a little more than a terminal more per time slot than the Quick algorithm. The effectiveness of the various algorithms when they attempt to place as many terminals as possible when 8 terminals (with randomized propagation delays) were assigned to a time slot while preserving the spreading gain is shown in Table 7.10. The unsorted Standard algorithm was able to find timing offsets to accommodate all 8 users most of the time, while the unsorted Quick algorithm was typically able to place 7 users. The sorted versions of both algorithms had the same typical performance, but was able to satisfy its upper limit of users about 12% of the time more, a noticeable performance increase over the unsorted versions. The Standard algorithm was able to place almost exactly one terminal more per time slot than the Quick algorithm. Finally, the effectiveness of the various algorithms when they attempt to place as many terminals as possible when 7 terminals (with randomized propagation delays) were assigned to a time slot while preserving the spreading gain is shown in Table 7.11. The Standard algorithm could usually accommodate all 7 users, while the Quick algorithm could usually accommodate 6 of the 7 users. As in the previous cases, the sorted versions of the algorithms performed better than the unsorted versions, though the performance benefit for sorting in this scenario is minimal. The limited benefit of the sorting in this scenario is due to the fact that the unsorted versions of both algorithms achieve
178
their upper limit virtually all of the time, so there isnt much room for improvement to be gained by sorting. The Standard algorithm was
able to place almost exactly one more terminal per time slot than the Quick algorithm. The initial study to see the potential of the timing synchronization algorithm studied the case where 11 random users were assigned to a time slot. The measurement of the number of users which were able
to be assigned chip offset adjustments to allow their uplink and downlink signals to achieve the spreading gain despite all users transmitting simultaneously showed the potential for the timing synchronization al-
gorithm. From these results, it was concluded that the sorting routine wasnt necessary, as it did not result in a significant performance benefit. After studying the potential of the timing synchronization algorithms, the performance of the algorithms was analyzed by setting the number of users in a time slot and measuring how many of those users were able to be placed in a slot while satisfying the requirements that each users chip offset for the downlink and the total chip offset for the uplink be unique, preserving the spreading gain for the uplink and downlink transmissions. These results were quite revealing, as the sorted versions of the algorithms slightly but noticeably outperformed their unsorted counterparts, with the exception of the 7 user case. The 7 user case didnt benefit much from sorting because the unsorted version already achieved its upper limit of users over 98% of the time, leaving only a small improvement possible for the sorted versions.
Physical Layer Simulation Results
179
A study of the results from Tables 7.4-7.11 shows that the Standard algorithm was able to place slightly more than one user per time slot more than the Quick algorithm. The gap in performance between the two algorithms increased as the number of users available for placement in a time slot increased, ranging from an average of 1.6 users for the 11 user case to just 1 user in the 6, 7, and 8 user cases.
7.6
Before designing the physical layer of the SWL system, a thorough study was conducted of the experimental results taken by the smart antenna testbeds in the Electrical Engineering Research Laboratory at
the J.J. Pickle Research Campus of the University of Texas at Austin. The beamforming algorithm used by SWL was selected to be the pseu-
doinverse spatial signature method because of the superiority of that algorithm for interference suppression in the computer simulation results shown earlier in this chapter and the experimental results shown in Section 6.11. A computer simulation was constructed via MATLAB to simulate the uplink and downlink of the physical layer of the Smart Wireless LAN system. A user-specified number of terminals transmit and/or receive data simultaneously as described in Chapter 4. The simulations used a 2000 bit data frame, with four users receiving frames at the same time. Each simulation ran for at least 10 frames, with the data, noise, DOAs, and spatial signatures randomized for all users each time, and the average bit error rate (BER) was computed. A typical downlink example is shown in Figure 7.8, where the base station used pseudoinverse spatial signature beamforming to minimize the interference to the other simultaneous users and it was assumed that the base station had perfect knowledge of each terminals spatial signature for the computation of the weight vectors by the beamforming algorithm.
There are a few factors that should be noted when we interpret
the results shown in Figure 7.8. First, the SNR in the figure does not include the interference in the system due to the other three users which
180
are receiving frames simultaneously with our desired signal, so the actual SINR of the system is worse, possibly much worse, than the SNR listed. Second, the spatial signature vectors were chosen via the FCFS method each simulation run so it was possible that two terminals were assigned highly correlated spatial signatures. It was shown in Section 7.3 that
this FCFS assignment of users has an inferior performance than that

