WIRELESS PERSONAL COMMUNICATIONS

Advances in Coverage and Capacity

THE KLUWER INTERNATIONAL SERIES IN ENGINEERING AND COMPUTER SCIENCE

COMMUNICATIONS A N D INFORMATION T H E O R Y Consulting Editor Robert Gallager

Other books in the series:

PERSONAL AND WIRELESS COMMUNICATIONS: Digital Technology and Standards, Kun II Park

ISBN: 0-7923-9727-4 WIRELESS INFORMATION NETWORKS: Architecture, Resource Managment, and Mobile Data, Jack M . Holtzman

ISBN: 0-7923-9694-4 DIGITAL IMAGE COMPRESSION: Algorithms and Standards, Weidong Kou

ISBN: 0-7923-9626-X CONTROL AND PERFORMANCE IN PACKET, CIRCUIT, AND ATM NETWORKS, XueDao Gu, Kazem Sohraby and Dhadesugoor R. Vaman

ISBN: 0-7923-9625-1 DISCRETE STOCHASTIC PROCESSES, Robert G. Gallager

ISBN: 0-7923-9583-2 WIRELESS PERSONAL COMMUNICATIONS: Research Developments, Brian D. Woerner, Theodore S. Rappaport and Jeffrey H. Reed

ISBN: 0-7923-9555-7 PLANNING AND ARCHITECTURAL DESIGN OF INTEGRATED SERVICES DIGITAL NETWORKS, A. Nejat Ince, Dag Wilhelmsen and Bülent Sankur

ISBN: 0-7923-9554-9 WIRELESS INFRARED COMMUNICATIONS, John R. Barry

ISBN: 0-7923-9476-3 COMMUNICATIONS AND CRYPTOGRAPHY: Two sides of One Tapestry, Richard E. Blahut, Daniel J. Costello, Jr., Ueli Maurer and Thomas Mittelholzer

ISBN: 0-7923-9469-0 WIRELESS AND MOBILE COMMUNICATIONS, Jack M . Holtzman and David J. Goodman

ISBN: 0-7923-9464-X INTRODUCTION TO CONVOLUTIONAL CODES WITH APPLICATIONS, Ajay Dholakia

ISBN: 0-7923-9467-4 CODED-MODULATION TECHNIQUES FOR FADING CHANNELS, S. Hamidreza Jamali, and Tho Le-Ngoc

ISBN: 0-7923-9421-6 WIRELESS PERSONAL COMMUNICATIONS: Trends and Challenges, Theodore S. Rappaport, Brian D. Woerner, Jeffrey H. Reed

ISBN: 0-7923-9430-5 ELLIPTIC CURVE PUBLIC KEY CRYPTOSYSTEMS, Alfred Menezes

ISBN: 0-7923-9368-6 SATELLITE COMMUNICATIONS: Mobile and Fixed Services, Michael Miller, Branka Vucetic and Les Berry

ISBN: 0-7923-9333-3 WIRELESS COMMUNICATIONS: Future Directions, Jack M . Holtzman and David J. Goodman

ISBN: 0-7923-9316-3 DISCRETE-TIME MODELS FOR COMMUNICATION SYSTEMS INCLUDING ATM, Herwig Bruneel and Byung G. Kim

ISBN: 0-7923-9292-2 APPLICATIONS OF FINITE FIELDS, Alfred J. Menezes, Ian F. Blake, XuHong Gao, Ronald C. Mullin, Scott A. Vanstone, Tomik Yaghoobian

ISBN: 0-7923-9282-5 WIRELESS PERSONAL COMMUNICATIONS, Martin J. Feuerstein, Theodore S. Rappaport

ISBN: 0-7923-9280-9 SEQUENCE DETECTION FOR HIGH-DENSITY STORAGE CHANNEL, Jaekyun Moon, L. Richard Carley

ISBN: 0-7923-9264-7 DIGITAL SATELLITE COMMUNICATIONS SYSTEMS AND TECHNOLOGIES: Military and Civil Applications, A. Nejat Ince

ISBN: 0-7923-9254-X IMAGE AND TEXT COMPRESSION, James A. Storer

ISBN: 0-7923-9243-4 VECTOR QUANTIZATION AND SIGNAL COMPRESSION, Allen Gersho, Robert M . Gray

ISBN: 0-7923-9181-0

WIRELESS PERSONAL COMMUNICATIONS

Advances in Coverage and Capacity

edited by

Jeffrey H. Reed Theodore S. Rappaport

Brian D. Woerner Virginia Polytechnic Institute

& State University

SPRINGER SCIENCE+BUSINESS MEDIA, LLC

ISBN 978-1-4613-7861-7 ISBN 978-1-4615-6237-5 (eBook) DOI 10.1007/978-1-4615-6237-5

Library of Congress Cataloging-in-Publication Data

A C L P . Catalogue record for this book is available from the Library of Congress.

Copyright © 1997 by Springer Science+Business Media New York Originally published by Kluwer Academic Publishers in 1997 Softcover reprint of the hardcover 1st edition 1997

Al l rights reserved. No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means, mechanical, photo­copying, recording, or otherwise, without the prior written permission of the publisher, Springer Science+Business Media, LLC.

Printed on acid-free paper.

TABLE OF CONTENTS

PREFACE ix

I NOVEL WIRELESS SYSTEMS AND HARDWARE

1. CellemetrySM - Telemetry Via Cellular 1 T. F. Evans

2. Smart Antenna Testbed for Mobile Wireless Systems 11

J. P. Kennedy and S. W Ellingson 3. A Practical Perspective on Cellular Digital

Packet Data (CDPD) Systems 23 N L. Romanowich

4. Low Cost BPSK 2.4 GHz Radio For Wireless LAN 33

W 0. Camp, Jr., A. Fleek, W Nunnery, R. Yeager and M Bracco

5. Error Floor of TI/4 DQPSK in Mobile Radio Channels 45

A. F. Molisch and Ernst Bonek 6. Real-Time Implementation of QCELP Vocoder

for Speech and Data in CDMA Cellular System Using TMS320C50 Fixed Point DSP Chip 57

J. W Kim, B. S. Yoon, S. J. Jang, W M Lee, M S. Lim and K. C. Han

II CDMA AND DSP TECHNIQUES

7. Spread-Spectrum Packet Radio Using Narrowband Headers 63

R. K. Morrow, Jr. 8. Congestion Relief on Power-Controlled

CDMA Networks 75 J. M Jacobsmeyer

vi

9. A Simulation Study of Interference and SIR in Integrated VoicelData Wireless

DS-CDMA Networks 85 K. Das and S. D. Morgera

10. Comparative Study of Adaptive Antenna Arrays in CDMA Communication Systems 97

R. B. Ertel and S. V. Schell 11. Cochannel Receivers for CPM Signals Based

Upon the Laurent Representation 109 P. A. Murphy and G. E. Ford

III PROPAGATION AND SYSTEM DESIGN ISSUES

12. Site Survey for Wireless Communications Evaluation for New York Mercantile Exchange 121

C. Panasik, D. Cone and T. Siep 13. Measurement of RF Propagation Into

Concrete Structures Over the Frequency Range 100 MHz to 3 GHz 131

C. D. Taylor, S. J. Gutierrez, S. L. Langdon, K. L. Murphy and w: A. Walton, III

14. Interference from Wireless Devices into Network Equipment Inside COs Between

450 and 914 MHz 145 M Hoque, S. Harbin, Z. Ma

15. Fading Phenomena in the Wireless Environment 153

F. Bohn, P. Major, L. Taylor 16. Transmitter Timing Optimization in Digital

Simulcast Networks 167 R. Petrovic

17. The Prediction of the Performance of Diversity Systems by Modal Analysis and FDTD 177

G. Dolmans 18. The Spectrum Management Process: How it is

Working to Support the New Wireless Services 189

D. J. Cohen

19.

20.

INDEX

Analysis of Eigenstructure Technique for DSSS Synchronization

N R. Mangalvedhe, J H. Reed Future Wireless Challenges for Integrated

Circuits D. E. Fague

vii

201

215

221

Preface

The wireless industry is growing at a phenomenal rate. Cellular subscribers are increasing at a rate of 45% per year, the market for wireless local loop service is growing at a rate of 42%, and the wireless local area network market is growing at a rate of 61 %. This growth and potential for future growth has motivated companies to commit $20 billion in obtaining 90 MHz of PCS spectrum during the recent FCC auctions in the United States. Obviously spectrum is a costly, but critical, resource. Efficient utilization of this resource is essential for profitable wireless service. To meet this challenge, researchers in wireless communications are tenaciously developing more spectrally efficient modulation techniques, planning tools for efficient communication system layout, and digital signal processing techniques for more robust communications.

The papers and lectures presented in this book were originally given at the Sixth Annual Virginia Tech Symposium on Wireless Personal Communications and cover a broad range of topics in wireless communications. The majority of the papers are relevant to creating higher capacity (spectrally efficient) systems with greater coverage. Topics include adaptive antenna array measurements and algorithm comparisons, Cellular Digital Packet Data deployment guidelines, speech coding techniques, wireless system design methodology, and propagation measurements in hostile or previously unexplored channels.

The annual Symposium on Wireless Personal Communications is held during the first week of June at the Virginia Tech campus in Blacksburg, Virginia. This conference, hosted annually by Virginia Tech's Mobile and Portable Radio Research Group (MPRG), was host to over 250 leading researchers from industry and academia around the world.

The MPRG was founded in 1990 to serve the educational and research needs of its Industrial Affiliates, a group of companies that provide technical direction, managerial advice and financial support to the group. Current MPRG affiliates include Bellcore, BellSouth, Ericsson, FBI, Grayson Electronics, GTE, Hewlett Packard, Hughes Electronics, Lucent Technologies, MCI, Motorola, National Semiconductor, Nortel, Southwestern Bell, STi-CO Industries, Tektronics, and Texas Instruments. The research presented here reflects the research interests of these companies. MPRG is grateful to these companies for their contribution to Virginia Tech's academic pursuits in wireless communications and to the symposium.

This book is organized into three sections. Section I covers novel wireless systems and hardware and focuses on adaptive arrays, wireless hardware methodology, and deployment guidelines for CDPD among other topics. Section II focuses on Code Division Multiple Access (CMDA) and DSP techniques, both of which are used to facilitate more efficient utilization of the spectrum under adverse channel conditions. This section includes techniques for mitigating interference and multi path. Section III

x

deals with propagation and system design issues. Propagation measurements are presented and include results for the channel characteristics at the New York Mercantile Exchange with closely packed people and hostile electromagnetic interference, to measurements of concrete slabs.

Section I. Novel Wireless Systems and Hardware. This section presents emerging technology for implementing wireless systems. Tom Evans from BellSouth Wireless, Inc., introduces a new and very inexpensive way for transmission of short messages using excess capacity on the cellular control channels. Joe Kennedy with E-Systems, Inc., discusses the smart antenna testbed deployed in the Washington, DC area, summarizes the channel characteristics observed, and compares the performance of switched fixed beam and fully adaptive antenna arrays. Nadine Romanowich, AT&T Wireless Services, relates her experience and provides guidelines for deploying Cellular Digital Packet Data (CDPD) Systems. William Camp and his team at the IBM Network Hardware Division, Research Triangle Park, North Carolina, describe their design methodology in creating a low cost wireless local area network system. Andreas Molisch from the Telechnishe Universitat Wien provides a refined estimate of the error floor for n/4 DQPSK in mobile channels. The design of a QCELP speech coder is presented by J.W. Kim with the Electronics and Telecommunication Research Institute, Taejon, Korea.

Section II. CDMA and DSP Techniques. The second section deals with code division multiple access (CDMA) and digital signal processing (DSP) techniques for wireless telecommunications. Both techniques are used to mitigate the impact of interference. Robert Morrow, of Morrow Technical Services, describes the network ramifications of using narrowband headers in conjunction with spread spectrum packet radio. Jay Jacobsmeyer at Pericle Communications Company describes a method for increasing the capacity of CDMA systems by adjusting both the power and data rate of users' signals. Kaushik Das, with McGill University presents a simulation case study of multiple access interference in an integrated voice/data wireless network. Rich Ertel from Pennsylvania State University provides a comparative study of the performance and complexity of several adaptive array algorithms applied to CDMA communication systems. Peter Murphy from the University of Califomia, Davis, presents a unique way to form an optimal receiver for continuous phase modulated signals based upon the Laurent series.

Section III. Propagation and System Design Issues. This section is concerned with measurements of propagation conditions that are used to create deployment guidelines for wireless systems. In this context, Carl Panasik from Texas Instruments, presents measurements made at the New York Mercantile Exchange (NYMEX). This is a particularly hostile propagation environment because of the dense packing of bodies and significant electromagnetic interference. Clayborne Taylor with Phillips Laboratory discusses the propagation characteristics of concrete structures over the range of 100 MHz to 3 GHz. Mahbub Hoque, of Bellcore, examines the electromagnetic compatibility of wireless transmitting devices with switching equipment inside a telephone company facility. Frank Bohn presents a review paper in propagation and

xi

examines research and experimental work to enable more accurate prediction of Rician propagation loss and bit error rates under different environmental conditions. Rade Petrovic, with the Center for Wireless Communications at the University of Mississippi, examines the impact of differential delay of simulcast signals on an FSK receiver. Guido Dolmans with Eindhoven University of Technology of the Netherlands presents a propagation model based on modal analysis which is used to provide prediction of bit error rates, signal to noise ratios and coverage inside buildings. David Cohen with the Graduate School of Management and Technology at the University of Maryland, provides a concise overview of the organization of U.S. spectrum management, spectrum allocation and assignment, wireless allocation policy and spectrum engineering and economics. This analysis suggests ways of optimizing a simulcast network. Nitin Mangalvedhe, a Ph.D. student at MPRG, presents a new technique for spread spectrum synchronization that provides a gradual or "soft" estimation of the spreading sequence. Dan Fague from Analog Devices, discusses the technical challenges in creating next generation integrated circuits for wireless communications and promising technical approaches to meet these challenges.

This book would not have been possible without the diligence of the MPRG staff. Special thanks goes to administrators Jenny Frank, Annie Wade, and Kathy Wolfe for handling many of the administrative details; Research Associates Prab Koushik and Francis Dominique for coordinating events during the symposium; and visiting professor Ivan Howitt for helping to review the submissions to the symposium. Lastly, we are grateful to the authors who have given their time to present their work. Their efforts have made this manuscript possible.

WIRELESS PERSONAL COMMUNICATIONS

Advances in Coverage and Capacity

Abstract

1 Cellemetry·m - Telemetry Via Cellular

Thomas F. Evans

Network Access Strategy

BellSouth Wireless, Inc.

1100 Peachtree Street NE, Room 808

Atlanta, Georgia 30309

Cellemetry,m, as its name implies, is the combination of telemetry with the cellular telephone

network. Cellemetry'm provides the means of sending short, telemetry-like messages over the

cellular telephone system in a manner that is virtually transparent to the cellular operator.

CellemetrySm can provide message transport for many different businesses. It can report alarm

panel status, utility meter readings, vehicle and trailer location, vending machine status, to name

afew applications. The list of potential applications is limited only by one's imagination.

Cellemetrysm uses a heretofore underutilized portion of the cellular system, the overhead control

channels, to convey the Cellemetrym messages. These control channels are used by the cellular

system to communicate information necessary for all call initiations (both incoming and outgoing)

between the cellular system and the cellular customer. The message handling capacity of these

control channels is far greater than is required by the cellular system, even during the busiest

times of the day.

The key to Cellemetry,m operation is the manner in which roamer telephones operate in the

cellular system. A roamer cellular telephone is defined as a cellular telephone operating in any

system other than its home system. When a roamer cellular telephone is turned on, it recognizes

the fact./hat it is not in its home system and accordingly sends its Mobile Identification Number

(MIN) and its Electronic Serial Number (ESN) to the cellular system via one of the control channels. The cellular system recognizes the roamer number and routes the MIN and ESN to the

roamer's home system for validation via a special network which links all of the cellular systems

J. H. Reed et al. (eds.), Wireless Personal Communications

© Springer Science+Business Media New York 1997

2

together across the United States. The Cellemetr/m radios respond exactly like roamer

telephones except the MINs are specially assigned so that the MIN and ESN are routed to a

Cellemetrysm Gateway connected to the same intra-cellular network. In the case of Cellemetr/m ,

the MIN serves to identifo the Cellemetr/m radio and the ESN is the data field which contains the

32 bit telemetry message.

The CellemetrySm Gateway processes, stores and routes the CellemetrySm messages according to

their type and customer requirements. Some applications may require immediate processing as is

the case with alarm monitoring while an application such as vending machine status may only

need the messages stored and transferred in a batch once a day.

I. Introduction

As one might gather from its name, Cellemetrysm is the marriage of cellular and telemetry

technologies. Cellemetrysm provides the conveyance for short telemetry messages via the cellular

telephone system. Unlike other recent similar systems, Cellemetrysm is virtually transparent to

the cellu.lar operators and is transparent to the regular cellular telephone customers. Cellemetrysm

also has another significant advantage over preceding systems in that it does not require that any

additional equipment be installed at the cellular base station. The only equipment that must be

installed is the Cellemetrysm Gateway which can be located either at the individual Mobile

Switching Center (MSC) or at a regional facility. More significant than the elimination of the

need for equipment installation at each cellular base station is the fact that Cellemetrysm will

cover the entire cellular system's coverage area from the first day of operation. There is no need

to concentrate users to make the system economically feasible.