of the First Fit method that SWL can use. Third, the SNR specified doesnt include the spreading gain from the use of spreading, which adds about 10 dB to our overall system performance. It also doesnt include the antenna diversity gain that achieved by using an antenna array [77], which significantly increases the effective SNR of our system (by as much as 9 dB for an eight-element antenna array like the one used in the simulation). Finally, more advanced beamforming techniques can achieve superior performance at a cost of additional computational complexity.
The results in Figure 7.8 are significantly better than the results in [30]. This is due to the improved timing and synchronization features with the addition of the SYNC and SFD fields in the header, use of
181
DBPSK/DQPSK instead of BPSK/QPSK, and an improved beamforming application [14].
The simulation assumed perfect knowledge of the spatial signature, where in a real system the beamforming routines would perform at a lower level due to errors in the previously obtained spatial signatures. With imperfect knowledge of the spatial signatures, the system performance decreases as in Figure 7.9, where each spatial signature was corrupted with noise before computation of the beamforming weight vectors. The results in Figure 7.9 were worse than expected, which was found to be due to the assumption that the spatial signature vector was corrupted with the same noise level as the system noise. Studies are currently underway to find the appropriate levels of spatial signature uncertainty for various scenarios, with comparisons to our experimental results used to select the appropriate uncertainty level parameters.
The performance of the simulations was unchanged when the chip offsets were modified, as expected. There was no noticeable change in
182
performance when the relative chip offsets between users was changed from 3-chips to randomly picked (nonzero) offsets. As long as the timing adjustments resulted in each terminals spreading code beginning at a different chip period than all the other terminals spreading codes, the systems performance was unchanged. This was as expected, as the Barker code is uncorrelated with any other part of itself, so as long as an offset is maintained the users are easily distinguished from each other. However, if multiple users begin transmitting on the same chip period, then much or all of the spreading gain is lost by the system, also as expected. This demonstrated the importance of the timing synchronization algorithms purpose of insuring that the multiple received signals all have different chip offsets when received by the base station. The performance of a system with no chip offset between the multiple simultaneously transmitting terminals under the assumption that the base station has perfect knowledge of the spatial signatures of all the terminals is shown in Figure 7.10. The beamforming routine is extremely effective in this perfect knowledge case, which means that the interference signals are significantly reduced. Consequently, much of the spreading gain is still achieved because of beamforming. In the ideal case when the beamforming achieves perfect nulls for all interference sources, we would not have to worry about a timing routine for chip offsets at all because the system would only effectively have one spreading code being active at a time for each terminal. Each point in Figure 7.10 was an averaged result of at least 100 independent simulation runs, much more than the minimum 10 runs for Figures 7.8-7.9. The increased number of runs was desired because of the increased volatility of the results when the chip offsets werent used. This is probably because the beamforming routine has a varying effectiveness (due to the randomized spatial signatures), which allows for a wide fluctuation in the results for the system when the SNR was kept constant. In this scenario, the effectiveness of the beamforming routine not only determined the interference that each terminal experiences, but also the amount of the 10 dB of spreading gain that the terminal is able to benefit from. A comparison between Figures 7.10 and 7.8 shows that
183
the average spreading gain achieved with no chip offset was between 3-5
dB, since the performance of the no chip offset scenario was between 5-7
dB worse than that of the scenario with chip offsets.
The beamforming routines are less effective when we remove the

assumption about perfect spatial signature knowledge. Hence, the lack
of chip offsets between the users results in a loss in spreading gain, as shown in Figure 7.11. Each non-offset terminal is in effect an interference source for the spreading gain, and the cumulative reduction in spreading gain reduces the overall SNR of the system. Because the overall SNR is
directly linked to the overall BER of the system, the BER in Figure 7.11
is significantly worse than the BER in Figure 7.9 for the same input SNR. Remember that the overall SNR of the system is the input SNR plus the spreading and antenna gains of the system. As in Figure 7.9, each point in Figure 7.11 was an averaged result of at least 100 independent
simulation runs.
184
A close study of Figure 7.11 reveals some interesting results.
When we compare the plot with Figure 7.9, we see that the average
spreading gain achieved with no chip offset seems to be again between 3-5 dB, since the performance of the no chip offset scenario was between 5-7 dB worse than that of the scenario with chip offsets. At first glance it is surprising that the same amount of spreading gain appears to be achieved by both the perfect and imperfect spatial signature knowledge simulations. However, the changing effectiveness of the beamforming algorithm must be taken into account. As the SNR is increased, the uncertainty in the spatial signature knowledge is reduced, which boosts the performance of the beamforming algorithm. Consequently, some of that 3-5 dB of apparent spreading gain is due to the improvement of the beamforming routine as the SNR was increased.
A comparison between the no chip offset case with imperfect spatial signature knowledge of Figure 7.11 and the case with chip offsets and perfect spatial signature knowledge of Figure 7.8 shows a 15-20 dB
Media Access Control Layer Simulation Results
185
change in system performance between the ideal and the unrealistically difficult scenarios. The scenario in Figure 7.11 is unrealistically difficult because it assumes that all of the users have no chip offsets between them (the worst possible case, and something with an extremely low probability even if the chip offsets were assumed to be random since there are 11 possible chip offset values) and that the spatial signature knowledge was extremely poor for all users. However, that worst-case scenario is useful for comparison purposes with the ideal and more realistic scenarios.
7.7
The experimental results given in [38] were used to select the proper parameters for a computer simulation comparison of the 802.11 standard with the SWL protocol [31]. A user-specified number of voice, file transfer, and electronic mail (e-mail) terminals (ranging from 1 terminal to 160 total terminals) submitted requesting packets to contend for access to the network. The requesting packet was transmitted in the contention period using the p-persistent slotted Aloha [78, 79, 80] protocol on a frame-by-frame basis and contained the terminal ID number, the priority level of the transmission, and the number of time slots requested per frame. The voice data was given a high priority designation and file transfer and e-mail data were given low priority designations. The parameter p for the slotted Aloha was set to 1 for high priority sessions and 0.8 for low priority sessions. The simulations started with an empty queue and all terminals in the OFF state. The simulations assumed that all transmitted messages reached the base station successfully and thus ignored physical layer effects such as the acquisition and updating of spatial signatures that would likely have increased packet delays. The simulations used a frame size of 10 ms. The voice sessions had an average ON period of 1.0 seconds, or 100 frames, and an average OFF period of 1.2 seconds, or 120 frames. A voice terminal requested one slot per frame. The file transfer sessions had average ON and OFF periods of 10 seconds and 60 seconds, respectively, and the e-mail sessions
186
had average ON and OFF periods of 30 ms and 100 ms, respectively. Email sessions also requested one slot per frame and file transfer sessions requested a random number of slots per frame, uniformly distributed from 1 to 5. A frame in the global scheduler had one row for the 802.11 case (since only one user can transmit at once in a frequency band) and four rows for SWL case (since it was assumed that four users could
transmit at once in a frequency band) in its buffer, the number of slots
that the global scheduler contained in each row was set to 40, and both the 1 and 2 Mbps transfer rates in the 802.11 standard were used. Each
simulation ran for 36,000 frames, simulating 1 hour of network time.

The simulations were run under identical conditions for both the 802.11 and SWL protocols and the utilization of the network, maximum buffer and queue sizes, maximum delay for high-priority terminals, and a few other important parameters were recorded. Note that the 802.11 simulation is actually not an exact simulation of the 802.11 standard, as the simulation was modified to allow for comparison purposes with SWL, adding such features as priority handling of data. The simulations also did not simulate the SWL system exactly, as they assumed that the contention period was able to have four successful contentions per time slot (which was the case in the original design) instead of just one contention per time slot. Hence, these results should be not be considered precise representations of the performance of the SWL and 802.11 systems, but instead benchmarks demonstrating the benefits of allowing multiple terminals to transmit per time slot and adding priority designations on data.
Table 7.12 shows a sampling of the simulation results, displaying for both the 802.11 and SWL networks the network utilization, maximum delay for voice data (in Time-Space frames), percentage of long (more than 5 frames) delays for voice data, and the maximum e-mail buffer size. For example, the last scenario in the table has 35 voice, 10 file transfer, and 35 e-mail terminals. The superiority of SWL over 802.11 for this scenario is shown in every category recorded, with the network utilization of 802.11 at 82.21% while it was only 18.34% for SWL (the smaller number is superior, as explained later), and the voice data failures for 802.11 and SWL contrasting sharply at 23.3% and 0.14%, respectively.
187
Network utilization was computed by
where , and are the number of voice, file transfer, and email sessions, respectively. The amount of information that a terminal generates when it is active is given by , , and ; the probability that a certain terminal is active is given by , , ; the number of users allowed to transmit simultaneously during the contention free period is ; and the steady state value of the number of reservation (contention free) slots in each frame is given by . Note that a smaller value for network utilization under this definition is superior to a larger value, as a smaller value means that the load on a network is less (in other words, less of the networks available capacity is being used).
The actual network utilization during the simulations was computed and was found to correspond closely to the theoretical utilization
188
found in Equation 7.1. The utilization for both SWL and 802.11 protocols is plotted in Figure 7.12 for e-mail terminals and is clearly linear as the number of e-mail terminals increase, as expected. It is clear from the figure that the utilization of the 802.11 network is approximately four times the utilization of the SWL network. This result was expected because the SWL network in this simulation allowed four users to transmit at a time while the 802.11 simulation allowed only one user to transmit at a time. Similar results were found for scenarios with only voice or file transfer terminals, and for various mixed traffic conditions. The scenario for voice terminals is shown in Figure 7.13.
A critical requirement of a network with delay-sensitive data is the percentage of delay-sensitive data that suffers long (defined as greater
than 5 frame buffers for our simulations) delays in transmission. Long delays for voice or video data could make the data unusable. A plot of
the percentage delays for the 802.11 network is shown in Figure 7.14. It is clear from the plot that the 802.11 network suffers unacceptably large delay failures as it moves into its moderate and heavily loaded re-
189
190
191
gions. Since it is difficult to see the results for the lightly loaded cases, Figure 7.15 shows the gradual increase in the percentage delays as the network operates in its lightly loaded region. This figure provides an interesting comparison to the long delays for the SWL network in Figure 7.16. It seems that the rate of increase in the lightly loaded regions for both networks are comparable. Note that the same number of terminals transmitting voice, file transfer, and e-mail data were used until the total number of terminals became greater than 30 (10 of each type), then the number of file transfer terminals was set to 10 while the number of voice and e-mail terminals were set equal to each other. The reason for setting a maximum number for the number of file transfer terminals is because it was found that the file transfer terminals have a disproportionate effect on the loading of the network and using more than 10 file transfer terminals obscures the performance effects for the voice and e-mail terminals.