Cellemetrysm can be integrated in many specific services that require one-way or two-way short

telemetry messages. Flexibility is further enhanced since the Cellemetrysm Gateway can process

the Cellemetrysm messages according to the specific needs of each individual application. For

instance, if Cellemetrysm is used to convey a message from a alarm panel, the Cellemetrysm

Gateway will process the message on a real-time, immediate basis and pass the message to the

Central Alarm Monitoring Service. On the other hand, if a soft drink vending company utilizes

Cellemetry"m to poll its machines each night for their stock status, the Cellemetrysm Gateway

will accumulate all of the responses from the individual vending machines each night and provide

them in batch form when requested from the vending company the next morning. In fact,

individual applications can have different responses from the same Cellemetrysm radio. While the

vending machine uses batch processing for its stock status, it could have an alarm message

3

conveyed to the vending company on an immediate basis. A similar scenario is applicable for

utility meter reading. Normal meter readings can be obtained on a batch basis during the night and

delivered to the utility company the following morning. However, real-time meter readings can

be made any time during the day for customers who desire to close out or open service and

require an immediate, current meter reading. In fact, Cellemetrysm can even be used to used to

turn on or tum off utility service remotely by the utility customer service representative. With

Cellemetrysm , waiting a day for the utility technician to travel to a residence and install a meter

to initiate service is a thing of the past.

II. System Description

First, let's look at the cellular telephone system and how Cellemetrysm works in harmony with it.

The AMPS cellular telephone system has a total of 832 channels, half of which are assigned to

each of the two competing cellular carriers in each market. Each cellular carrier uses 21 of its 416

channels as control channels. Each control channel set consists of a FOrward Control Channel

(FOCC) and a REverse Control Channel (RECC). The FOCC is used to send general

information from the cellular base station to the cellular telephone. The RECC is used to send

information from the cellular telephone to the base station; and hence, the cellular system. The

control channels are used to initiate a cellular telephone call. Once the call is initiated, the cellular

system directs the cellular telephone to a voice channel. Once the cellular telephone has

established service on a voice channel, it never goes back to a control channel. All information

concerning handoff to other voice channels and termination of the telephone call are handled via

communication over the voice channels.

The cellular control channels are more robust that the voice channels for several reasons. First,

the control channels are digital by design and utilize majority voting as error detection for all

messages sent over the RECC. Each message is transmitted five times via the RECC. If the

cellular base station receives the same message for three of the five transmissions, it deems the

message as correct. Second, the frequency reuse plan for control channels is different than the

reuse plan for voice channels, 12 to I for control channels and 7 to I for voice channels, which

reduces interference on the control channels. Finally, most cellular operators operate the control

channels at the maximum permitted transmit power and slightly reduce the voice channel transmit

power. All of these factors make Cellemetrysm a reliable medium.

The data flow of the Cellemetrysm service is based on the data flow for roaming cellular telephone

customers. A roaming cellular customer, or roamer, is defined as a customer who is operating

4

outside of his or her local, or home, system. The FOCC broadcasts information concerning the

System IDentification (SID) of the cellular system on a frequent basis. Each cellular telephone

when it is initially turned on, compares the SID of its home system, which is stored in its non­

volatile memory, to the SID being sent over the FOCC. If a match is obtained, indicating that the

telephone is operating in its home system, nothing occurs. If a match is not obtained, indicating

that the cellular customer is roaming outside of the home system, the cellular telephone

illuminates its ROAM light to alert the customer that roaming cellular rates will apply. The

cellular telephone will then check other portions of the FOCC message stream for instructions

pertaining to roamers. In particular, each cellular system operator may specify how often the

roamer must transmit its identity via the RECC to the cellular system. Depending on the cellular

operator's preference, roamers may be required to register as often as each call or as infrequently

as once a day.

As shown in Figure 1, when the roamer registers with the visiting cellular system, it sends its

Mobile Identification Number (MIN) and its Electronic Serial Number (ESN) via the RECC to

the Visiting MSC. In this example let's assume a cellular customer from southern Florida is

roaming. in the Atlanta cellular system. All of the MSCs in the USA are connected via the

Intersystem Signaling Network, IS-41, which provides a means of verifying the identity of

roamers. The Visiting MSC recognizes that the MIN is not one of its Atlanta cellular customers

since its first six digits do not correspond to Atlanta cellular telephone numbers. The Visiting

MSC looks in its data base to determine which Home MSC corresponds to the first six digits of

the roamer's MIN in order to route the MIN and ESN information. Once the routing is

determined, a request for validation is routed to the Home (southern Florida) MSC. The Home

MSC checks its data base and determines that indeed the MIN and ESN are valid, that the

customer's bill is current and what, if any, custom calling features this customer has. This

information is sent as a Registration Notification back to the Visiting MSC. The roamer now

enjoys the same level of cellular service in the Visiting MSC as the Home MSC while the Visiting

MSC has a level of confidence that the roamer is not fraudulent.

Inter System Signaling Network (IS-41)

Figure 1

305!t1~7' ESN 32 Bit Number

Roamer Registers As a Roaming Cellular Uaer

When Phone Is Turned On

5

If we connect the Cellemetrysm Gateway to an IS-41 port on the MSC, we can capture the

validation request messages exactly in the same manner as a roamer validation request. The

Cellemetrysm Gateway is a hardware and software redundant platform operating on UNIX which

occupie~ approximately one 19 inch equipment rack. Reliable operation of the Cellemetry=

Gateway can survive link, hardware and software failure. In Figure 2, the roamer is replaced by a

Cellemetrysm radio. The MIN now becomes the 10 digit equipment ill and the ESN becomes the

data payload. The Gateway looks exactly like another MSC to the Visiting MSC. Information is

added to the data base of the Visiting MSC to direct the special MINs assigned to the

Cellemetrysm Gateway. As described earlier, the Cellemetrysm Gateway processes the

equipment ill (MIN) and data payload (ESN) as determined by the particular Cellemetrysm

application. The Gateway uses dial-up or dedicated lines to communicate with the Cellemetrysm

Service Provider (e.g. the security company, electric utility company, vending company, etc.)

6

Gateway

To Service Provider via Dial Up or Dedicated

Lines

Figure 2

Cellemetry Device

Equipment /0 000-1234567

Meter Reading 32 Bit Number

In order to make the Cellemetrysm system appear transparent to the cellular system, the

Gateway must send the proper validation response back to the MSC, exactly the same as the

Home MSC would. However, in the case of Cellemetrysm , the Gateway sends back a validation

response that reports the MIN and ESN as being valid, but that the customer cannot make any

outgoing telephone calls. This validation is an addition safeguard against fraudulent calls. Even if

someone intercepted the Cellemetrysm unit's MIN and ESN, they could not use this MINIESN

combination to make cellular calls.

After a preset period oftime, the Gateway sends a Registration Cancel message for each

Cellemetrysm message. This action is necessary in order not to load up the Visitor Location

Register (VLR) of the Visiting MSC. A roamer registration will remain in the VLR until the time

for re-registration occurs, which could be as long as 24 hours, or until the Home MSC informs the

Visiting MSC that the roamer has moved to another MSC's system. In this manner, the

Cellemetry message registration remains in the Visiting MSC's VLR only for a few seconds.

Cellemetrysm communications in the opposite direction via the FOCC are slightly different. As

shown in Figure 3, the Service Provider provides an input to the Gateway via a landline. The

Gateway, in turn, places a "call" to the Cellemetrysmradio via the Roamer Access Port of the

Visiting MSC. The MSC must be programmed to accept the special MINs associated with

Cellemetry service is its market area. The MSC is also configured to accept outbound

Cellemetrysm messages only from a specific telephone line or trunk group, further insuring that

no one can send an outbound Cellemetrysm message except through the Gateway. In the

7

outbound message the ESN is not transmitted, so some other means of coding must be employed

to provide information to the Cellemetrysm radio and its host equipment. The Cellemetrysm radio

has the ability to look for ten separate MINs, including its own Equipment ID, and report the

occurrence of any of these MINs should they be transmitted via the FOCC.

Roamer Access Port

Gateway

Trigger Message Received From

.- Service Provider

Figure 3

111. Modes of Operation

There are two modes of operation of the Cellemetrysm radio, the modem mode and the standalone

mode. In the modem mode, the Cellemetrysm radio acts exactly like a modem, passing

information in both directions without modification. A host controller will be required in the

modem mode to direct the action of the Cellemetrysm radio to initiate the transmission of the

Cellemetrysm messages via the RECC and to encode the information to be transmitted in the ESN

field. The host controller also interprets the MINs transmitted via the FOCC to the

Cellemetrysm radio and takes appropriate action.

In the standalone mode, the Cellemetrysm radio uses resident software to provide a definite

functionality without the need of an external host controller. This mode is designed for

applications where size, power consumption and cost must be minimized. In the standalone

mode, the Cellemetrysm radio can be used to respond with a pulse count, such as a utility meter

reading or copier counter reading, either on an immediate basis or on a delayed basis (e.g. 0-3

hours) depending on the MIN transmitted via the FOCC. The Cellemetrysm radio can also be

8

used to turn on or turn off a remote device in the standalone mode as well as report a contact

closure.

IV. Applications

First, let's look at an electric meter reading application. A Cellemetrysm radio is installed in the

electric meter. Operating in the standalone mode, circuitry within the electric meter sends pulses

to the Cellemetrysm radio, reflecting the power usage at the residence or business. During the

night at the desired interval, typically once a month. for residential customers, the Cellemetrysm

Gateway sends out a page over the cellular system using a special MIN that is assigned to a

number ofCeliemetrysmradios in the system. This special MIN, called the group MIN, causes

the Cellemetrysm radios to read the electric meter immediately, but reply randomly over the next

three hours. In this manner, all of the CRFMs will not try to respond at once, reducing

collisions. The Cellemetrysm Gateway will accumulate all of the meter readings and report them

to the electric utility company when requested. If an immediate meter reading is required, such as

with closing out or opening an account, a MIN page corresponding to the Equipment ID of the

meter is ,sent out over the cellular system. The Cellemetrysm radio responds almost immediately

with its current meter reading. There are other special MINs, which, when transmitted via the

FOCC, causes a switching signal to be delivered to the electric meter, providing a means of

controlling a device such as a contactor to disconnect/connect electric service or an external

contactor to shed certain load devices such as a water heater or air conditioning unit.

A second Cellemetrysm application is security alarm panel reporting. In this situation, the

Cellemetrysm radio operates in the modem mode connected to the host controller in the security

panel. Following industry established protocol, the Cellemetrysm radio sends a message when

directed by the host controller. The MIN identifies the account or customer while the ESN

identifies the type of alarm, such as intrusion, panic, fire, etc. Once the Cellemetrysm message is

received at the Cellemetrysm Gateway, the Cellemetrysm Gateway places a call to the alarm

central monitoring facility. The alarm central monitoring facility handles the Cellemetrysm

reported alarm exactly as it would an alarm reported over telephone lines. The alarm central

monitoring facility returns a call to the Cellemetrysm Gateway, indicating that the alarm situation

has been received and is being handled by the alarm company. The Cellemetrysm Gateway in

turn initiates action for the cellular system to page the Cellemetrysm radio which originally sent

the alarm via the FOCC. Once the page is received by the Cellemetrysm radio, the page is

reported to the security panel, thus completing the round trip, indicating that the alarm has been

properly received and reported.

Measurements have shown that the capacity of the control channels far exceeds the needs of the

cellular system. In fact, a single control channel can accommodate 35,000 registrations per hour.

Even on the busiest cell sites, there is more than enough capacity for Cellemetrysm. To further

eliminate the possibility of having any impact on the cellular system, the Cellemetrysm radio

utilizes the busy-idle bit which is sent over the FOCC. The busy-idle bit is one bit multiplexed

in the FOCC data stream which indicates that the cellular base station is communicating with a

cellular user. A cellular telephone will not attempt to register with the cellular base station if the

busy-idle bit is set high. Only if it is set low will the cellular telephone attempt a registration.

The Cellemetrysm radio looks at the busy-idle bit over a multi-second window. If the busy-idle

bit is set high for greater than a certain percentage of the time, the Cellemetrysm radio will defer

its registration until the busy-idle bit activity is reduced. In this manner, regular cellular

customers always will obtain the control channel and service first.

V. Summary

9

Cellemetrysm provides a low cost solution for telemetry messages, providing cove rage over the

entire footprint of a cellular system from the first day of installation. No additionai equipment is

required at the cellular base stations. Economical service can be provided without having to

concentrate Cellemetrysm customers. Finally, Cellemetrysm service can be provided by a cellular

carrier with no impact on the cellular system's normal voice operations.

© 1996 BellSouth Wireless, Inc.

Abstract

2 Smart Antenna Testbed for Mobile Wireless Systems

J. P. Kennedyand S. W. Ellingson

E-Systems, Inc. 1595 Springhill Rd. Vienna, VA 22182

A software radio-based test bed has been built and operated to collect experimental data related to smart

antenna performance. Environmental data on radio channel characteristics as well as comparative data

on different types of smart antenna algorithms were gathered and analyzed. To make realistic

predictions of "smart" antenna performance, statistical characterizations of multipath propagation and

interference were performed in the suburban Washington. D. C. area using data collected on coherent

eight-channel software radio equipment built to geolocate mobile cellular phones. Analysis of the data

yields estimates of the probabilities of encountering a given number of co-channel interferors or

multipath components. The distributions of interference and multipath components as a function of

relative power and angle are also presented. Preliminary results suggest that multipath fading, and not

interference,. is a dominant mode of signal degradation. Another variant of the software radio was

constructed to evaluate different C/I measurement algorithms and to compare different smart antenna

algorithms. Comparisons of performance for fully adaptive algorithms vs. switched fixed beams in an

suburban/urban environment is presented.

Background

This paper presents two results. Both are based on data collected through multichannel wide band

software radio equipment. The first result pertains to a detailed analysis of the radio propagation channel

as it relates to smart antenna performance. Specifically, the likelihood of multipath and interference are

investigated and characterized with respect to relative power, angle of arrival and time of arrival. These

results were collected during a field trial on geolocation equipment in the Washington DC area in 1994

and 1995. The second result uses the same hardware architecture, loaded with smart antenna software

algorithms, and compares the performance of different antenna algorithms to that of a reference sector

antenna. True "apples to apples" comparative data can be collected by tuning multiple software radio

channels to the same cellular user channel, and then programming different smart antenna algorithms into

each of the software radio channels. In this way, different diversity schemes, beam/null steering

techniques, or fully adaptive algorithms can be directly computed on the same live source data. The

comparative data presented was collected in the Tyson's Comer, Virginia area in 1995.

J. H. Reed et al. (eds.), Wireless Personal Communications

© Springer Science+Business Media New York 1997

12

Radio Channel Effects on Smart Antennas

The ability of "smart" antennas to mitigate multipath fading and to null co-channel interference can be

exploited in cellular wireless communications systems to increase system capacity by increasing the

frequency reuse factor. Evaluating the economics of deploying such antennas over a network requires an

estimate of the capacity increase, and this estimate depends upon assumptions regarding the radio

environment as well as some measure of antenna performance. While antenna performance is easily

measured by simulation or in the lab, in practice the radio environment often remains uncharacterized

until the beginning of field trials, a practice which can lead to unpleasant surprises.

An ideal tool for characterizing the radio environment would be co-located with existing base station

antennas, would exploit ordinary cellular traffic instead of special test signals, and would automatically

measure the angle and power of each multipath or interference component present on a channel. The

CAPITAL system, originally developed to monitor traffic congestion in the suburbs of Washington, D. C.

Figure 1. Map Showing Time Difference of Arrival and Line of Bearing Curves from Three Sites

Intersecting at a Mobile Emitter

CAPITAL instrumented selected base station towers with an eight element linear array. The array is

followed by a software radio-based signal receiver/signal processor as shown in Figure 2. The same

hardware architecture may be configured for geolocation or for real-time operation as a adaptive

beamformer and optimal diversity combiner. The signal from each element is downconverted to a

13

wideband IF and then digitized by a high dynamic-range converter as shown in Figure 3. Digital tuners

convert the signal in the desired channel to a complex baseband format at a sample rate commensurate

with the channel bandwidth.

RF8

Timing Reference

Frequency Reference

Figure 2. Hardware Block Diagram

The multichannel digital data is subsequently broken down into signal components using the alternating

projection technique [1]. A byproduct of the analysis is a set of beamforming coefficients, one for each

component. .The components are separated using these coefficients and then the Supervisory Audio Tone

(SAn frequency is used to discriminate between multi ath com onents and interference com onents.

Angle&TOA

Figure 3. Functional Block Diagram of CAPITAL

14

The results of the analysis are conveniently expressed in a "PDAgram" (Power-Delay-Angle diagram) as

shown in Figure 4. The peak on the zero delay axis corresponds to the direct path while the other pair of

peaks represent multipath components. In the second example shown in Figure 5, six signal components

were identified, four from one source with a SAT frequency of 6000 Hz., and two components (shown in

a different color) from another source with a SAT frequency of 6030 Hz. The CAPITAL system can

produce thousands of these results every hour so that the interference and multipath of each channel can

be statistically characterized b site and time of day.

Relative Power (dB)

50

40

30

20

10

Figure 4. PDAgram of a Signal With Three Components

40

30

20

10

o o

Figure 5. PDAgram of a Signal With Six Components.

Radio Channel Characterization Results

15

The CAPITAL system was used to collect over 1400 signals in the "mixed suburban/urban" area in the

vicinity of Interstate 66 and Tyson's Cornt:r, Virginia. This area is characterized by a mix of residential

and commercial buildings up to 20 stories high, grid oriented side streets and wide thoroughfare

roadways, and moderate to high cellular traffic. The antenna arrays were located on a variety of

structures including roof tops, sides of buildings, sides of water towers, and inside offices operating

through windows. Only one site was a "true" antenna tower. These mounting locations are consistent

with the mounting sites expected for smart arrays. The relative incidence of multipath and interference

encountered is shown in Figure 6. At least one multipath component was present about half the time, but

significant interference was only observed in less than five percent of all signals measured. For this

experiment, the equipment was set up to measure coherent energy received down to 20 dBc from the

largest signal in the sample set. Measurements beyond this level are not as significant to the smart

antenna performance assessment, and more difficult to make since the number of multipath components

at lower power levels become very large.

16

4 Number of Components

Figure 6. Relative Likelihood of Multipath and Interference Components.