It is clear from the simulation results that the 802.11 standard works well for a small number of users, as expected, but it becomes rapidly overloaded when too many users are attempting to use the network. Hence, the network would refuse a connection or become unstable. Either way, the maximum delay for a user attempting to send data would become unacceptably long. The SWL results were clearly superior in each case to the 802.11 results, with increasing benefits as the number of users increase. Both of these results were as predicted from throughput performance evaluations. It should also be pointed out that 802.11 systems are able to use more than one frequency band at a time, as described in Section 3.4. For example, DSSS systems such as the ones simulated here have the choice of 11 center frequencies (in North America), from which they can select 3 without overlap (except in Spain, France, and Japan). Hence, the capacity of the 802.11 and SWL systems would triple (ignoring multiplexing, handoff, and other effects) because we would have one of these Time-Space frames and simulation scenarios for each of the center frequencies chosen by the network.
Chapter 8
Conclusion
The use of wireless LAN systems has been increasing rapidly, with worldwide sales projected to reach $1 billion per year soon. The proliferation of wireless LAN systems has been aided by the worldwide adoption of the IEEE 802.11 wireless LAN standard. However, the 802.11 protocol has many critical limitations that limits the flexibility and usefulness of the wireless LAN systems. Because the bandwidth is limited and only one user is allowed to transmit at a time in a frequency band (which are also limitations of the wireless LAN systems for the OpenAir Standard ), the throughput for each user is limited and users can experience long delays when the network gets loaded. In addition, the 802.11 protocol does not provide specifications for multimedia or real-time applications. Not only may the throughput be too low for applications such as voice, but the protocol does not include the priority designations that are required for real-time applications.
The typical wireless LAN scenario, with stationary or slowly moving terminals, is particularly ideal for smart antenna systems. Experimental results show that a stationary terminals spatial signature will remain virtually constant over long periods of time, reducing overhead on the network and demonstrating the feasibility of utilizing beamforming techniques. The experimental results also demonstrated how the spatial signatures of terminals in different locations, even terminals which are placed close together, vary significantly and are uncorrelated, also enabling the beamforming techniques to be effective. Beamforming and dynamic slot assignment algorithms are critical to enable multiple users to transmit simultaneously in a frequency
194
Conclusion
band. Computer simulation results, when coupled with the experimental results found in this and other research, showed that the beamforming algorithms were able to minimize the interference a terminal experiences from the other simultaneously transmitting terminals. The pseudoinverse spatial signature beamforming method was shown to be particularly effective in terms of SIR. Hence, that beamforming method was selected for the dynamic slot assignment algorithms, where it was effective in enabling up to nearly 8 terminals per time slot (which is the maximum for the 8-element antenna array currently assumed for the SWL system). The FCFS algorithm performed adequately, but the First Fit algorithm demonstrated superior performance. The First Fit algorithm performed
better when all terminals in a time slot were taken into account before
a new terminal was added to the time slot when the SNR was low, but only the new terminal needed to be checked when the SNR was above 15 dB. In all cases, sorting the terminals in order of ascending power improved the performance of the algorithms, with the improvement being more pronounced at low SNR levels.
Offset coding and timing synchronization are required to allow the terminals and base station to achieve the 10 dB of spreading gain when all the simultaneous transmissions using the same spreading code. The downlink transmission merely requires the base station to set the offset spreading codes, but the uplink transmission requires the timing synchronization algorithm. The timing synchronization algorithm adjusts the starting chip times by the base station for its downlink transmission in order to take advantage of the previously obtained round trip delays to insure that the uplink signals are received with the necessary chip offsets. Simulation results show that the standard SWL algorithm can find appropriate timing offsets for over 9 terminals on average when it has just 11 terminals to choose from for a time slot. The results also
show that the algorithm is extremely effective in placing most, if not all,
of the terminals that were assigned to the time slot. A quick algorithm was also shown to be fairly effective in synchronizing terminals, and this
algorithm was chosen to serve as a first pass for the dynamic slot alloca-
tion. In the event that the quick algorithm cannot place enough of the terminals in a time slot, the standard algorithm can then be utilized with
Conclusion
195
none of the quick algorithms computations wasted because the standard

algorithm can start at the point that the quick algorithm terminated.
The overall SWL physical layer simulation showed the excellent performance of the system. The ideal case with perfect spatial signature knowledge and the difficult case with an extremely noise corrupted spatial signature used by the beamforming routine were tested, with both scenarios performing well. The worst case when no chip offsets were used by the terminals was also tested, and it was found that 5-7 dB of the spreading gain was lost in that situation. The 3-5 dB of spreading gain still achieved by the system was due to the beamforming routine reducing the interference from the other users. Interestingly, this 3-5 dB of spreading gain appeared to be achieved by the simulation with perfect spatial signature knowledge and also by the simulation with imperfect spatial signature knowledge. However, part of the performance gain in the imperfect spatial signature knowledge case is actually due to the improved performance by the beamforming algorithm as the spatial signature uncertainty is reduced. The Smart Wireless LAN system was created to alleviate certain limitations of the IEEE 802.11 standard and utilize the significant benefits that smart antenna systems can provide wireless communications systems. SWL is an adaptation of smart antenna systems for wireless LAN systems and is compatible with the IEEE 802.11 protocol, so only the base stations, not the individual terminals, will have to be changed. The utilization of a smart antenna system at the base station allows for uplink and downlink of data by multiple terminals at the same time, significantly increasing the throughput of the network. In addition to the increase in throughput, this new protocol has additional features such as simple implementation, adaptability to multimedia traffic with diverse bandwidth requests, network security, guaranteed fairness in bandwidth sharing, and ease of adaptation to the 802.11 wireless LAN standard. Also, this protocol adds priority level designations to the packets to allow for delay-sensitive communication links for multimedia applications such as voice or video. The only new requirements of this system are an array of multiple antennas installed at a base station and advanced signal processing software to process the received data.
Bibliography
[1] T. Rappaport, Wireless Communications: Principles and Practice, Prentice Hall, Englewood Cliffs, NJ, 1996.
[2] W. C. Y. Lee, Mobile Communications Design Fundamentals, John Wiley & Sons, Inc., New York, NY, 1993.
[3] J. D. Parsons, The Mobile Radio Propagation Channel, Pentech Press, London, UK, 1992.
[4] R. Steele, Mobile Radio Communications, Pentech Press, London, UK, 1992.
[5] W. C. Y. Lee, Mobile Cellular Communications, McGraw Hill, New York, NY, 1989.
[6] J. D. Parsons and J. G. Gardiner, Mobile Communication Systems, Blackie, London, UK, 1989.
[7] W. C. Jakes, Microwave Mobile Communications, John Wiley & Sons, Inc., New York, NY, 1974.
[8] J. D. Parsons, M. Henze, P. A. Ratliff, and M. J. Withers, Diversity Techniques for Mobile Radio Reception, IEEE Transactions on Vehicular Technology, pages 7584, August 1976.
[9] W. C. Jakes, A Comparison of Specific Space Diversity Techniques for Reduction of Fast Fading in UHF Mobile Radio Systems, IEEE Transactions on Vehicular Technology, pages 8192, November 1971. [10] P. Balaban and J. Salz, Optimum Diversity Combining and Equalization in Data Transmission with Application to Cellular Mobile
198
Bibliography
Radio-Part I: Theoretical Considerations, IEEE Transactions on Communications, pages 885894, May 1992.
[11] S.-S. Jeng, G. Okamoto, G. Xu, H. P. Lin, and W. J. Vogel, Experimental Evaluation of Smart Antenna system Performance for Wireless Communications, IEEE Transactions on Antennas and Propagation, pages 749757, June 1998.
[12] A. Naguib, A. Paulraj, and T. Kailath, Capacity Improvement with Base-Station Antenna Arrays in Cellular CDMA, IEEE Transactions on Vehicular Technology, pages 691698, August 1994.
[13] R. Kohno, H. Imai, and S. Pasupathy, Combination of an Adaptive Antenna Array and a Canceler of Interference for Direct-Sequence Spread Spectrum Multiple-Access System, IEEE Journal on Selected Areas in Communications, pages 675682, May 1990. [14] G. Okamoto, S.-S. Jeng, S. Tseng, and G. Xu, Evaluation of Beamforming Algorithm Effectiveness for the Smart Wireless LAN System, Proc. of VTC98, May 1998.
[15] L. Goldberg, Wireless LANs: Mobile-Computings Second Wave, Electronic Design, pages 5572, June 1995.
[16] A. Juodikis, Wireless LANs: Considerations and Implementations, Communication Systems Design, pages 4853, September 1995. [17] C. Leidigh, An 802 LAN Standard Primer, Communication Systems Design, pages 3040, September 1995.
[18] R. O. LaMaire, A. Krishna, P. Bhagwat, and J. Panian, Wireless LANs and Mobile Networking: Standards and Future Directions, IEEE Communications Magazine, pages 8694, August 1996.
[19] J. Abramowitz, Wireless LANs: A Status Report, Telecommunications, page 70, March 1997.
[20] M.Leonard, Wireless Data Links Broaden LAN Options, Electronic Design, pages 5155, March 1992.
Bibliography
199
[21] P. Dunn, Wireless LAN Suppliers Foresee Growth As User Conditions Fade, Chiltons Electronic News, page 18, February 1991.
[22] J. Ubois, Wireless LANs Are Due For An Encore, MacWEEK, pages 2425, March 1993.
[23] R. Lauriston, Wireless LANs: Your Next Network?, PC World, pages 224228, September 1992.
[24] L. Goldberg, MAC Protocols: The Key To Robust Wireless Systems, Electronic Design, pages 6374, June 1994. [25] Electronic News Publishing Corporation, Triquint, Proxim eye
2.4GHz markets, Electronic News, pages 6263, August 1994.