The data collected through the equipment mounted at each tower is a combination of mobile emitted

signals which are connected to the host tower for cellular service, and mobiles connected to other towers

within radio line of sight for cellular service. This mix of data, as opposed to only data on mobiles

connected to the host tower, is more likely to exhibit the characteristics of the environment in which a

smart antenna will operate.

The statistics of the measured multipath angles and powers (relative to the line-of-sight path) are shown in

the histograms presented in Figures 7 and 8. The angular distribution exhibits a strong peak around zero

degrees and the power distribution exhibits a peak near -10 dB. The joint distribution shown in Figure 9

suggests that multipath components with angles of arrival close to the line-of-sight component tend to be

stronger than components at wider angles.

0.18,..----------------------------,

0.16

0.14

0.12

0.06

0.04

0.02

-180-162-144-126-108 -90 -72 -54 -36 -18 0 18 36 54 72 90 108 126 144 162 Relative Angle (Degrees)

Figure 7. Angular Distribution of Multipath

17

The angular distribution and power profile for the multipath matches an intuitive model of how the

multipath is generated. Specular reflectors close to the source mobile will create close-in-angle, relative

close in power, and relative short time delay components. The statistics in Figures 7 and 8 match this

model, as does the delay spread data presented in Figure 10. These results are significant in that they

indicate that the beams created by "smart" antenna arrays operating in this environnIent will not generally

attenuate the highly correlated multi ath arriving close in angle to the direct ath signal.

0.45-r-------------------------, 0.4

0.35

0.3

~0.25 .g l 0.2

0.15

0.1

0.05

0----, -20 -16 --12 -8 -4

Relative Power (dB)

Figure 8. Power Distribution of Multipath.

12 16

18

Figure 9. Joint Distribution of Power and Angle.

0.09-,-::;-------------------------

0.08

0.07

0.06

f:::: 0.03

0.02

0.01

§ ~ ~ ~ ~ ~ R ~ ~ § ~ ~ ~ ~ ~ § ~ @ ~ ~ ~ ~ ~ ~ ~ Delay (Nanoseconds)

Figure 10. Distribution of Multipath Delay

19

Smart Antenna Algorithm Comparisons

Another capability of the software radio instrumentation is to collect raw data that can then be used to test

different types of smart antenna algorithms. The software radio architecture configured to perform this

evaluation is shown in Figure 11. It consists of a tower-mounted antenna unit and an electronics cabinet

that is installed in the base station enclosure. The system interfaces to the existing base station as an

applique. From the base station's point of view, the smart antenna is simply a replacement for the existing

antennas for a sector.

The antenna unit consists of a four-column linear broadside array with half-wavelength spacing.

Commercial off-the-shelf static sector beam antennas were selected to provide 14 dBd gain, 83°

beamwidth, and> 20 dB front-to-back ratio for each column. Each column consists of two vertically

stacked elements to rovide additional gain in response to many market's requirements.

ADAPTIVE DIGITAL SIGNAL PROCESSING

DIGITAL UPCONVERSION

O/A CONVERSION

RF COMBINING & UPCONVERSIDN

NARROWBAND CHANNEL RECEIVERS FOR SECTOR

Figure 11. Smart Antenna Reverse Link Processing

The operation of the test fixture is as follows. First, each of the four antenna outputs is individually

downconverted and digitized." The data is distributed to a bank of 10 channel processing "streams", each

consisting of digital downconverters, a digital signal processor (DSP), and a digital upconverter (DUC).

20

The digital downconverters output a single 30 kHz cellular channel as complex data at baseband. The

DSPs (one Texas Instruments TMS320C40 per channel) run an adaptive signal processing algorithm, and

output a single data stream to a DUe. The DUC modules perform upconversion and digital-to-analog

conversion. DUC outputs are passed to an RF combinerl up converter stage. The output is a single feed to

the base station's receive multicouplers.

The adaptive signal processing algorithm is a specialized variant of the Constant Modulus Algorithm

(CMA). The algorithm is "blind"; meaning that the array geometry is not critical to the design. This

algorithm directs gain toward desired signals, steers nulls onto signals identified as interferers, and strives

to optimally combine multipath to further enhance the strength of the desired signal.

Smart Antenna Algorithm Comparative Results

For this testing, the test antenna unit was positioned on the roof of a 7-story office building in Vienna,

VA, located about 1 mile from the 1-495 Beltway surrounding Washington, DC. This area is urban, with

a dense concentration of tall office buildings and a high subscriber density. The majority of reverse-link

signals originate from nearby Route 7, a highly congested 6-lane divided highway. Due to a relatively

low antenna height, and because the array was not located at a base station site, the ambient reverse-link

C/I was relatively low, typically 10-15 dB. As shown earlier, wide-angle multipath tends to be the

dominant factor limiting CII performance in the test area, although co-channel interference and other

fading mechanisms were also prevalent.

In this testing, direct combining of the four antenna outputs was used to generate the reference antenna

output. With half wavelength spacing, this method gives a good approximation of the single-element

performance, and does not require any special calibration procedures. The FSB algorithm used in this

case samples the power on four narrow, equally spaced beams and selects the strongest one as the output.

It is implemented in software as described above, has a maximum array factor of 6 dB, 4 dB crossover

points, and a 40 kHz switching rate.

Figure 12 shows comparative performance data for CII enhancement for space diversity, fixed switched

beams, and fully adaptive array (AA) combining as compared to a single sector beam. The fully adaptive

combining algorithm performance advantages are principally due to its ability to constructively combine

the ample multipath in the environment.

Insight into the operation of the fully adaptive algorithm as a function of the instantaneous operating

environment is shown in Figure 13. The C/I probability distribution based on real-time signal to

interference plus noise ratio (SINR) analysis and an upper and lower bound estimates based on a

statistical analysis of power levels using the "C+I vs. I" approximation [3] are shown. The ~C/I

distribution between these bounds is due primarily to steering of the maximum array factor and optimal

combining of multipath. ~C/I above the upper bound is primarily due to nulling of interferers. ~C/I

21

below the lower bound is due primarily to high-quality incident signals with SINR> 25 dB or so. All the

AA can do with these si nals is increase gain, which has little effect on the signal quality.

Conclusions

:::: u )C

... " 0 ... ..... ..... = E

.~ .. ... ~ .. 0 ..... o ... ~! lit

-4 -2

80

60

40

2

---Diversity Adaptive Array

-------- Fixed Switched Beams

--- - - Space Diversity

o t 2 3 4 5 6 7 8 910111213141516 17 18

x (dB Relative to Static Sector Beam)

Figure 12. Performance comparison of three techniques.

MOSTLY OPTIMAL MOSTLY SINR > 25 dB COMBINING MOSTLY INTERFERENCE REJECTION

o 2 4 6 8 10 12 14 16 18 20 L1C/l(dB)

Figure 13. PDF of AA <le/I performance

Software radio-based testbeds are a valuable tool in assessing the performance of smart antennas in

complex propagation environments. The two principal causes of capacity limitation and poor voice

quality, multipath and co-channel interference, can be statistically characterized with respect to relative

power, angle of arrival and time of arrival. Fully adaptive smart antenna algorithms which emphasize

constructive combining of the multipath, as opposed to either avoidance of multi path (space diversity), or

maximizing beam power (fixed switched beams), provide superior ell margin increases.

22

References

[1] 1. Ziskind and M. Wax, "Maximum likelihood localization of multiple sources by alternating

projection," IEEE Trans. on Acoustics, Speech, and Signal Processing, vol. 36, no. 10, pp. 1553-1560,

Oct. 1988.

[2] William C. Y. Lee, Mobile Cellular Telecommunications, Second Edition, New York: McGraw-Hill,

1995.

[3] S.W. Ellingson and J.P. Kennedy, "Design and Testing of an Adaptive Array for Analog Cellular",

Proc. AMTA 17th Meeting and Symposium, Williamsburg, VA, Nov. 1995.

3 A Practical Perspective on

Cellular Digital Packet Data (CDPD) Systems

Abstract

Nadine L. Romanowich

AT&T Wireless Services

15 East Midland Avenue

Paramus, NJ 07652

Early Cellular Digital Packet Data (CDPD) deployments have been a valuable learning experience for

the cellular carriers that established them. The purpose of this paper. is to discuss the practical aspects

ofCDPD systems, based on such experience. Guidelinesfor the planning, deployment and management

of cellular d{Jta networks are presented. This paper is intended for cellular service providers but may

also benefit application developers who wish to understand the issues faced by carriers with respect to

wireless data.

The specific topics covered in this paper include deployment strategies, frequency planning and

customer engineering. The conclusion emphasizes the importance of a well-developed strategy. coupled

with the adaptability to meet the changing needs of an evolving customer base.

1.0 Introduction

There has been controversy surrounding CDPD ever since the first specification document was released

in June of 1993. From the beginning, many knew that cellular packet data would be a significant

contribution to the wireless data industry. The justification was its open architecture, low cost and rapid

nationwide availability. Cellular carriers pledged their commitment to CDPD, with roll-out schedules

that pleased the waiting public. Early expectations for this new technology were extremely high. As

time went on, it became clear that availability would take longer than expected. The delays were often

interpreted as a sign the technology was in trouble. This lead t~ general confusion regarding the viability

of cellular data services.

J. H. Reed et al. (eds.), Wireless Personal Communications

© Springer Science+Business Media New York 1997

24

The deployment aspect of CDPD services has often been oversimplified and misunderstood. It is

important to recognize the full scope of responsibilities associated with entering the cellular data market.

CDPD is not just a feature of the cellular system, but a whole new business. Carriers must do more than

add equipment to each site, they must adjust to the dual role of phone company and data network service

provider. Market pressure and hype motivated carriers to attempt unrealistic deployment schedules,

resulting in certain delays. Despite many challenges, cellular carriers have made excellent progress

regarding CDPD services. The following document provides insight to the deployment process.

2.0 Deployment Strategies

There are two opposing schools of thought when building a CDPD network. One approach is to proceed

with caution, build a limited system and expand as required when revenue comes in. The other approach

is to deploy aggressively, invest a lot and work towards creating the market. This is the first decision a

cellular carrier must make for their CDPD system.

2.1 The Turtle vs. the Hare

The conservative deployment plan is often considered to be the sensible choice. It follows the classical

model for cellular system design, in which expansion is driven by demand. Voice systems today are

engineered with respect to collected statistics and predicted traffic values. To build a system with

capacity that far exceeds expected traffic, would have a negative impact on profitability. The same logic

is often applied to CDPD, because experienced cellular engineers use the concepts they are most familiar

with.

Systems that have been implemented by this approach have specific characteristics. Either deployment

is limited to a densely populated core area, such as a large city, or the entire footprint is serviced by a

subset of sites that are strategically spaced throughout the system. The core area deployment represented

in Figure I works well, but soon leads to coverage issues for early users. The deployment in Figure 2

shows CDPD sites spaced far apart, which must be powered higher than voice in order to cover nearby

cells without CDPD equipment. These "boomer" sites require special RF planning and dedicated

channels. Many cellular providers will use a combination of the two systems shown.

Figure 1. Core Sites Deployment Figure 2. "Boomer" Sites Deployment

• voioe siteS' wI CDPD o voice sites wlo CDPD

25

The benefits of a conservative deployment are lower costs because fewer sites are requ:red, and rapid

deployment which is a competitive advantage. However, these systems have increased engineering

costs, limited flexibility and less redundancy. Since the advantages are immediate and the drawbacks are

long term, this type of system should only be considered as an interim phase of deployment.

2.2 If You Build It, They Will Come

Although many cellular engineering teams tend to model CDPD after the classical voice system, it is

important to consider differences between the two. The data system should be recognized as a separate

entity with distinct engineering requirements. The wireless data has not yet evolved to the maturity of

cellular voice, therefore the expansion by demand rule does not necessarily apply. As a start up business,

it is understandable that deployment costs will far exceed revenue income. What if the marketing

department chose to be conservative, and were to advertise only after there was enough revenue to

support marketing expenses? Certainly, aggressive and strategic marketing can influence the growth of

revenue income. Likewise, the proper deployment of wireless data services can help to create the

market. The cellular data infrastructure is the foundation on which the business is built. The best data

system from a technical point of view, is that which mirrors the existing voice system. Therefore, the

deployment of CDPD equipment at all sites in the system can be considered a strategic investment.

The complete deployment of CDPD has many advantages with respect to system quality. More sites

equate to better coverage and more capacity. System reliability also improves, with overlapping cell

coverage. If one site fails, neighboring cells can provide service until repairs are made. Redundancy is

26

critical when working with a new technology, due to the elevated failure rates and limited fault

management tools. Implementing CDPD at every site also means that the data system can adopt the RF

design of cellular voice. While this reduces the effort associated with RF engineering, the Field

Operations teams must work harder to install and maintain additional sites, Another drawback, is that

system wide service takes longer to achieve. In order to satisfy early users during phases of partial

coverage, customer locations must be given priority. The engineering team must partner with the sales

team to anticipate coverage needs.

2.3 Consistency

Cellular service providers are likely to evaluate all strategies during the planning 'stage of CDPD

deployment. The complete approach may not be defined from the beginning, but instead evolve as the

buildout is in progress. No matter how a carrier arrives at their strategy, it is essential that consistency is

maintained in all elements of the business. For example, if the plan is to engineer and deploy

aggressively, the sales organization should pursue customers in the same manner. This balance is further

required within the technical organization itself. Although it is tempting to apply all resources toward

the implementation of new sites, equally important is the infrastructure to manage and support those

sites. Each level can be broken down yet again to compare hardware versus software, etc.

Consistency and balance are an obvious need for any business, but attaining them is especially difficult

for a new player in the cellular industry. Excessive growth and lack of resources prohibit one team from

keeping up with the others quite often. The struggle to keep consistency is further complicated when

goals are not carefully chosen. Setting first year goals that only pertain to the number of sites deployed,

places less emphasis on the quality of those sites. Second year goals that only pertain to site stability

will subsequently displace efforts elsewhere. Anyone managing CDPD roll-out should be aware of the

importance of consistency and the trade-off's associated with it. Both the strategy selected and the

manner in which it is applied, influence decisions made further down the line.

3.0 Frequency Planning

The basic idea behind CDPD is to send packet data on unused analog voice channels of an existing

cellular network. The unused channel can either be hopped for short intervals between voice calls or be

27

dedicated to data service full time. Discussion of CDPD often leads to the hopping vs. dedicated channel

debate. Although this topic can be approached from many different angles, most arguments tend to be

more philosophical than practical. The real decision concerning a frequency plan is based on the

characteristics of a specific cellular system and has little to do with general theory. This section will

identifY the limitations of economical analysis and focus on the technical aspects that influence

frequency planning.

3.1 Limitations of Economical Analysis

The primary argument in support of the hopping channel scheme, addresses the cost associated with

eliminating voice channels. This argument has some weaknesses due to the subjectivity of its supporting

economical analysis.

This cost of dedicating channels to CDPD is usually calculated by multiplying the voice revenue

associated with each channel during busy hours, for every sector in the cellular system. For example, a

cellular system consisting of 100 sites with 3 sectors per site will dedicate one channel per sector to

CDPD services. Each channel is responsible for approximately $ .50 in revenue per minute of use, for IS

minutes of each busy hour. There are two busy hours per work day and 250 work days a year. The

calculated impact on voice revenue is:

Impact on Voice Revenue 100 x 3 x I x $.50 x IS x 2 x 250 $ 1,125,000 per year (1)

There are too many assumptions made in this calculation. Realistically, only a subset of sites will need

to take a voice channel out if service. The possibility that CDPD will generate revenue has also been

overlooked. Finally, there are additional economic factors associated with the blocking of both voice

and CDPD that are not easily quantified, yet need to be considered.

The percentage of sites required to remove voice channels from service, will vary for each cellular

system with respect to frequency group utilization. In a standard system, each sector is assigned a

frequency group. A frequency group is a subset of the cellular band, which has a limited number of

voice channels. On average, there-~re 17 analog voice channels associated with each group. Because

voice traffic is not uniformly distributed throughout the system, some sectors will partially utilize the

channel set while others need to exceed it. Sites with channels available in their frequency groups have

room for future growth. In this case, no voice channels are removed from service in order to support

28

dedicated channel CDPD. Sites that are unable to support existing traffic while using the complete

frequency group, experience voice call blocking. In this case, a voice channel must be taken out of

service for dedicated channel CDPD. Excessive blocking also indicates that an additional cell site must

be built in that area. If blocking is foreseen, plans have already been made for the new site. There is

typically an interim period of blocking before the new site is available to offload traffic. After the new

site is on air, it is possible that there will be excess channels until traffic grows beyond capacity once

again. When zoning requirements delay a new site, call blocking becomes difficult to manage. Isolated

events such as a major sports championship or a visit by the President also contribute the call blocking.

Many factors influence channel capacity which will vary by both time and geographic area.

If channels are available throughout seventy-five percent (75%) of the system in this example, then only

the remaining sites will sacrifice capacity and voice revenue for CDPD. Therefore, the economic model

should include the following:

Impact on Voice Revenue 100 x 3 x 1 x $.50 x 15 x 2 x 250 x.lS = $ 281,250 per year (2)

As shown here, there is a great difference between the result of this formula and the first. Some cellular

systems are also confined to a partial frequency group by design. In this case, the only option may be to

redesign the existing voice frequency plan to accommodate dedicated channel. The cost then will be

associated to reengineering the system rather than removing voice channels from service. Each cellular

provider will vary significantly with respect to frequency group utilization, therefore, the cost associated

with dedicating channels to CDPD will also vary.

The term "investment" is more appropriate than "cost" in this discussion because revenue is not actually

lost, it is deferred. CDPD is capable of generating revenue with these channels. In the short term, data

revenue may not offset that which would have been from voice. However, the potential revenue of

CDPD is greater than analog voice. Even at low utilization levels, a CDPD channel can generate a few

dollars per minute of use. Unlike analog voice technology, CDPD is not subject to fraud.