[26] J. Edney, IEEE 802.11 Standards: New Options, More Freedom in Wireless LAN Connectivity, Commercial Wireless Systems Supplement, pages 2326, September 1995.
[27] J. McGarvey, Lucent Upgrades WaveLAN Gear, Interactive Week, page 31, April 1998.
[28] C. Mason, Raytheon Advances Into WLAN Market, Wireless Week, pages 1,8, April 1997.
[29] G. Okamoto, S.-S. Jeng, and G. Xu, Adapting SDMA for Wireless LAN Systems via the SWL Protocol, Submitted for publication in the IEEE Transactions on Vehicular Technology, 1998.
[30] G. Okamoto and G. Xu, The Smart Wireless LAN System: Physical Layer Design and Results, Proc. of VTC97, May 1997.
[31] G. Okamoto and G. Xu, Throughput Multiplication of Wireless LANs: Spread Spectrum with SDMA, Proc. of VTC96, April 1996.
[32] G. Okamoto and G. Xu, Multimedia Communications over Wireless
LANs via the SWL Protocol, Proc. of HICSS97, January 1997.

[33] G. Okamoto and G. Xu, The Smart Wireless LAN System: Adapting SDMA for IEEE 802.11, Proc. of Austin Telecommunications Conference, November 1996.
200
Bibliograghy
[34] G. Okamoto, The Smart Wireless LAN System: Physical Layer Design, Proc. of University of Texas Student Research Conference, October 1996.
[35] G. Okamoto, Spread Spectrum with SDMA for Wireless LANs: Simulation and Results, Proc. of University of Texas Student Research Conference, November 1995.
[36] H. L. Bertoni, S. Kim, and W. Honcharenko, Review of In-Building Propagation Phenomena at UHF Frequencies, Proc. of Asilomar95, pages 761765, November 1995.
[37] W. Honcharenko, H. L. Bertoni, J. L. Dailing, J. Qian, and H. D.
Yee, Mechanism Governing UHF Propagation on Single Floors in Modern Office Buildings, IEEE Trans. Vehicular Technology, pages
496504, November 1992.
[38] S.-S. Jeng, G. Xu, H. P. Lin, and W. J. Vogel, Experimental Studies of Spatial Signature Variation at 900 MHz for Smart Antenna Systems, Accepted for publication in the IEEE Transactions on Antennas and Propagation, 1998. [39] A. Paulraj, R. Roy, and T. Kailath, A Subspace Rotation Approach to Signal Parameter Estimation, Proceedings of the IEEE,
74(7):10441045, July 1986.
[40] T. J. Shan, A. Paulraj, and T. Kailath, A Subspace Rotation Approach to Signal Parameter Estimation, IEEE Trans. on Acoustics, Speech, and Signal Processing, 33(10): 13771385, October 1987. [41] S.U. Pillai and B.H. Kwon, Forward/Backward Spatial Smoothing Techniques for Coherent Signal Identification, IEEE Trans. ASSP,
37(1):815, 1989.
[42] J. H. Winters, J. Salz, and R. D. Gitlin, The Capacity of Wireless Communication Systems can be Substantially Increased by the Use of Antenna Diversity, Proc. of the 1st International Conference on Universal Personal Communications, pages 2832, September 1992.
Bibliography
201
[43] D. Johnson and D. Dudgeon, Array Signal Processing Concepts and Techniques, Prentice Hall, Englewood Cliffs, NJ, 1993.
[44] R. O. Schmidt, A Signal Subspace Approach to Multiple Emitter Location and Spectral Estimation, Ph.D. Thesis, Stanford University, 1983.
[45] G. Zelniker and F. Taylor, Advanced Digital Signal Processing, Theory and Applications, Marcel Dekker, Inc., New York, NY, 1994.
[46] C. Nobel, Wireless LANs Meet IEEE Standard, PC Week, page 10, April 1998.
[47] G. Miller, Wireless LAN Standard Ratified, Wireless Integration,

pages 4,10, July/August 1997.
[48] J. Shandle, Wireless LANs: Welcome to the Virtual Workplace, Electronics, pages 2628, April 1992.
[49] A. Champness, 802.11s Step Toward Standardization, Wireless World, pages 4649, April 1997.
[50] B. Tuch, An IEEE802.1l WLAN Primer, Communication Systems Design, pages 2733, September 1997.
[51] L. Kleinrock and F. A. Tobagi, Packet Switching in Radio Channels: Part I - Carrier Sense Multiple-Access Modes and Their Throughput-Delay Characteristics, IEEE Trans. on Communications, pages 14001416, December 1975. [52] F. A. Tobagi and L. Kleinrock, Packet Switching in Radio Channels: Part II - The Hidden Terminal Problem in Carrier Sense Multiple Access and the Busy Tone Solution, IEEE Trans. on Communications, pages 14171433, December 1975. [53] R. Garces and J. J. Garcia-Luna-Aceves, A Near-Optimum Access Protocol Based on Incremental Collision Resolution and Distributed Transmission Queues, INFOCOM98, pages 158165, March 1998.
202
Bibliography
[54] K. C. Chen, Medium Access Control of Wireless LANs for Mobile Computing, IEEE Network Magazine, pages 5063, September/October 1994. [55] et al J. Chen, A Comparison of MAC Protocols for Wireless Local Networks Based on Battery Power Consumption, INFOCOM98, pages 150157, March 1998.
[56] P802.l l , Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications, IEEE Press, May 1997.
[57] J. Stine and G. de Veciana, Tactical Communications Using the
IEEE 802.11 MAC Protocol, MILCOM98, October 1998.

[58] J. Sobrinho and A. S. Krishnakumar, Real-Time Traffic Over the
IEEE 802.11 Medium Access Control Layer, Bell Labs Technical Journal, pages 172187, Autumn 1996.
[59] M. Fahey, Lucent Delivers Standard-Compliant Products, Wireless
Integration, pages 6,11, May/June 1998.