3.2 Hopping Channels

The ability to borrow channels from the cellular voice system is considered by many to be the most

attractive feature of CDPD. It is wonderful that data systems can use idle voice channels, without

29

restricting call capacity. CDPD can be implemented even at busy sites and service is good except when

voice activity is at its peak. Another advantage of hopping is that channel assignments are already taken

care of. By using the same channels, the cellular data system can benefit from the thought and effort that

went into frequency planning the voice system. With channel hopping, there is also less chance of co­

channel interference because channels are distributed in a somewhat random fashion.

Despite the advantages, a hopping scheme will not be practical in most environments. It is unnecessary

to hop channels at a given site if the frequency group is not fully utilized. In addition, hopping cannot be

used for a "boomer" site deployment because it restricts CDPD to the transmit power level of voice

service. High blocking rates also create problems, due to the unpredictable availability of voice

channels. The only environment that is suitable for hopping has very specific characteristics. Call

blocking is neither too low or too high, and coverage outside the site voice footprint is not required.

Even if all these conditions are met, there remains the operational challenge of coordinating all RF

changes at the site for both voice and data. Interference is inevitable when either service deviates from

the shared RF design.

3.3 Dedicated Channels

From the CDPD perspective, dedicated channels are best. This scheme allows for the highest throughput

rates. Using dedicated channels also helps to isolate CDPD services from our busy voice network,

thereby ensuring the integrity of each system. Furthermore, data coverage is not limited to the voice

footprint. RF engineering can compensate for sites without CDPD equipment. Coverage flexibility

helps to accommodate data users with unique coverage needs. Another important benefit is that, users

have less to worry about. Many customers are concerned about the availability of CDPD when voice is

very busy. Many perceive hopping as unreliable, even when the probability of blocking is low.

The biggest disadvantage of dedicated channels is the impact on analog voice capacity. A lot of effort is

required to migrate voice service off the channels reserved for CDPD, without increasing call blockage.

There is also greater potential for CDPD co-channel interference, although this particular problem can be

managed.

30

Table 1. Comparison of Frequency Plans

Features of the Dedicated Channel Scheme Features oft~e Channel Hopping Scheme "',,, "<,:'

+ High data throughput + No impact on analog voice capacity

+ Isolation from analog voice + Channel assignments are easy

+ Allows for RF overlays (for sites wlo CDPD) + Minimal CDPD co-channel interference

- Impacts analog voice capacity - Unpredictable availability

- Effort required to free up analog channels - Reduces overall data throughput

- Potential for CDPD co-channel interference - Potential for interference with AMPS

3.4 The Hybrid Channel Scheme

A cellular carrier may decide to implement a hybrid of dedicated and hopping channels. Carefully

implemented, the hybrid scheme with a strong bias toward dedicated channels is very practical. While

the majority of sites should use dedicated channels, hopping mode can be used to temporarily alleviate

call blocking. This is a great solution for areas that need a new cell site. Hopping can provide relief to

the voice system in the interim period while waiting for the new site to be built. Used in this fashion, the

list of sites that hop will change over time. Carriers must effectively manage the transition from one

mode to another so that all related requirements are met. Typical· changes include software

configurations and transmit power level adjustments. Although additional effort is required to maintain a

hybrid system, the carrier will benefit from increased channel efficiency.

4.0 Customer Engineering

Customer engineering plays a critical role in the evolution of wireless data services. The CDPD

customer base though initially small, is representative of the larger untapped wireless data market. To be

successful, cellular carriers must gain an understanding of what these data users want. In return,

customers rely on the carriers for guidance to achieve maximum benefit from wireless data services.

CDPD customers are more technical than the average cellular customer and have additional needs.

Although many potential users recognize the value of wireless data, few are prepared to face the

challenge without assistance. Beyond coverage, they require hardware, software, training and integration

31

support. Even existing applications that run on wire line networks, must be modified to operate over the

wireless network. Cellular application engineers team up with the sales organization to provide

customers with a complete wireless data solution.

The customer-engineering relationship is important in other ways. To best use deployment resources,

cellular carriers prioritize system growth around the customer. The carrier must work closely with users

to anticipate coverage needs which often change over time. Information concerning customer location,

application type, data throughput requirements and traffic patterns is tracked with great detail. This data

will often influence the deployment, frequency plan and optimization schedules.

5.0 Conclusion

Cellular carriers have a responsibility to wireless data users and the industry, to provide quality CDPD

services. Delays in service availability often overshadow the accomplishments of these carriers. Well

developed strategIes have helped them to overcome many technical and operational challenges.

Experience with the existing voice systems and customer needs also contribute to quality of the data

system. Early customers are just starting to realize that CDPD was worth waiting for.

Acknowledgments

The content of this paper is largely derived from the concerted efforts of AT&T Wireless Data Division

and AT&T Wireless Services of New York, Engineering & Network Operations teams. Special thanks

to Jeff Baker for assistance with the maps and Ron Comella for the editing.

References

[1] Cellular Digital Packet Data System Specification, Release 1.1, January 1995.

[2] William C. Y. Lee, Mobile Cellular Telecommunications Systems, McGraw-Hili, New York, 1989.

[3] Nathan J. Muller, Wireless Data Networking;, Archtech House, 1995.

Abstract

4 Low Cost BPSK 2.4 GHz Radio For Wireless LAN

William O. Camp, Jr., Arthur Fleek, William Nunnery, Ralph Yeager, Michael Bracco

IBM Network Hardware Division Research Triangle Park, NC 27709

We describe a radio architecture for the IBM Wireless LAN Entry commercial product that

achieves low cost and small size by using direct modulation of the master oscillator signal by a

vector modulator at the transmittedfrequency, and by using a single downconversion to a

relatively.!ow IF frequency in the receiver. The transmitter requires only filtering at its output

to control harmonic radiation. Spectral control is accomplished by filtering the baseband data

signal to the vector modulator ports. The amplified and limited IF signalfrom the single

down conversion is processed digitally for carrier sense, frequency compensation and

demodulation. We will describe how this is done with, predominantly, high resolution counting

techniques. The receiver passband is determined with an IF filter consisting of lumped Land C

elements. The modulation is aform ofBPSK with unusual modifications to meet the FCC

requirements for the ISM band and still meet the requirements of the simple digital

demodulator. We will describe the limits on input signal to noise ratio of these techniques. The

data rate is 0.5 Mb/sec. This simple architecture along with only one bandpass filter in the

radio, makes for the simplest possible half duplex transceiver, and readily fits into a Type II

PCMCIA package with extension for integral antenna, using only off-the-shelf standard (and

relatively large) components.

J. H. Reed et al. (eds.), Wireless Personal Communications

© Springer Science+Business Media New York 1997

34

Introduction

This paper will show how we achieved a low cost wireless LAN with 0.5 Mb/s airlink data rate in the

2.4 GHz ISM band. The driving factors for this project were:

• Low cost architecture

• FCC part 15 requirements for ISM band compliant

• PCMCIA Type II format with smallest possible extension

• Built-in antenna.

We postulated that the simplest possible architecture would revolve around a transmitter with direct

modulation of the carrier; i.e., a transmitter with no upconversion. This would present the need for the

least amount of filtering at the microwave frequencies, eliminate mixers, and spurious products associated

with them. Meeting FCC requirements would be easier. We also postulated that the receiver could be

made simple by using a single down conversion to a low to moderate Intermediate Frequency which could

be processed directly with a digital ASIC. Here too, the filtering would be easy to achieve with normal

inductors and capacitors, avoiding expensive and large SAW filters.

The choice of modulation was then influenced by these prior assumptions. All modulations are basically

forms of amplitude and phase variation of the carrier, such as FSK (Frequency Shift Keying), PSK

(Phase Shift Keying), etc. They can all be generated by using a vector modulator between an oscillator at

the carrier frequency and the antenna. Because BPSK (Binary Phase Shift Key) modulations can be

generated with simple baseband waveforms on a variety of phase modulators we focused on them first.

As will be seen later, this type of modulation also presented itself as a particularly easy form of

modulation to decode with digital circuits. The data rate finally chosen was 0.5 megabits per second, and

the particular amount of phase shift used was 135 degrees.

With the parameters selected, we were able to:

• Meet the FCC regulations for occupied bandwidth,

• Reduce the amplitude variation and successfully use a nonlinear power amplifier,

• Provide an easily detectable amount of variation in received IF period.

Figure 1 shows the block diagram of the RF microwave and digital functions of this radio. In the

RF microwave section, a single VCO (\I oltage Controlled Oscilhtor) is shared between receiver and

transmitter, being shifted in frequency by the receiver IF frequency when going between receive and

transmit The IF passband is set by LC lowpass and highpass filtering. The IF frequency is chosen to be

2 Mhz for easy processing in the custom digital logic. The IF signal is amplified, limited and put through

a comparator to create a I bit signal (shown in Figure 5). Transition time spacings of this limited IF

signal are measured with digital logic driven with a 54 MHz clock Additional logic, not shown in the

Figure 1, is used to:

• Control the frequency hopping and coordination

• Package the data into frames

• Generate and check CRC

• Send the data through the PCMCIA interface

• Buffer data in memory

• Store constants in non-volatile memory

• MAC and collision avoidance

Figure I -- Block diagram ofRF microwave and digital functions

Vector Modulator

Data Out

Clock Out

Carrier Sense

Data In

35

36

Modulation

The particular fonn of BPSK we are using is non-standard in that the phase shift is 135 degrees rather

than the normal 180 degrees. The original intention of this was to select a sufficiently small phase shift to

drive the spectrum sidelobes down to below -20 dBc at the 1 MHz band edges. This was possible to do

with 90 degrees phase shift, and the first modulator used a delay line between 2 RF switches. However,

this precluded any premodulation filtering, and made the total transmitted spectrum was unnecessarily

wide.

When a 2.4 GHz vector modulator came available during the development phase, we used it in place of

the delay line modulator, and, therefore, could readily add premodulation filtering. We found that we

could then increase the phase shift to 135 degrees and still meet the FCC occupied bandwidth

requirements. This constellation is shown in Figure 2. It is particularly easy to generate, in that the

modulation signal is impressed on the I input to the vector modulator and a constant DC bias is impressed

on the Q input.

Figure 2 - Modified BPSK Modulation Constellation with 135 degree phase shift

Q

_..--_____ ./ Traeoctory after Saturatil1c Amplifier

o Traeo<tory _ ~--il---~ after Vector Modulator ;;t

There were two further restrictions that existed on this modulation. First, the demodulation scheme we

were using limited the amount of lowpass filtering to that which would not make the transition time of the

modulation wavefonn longer than one cycle of the receiver IF frequency. This was necessary in order to

force all of the phase shift for the symbol transition into one cycle ofIF. Second, the first sidelobe

was very close to the FCC limit of -20 dBc, and required additional filtering without violating the

previously mentioned limit on the lowpass cutoff frequency. This was accomplished by the brute force,

but expedient method of putting a controlled Q-factor notch at 750 kHz on the I port input to the vector

modulator. The controlled Q had the effect of reducing that sidelobe by only 4 dB, and minimally

degrading the phase waveform. The resulting filtered modulation waveform to the vector modulator is

shown in Figure 3. The spectrum of this baseband signal is shown in Figure 4a. The spectrum after

limiting amplification in the severely overdriven power amplifier is shown in Figure 4b. Note the small

amount of spectral regrowth that has occurred, because the phase shift has been limited to 135 degrees.

V 1 o 0 o

TIME

Figure 3 - Filtered Modulation Waveform at I port input to Vector Modulator

dBc

o

-20

-40

-60

-2.0 -1.0 0 1.0 2.0 MHz

2336 2440 MHz

2444

Figure 4a - Baseband Spectrum Figure 4b - Passband RF Spectrum after Saturating Amplifier

37

38

One further precaution had to be taken for this type of direct modulation. Since the strong output signal

from the transmitter is at the same frequency as the yeO, feedback can cause uncontrollable phase

variations in the yeO phase, and therefore, output phase. Thus, all the yeO ports must be well shielded.

This is done by using an attenuator on the output port and by shielding the tuning line.

Demodulation

The demodulation for this radio is an attempt to break away from the viewpoint that phase modulation

requires a complicated demodulator as well as a strictly linear system. We have already seen that the

transmitter can be nonlinear without severe spectral degradation. We note that the same can be done in

the receiver. If the spectrum remains unchanged after the nonlinear amplification of the transmitter, we

reasoned that the receiver could also use a limiting IF amplifier, eliminating the need for AGe throughout

the receiver chain. We further reasoned that the demodulation process could occur in digital eMOS logic

if the IF frequency was kept sufficiently low. Since the IF was being limited, any more resolution than 1

bit would be unneeded as well.

The choice ofIF frequency was critical. Since we would be measuring phase shift by measuring the time

between the edge transitions of the IF signal, we reasoned that there should be a minimum amount of

perturbation due to the beat note between the data rate and the IF frequency. In other words, the IF had to

be a harmonic of the data rate. We chose 4 times the data rate, or 2 Mhz, as it was low enough to

measure accurately with a 54 MHz clock we had available, and yet high enough to avoid the significant

interference from spectral spillover from the negative frequencies of the downconversion. A typical IF

waveform with modulation is shown in Figure 5. Note the elongation in IF period that occurs for 1 to 0

data transitions, and the shortening in IF period for 0 to 1 data transitions. This stretching and shortening

becomes the basis for our demodulator-- we simply measure the time between similar polarity transitions

and test whether that time is more than or less than the normal time for a 2 MHz squarewave.

1 DATA

o

IF

o

TIME

___ -- Time between __ ------- rising edges

TIME

Figure 5 - IF Wavefonn and time bars showing delta time between rising edges ofwavefonn. The

resulting data output is shown in the upper graph.

As straight forward as this method of demodulation might seem, there is a complication-- the oscillators

of two radios will not be synchronized, and therefore, the receiver IF frequency will not be exactly 4.0

times the data rate. This means that periodically, depending on the deviation from exact integer

relationship, either a transition will be lost or an extra transition will be added to account for the phase

slip occuring at the beat frequency between the two rates. We overcame this problem by simply

measuring the time between negative transitions AND positive transitions and OR'ing the results of the

two measurements. The histograms in Figure 6 show the results of this measurement of the IF wavefonn

for 3 power levels, representing strong, weak, and marginal input signal level conditions. The top graphs

show the time counts (in 20 nanosecond bins) for the periods when no data transitions occur. The bottom

graphs show the time counts seen when a data transition occurs. There are 354 data transitions is this

data. Only the largest or smallest time measurement (as appropriate for the polarity of the transition) is

used for each transition. As expected, the histograms become dispersed as the signal to noise ratio

decreases. For proper operation, these 3 regions cannot overlap, or an error will occur. From just such

histrograms we chose thresholds of 574 nanoseconds and 426 nanoseconds for the respective transition

criteria. (These are the actual numbers used in the product because of the 54 MHz clock; the data in the

figure is measured with a 50 Megasample per second oscilloscope.)

39

40

2000 Count

1000

100 Count

50

-25dBrn

300 400 500 600 700 lIS

300 400 500 600 700 lIS

2000 Count

1000

100 Count

50

-75dBrn

300 400 500 600 700 lIS

300 400 500 600 700 lIS

2000 Count 1000

100 Count

50

o

-85dBrn

300 400 500 600 700 lIS

300 400 500 600 700 lIS

Figure 6 - Histograms showing nontransition and transition time measurement occurrence counts for

3 different receiver input signal levels (top of each graphs).

Further embellishments on this digital demodulation method were required to make it robust. Transition

times less than 300 nanoseconds are not counted as valid, and symbol transitions are not permitted for

two IF periods after the detection of a transition. Similarly, like polarity transitions in succession are

blocked. A simple bit stuffmg algorithm that limited a run ofO's or l's to less than 6 was used to

guarantee a transition often enough to be used for clock recovery, and served as a reliable end of frame

delimiter.

Because of the variations in frequency in the crystal oscillator clocks, there will be frequency differences

between radios up to 6% of the 2 MHz IF frequency. In this type of digital demodulator, it was

particularly easy to compensate for this frequency difference. The net effect of a shift in IF will be that

the histograms shown in Figure 6 will shift, en masse, up or down as the IF frequency goes down or up.

We have built into the MAC an unmodulated period at the beginning of the transmitted burst, during

which we measure the IF period. We do this over 64 IF periods in order to obtain an accurate value, and

use the difference in time relative to 500 nanoseconds to directly shift the transition thresholds for

determining whether a transition has occurred or not. With this method, we accomplish an automatic

frequency compensation (AFC) of the demodulation process.

Carrier sense is a key function for the CSMAlCA (Carrier Sense Multiple Access! Collision A voidance)

MAC (Media Access Control) that we use. When looking for the presence ofa carrier, we use the period

time measurements to detennine whether the IF signal is a coherent signal or noise. A series of 64 period

time measurements is tested for unifonnity, and deemed a signal if within a set threshold. During a

transmission, the time periods between transitions are tested for a similar criterion.

This method of carrier detection can be quite sensitive. A carrier can be detected at a signal level far

below which it can be successfully decoded. One can appreciate that the hidden node problem can be

greatly diminished with this extra sensitivity carrier sense. However, a disadvantage of this extra carrier

sense sensitivity was to cause the radio to respond to the signal before the VCO frequency has stabilized.

To counter this problem, a spoiler modulation of single transitions every bit time was added at the start­

up of each transmission, then stopped for the frequency compensation period.

The one adverse aspect of this simple demodulation scheme is that the IF bandwidth, set by a lowpass

filter and a highpass filter, must be kept sufficiently wide in order that the phase transition rate is

preserved. For this set of parameters we kept the lowpass filter cutoff at 3.5 MHz. This does increase the

noise bandwidth over that which is needed for the data rate and a more standard demodulator.

Performance

The perfonnance for this radio design is shown in Figure 7. The packet error rate versus input signal level

at the receiver antenna port is shown in the frrst graph. The packet length is 64 bytes. From this data and

knowing the packet length we get the Bit Error Rate (BER) versus receiver input signal level shown in the

second graph. This data is shown for the transmitter at a fixed frequency (2440 Mhz), and for the

transmitter hopping as it would in actual usage. The typical benchmark is BER = I x 10.5, and the

perfonnance for our receiver at this value is -80 dBm receiver sensitivity.

41

42

Packet Enor Rate

10-1 • 10-2

+ • +

10-3 • +

10-4 • 10-5

+

Bit Enor Rate

-2 10

* 10-4

-6 10

10-8

• Hoppizlg

+ Not Hoppizlg

• + • + • + •

+

-85 -83 -81 -79 -77 -85 -83 -81 -79 -77 Received Signal Level (dBm) Received Signal Level (dBm)

Figure 7 - Packet Error Rate and BER versus Received Input Signal Level (dBm) for hopping and non­

hopping cases.

The expected performance can be estimated in the following manner. The formula for MDS is:

Minimum Detectable Signal =-174 dBm + BW + NF + (SIN)input [IJ

The receiver noise figure is 9 dB. The noise bandwidth is 8 MHz (N.B. that the image doubles the noise

bandwidth.) For the non-hopping measurement, I x 10.5 BER is obtained at -S2 dBm. From equation [I]:

-82 dBm = -174 dBm + 69 dBHz + 9 dB + (SIN)input [2J

or the input SIN = 14 dB measured. Since ordinary Differential BPSK requires an input SIN = 11 dB, the

reduced phase shift (135 versus ISO degrees) and digital time resolution have resulted in an

implementation loss of3 dB. The extra 1 dB loss when hopping represents NF variation across the band.

With + IS dBm output power from the transmitter, we reliably achieve 150 feet in building distances with

this level of performance. Other pertinent numbers are: 5 volt operation, 1.2 watts receive, 1. 7 watts

transmit input power; the size in PCMCIA form factor with an extension containing a built-in shortened

monopole antenna.

CONCLUSIONS

We have demonstrated a very small and low cost radio in the 2.4 GHz band that is suitable for a

wireless local area network. Its PCMCIA Type IT extended form factor with integral antenna is

very convenient for use in Laptop PCs. The receiver demonstrates state of the art sensitivity,

achieved through the use of a modified BPSK modulation. The transmitter is simplified through

the use of direct phase modulation of the carrier frequency. The in building range is consistently

150 feet. The cost has been kept very low by shifting all of the modulation and demodulation

function into the ASIC along with other digital functions associated with the MAC (Media

Access Control). This shift to digital function was enabled by use of new approaches to the

demodulation, carrier detection, and automatic frequency compensation.

43

5 Error floor of 1d4-DQPSK in mobile radio channels

Andreas F. Molisch and Ernst Bonek

Institut fUr Nachrichtentechnik und Hochfrequenztechnik, Technische Universitat Wien GuBhausstralle 25/389. A-I 040 Vienna. Austria

FAX: +43 I 587 05 83 EMAIL: amolisch@et.tuwien.ac.at

Abstract

We analyze the error floor of rrl4-DQPS K in mobile radio channels due to the intersymbol intetjerence

caused by the time dispersion of the channel and compare it to results for GM SK. We first analyze a two­

delay channel with fixed (but arbitrary) sampling instant;foroptimum sampling, the error probability BER

depends only on the mean rms delay spread S normalized to the symbol length T: BER=k·(S/T;Z, where

the proponionality constant k depends on the filtering. For two-branch selection diversity, the BER goes

with (S/T)4, and is aboat a factor 3 better for BER-driven diversity than for RSSI-driven diversity. In a

next step, we analyze the case at adaptive sampling, and find that the error probability can be reduced

as compared to fixed sampling. However, in contrast to MSK, it is not possible to completely eliminate

the error floor even for "basic" rrl4-DQPSK. Finally, we analyze the N-path channel with fixed sampling

tim e. We find that here, the B ER depends /lot only on the delay spread, but also on the shape of the delay

power profile.

1. Introduction

Error mechanisms in digital mobile communications systems, apart from thennal noise and co­

channel interference, are random FM [1] and intersymbol interference (lSI) due to the time dispersion

of the channel [2]. The latter two mechanisms cannot be eliminated by simply increasing transmitter

power, and are thus often called "irreducible errors" or "error floor". In the high-data rate systems that

are common today, the random FM becomes negligible, and the error floor is dominated by the lSI [3];

we will henceforth denote only these types of errors as "error floor". The computation of the error floor

is of special interest for cordless systems: there, the noise and co-channel interference are low, and

equalizers are usually avoided to decrease costs. In that case, the perfonnace is detennined by the error

floor.

The two modulation fonnats that are the most important for mobile radio applications are

(G)MSK, which is used in the GSM and DECT systems, and 1tI4-DQPSK, which is used in the DAMPS

J. H. Reed et al. (eds.), Wireless Personal Communications

© Springer Science+Business Media New York 1997

46

and ADC systems [4]. The enm floor of nI4-DQPSK has been investigated both by numerical and

analytical techniques. Computer simulations (e.g. [5]) have provided valuable results, but require a large

amount of CPU time and it is difficult to draw general conclusions. The elegant analytical investigation

by Liu and Feher [6] includes noise and co-channel interference, but is restricted to two-delay channels

with fixed sampling on the first arriving path and allows no line-of-sight component. Furthermore, it

models the lSI only in an approximate way. Other theoretical investigations have similar restrictive

assumptions. It would thus be desirable to have a more general analytical formulation. In a recent series

of papers [7], [8], [9], [10], we have developed new mathematical methods for the computation of the

error floor of (G)MSK. In this paper, we show how to modify these methods to 1tI4-DQPSK, and give

results and comparisons between 1tI4-DQPSK and (G)MSK.

In a first step, we analyze the error-floor in a two-delay channel with arbitrary fixed sampling time

and arbitrary amplitude statistics for each path. We identify the channel constellations that lead to errors

and find that there are certain "enm regions" in the complex phasor plane. We then show that the error

probability is proportional to k·(S/T)2, where S is the delay spread, T is the symbol duration, and the

proportionality constant k is on the order of unity. We also analyze selection diversity, and show that it

is quite effective for combatting the error floor.

·In a second step, we generalize the concept of the "error regions" to deal with adaptive sampling.

In that case; the evaluation of the error probability must be done semi-numerically. However, the

numerical efforts for this method is orders of magnitude smaller than for MC simulations. Using this

method, we show that in contrast to MSK, the use of adaptive sampling leads only to a small reduction

of the error probability.

In a third step, we deal with an N-delay fading model, which is a more realistic model for actual

mobile radio channels. For fixed sum piing, the problem can be reduced exactly to an equivalent two-delay

problem. We give comparisons between GMSK and 1t/4-DQPSK for all three models.

2. Error floor in the two-delay channel

The structure that we analyze is sketched in Fig. I. Let us first consider the transmitter. A data

source generates binary data, which are then grouped into two-bit symbols by a serial/parallel (SIP)

converter. The resulting symbol sequence is then sent into a nI4-DQPSK modulator. The symbol

constellation of 1t/4-DQPSK is shown in Fig. 2. There are two sets of symbols, one comprises the phases

n·1tI2, n=O, I ,2,3, and the other set comprises the phases 1tI4+n·1tI2. Two subsequent symbols are always

from different sets. If the symbol is 00, then the phase is increased from the current value by 1tI4, for 01,

it is increased by 31t/4, for 11, by 51t14, and for 10 by 71t14 (Gray coding). The resulting signal is then

47

a) Transmitter

a2(t)

1

b) Channel

~"--:"-+-K J====~ ~ampllng -Sin O)ot at T s + kT

c} Receiver

Fig. 1 Block diagram of the system.

sent through a low-pass filter LPF with impulse response hT(t). This signal is then mixed with a local

oscillator, so that is shifted into the RF range. Mathematically, we can write the transmitted signal as

L cos( Iilot + () .. ) h.f.,t-m 1) (1) .. The model for the channel is sketched in Fig. 1 b. The two-delay fading channel is a very popular and

widely-used model for the mobile radio channel (see, e.g., Refs. [II], [12) and references therein). It gives

the most important features of the channel (fading, time dispersion) but is still reasonably simple. The

impulse response h(t,'!) consists of two delta pulses, whose amplitudes a, and ~ and phases <1'1 and <1'2

are independent stochastic variables, so that the impulse response (in complex baseband representation)

is

48

Here, 1:1 and 1:2 are the delays of the first and second pulse; to

simplify notation, we hencefOlth set 1: I =0 (no restriction of

generality). The phases <PI and <Pl are uniformly distributed; the

statistics of the amplitudes are arbitrary. In most cases, a two-

delay Rayleigh-fading channel is assumed so that a l and az are

Ray leigh distributed with variances cr I and cr2. However, other

statistics, e.g. Rician or Nakagam i, are also of interest. In our

computations, we assume that the excess delay 1:2-1: I =1:2 is

smaller than the symbol length. Another assumption that we will

Fig. 2 Signal space representation of 7tl4-DQPSK.

use in the remainder of the paper is the slowly time-varying channel [13]. This means that the amplitudes

and phases of the impulse respunse remain unchanged for several bits. This is fulfilled in most practical

situations. Channel characteristics typically change within Al4, i.e. 3-8cm for 900-l800MHz systems. Even

if we move with 200km/h, we need on the order of I ms to cover such a distance, while symbol durations

are on the order of illS .

. The receiver is sketched in Fig. I c. The signal is shifted to baseband, low-pass filtered, and

sampled. We then multiply the sample at time k with the complex conjugate of the sample at time (k-I),

and see whether real and imaginaty parts are smaller and larger than zero. Appropriate combination of

the results of in-phase and quadrature-phase components give the output symbol.

3. Error floor in two-delay channels

In order to find the error probability, we compute the received signal for all transmitted symbol

combinations and channel constellations. We first define a function gl.2(t), which is the convolution of

the LPF impulse response hLPF(t) with delta pulses at 1:1 and 1:2,

(3)

and hLPF denotes the convolution of the impulse responses of the low-pass filters at the transmitter and

the receiver, hLPF(t)=hT(t)*hR(t). Typically, this convolution is a raised-cosine pulse, i.e.

sln( 1t!..) .. nJ 1t«!.. )) h (t)=_~_\T_.~

LPP' t ()2 1t"T 1-4«~

49

(4)

where the "roll-off" factor a detennines the width of the spectrum. The case of pure 7tl4-DQPSK is

recovered if hLPp(t) is a triangular pulse (i.e. tranSmit and receive filter each have a rectangular impulse

response).

The received signal can be written in tenns of ai' "2, simp, cosq>, and ck' where q>=q>z-q>l' and the

constants c are given as

(5)

The variable that determines the decision for the in-phase component is (see Fig. Ic)

(6)

which after some manipulations becomes

2 U = Q 1 [C I Cs +C2 C6] +QI azCOS'P[CI C7 +C2 Cs + C3 CS +C4 C6 ] (7)

Q1 Q2 Sin'P[ -C1 Cs +C2 C7 + C3 C6 -C4CS] +azlC3 C7 +C4 Cs]

For the quadrature-phase component, we need the variable

(8)

which can be shown to be

2 V= Q 1 [C 1 C6 -C2 CS] +QI azCOS'P[CI Cs - C4 Cs +C3 C6 - C2 C7 ]

(9)

Q 1 azSin<P[CI C7 -C3 CS +C2 Cs -C4 C6] +azlC3Cs -C4 C7]

Equations (7) and (9) are the set of equations that determines which symbol is actually chosen

at the receiver. Thus, we can use these equations to determ ine the error probability. For this, we use the

"error-region" method, which is especially suited for the two-path model (of course, evaluation of these

50

equations by a Monte Carlo method is equally possible, but requires more computer time}. We first divide

Eq. (7) by a12, and define r=~/al' It then becomes

(10)

where the dj are the terms in brackets in Eq. (7). The variables r and q> can be interpreted as the polar

coordinates in a "normalized phasor plane": each realization of the channel corresponds to one point in

this plane. Errors occur if the phasor falls into certain regions of the phasor plane, the so~alled error

regions. Obviously, these regions are bounded by the condition U(r,q»=O, since this is exactly the case

when our decision at the receiver switches. Comparing Eq. (10) with the equation for a circle in polar

coord inates

(11)

we see that the error regions are circles, and can easily determine the origin and radius of this circle. The

error probability is then the probability that the normalized phasor falls into this circle, i.e. approximately

(12)

where p,(r} is the probability density function (pdf) of r=~/al' For Rayleigh variables,

(13)

where b=cr/cr2, and cr l and cr2 are the average amplitudes in the first and second paths, respectively. The

analoguous expression for Rician distribution of the first path and Ray leigh distribution of the second path

is given in [7].

The equations for the error circles must be evaluated for all possible symbol combinations; since

(for "basic" 1tI4-DQPSK and 0<T,<'t2) 4 symbols can influence one decision, we have to check 44=256

symbol combinations. Similarly, we determine the error regions for the quadrature components. Since we

assume Gray coding, double errors occur only of the phasor falls into the overlap region of error regions

for in-phase and quadrature-phase components; single errors occur if the phasor falls into only one error

region. We thus can just add up the errors from U and V.

The actual error probability depends on the filtering, i.e. on the roll-off factor of the raised~osine

pulse. This dependence can be seen in Fig. 3. Changing from a triangular hLPF(t} to a raised~osine

approximately triples the error probability. This is a stronger dependence than e.g. for GMSK with the

usual filter bandwidths (BT=0.3-0.5), which change the error probability by less than 50%). Another

interesting point is that for the "basic" 7tl4-DQPSK scheme (i.e. with a triangular hLPF) the error

51

10-1

/' I

"' 10-2 1/ triangular ,

>.Ll ~

E :E 10-3

"

raised ,cosine Ft= //

a 1.0 r+ ~ raised cosine

tz' __ ::-",0.35 ,

.n F' '0 - -0 .... 0.. / MSK I .... 0 t:: 10-· 1/ V V- I

,

<l) _.

[;Z // I \--'-

10-5 ~ , i 0.01 0.1 1.0

nonnalized excess delay ,iT Fig. 3 Error probability as a function of the nonnalized excess delay '2IT for various filters_ Channel is a two-delay Rayleigh-fading channel, cr ,=cr2.

lO-1

I

"' 10-2 >.Ll ~

~ :0 lO-l

" ,b=O

b--==2 j .n _. 0 .... 0.. .... 0 t:: lO-' Ff

, 1/ 1/ !

<l) Ie

b 3 17/ -- - -liHI- -- -- - --

10-5 V II II I I 0.01 0.1 1.0

normalized excess delay '2fT

Fig. 4 Error floor as a function of the nonnalized excess delay '2/T for various Rice factors; crl=crZ' hLPF(t) is triangular.

probability is about the same as for GMSK if we have the same bit rate in the two systems (i.e. the

symbol period in the lt/4-DPSK system is twice as long). The error floor for various Rician factors is

shown in Fig. 4.

The error region method also allows the analysis of diversity. For RSSI-driven selection diversity,

the BER is the probability that one nonnalized channel phasor falls into an error region and is

simultaneously larger than a second (statistically independent) phasor. For BER-driven diversity, errors

52

10-2 §;

10-3 I no diversity

~ ~

10-4 ~ ~

~ 10-5

~ .0 10-6 0 .... 0..

P? iiim m

1/ ~

17V ~

.... 0 10-7 .... .... " 10-8

/ BER-driven selection diversity

/

10-9 0.01 D.! 1.0

normalized excess delay 121T

Fig. 5 BER as a function of the nOlmalized excess delay 't/f for selection diversity. Channel is two-delay Rayleigh fading channel, 0,=crz. hLPF(t) is triangular.

occur if one phasor leads to a finite error probability, and the second (independent) one to an even larger

error probability. Figure 5 shows the BER as a function of the delay spread. As anticipated, the BER goes

with the fourth power of the delay spread. We also see that the (more complicated) BER-driven diversity

is better by about a factor 3; this is very similar to the results for MSK.

4. Adaptive sampling

Previous investigations of rr/4-DQPSK have assumed that the sampling time is fixed, e.g. at the

average mean delay. However, we have shown [9) that for MSK, adaptive sampling can lead to a drastic

reduction of the error floor. It thus seemed reasonable to anticipate a similar gain also for rr/4-DQPSK.

For the theoretical treatment, we first have to generalize the concept of the error region.

Specifically, we define that a point in the complex phasor plane belongs to an error region if the relevant

channel constellation leads to an error regardless of the sampling time. In that context, we have to treat

all symbol combinations simultaneously, since of course we cannot choose different sampling times for

different symbol combinations. The error regions are thus the overlap of the error circels for the individual

sampling times. While it is possible to formally write them down in closed form, the large number of

possible symbol combinations enforces computer aid for the evaluation in practice. Once we have

computed the error regions, computation of the error probability is straightforward: we just have again

to integrate the pdf of the amplitude ratio of the phasor over the error region. The CPU-time requirements

for this method is several orders of magnitude smaller than for Monte Carlo simulations.

53

10-1

1

~ 10-2 / I !

~ iJ4 triangu1ar

~ :0 10-3

'"

I------- raised cosine / ,

a;0.35 / ..0 2 0. .... 0 .... 10-4 .... 0)

V GMSK, + / VI/ BT;0.5 ~ raised cosine ~. receiver filler ~ -------;7 1-/ !X-l.0 BT;0.625

/ I I I I I V/ I I IJIlI i

0.1 l.0

normalized excess delay '2rr Fig, 6 Error probability for adaptive sampling for various filters a~ a function of the normalized delay spread.

We analyzed both "basic" and raised-cosine filtered 1tf4-DQPSK. In neither case did the adaptive

Sarripling lead to a large decrease in the error probability, In the most favorable case (i.e, "basic" 1tf4-

DQPSK), the adaptive sampling decreased the error probability by about a factor 2; for raised cosine with

a=0.35, the BER was decreased by about a factor 1.5 (see Fig, 6), This is an important difference to

(G)MSK, where we can completely eliminate the error floor for unfiltered MSK, and (depending on the

filter bandwidth) can decrease the error floor by several orders of magnitude for GMSK.