[60] A. Emmett, Proxims Product Suites, Wireless Integration, page 9,
May/June 1998.
[61] G. Xu and S.Q. Li, Throughput Multiplication of Wireless LANs for Multimedia Services: SDMA Protocol Design, Proc. of Globecom94, August 1994.
[62] W. Chen, S. Q. Li, and M. Schwartz, A New Voice Scheduling Scheme for Broadcast Bus Local Area Networks, Proc. IEEE Infocorn, pages 448457, 1989.
[63] S. Q. Li and M. El Zarki, Dynamic Bandwidth Allocation on a Slotted Ring with Integrated Services, IEEE Trans. Communication, pages 826833, July 1988.
[64] S. Sigmond, Priority SDMA for Wireless LAN, Master Thesis, University of Texas at Austin, June 1995.
Bibliography
203
[65] K. Sakakibara, Performance Approximation of a Multi-Base Station Slotted ALOHA for Wireless LANs, IEEE Trans. on Vehicular Technology, pages 448454, November 1992.
[66] W. Yang and G. Xu, A New Method for Designing Smart Antenna Downlink Weighting Vectors Based on the Filter Bank Concept, Proc. of Conference on Advanced Signal Processing Algorithms, Architectures, and Implementations VII, July 1997. [67] R. Sinha, T. Todd, V. Kezys, and J. Litva, Smart Antenna Basestations in Indoor SDMA Networks, Proc. of Wireless97, July 1997.
[68] F. Shad, T. Todd, V. Kezys, and J. Litva, Indoor SDMA Capacity Using a Smart Antenna Basestation, Proc. of ICUPC97, October 1997. [69] S.-S. Jeng, G. Okamoto, and G. Xu, Experimental Evaluation of Smart Antenna System Performance for Capacity Improvement, Proc. of Globecom97, November 1997.
[70] S. Andersson, M. Millnert, M. Viberg, and B. Wahlberg, An Adaptive Array for Mobile Communication Systems, IEEE Trans. on
Vehicular Technology, 40(l):230236, 1991. [71] J. H. Winters, Signal Acquisition and Tracking with Adaptive Arrays in the Digital Mobile Radio System IS-54 with Flat Fading, IEEE Trans. on Vehicular Technology, 42(4):377384, November 1993.
[72] B. Veen and K. Buckley, Beamforming: A Versatile Approach to Spatial Filtering, IEEE ASSP Magazine, pages 422, April 1998. [73] G. Strang, Linear Algebra and Its Applications, Harcourt, Brace, Jovanovich, San Diego, CA, 1988.
[74] R. J. Leopold and A. Miller, The Iridium Communications System, IEEE MTT-S Digest, pages 575578, 1993.
204
Bibliography
[75] J. Berkebile, A Two-Element Phase-Array Antenna for Reducing Multipath Effects in Hand-Held Satcom Units, Microwave Journal, pages 2436, January 1996. [76] R. H. Roy, ESPRIT, Estimation of Signal Parameters via Rotational Invariance Techniques, Ph.D. Thesis, Stanford University, August 1987.
[77] S.-S. Jeng, G. Okamoto, and G. Xu, Experimental Evaluation of Fading Reduction and Diversity Gain for Smart Antenna Systems, Proc. of VTC98, May 1998.
[78] N. Abramson, The Aloha SystemAnother Alternative for Computer Communications, Proc. Fall Joint Comput. Conf, AFIPS
Conf., page 37, 1979.
[79] L. G. Roberts, Aloha Packet System with and without Slots and Capture (ASS Note 8), Stanford Research Institute, Advanced Research Projects Agency, Network Information Center, Stanford, CA, 1972.
[80] D. Bertsekas and R. Gallager, Data Networks, Prentice Hall, Englewood Cliffs, NJ, 1992.
Index
Aloha protocol, 20, 185 collision avoidance, 23 contention resolution via splitting, 24 CSMA, 21
results, 164, 167 first fit new design, 68 results, 161 sorting benefits, 69
four-way handshaking, 24 slotting, 21 Array response matrix, 10

Beamforming complex conjugate SS, 14 definition, 11 dominant DOA, 13 feasibility, 130 pseudoinverse DOA, 13 pseudoinverse SS, 14, 67 SIR performance experimental results, 141 simulation results, 157 Data scrambler/descrambler, 66 Diversity gain definition, 134 experimental results, 133
EERL, 79, 98, 126, 133, 179 ESPRIT, 11, 13, 128, 149, 153 Experimental environment, 98
Experimental results summary, 153
Fading reduction experimental results, 133 peak-to-peak definition, 134 First fit SDMA/TDMA, 68
DOA based beamforming, 12 estimation, 11 Dynamic slot assignment FCFS design, 69

results, 161 first fit all design, 164
IEEE 802.11, 17 and smart antenna systems, 44,67 background, 17 DCF, 32, 60, 68 operation, 42 design objectives, 28 DSSS systems, 47 Lucent, 50 fast version, 17, 49, 54 FHSS systems, 47 Proxim, 53 IR systems, 46, 48 limitations, 4, 193 scope, 18
MAC Layer, 19
network operation, 39 interframe spaces, 39 transmitting data, 40
206
transmitting real time traffic, 41
Index
RangeLAN2, 3, 52 RangeLAN802, 53
PCF, 33
operation, 43
PHY Layer, 46
DSSS systems, 47 FHSS systems, 47 IR systems, 48
RadioLAN narrow band system, 54

Relative amplitude change, 106
Relative angle change, 106 SDMA, 5, 88, 105 SHARC DSP chip, 88
computational advantages, 93
PLCP frame format, 64 power saving mechanism, 34 RadioLAN alternative, 54

WEP encryption, 63 Iridium, 147, 152, 155
JVC Office LAN, 56 LEO satellite emulation

experimental results, 147 experimental setup, 99
DMA capability, 93 I/O port, 93 photograph, 94 SHARC board, 92 Singular value decomposition, 106 Smart antenna systems
and moving users, 46, 120, 133,141,157
Lucent
IAPP, 18
and the 802.11 MAC, 44

benefits, 1, 7, 153 Smart antenna testbeds, 79
WaveLAN, 3, 50 WavePOINT-II, 52 MATLAB, 63, 74, 157159, 179