5. N-delay channels

For the computation of the N-delay channel, we employ the two-path equivalent matrix (TPEM)

method recently introduced by us [I OJ, The basic idea is the following: the decision at the receiver is

based only on the real and imaginary parts of the signal at two subsequent sampling instances, i.e. on 4

real variables, Since the variables are Gaussian, they are completely described by their correlation matrix,

For the TPEM method, we thus first find a two-delay channel that results in the same correlation matrix

as the actually considered signal. We then compute the error probability in this two-delay channel by the

method of Sec, 3, One interesting feature of this method is that noise can be transformed into equivalent

time dispersion, so that A WGN could also be included in our computation, The main limitation of this

method is that it is only valid for fixed sam piing.

When we transmit a 1tf4-DQPSK signal through an N-path channel, the phasor at sampling time

k is described mathematically by

54

N

bk = I: aiI: eXP[i(lPi+itm)]hLPP(Ts+kT-mT-'ti) + 11k i=1 rn

(14)

where t1m is the modulation phase of the m-th symbol (a multiple of 7tl4); nk is the noise value at time

k. The phasor at time k-l is defined similarly. Since ai·cos(<p) and ~·sin(<p) are independent Gaussian

variables, bk can be wlitten in the fonn

(15)

(and analoguously for bk), where the xi are Gaussian variables and the dj,i can be easily computed from

the (known) t1m, hLPF(t), and "ti. The variables ej signify the line-of-sight component. The variables xi have

the probability density function

oi 0 0 0 0

o 0; 0 0 o c= x

(16)

o 0 0 0 ... 02N

Since the xi are Gaussian variables, this also holds true for any linear combination of them. We can thus

define the new variables

2N+4

Yj = I: dj,iXi (17)

i=l

The pdf of y=(y I' Y2' y" Y4)T is detennined by a matrix Cy, which contains the correlation coefficients

between all the yj.

(18)

where Cy is the correlation matrix of the Yj' and D is a 4*(2N+4) matrix containing the dj .i elements.

55

In tenns of the yj' the condition that an error occurs is for the real part of 1tI4-DQPSK

(19)

and for the imaginary part

(20)

Real and imaginary parts of the quadrature modulation have thus been reduced to the same fonn

as in [101, we only deal with the variables yj' which do not "know" by which modulation fonnat they

were created. Thus all further computation methods, especially the computation of an equivalent two-delay

channel, can be done in the same way as in this reference (note, however, that C""c33)' Figure 7 shows

the error floor for various delay power profiles. We see that it depends essentially on the delay spread,

and very little on the shape of the profile. The difference e.g. between rectangular and two-delay profile

is about 20%; for MSK, this difference is about 50%.

~ 10" ~ III

~ :E 10-2

'" e c..

~ 10-3

10--1

0.01 0.' 1.0

normalized delay spread SIT

Fig. 7 BER for various delay power profiles exponential (solid), rectangular (dashed), and two-delay (dotted). hLPF(t) is triangular. Fixed sampling on the average mean delay.

6. Summary and conclusion

We have investigated the error-floor of nI4-DQPSK in mobile radio channels. For fixed sampling,

the results for this modulation fonnat differ very little from the results of (G)MSK; the error probability

is essentially proportional to (Srr)2, and the proportionality constant depends on the filtering (we usually

assumed raised-cosine filtering in our computations). For adaptive sampling, nI4-DQPSK perfonns

56

considerably worse than (G)MSK, naJ11ely that (in contrast to GMSK), the error floor cannot reduced by

orders of magnitude,

Acknowledgements: We thank Prof. Dr. Johannes Huber, Dip\. lng, Mag, Josef Fuhl, and Dip\. Lng. Hans­

Peter Bernhard for stimulating discussions. The financial support of the Austrian Fonds zur Fiirderung der

wissenschaftlichen Forschung is gratefully acknowledged.

References

[I] W. C. lakes (ed.), Microwave.Mobile Communications, IEEE Press, 1993.

[2] J. C. 1. Chuang, "The effects of time delay spread on portable radio communications channels with digital

modulation", IEEE 1. Select. Areas Comm. vol. SAC 5, 879-889, 1987.

[3] 1. Crohn, G. Schultes, R. Gahleitner, and E. Bonek. "Irreducible error performance of a digital portable

communication system in a c:ontrolleu time-dispersion indoor channel", IEEE 1. Select. Areas Comm., vol. SAC-It,

1024-1033,1993.

[4] T. S. Rappaport, Wireless Communications, Principles and Practice, IEEE Press (1996).

[5] V. Fung, T. S. Rappaport, and B. Thomas, "Bit error simulation of 1t/4-DQPSK mobile radio communications

using two-ray and measurement-based impulse response model", IEEE 1. Set. Areas Comm. vol. SAC-H, 393-405,

1993.

[6] C. L. Liu and K. Feher, "Bit error rate performance of 1t/4-DQPSK in a frequency-selective fast Rayleigh fading

channel", IEEE Trans. Vehicular TedllZ. vol. VT-40, 558-568, 1991.

[7] A.F. Molisch, J. Fuhl, and P. Prok.<eh;--"Error floor of MSK modulation in a mobile-radio channel with two

independently-fading paths", IEEE Trans. Vehicular Techn., in press.

[8] A.F. Molisch, J. Fuhl, and P. Proksch, "Bit error probability of MSK modulation with switched diversity in a

mobile radio channel with two independently fading paths", Proc. Personal Mobile Comm. Conj. 95, Toronto 1223-

1227, 1995.

[9] L. B. Lopes, A.F. Molisch, M. Paier, and J. Fuhl, "On the error !loor in DECT-like systems", Proc. Europ.

Personal and Mobile COI1lI1l. Conj. Bologna 95, 170-176, 1995.

[10] A.F. Molisch and J. Fuhl, "Bit error probability of differentially detected (G)MSK in indoor mobile radio

channels", 46th IEEE Vehicular Tedlll. Conj., Atlanta 96, in press, 1996.

[II] c. L. Liu and K. Feher, "An asymmetrical pulse shaping technique to combat delay spread", IEEE Trans.

Vehicular Techn. vol. VT-42, 425-433, 1993.

[12] A. F. Molisch, "Statistical properties of the RMS delay spread of mobile radio channels with independent

RaYleigh-fading paths", IEEE Trans. Vehicular Techn .. vol. VT 45,201-205, 1996.

[13] G. A. Halls, "Modelling of the HIPERLAN radio channel", Proc. ICCC W CN 94, The Hague, 954-958 (1994).

6 Real-Time Implementation of QCELP Vocoder for speech and data in

CDMA Cellular System Using TMS320C50 Fixed Point DSP Chip

JW kim, BS Yoon, WM Lee, SJ Jang, MS Lim, KC Han

Signal Processing Section, Mobile Commun. Division

Electronics and Telecommunications Research Institute

161 Gajeong-Dong, Yusung-Gu, Taejon, KOREA, 305-350

Abstract

An variable rate QCELP speech coder was proposed for CDMA digital cellular system and QCELP algorithm has been chosen as the speech coder of digital cellular system in North America and Korea.In this paper, real-time implementation of variable rate QCELP speech coder algorithm (IS-96) is presented. Before programming by assembly code for real time operation, a fixed point simulation of QCELP coder is performed in order to find the the optimum data size and scaling factor. Thereafter a real-time implementation of the QCELP algorithm is achieved using a 16 bit fixed point TMS320C50 DSP Chip.. The QCELP codec implementation is evaluated in a DSP evaluation board and verified within our real CDMA digital cellular system for speech and data service.

1. Introduction

As the mobile cellular system is getting popular, the capacity is required to be increased. The current analog mobile system is almost on the edge of the capacity limit and new digital mobile systems are developed to solve the service capacity problem. To meet the increasing capacity and various service demands, operating companies are moving from the analog systems to the digital systems.A1so the digital cellular system requires efficient speech coding algorithm. In the North America, Time Division Multiple Access (TDMA) system and Code Division Multiple Access (CDMA) system are developed and in Europe, Group Special Mobile (GSM) system is

developed. In Korea, mobile cellular service companies has adopted CDMA system as digital cellular system standard and started the mobile communication services since the beginning of 1996.

All of digital cellular systems have their own digital speech compression techniques. The TDMA system uses 8 kbps Vector Sum Excited Linear Prediction (VSELP) coder[l ][2], the GSM system uses 13 kbps Regular Pulse Excited-Long Term Prediction (RPE-LTP) coder[3] and the CDMA system uses Variable Rate Qualcomm Code Excited Linear Prediction (QCELP) coder[ 4]. It has been chosen as an interim standard vocoder (lS-96) for CDMA digital cellular system in North American and Korea. The main advantage in use of variable rate QCELP coder can reduce the average date rate required to transmit conversation speech by a factor of two or more, while the speech quality of variable rate coder maintains same as that of full rate coder. The use of variable rate coder leads to increase the capacity of a CDMA system.

In this paper, we present a real-time implementation of QCELP using a TMS320C50 fixed point DSP chip for speech and data service. The special efforts are required to reduce the finite word­length effects in implementing QCELP coder with TMS320C50 because it is implemented with a 16 bit fixed point DSP chip. The finite-word effects are carefully examined in fixed point simulation before programming in Texas Instruments C50 assemble language. Through the fixed point simulation of QCELP algorithm, the optimum value of each variable is determined and the effect of finite word­length such as overflow, underflow, and quantization error is examined. The effect of use the variable

J. H. Reed et al. (eds.), Wireless Personal Communications

© Springer Science+Business Media New York 1997

58

scaling method is investigated in routines with wide dynamic ranges. The performance of QCELP with fixed point implementation is evaluated in the C fixed point simulation in terms of the Signal-to-Noise Ratio(SNR). Furthermore, the hardware architecture and software structure of QCELP vocoder merged with COMA system and an implemented hardware platform for testing are described In addition to speech service, data service implementation using interrupt routine within COMA system is also described.

2. QCELP Algorithm

The QCELP vocoder developed by QUALCOMM Inc. is a variable rate speech coder applicable to COMA digital cellular system. This variable rate encoding algorithm operates at rates of 8, 4, 2, or I kbps, depending on the level of voice activity. The QCELP algorithm is based upon the general CELP structure. The· overall block diagram of encoder is shown in Fig. I. The important parameters are described in Table I.

Fig. I. Block Oiagram of QCELP Encoder

The input speech is sampled at 8 kHz and is broken down 20 ms speech frames consisting of 160 samples. The linear predictive coefficients (LPC) with lOth order are calculated regardless of data rate selected. The LPC parameters are transformed into line spectral pairs frequency (LSP) for more finely quantization, interpolation, and easily stability check.

The LSPs are updated in each subframe through interpolation of LSPs in neighbor frames. A weighting filter is used to reduce the loudness of the quantization noise.

In pitch search, pitch gain and pitch lag are determined by standard analysis-by-synthesis error minimization procedures. The pitch lag is quantized

Table 1. Parameters for Each Rate

Parameter Rate 1 Rate2 Frame for LPC, LA 160 160 Bits per LPC Update 40 20 Pitch Subframe, Lp 40 80 Bits per Pitch Update 10 10 Codebook Sub frame, Lc 20 40 Bits per Codebook Update 10 10

Parameter Rate Rate 114 118

Frame for LPC, LA 160 160 Bits per LPC Update 10 10 Pitch Subframe, Lp 160 -Bits per Pitch Update 10 -Codebook Subframe, Lc 80 160 Bits per Codebook Update 10 6

from 17 to 143 samples using 7 bits for each pitch update. pitch gain is scalar-quantized with range from o to 2 using 3 bits per pitch parameter update. To determine the optimal pitch gain and pitch lag, the global search is performed over all allowable q)1antized values. The codebook gain and codebook index are determine once for each codebook update. As in the pitch search procedure, the codebook index and the codebook gain are chosen using the analysis­by-synthesis procedure. The codebook is organized as an overlapping code book such that each code vector differs from the adjacent code vector by one sample[5]. The circular codebook consists of the 128 elements including 10 I zero elements and 27 nonzero elements. The use of overlapping codebook provides several advantages such that the storage requirements for codebooks of various rates can be reduced significantly and fast codebook search can be applied by using the dependency of the neighbor candidate code vector. Moreover, the extra complexity reduction without any performance degradation can be achieved

by using the recursive form in calculation of energy term in pitch parameters search and codebook parameters search[ 6]. In lower transmission rates the complexity reduction by the recursive calculation of energy term is increased.

3. Fixed Point Simulation of QCELP Coder

We evaluated the QCELP performance through the C fixed point simulation before implementing the QCELP algorithm on a fixed point DSP chip. We investigated all of blocks for protecting overflows, underflows, and large quantization error. The flow diagram of the QCELP encoder and decoder is shown in Fig 2. Each routine in a function block uses different operations because the effect due to use of the finite word-length variable varies in each function block. The dynamic ranges of all variables in each function block are examined for finding the optimum precision of the variables.

The double precision operations may be used in routines which have the wide dynamic ranges and degrade severely the overall performance of coder due to the use of single precision operation. However, the double precision operation is not performed by the direct double multiplication, but the use of variable scaling method in each speech subframe. The scaling factors of important variables in each function block are variably decided in every subframe by checking the dynamic range of variables in that subframe. The direct double precision multiplication can be avoided by the use of variable scaling factor. The single precision multiplication(l6 x 16 bits) is performed in only magnitude part and the normalization factor of output variable is adjusted according to input variable scaling factors. This variable scaling method can significantly reduce the complexity in routines such as pitch parameter search and codebook parameter search function block in which they require extensive multiplications. The special efforts are required to reduce the instruction in pitch and codebook parameter search routine.

As a result of fixed point simulation of QCELP, we evaluated the variable rate QCELP algorithm in terms of signal-to-noise ratio(SNR). The Table 2 shows the SNR of variable rate QCELP coder. The performance degradation due to fixed point implementation was about 0.15 dB in SNR. However,

59

the degradation of speech quality could not be perceived in an informal subject listening test.

Input Speech

From Receiver

To Transmitter

Fig. 2. Flow Diagram of QCELP Encoder and Decoder

Table 2. Performance Comparison(SNR) of Floating Point QCELP and Fixed Point QCELP

Sentence Floating Point Fixed Point Simulaiion Simulation

English 12.37 dB 12.24 dB Sentences Korean 15.23 dB 15.07 dB

Sentences

60

4. QCELP Implementation on TMS320C50 DSPChip

To implement the QCELP algorithm with TMS320C50 DSP chip, we used the C50 evaluation board. The DSP board configuration is shown fig. 3. The board has TMS320C50 DSP chip of 40 MIPS(Mega Instruction Per Second) and l28k word external SRAM with 10 nsec access time for executable code downloading and 14 bit AID, D/A Analog Interface Circuit(AIC) and RS-232 module for communication.

RS-232

MIC

SPK

Fig. 3. The DSP board configuratipon.

All of instructions that operate QCELP algorithm are consists of TMS320C50 assembly language. This program is composed of an encoding routine, decoding routine and interrupt handler routine. The voice signal from microphone is converted into 14 bit PCM data and it is sent to DSP chip through the serial port. At encoding routine, the PCM data is converted to transmission packet data after the QCELP encoding procedure. The transmission data is variable from 1 kbps to 8 kbps. The output PCM data is reconstructed from transmitted speech packet data at decoding routine. Then the PCM data is converted into analog signal to output to speaker. The architecture of real­time QCELP implemen- tation is show in Fig. 4. The architecture of real-time QCELP implementation is divided by three part. The encoding Idecoding part contains the QCELP coder algorithm. The serial port processing part is responsible for transmission and receiving of PCM sample and it also controls the start of encoding and decoding using counting of serial interrupt occurence. The external command processing part interprets commands, transmits/receives the encoded packets with the

vocoder controller for verifying QCELP algorithm. The several external commands are executed using interrupt pins (INT I, INT 3) and XF pin of TMS320C50 processor for hardware, software debugging purpose.

INTl

INTJ

XF

peM In

PCM Out

External command processing

part

Tx. Buffer

Rx. Buffer

Fig. 4. Architecture of Real-time QCELP Implementation

The implemented QCELP algorithm by TMS320C50 requires 4K word data memory for internal process, 10K word program memory for source code downloading, it needs about 33 MIPS in execution complexity. The execution complexity required in each function block is shown in Table 3.

Table 3. Complexity Required Each Function Block.

Routine No. of no. of MIPS instruction runsl

frame LPC Analysis 23,000 I l.l LPC to LSP 26,000 I 1.3 LSPQuan. 450 I 0.02 Rate Dec. 630 I 0.03 Pitch search 60,000 4 12 CB search 35,000 8 14 LSP to LPC 3,520 4 0.7 Recon.(Enc.) 4,560 8 1.8 Recon.(Dec.) 44,000 I 2.2

PCMIIO

PCMIIO

Host A

Trasferring of Encoded Data

HostB

Fig. 5. Full-duplex Hardware Platfonn Using Evaluation Boards

The full-duplex hardware platfonn to simulate the operation of the real-time QCELP implementation is shown in fig: 5. Here, host A and B are made by Personal computers that are equipped by Tiger 5XF evaluation board. The telephone interface module of Tiger 5XF is used for PCM input and output, the RS-232C module is used for transmission of encoded data. In other words, we replaced the wireless channel with RS-232C line. In addition to test the QCELP coder in evaluation boards, the real-time QCELP implementation is verified in our CMS (CDMA Mobile System) that have been developed by ETRI of Korea. The software structure of QCELP vocoder in CDMA digital cellular system is shown in Fig. 6. After power up and initialization, the vocoder waits for the Vocoder Reference Strobe signal from the interleaver of channel card to synchronize the transmit and receive frames with system reference time. The vocoder waits for 160 samples before the encoding routine begins. When the interrupt handler receives the received packet data from control DSP and it is decoding time, the encoding task is suspended and starts the decoding routine. After the completion of decoding routine, the encoding task is resumed. Tx, Rx "'packet commands are implemented for receiving/transmission with control DSP and another several commands(diagnostic,pcm- loopback, packet loopback etc.) were implemented for debugging of system and vocoder operation. it is also required

61

conversion between u-Iaw and linear PCM within serial PCM interrupt routine in CDMA digital cellular system.

Fig. 6. Vocodet Operation Procedure in CDMA Digital Cellular System

5. Data Service

Data service below 8.55kbps is implemented using interrupt service routine of QCELP assembly codes within CMS digital cellular system. Data service routine is as follow. If call type is data service call at call connection state, QCELP vocoder algorithm is disabled and data service routine is activated. This mode is called by bypass mode. Because the rate of packet transmission between Vocoder and control DSP is 8.55kbps(171bits/20msec) and that of sample transmission through PCM port is 64kbps, data rate adaptation is required. Therefore, the only first subframe of four decoder subframes has valid data packet at serial PCM transmission. Preamble bytes are used for valid data packet classification. The packet format between vocoder and control DSP is the

62

same in speech service. The fonnat between vocoder and IWFS(Inter-Working Function Sub-system for data service) is composed of preamble(6byte), length byte(lbyte) and data packet (maximum 22bytes). We have evaluated data service through character and file transmission in CMS system.

6. Summary

In this paper, we presented the real-time QCELP implementation on the TMS320C50 fixed point DSP chip. The principles of the QCELP algorithm, fixed point simulation of QCELP coder for performance optimization and the method of real-time QCELP implementation and data service below 8.55kbps have been described. The implemented QCELP on 16 bit fixed point DSP chip was evaluated in the evaluation board and real CD MA digital cellular system. The speech quality of implemented QCELP speech

codec has been showed good in both of two different environments and data service has been operated well in CMS cellular system. The required computing power of impl~mented QCELP on TMS320C50 was about 33 MIPS and its program size was 10 kwords. It is necessary to make continuous efforts for shorter execution by code optimization and for improvement of speech qUality.

References

[I] 1. A. Gerson and M. A. Jaisuk, "Vector sum excited linear prediction(VSELP) speech coding at 8 kbps," in Proc. IEEE Int. Conf. Acous., Speech, Signal Processing, pp. 461-464, Albuquerque, USA, April, 1990. [2] M. H. Sunwoo and S. Park, "Real-time implementation of the VSELP on a 16-bit DSP chip", IEEE Trans. Consumer Electronics, pp.772-782, Vol. 37, No.4, November, 1991. [3] 1. E. Natig, "Evaluation of six medium bit-rate codes for Pan-European digital mobile radio system," IEEE Journal on Selected Area in Commun., pp.346-352, Vol. 6, No 6, June, 1988. [4] Qualcomm Inc, TWEIA Interim Standard-96, "Speech option standard for wideband spread spectrum digital cellular system", April, 1993. [5] D. Lin, "Speech Coding Using Pseudo-Stochastic Block Codes", in Proc. Int. Conf. Acoust., Speech,

Signal Processing, pp. 1354-1357, Dallas, USA, April, 1987. [6] K. Han, B. Yoon, 1. Lee, and S. Kang, "Implementation of variable rate QCELP codec using TMS320C30 DSP," in Proc. Int. Conf. Signal Processing Application Technology , pp 203-208, Santa Clara, USA, Sep., 1993.

7 Spread-Spectrum Packet Radio Using Narrowband Headers

Robert K. Morrow, Jr. Morrow Technical Services

6976 Kempton Rd. Centerville IN 47330

317 -855-2348, rkmorrow@aoJ.com

Abstract: Transmitting the header of a packet using narrowband signaling coupled with an

associated channel access protocol such as ALOHA or carrier-sense multiple-access (CSMA),

followed by the data portion of the packet as a spread-spectrum signal with a spreading

sequence that is common to all nodes in the network, provides several advantages over using

either narrowband or wideband techniques exclusively. For example, all network nodes can

obtain channel traffic loading and destination information by examining only the narrowband

headers. Wide band synchronization is facilitated by using the end of the header transmission as

a time index for wideband signal acquisition. Long signature sequences minimize the probability

that two or more wide band packets will be transmitted with their sequences aligned, while

avoiding the requirement that each receiver search for several different sequences as in a

transmitter-oriented protocol. A common signature sequence facilitates packet broadcasting and

simplifies transmitter and receiver design. Flow control can be implemented by monitoring the

CSMA narrowband channel and insuring that the network offered rate remains below the

saturation point. Finally, adaptive interference cancellation is simplified. The analysis shows

that the system has similar throughput performance to a wideband-only network provided that

the headers are short compared to the total packet length. This system was granted u.s. Patent

5,022,046.

1. INTRODUCTION

Packet radio systems have become widely

used for many forms of digital

communication. Although various channel

access protocols exist, the underlying trait of

such systems is a shared communication

channel, and stations are allowed access to this

channel when they have something to transmit.

Spread-spectrum signaling may be used to

reduce the effect of the inevitable packet

overlaps (collisions) by deliberately increasing

the bandwidth of all signals. Now multiple

users can transmit together without necessarily

causing mutually destructive multiple-access

interference (MAl). Packet radio networks

can be coupled with spread-spectrum signaling

in an attempt to gain the advantages of both

[1]. This paper presents and analyzes a packet

radio system that combines certain features of

J. H. Reed et al. (eds.), Wireless Personal Communications

© Springer Science+Business Media New York 1997

64

both narrowband and wideband signaling tech-

niques to allow a number of advantages over

systems that utilize narrowband or wide band

methods exclusively. This system was granted

US Patent 5,022,046 on June 4, 1991 [2].

network, and is independent of propagation

delay. The "blurring" effect of multipath on

the narrowband signal and receiver bandwidth

limitations and timing errors are the sources of

uncertainty which prevent the receiver from

knowing exactly when the wide band signal

begins. The unslotted packet transmission

2. SYSTEM OVERVIEW scheme precludes the need for a master syn-

chronization clock among the nodes. Nodes

In this network, packet preambles and headers can monitor the narrowband channel for

are transmitted on a narrowband channel, header traffic and any associated

using one of the established protocols such as acknowledgment packets to dete:rnine which

ALOHA or CSMA, while the body of the nodes are in range, which of these are busy,

packet is sent using spread-spectrum tech- and how heavily the channel is loaded.

niques (Fig. 1). Furthermore, a node can use the relative power

~ Narrowband . -tI.~ . .--- Wideband ~

Figure 1. Packet structure.

The spreading sequence is common to all

nodes in the network and does not repeat

throughout the entire packet. Every node in

the network uses the same narrowband

channel for packet headers and identical sig­

nature sequences on the wideband channel.

Packet broadcasting is done by simply listing

multiple destination node identifiers in the

header; all specified receivers can then acquire

the data portion of the packet together. Receiver synchronization is facilitated because

the time between the end of the narrowband

header and the beginning of the wideband data

is fixed and known by all nodes in the

of a received narrowband header to set its own

power to a level that will provide reliable

communication with a desired node while

mInimIZing interference to others.

Interference cancellation is facilitated by using

the narrowband header for power information

and the common signature sequence allows

simple implementation of a signal subtraction

algorithm. A CSMA narrowband channel

effectively controls access to the spread­

spectrum channel, so wideband flow con trol is

easily accomplished.

3. ANALYTICAL MODEL

The process of developing throughput performance figures for a traditional

narrowband packet radio network centers

around finding the probability that two or

more packets will overlap [3]. If receivers in

65

the network have no signal capture capability, thermal noise on packet performance is small

then the communication channel is considered compared to degradations caused by collisions

useless during these periods of overlap and all on the narrowband channel and MAlon the

affected packets are presumed destroyed. The wideband channel. We also assume that a

multiple-access characteristic of spread- receiver will always be able to synchronize on

spectrum means that packet overlap no longer the wideband signal after successfully

results in mutual destruction, but instead acquiring its narrowband header.

produces a gradual degradation of bit error

performance as the number of simultaneous 3.1 Packet Throughput, Narrowband Channel

users increases.

Like their narrowband counterparts, spread­

spectrum packets on a network can also ap pear

in either a slotted or unslotted ALOHA

configuration. Unslotted ALOHA nodes

transmit packets immediately upon arrival,

while slotted ALOHA systems require that all

packets arriving. during a particular time slot

be transmitted together during the next time

slot. Although slotted ALOHA systems are

easier to analyze, it is often more practical to

implement the unslotted scheme, since no time

synchronization among nodes is required.

Since only the header of each packet is carried

on the narrowband channel, we will examine

narrowband network performance when short

packets. are transmitted with either unslotted

ALOHA or CSMA channel access protocols.

For the wideband analysis, we use a heuristic

approximation based up~n an established

slotted analysis to obtain unslotted

performance results similar to those in [4]

and [5], but at a greatly reduced computational

complexity. We assume that the signal-to­

noise ratio between each pair of nodes in the

network is high enough that the effect of

By equating the overlap of two or more

packets to their mutual destruction, traditional

narrowband analysis has avoided complex bit

error event calculations while still providing a

reasonably good approximation to network

performance. It is necessary only to find the

probability that a tagged packet will be

transmitted alone over the channel. As shown

in [3], if packets in the network arrive

according to a Poisson process with arrival

parameter Iv packets per second, and if each

packet lasts for 1" n seconds, then the packet

arrival rate can be normalized to Gn = A1"n. In

an unslotted ALOHA network, some of these

packets will be destroyed through collisions,

producing the well-known normalized

throughput result

(I)

Since our narrowband network contains only packet headers, the packet length 1"n is short

compared to networks that transmit the entire

packet on the narrowband channel, so our network's offered rate Gn will be smaller for a

given A, reducing the collision probability.

66

For CSMA analysis, we select the unslotted

non-persistent protocol [6], where the

transmitting node first examines the channel

and transmits the packet if the channel is

sensed idle. Otherwise, the node will wait a

random time before once again checking the

channel and proceeding as before. Due to

non-zero signal propagation and processing delays r p' it is possible for a busy channel to

be sensed idle and for an idle channel to be

sensed busy. In either case, network per­formance is diminished. Now let a = r /r p n'

the ratio of propagation and processing delay

to narrowband packet length. Then the

throughput of this network is given by

. G e-aG, S = n

n2 Gn(2a+l)+e-aG, (2)

Fig. 2 shows a plot of Gn and Sn for non­

persistent CSMA with various values of a, and

for pure ALOHA.

When a is small, non-persistent CSMA

approaches perfect channel utilization (Sn2 = 1) for large Gn. When a = 1, however,

0.6

0.5

'$ 0.4 ]' ~ 0.3 l: ~ 0.2

O.l

2

Offered Rate

Figure 2. Narrowband packet throughput.

One implication of CSMA is that, unlike

ALOHA, packets "offered" to the CSMA

narrowband channel when it is sensed busy are

not transmitted immediately. Offered packets

fall into three categories: successfully

transmitted, unsuccessfully transmitted, and

not transmitted but rescheduled instead.

Equation (2) gives the rate of successfully

transmitted packets, but packets in the first

two categories will be transmitted on both the

narrowband and wideband channels. If this occurs with rate Go' then Sn2 < Go < Gn for

maximum normalized throughput is only

0.144 at an offered rate of 0.46, compared to

the ALOHA figures of 0.184 at 0.50, respectively. Since our packet headers are realizable networks. By combining some

short, it is possible that a may be significant, results in [6] we have

so the added complexity of implementing

CSMA instead of pure ALOHA must be

weighed against a performance gain which

may be modest (or even negative).

G= Gn (aGn +l) o Gn (2a + 1) + e-aG,

(3)

Packet arrivals on the wideband channel from

the non-persistent CSMA process are clearly

not Poisson distributed, but retaining the

Poisson assumption provides analytical

simplicity and matches simulation results

closely when calculating CSMA performance

figures [6].

3.2 Bit Error Probability, Wideband Channel

The spread-spectrum system model used for

the wideband channel is similar to that in [8].

The "desired" signal is transmitted by user 0,

and there are K interfering users. The k-th

user's transmitted signal has the form

xkCt - r.k) = -ffPbk(t - "k)ak(t - "k)

·cos(mct + ¢k) (4)

where bk(t)and ak(t)are the data and spectral­

spreading signals, respectively, P is the

received signal power, and the carrier frequency is mo' The differences in

propagation and message start times are incorporated into the "k' and ¢k represents the

phase parameter in the carrier. Since these

parameters are taken with respect to the desired signal, we set "0 = ¢o = O.

Since the signature sequence of each user is

modeled as random from bit to bit, the delays "k in (4) can be completely characterized by

the relative chip delays Sk of the interfering

67

magnitude is N, the variance '¥ of the

multiple-access interference for any number of interfering users K is a function of the S k and

¢k relative to the desired signal [9], [10] and is

given by

'¥(Sl" ",SK'¢I'''''¢K) = K (5) I,N(2s; -2sk +1)cos2¢k k=l

When the chip delays are uniform on (0,1) and the carrier phases are unifoim on (0,21.), then the density function f'¥ (z) of this

variance is the (K -I)-fold convolution of the

density fz(z) of the variance of a single

interfering user, given by

1 1~2(N -z) +-INI fz(z) = lC-J2Nz log ~2(N _ z) _ -IN (6)

for 0 < Z ~ Nand Z 7:. N /2. N ow the

probability of wideband data bit error p,,(w) can

be accurately calculated using

where

1 - ( 2t Q(x) = -f2ir I exp -~ ru (8)

signals [8]. After normalizing the receiver Equation (7) is the improved Gaussian

decision statistic so that the desired signal approximation introduced in [9], and is shown

68

to be extremely accurate when measured against bounds on p,,(w) given in [8] for ran-

dom signature sequences.

3.3 Wideband Packet Success Probability

If a packet of length L bits is transmitted over

a memoryless binary symmetric communi­

cation channel with average probability of data bit error p" throughout the packet, and if the

packet includes block error control capability

that can correct t or fewer errors, then the

condi tional probability of packet success QE!P" = g(P,,;L,t) is a function of p" for each

Land t; that is,

tagged packet, we can condition on the delay

and phase values, use (9) to find the

probability of packet success for each of these

values, and average the results, yielding an

overall packet success probability of

QE = E[QE!P,,] = J g( Q( Jy }L,tYAY)dY.

(10)

Equation (10) incorporates the effect of bit-to­

bit error dependence within a tagged packet,

and assumes that the number of interfering

users K remains constant throughout the dura­

tion of the tagged packet. Also, since the function f",(y) depends on K, QE may also be

t (LJ. L . g(p,,;i,t) =~ i (p,,)'(l- p") -/ (9) expressed as a function of K. However, (10)

This corresponds to a situation where the

number of interfering users, and hence the

average bit error probability, remains constant

throughout a desired (tagged) packet, which is

characteristic of a slotted network. However,

both slotted and un slotted CDMA signal

analysis is made more complicated by the fact

that each interfering user has a relatively

constant chip offset from bit-to-bit during its

period of overlap with the tagged packet, and

carrier phase may be constant, or change only

slowly, from bit-to-bit as well. These factors

produce positively-correlated error events

between adjacent bits within the tagged packet

[9]. If we assume fixed chip delays and

phases for all interfering signals throughout a

does not account for the random changes in K

that are likely to occur in an unslotted

network, nor does it consider the distribution

of packets in the wideband channel whose

headers were successfully received on the

narrowband channel. We examine these issues

next.

3.4 Packet Throughput, Wideband Channel

If the narrowband channel uses the pure

ALOHA protocol, then packets entering the

wideband channel can be modeled as the narrowband arrival process delayed by 'Z'n;

thus wide band packet arrivals also form a Poisson process with arrival parameter A. By

letting r = 'Z'w / 'Z'n' the ratio of wideband to

narrowband packet transmission times, the

69

normalized wideband packet arrival rate Gw is 3.5 Error Control in the Wideband Network

equal to rGn • For this offered rate, slotted

CDMA network throughput Sw is [11] Packets transmitted using multiple-access

CDMA will benefit from an error control code ~ Gk

S = G Q"(w) = G e-Gw L ---l!:.. Q(w)(k) (11) that can correct the bit errors due to MAl [10], W w E w k=O k! E but fewer than L bits will now be available for

In unslotted CDMA networks, the number of

interfering users K may change randomly

several times during the transmission of a

tagged packet, greatly increasing the

complexity of network throughput models

compared to those used for slotted analysis.

Convexity of the Q function in (8) supports the

conjecture that a particular unslotted CDMA

system will experience a lower throughput

compared w that of a slotted system for a given offered rate Gw • In other words, during

un slotted transmission, increases in the

number of interfering users above the average

for portions of a tagged packet will hurt it

more than decreases during other portions will

help. In networks using CDMA only, the

unslotted ALOHA peak throughput decrease

from the slotted peak was shown to be about 15% when using either a A-channel model [4]

or an improved Gaussian approximation

Markov model [5]. Unfortunately, both

methods have high computational complexity.

However, if we let [12]

the message. For this analysis, we select a

block BCH code that can correct t or fewer

errors in an L-bit packet. If we let L = 2; -1 for a particular i E {2,3,4, ... }, then the

number of message bits in the packet will be at

least L = ti. By using equality as a lower

bound on the number of message bits, we can

define the code rate R as the ratio of message

bits to packet length; thus

L-ti R=-­

L (13)

For example, if L = 1023 and t = 51, then

R", 0.5 for this BCH code implementation.

Networks which employ packet error control

will have an effective packet throughput equal

to the actual throughput times the code rate.

We will use effective throughput for all

subsequent narrowband/wideband system

throughput calculations.

3.6 Composite Network Throughput

Gw =: (1.2r -1 )Gn

Now that the analytical foundation is in place,

(12) we can find effective network throughput

figures for a given offered rate by combining

narrowband and wide band throughput we can produce un slotted throughput results very similar to those in [4] and [5], but without

the associated high computational complexity.

equations. For our network using pure

ALOHA for packet header transmission,

70

Equations (1) and (11) can be com bined (using G = Gn ) to give throughput Sl as

(14)

where Qkw)(k) is the wideband packet success

probability given by (10) as a function of k.