Multipath angle spread, 125
1.5 GHz testbed, 79, 119 1.8 GHz testbed, 79, 87

backplane, 94 base station, 88
MUSIC, 13
Network utilization comparison between SWL and 802.11, 188 definition, 187
basesite handsets, 94 basesite T/R board, 89 computational units, 93 DMA Capability, 93

frequencies, 89 handset antenna, 98 handset T/R channel, 96
OpenAir Standard, 18, 52, 193
Proxim OpenAir Standard, 52
portable handsets, 96 sampling rates, 90 SHARC board, 92
Index
207
vocoders, 93 900 MHz testbed, 79 antenna array, 82 distribution box, 83 MUX/DEMUX box, 85 RF/IF unit, 83 stability of testbed, 87 transmitting tower, 99 Smart antennas and FDD systems, 113, 130, 154 Smart wireless LAN system, 59 dynamic slot allocation
capacity results, 166 design, 68
with people walking around the transmitters, 112 variation due to small displacement, 120 variation with displacement, 113 1.8 GHz testbed, 115 900 MHz testbed, 115 variation with frequency shift, 113 SS-SDMA, 69 Timing synchronization algorithm design, 72 background, 70 combined algorithm, 173 effectiveness, 174 quick algorithm, 76 results, 171 sorting, 171 standard algorithm, 74 Voice delay failure comparison for SWL and 802.11, 188 Wireless LANs adapted for smart antenna systems, 59 advantages, 2 adaptability, 3 installation speed, 3 mobility, 2 no wiring costs, 3 aloha protocol, 20 challenges, 19 hidden node problem, 26, 32 rapid growth, 4
worst-case SINR results, 161 MAC design, 60 MAC results, 185 motivation, 5 PHY design, 63 PHY results, 179 timing synchronization design, 70 results, 171 Spatial diversity, 5,105,113,120, 124, 130 Spatial signature based beamforming, 12 calculation of variations, 106 correlation for closely spaced transmitters, 119 definition, 7, 10 stability for stationary users, 107 1.8 GHz testbed, 109 900 MHz testbed, 108

Smart Antenna Systems and Wireless Lans The International Series in Engineering and Computer Science

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Smart Antenna Systems and Wireless Lans The International Series in Engineering and Computer Science

Uploaded by

Copyright:

Available Formats

SMART ANTENNA SYSTEMS AND WIRELESS LANS

THE KLUWER INTERNATIONAL SERIES IN ENGINEERING AND COMPUTER SCIENCE

SMART ANTENNA SYSTEMS

AND WIRELESS LANS

KLUWER ACADEMIC PUBLISHERS

eBook ISBN: Print ISBN:

All rights reserved

Created in the United States of America

Visit Kluwer Online at: and Kluwer's eBookstore at:

This book is dedicated to my parents, Matsuo and Betty Okamoto

This page intentionally left blank

3.3 MAC 3.3.1 3.3.2 3.3.3

3.3.4 3.3.5 3.3.6 3.3.7 3.3.8 3.3.9

3.4 PHY Layer of the 802.11 System . . . . . . . . . . . . . . . . 46

3.5 Examples of Current Wireless LAN Products . . . . . . . . .

4.2 SWL Media Access Control Design . . . . . . . . . . . . . . .

4.3 SWL Physical Layer Design . . . . . . . . . . . . . . . . . . . 63

5.2.3 Distribution Boxes . . . . . . . . . . . . . . . . . . . .

5.2.4 MUX/DEMUX Boxes . . . . . . . . . . . . . . . . . . . 85

5.4 1.8 GHz Handset . . . . . . . . . . . . . . . . . . . . . . . . . 96

5.4.1 Transmit/Receive Channel . . . . . . . . . . . . . . . . 96

5.4.2 DSP Chip . . . . . . . . . . . . . . . . . . . . . . . . .

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105

6.4 Spatial Signature Variation with Frequency Shift . . . . . . . 113

6.5 Spatial Signature Variation with Displacement . . . . . . . . . 113

6.10 Fading Reduction and Diversity Gain . . . . . . . . . . . . . . 133

Computer Simulation Results

Chapter 8. Conclusion Bibliography Index

193 197 205

at Austin. However, the material included here is broader in scope and

topic in wireless communications, but this book was written so that it

Shiann-Shiun Jeng for his invaluable contribution to our collaborative

agement over the years, none of my accomplishments would have been

possible. I am truly blessed to have them in my life.

Smart Antenna Systems

Rapid Growth of Wireless LAN Systems

in North America (many of whom have an 802.11-compliant product line

LANs. The first, and most obvious, advantage of a wireless system is

Rapid Growth of Wireless LAN Systems

IEEE 802.11 and the Smart Wireless LAN System

IEEE 802.11 and the Smart Wireless LAN System

This page intentionally left blank

Spatial Signature Definition

Background for Smart Antenna Systems

is the sequence of complex data symbols, p(t) is a real pulse,

is the propagation delay between the reference sig-

nal and the antenna element,

is the received signal without noise,

Spatial Signature Definition

Background for Smart Antenna Systems

where a is the spatial signature associated with the emitter defined as

If we now assume that we have time, our received signal is

terminals transmitting at the same

k. Written in vector form, the received signal is

denotes the array re-

Beamforming Techniques for Downlink

Backgroundfor Smart Antenna Systems

back to the mobile users, where

Beamforming Techniques for Downlink

Background for Smart Antenna Systems

is not simply and all it

Beamforming Techniques for Downlink

This page intentionally left blank