Although the infinite summation cannot be

evaluated exactly, the fact that Qkw)(k) decreases monotonically toward 0 as k

increases, along with rapidly decreasing

probabilities that k» rG, allow us to

truncate the summation arbitrarily close to the

final throughput value. For the numerical

results presented in the next section, we

truncate the summation at the point where

Qkw )(k)<1O-3 , or when the packet success

rate drops below one in one thousand.

When the non-persistent CSMA narrowband

protocol is used, only some of the packet

arrivals on the narrowband channel will be

transmitted immediately, so we have

( -aG -(1.2r-l)Go J

S = RrG ....:,e-;-:-_·_e-;:-_--::rc 2 G(2a+l)+e aG

~ [(1.2r-l)Got (w)() . £.., k! QE k

k=O

(15)

where Go is defined in (3).

The number of interfering users changes

slowly relative to the packet bit rate for

reasonable values of G, so our calculations

retain the effect of bit-to-bit error dependence

by using (10) to generate packet error

pi:obabilities. If desired, the procedure in [5]

can be used to calculate accurate average

packet success probabilities (without ac­

counting for bit-to-bit error dependence) for

use in (14) and (15) at the expense of higher

computational complexity.

4. NUMERICAL RESULTS

In this section, we examine some of the

implementation possibilities for this packet

radio network and develop some throughput

results. Network topology issues are avoided

by assuming an infinite user model where each

user's signal has equal power at each node.

Unless noted otherwise, we use a wideband

packet length L = 1023 bits and an error

control code rate R = 1/2. Also, we avoid

unnecessary complication by assuming that

wideband signals produce negligible

interference on the narrowband channel and

vice versa. (Co-located narrowband and

wide band carriers are examined briefly in

[12].) The plots show offered rates and

throughputs both normalized to the wideband

packet length to provide an indication of

relative data rates.

Fig. 3 shows overall effective network

throughput when the narrowband headers are

transmitted using unslotted ALOHA. Channel

capacity (peak throughput) is enhanced by

making r as large as possible, increasing the

proportion of traffic on the wideband channel

and hence reducing the probability of

narrowband collisions. For example, when

r = 1000 in Fig. 3, the probability of header

collision is only 0.04 at the offered rate at

which channel capacity occurs.

Offered Rate

Figure 3. Network throughput, ALOHA narrow band.

As expected, throughput performance for non­

persistent CSMA headers is slightly below that

of ALOHA when signal propagation and

processing delays are equal to the narrowband

packet length (Fig. 4), and shows a similar

sensitivity to variations in r.

For an existing network, adjusting r can be

done either by changing narrowband packet

transmission times with a corresponding

change in a (Fig. 5), or by adjusting L (Fig. 6).

71

10

:; 6 co. ..c ..

::I 0 .. 4 ..c ....

2

0

0 10 20 30 40 50

Offered Rate

Figure 4. Network throughput, CSMA narrowband (a = 1).

10

6

4

2

o 10 20

r=100

0=0.1 r=50

"\~-­, \

30 40

\ , \

50

Offered Rate

Figure 5. Adjusting narrowband packet length.

72

10 r=100 L=1023

r=50

r=200 -, ___ !-~511

8 L=2047 iI' .... , - .... iI',' \

:; 'I ", \

6 'I .'

"" I ..c . ........- \ .. . = I

/r=25 \ 0 . .. 4 ./ \ ..c

E-< :/ L=255 , 2

:;I

a=O.1

0

0 10 20 30 40 50

Offered Rate

Figure 6. Adjusting wideband packet length.

Once r is large enough that narrowband

collisions are rare and most packet losses are

due to wide band MAl, channel capacity is

affected only slightly by further increases in r.

On the other hand, for a given r and a, ca-

channel saturation and instability due to an

offered rate beyond that which produces

maximum throughput.

10

:; 6 "" ..c ..

= Q .. 4 ..c E-<

2

~.-. I .\ ' .

.' \ '. , • t=50 \ \ ,t=30 ..

\ \ ,

\ \ ,

. ,

,

.

a=O.1 " 0

0 10 20 30 40

Offered Rate

Figure 7. Throughput for different error control code rates.

50

pacity is significantly affected by changes in 5. INTERFERENCE CANCELLATION

the maximum number of correctable data

errors t, as shown in Fig. 7.

Since the CSMA narrowband channel controls

access to the wideband channel, it is relatively

easy to implement a flow control scheme in a

finite-user network. Unlike a packet system

using only CDMA, nodes can estimate

network activity by simply monitoring

narrowband headers. Now they can adjust

their own packet transmission and

retransmission rates to insure that throughput

requirements are met with minimum

transmission delay, but without causing

Although our system model and numerical

results are based upon the assumption that all

signals have equal power at the receiver, this

is usually not the case in practical wireless

systems. DS/SSMA can suffer severe

performance degradation if the desired signal

has significantly less power at the receiver

than one or more interfering signals.

Performance can be greatly improved,

however, by using adaptive interference

cancellation [13], where a receiver

synchronizes on several interfering signals

along with the desired signal and attempts to

73

subtract these interfering signals from the 6. CONCLUSION

composite.

As pointed out in [13], synchronizing on and

canceling the strongest interfering signals not

only provides the greatest benefit, but is also

the easiest to do. For cancellation to be

reliable, however, the receiver must closely

estimate the interfering signal powers and be

able to successfully synchronize on the

signals. Both of these requirements are

facilitated by using DS/SSMA with

By transmlttmg packet headers on a

narrowband channel, followed by packet data

on a wide band channel, we gain several

advantages over using either narrowband or

wideband techniques exclusively. All network

nodes can eas ily obtain channel traffic loading

and destination information by examining only

the narrowband headers. Wideband

synchronization is facilitated by using the end

of the header transmission as a time index for

narrowband headers. The header itself wideband signal acquisition. Long signature

provides a simple means to determine signal

strength, and the use of identical signature

sequences beginning at a fixed time after

header transmission aids the synchronization

process.

Suppose, for example, that a receiver has the

ability to simultaneously generate M copies of

its signature sequence. Then it has the ability

to synchronize on up to M simultaneous

transmissions, most of which are probably

interferers. If more than M signals are present,

then it can use the signal strength of (and

information in) the (M + 1 )th and subsequent

headers to determine if it should break sync on

the weakest of the interfering signals and

instead cancel the new sign!!l (or receive it, if

it is a desired signal). Unlike the successive

cancellation method presented in [13], which

requires the incoming signal to be stored in

memory and requires delays for the successive

cancellation process to occur, this cancellation

algorithm can take place in real time.

sequences minimize the probability that two or

more wideband packets will be transmitted

with their sequences aligned, while avoiding

the requirement that each receiver search for

several different sequences as in a transmitter­

oriented protocol. A common signature

sequence facilitates packet broadcasting.

Finally, real-time interference cancellation is

easy to implement.

REFERENCES

1. M. B. Pursley, "The role of spread­spectrum in packet radio networks," Proc IEEE, vol 75, pp. 116-143, Jan. 1987.

2. R. K. Morrow, Jr., "Narrowbandlwideband packet data communication system," United States Patent 5,022,046 dated June 4, 1991.

3. N. Abramson, "The throughput of packet broadcasting channels," IEEE Trans. Commun., vol COM-25, pp. 117-128, Jan. 1977.

74

4. J. M. Musser and J. N. Daigle, "Throughput analysis of an asynchronous code division multiple access (CDMA) system," Proc. ICC, pp. 2F.2.1-2F.2.7, Jun. 1982.

5. P. W. deGraaf and J. S. Lehnert, "Performance comparison of slotted and unslotted DS/SSMA networks," submitted to IEEE Trans. Networking.

6. 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. Commun., vol. COM-23, pp. 1400-1416, Dec 1975.

7. M. B. Pursley, "Performance evaluation for phase-coded spread-spectrum multiple-access communication-Part I: System analysis," IEEE Trans. Commun., vol. COM-25, pp. 795-799, Aug. 1977.

8. J. S. Lehnert and M. B. Pursley, "Error probabilities for binary direct-sequence spread-spectrum communications with random signature sequences," IEEE Trans. Commun., vol. COM-35, pp. 87-98, Jan. 1987.

9. R. K. Morrow, Jr., and J. S. Lehnert, "Bit­to-bit error dependence in slotted DS/SSMA packet systems with random signature sequences," IEEE Trans. Commun., vol 37, pp. 1052-1061, Oct. 1989.

10. -, "Packet throughput in slotted ALOHA DS/SSMA radio systems with random sig­nature sequences," IEEE Trans. Commun., vol 40,pp. 1223-1230,Jul 1992.

11. D. Raychaudhuri, "Performance analysis of random access packet-switched code di­vision multiple access systems," IEEE Trans. Commun, vol COM-29, pp. 895-901, Jun. 1981.

12. R. K. Morrow, Jr., "A Narrowbandl Wideband Packet Radio System", Proc. 1994 Tactical Communications Conference (AFCEAlDARPA), Fort Wayne, IN, May 10-12,1994.

13. J. Holtzman, "DS/CDMA successive interference cancellation," Code Division Multiple Access Communications, Kluwer Academic Publishers, 1995, pp. 161-180.

8 Congestion Relief on Power-Controlled CDMA Networks

Jay M. Jacobsmeyer

Pericle Communications Company P.O. Box 50378

Colorado Springs, CO 80949 (719) 548-1040

73251.1500@compuserve.com

Abstract-A digital cellular radio Code Division Multiple Access (CDMA) system can only support a finite number of users before the interference plus noise power density. 10. received at the cellular base station causes an unacceptable frame-error rate. Once the maximum interference level is reached. new arrivals should be blocked. In a power-controlled CDMA system. the base station can direct mobiles to reduce their power and data rate to reduce interference and allow more users on the system. This approach is employed in TIA IS-95 with respect to the time-varying voice activity on cellular voice channels. In this paper. we investigate an alternative technique where we adjust the power and data rate of mobile data users to the time-varying interference level to allow more users on a congested system. This scheme was simulated for various proportions of voice and data users and offered traffic levels. Blocking probabilities are reduced in some cases by two orders of magnitude. Message wait time. now a random variable. may exceed the wait time for a constant rate system at high traffic levels. If the cellular carrier has a maximum blocking requirement. an adaptive rate/power system can increase capacity. For example. a base station that normally supports 26.4 Erlangs offered traffic with 2% blocking can support 33.5 Erlangs with the same blocking probability if adaptive rates and power control are used. Thus. the adaptive rate system increases capacity by 27%.

1.0 Introduction

Cellular radio Code Division Multiple Access (CDMA) systems are capacity limited by the maximum tolerable level of interference-plus-noise power density, 10, received at the cellular base station. Although there is no hard limit on the number of mobile users served, there is a practical level of 10 that should not be exceeded if we wish to maintain low error rate communication. Once this maximum level is reached, new arrivals should be blocked. In a power-controlled CDMA system, the base station can direct mobiles to reduce their power and bit rate to reduce interference and allow more users on the system. This approach is used today in systems that detect voice activity and adjust the vocoder rate and mobile transmit power downward during pauses in speech. Unfortunately, this method is at the mercy of the statistics of the voice users and can only increase capacity in an average sense. Also, the advantage of voice activity detection diminishes as the proportion of cellular data users grows. However, unlike voice users, data users rarely require

'This material is based upon work supported by the National Science Foundation under award number 9361690.

J. H. Reed et al. (eds.), Wireless Personal Communications

© Springer Science+Business Media New York 1997

76

real-time two-way communications. Thus, we can reduce the power and bit rate of data users (maintaining a constant energy per bit, Eb) to temporarily relieve congestion and allow more users on the system. This approach can be applied to power-controlled CDMA systems like the North American standard, TIA IS-95 [2].

The remainder of this paper is organized as follows: Section 2 introduces the multiple access model. In Section 3, we address the models for the arrival and server processes. An analytical expression for blocking probability in a simplified 4-server model is derived in Section 4. Section 5 presents simulation results for blocking probability and message wait time for a larger, 35-server system. Section 6 concludes the paper.

2.0 CDMA Model

The wideband system under investigation employs code division multiple access (CDMA) in a cellular radio environment. Mobile users are perfectly power-controlled to ensure equal received power at the base station regardless of the position of each mobile or the mean propagation conditions. The system can tolerate a maximum interference power level before service quality drops below an acceptable level. This maximum interference power threshold is a constant.

There are two types of mobile users: voice and data. Voice users require a full-rate channel at all times. Data users operate at a time-varying rate designed to maximize throughput and minimize blocking. The data rate and transmitted power level are adjusted in the same direction to maintain a constant energy per bit, Eb' It is not practical to allow the data rate to be lowered to an infinitely small value, so the minimum data rate is set to r min' a constant. The number of busy channels at time t is given by k = kv + kd, where kv and kd are the number of active mobile voice users and data users, respectively. The maximum interference power threshold corresponds to the maximum number of full rate (and full power) users, m, the base station can support. If an arrival occurs when all full-rate channels or their equivalents are busy and the data rate is at its minimum, the arrival is blocked and cleared from the system. No queuing is allowed at the base station.

In [1], Audrey Viterbi and Andrew Viterbi develop a model for outage probability as a function of the multiple access interference power. Viterbi's model is a MIMloo queue with voice activity factor, p (p == .4). They did not consider the effects of data users. Because the capacity of a CDMA system is soft, Viterbi and Viterbi prefer outage probability to blocking probability. The outage probability is defined as the probability that the interference plus noise power density, 10,

exceeds the noise power density, No by a factor, 1/11, where 11 takes on typical values between 0.25 and 0.1 [5, pp. 204]. The resulting expression for outage probability is bounded by the tail of the Poisson distribution [5, pp. 205],

77

P e -pAJIl out <

(1)

where ;VI! is the offered traffic in Erlangs and Ko' satisfies the outage condition

m

~ v). < (W IR )(1 - Til = Ko' £.. Eb 110 ;= 2 (2)

where W is the spread bandwidth and Vj is the binary random variable indicating whether the jth voice user is active at any instant. For example, for a process gain of WIR = 128,11 = 0.1, and EblIO = 5, Ko' = 23. If voice activity is 100%, the maximum number of users supported is

m = Ko' + 1 = 24.

In contrast, our system model is a variant of the MIMlm Erlang loss system. The blocking probability for voice users is given by the Erlang B formula [3], rather than the Poisson distribution? but for m > 20 and blocking probabilities of a few percent or less, the Poisson distribution and Erlang B are practically identical. From (1) we see that the net effect of the voice activity detection is to reduce the offered traffic by a factor p. Thus, if we are comparing like systems, we can ignore voice activity detection by letting p = 1 and Vj = 1 for all j. Because our power control is perfect, the system performance is a function of the total interference power, not the total number of users. Thus, we can increase the number of users beyond the limit of Viterbi [5] as long as we reduce the power and bit rate so the maximum interference power level is not exceeded.

3.0 Arrival & Server Process

We shall compare our results to a conventional multi server (Erlang) loss system where all users operate at a constant rate. Users arrive at the base station according to a Poisson random process with rate A.. There are m servers (channels), each with independent, identically distributed service times. Arrivals who find all m servers busy are turned away and lost to the system. (They get a busy signal). No queuing of arrivals is allowed. Service times (call durations) are exponentially distributed with mean III!. For an m server system, the probability that k servers are busy is given by the Erlang B formula [3]:

78

(~)k [m (~f]-l Pk = -k-'- L -. ,-. ) .

j;O (3)

and the mean number of busy servers is

E[K] (4)

where Pm is the probability that m servers are busy. The probability Pm is also known as the blocking probability, Ph'

In the adaptive ratelpower system under investigation, voice users arrive according to a Poisson process with rate A." and remain on the system for an exponentially distributed call duration with mean (lIJlv), independent of the arrival process. Data users also arrive according to a second, independent Poisson process with rate Ad, and have exponentially distributed message sizes of mean N bits. The data user arrival process is independent of the voice user arrival process. The call duration for data users is a function of the normalized data rate, R, a random variable. The mean call duration at full rate is (lIJld) = Nlrf seconds where rf is the full rate in bits per second. The total offered traffic is A/Jl = VIlv + AiJld'

Data users may operate at fractional normalized rates, R, where R ;::: r min- All data users operate at the same rate at time t. The normalized (to rf) operating rate of data users is given by the following expression:

r- k

'

m -kv ;:::

--;C;;-' k;;- rmin

R = m -kv

rmin' k;;- < rmin (5)

where kv is the number of voice users and kd is the number of data users. The variables kd and kv are non-negative integers. The number of active equivalent full-rate channels is given by

(6)

If there are any data users in the system, all of the capacity is used, and kf = m. When a user (voice or data) attempts to access the system when the equivalent number of full-rate channels is

79

m, all data users lower their power levels and data rates proportionally. This process creates more full-rate channels and avoids blocking. If all data users are already at the lowest allowable data rate, r min' the call is blocked. We show in the next section that the probability of blocking is much lower than for a conventional constant rate system.

4.0 Analytical Results

To illustrate the server process, consider a simplified example with m = 4 and r min = 1/2. The state diagram for this system is shown in Figure 1. The state probabilities for the system are found by solving the system of linear equations formed by the balance equations (e.g., see [4, pp. 351]) plus the expression:

24

LPj = i=O (7)

Note that there are nine blocking states: 7, 8, 14, 15, 19,20,22,23, and 24. States 8, 16,20,23, and 24 are blocking states to both voice and data arrivals. States 7, 14, 19, and 22 are blocking

states to voice arrivals only. The blocking probability is given by

(8)

where Pv is the probability that a voice arrival occurs before a data arrival. Because the voice and data interarrival times are exponentially distributed with rates A.v and A.d' respectively, the

probability Pv is given by [4, pp. 195]

!..v Pv = !..v + !..d (9)

Equation (8) can be used to compute the system blocking probability for various arrival rates and

service times. For example, consider a total offered traffic of AlIl = 2 Erlangs with A.v = .01

arrivals/s, lIllv = 100 s, A.d = 0.1 arrivals/s, and l/Ild = 10 s (50% voice traffic, 50% data traffic). The blocking probability for these parameters is 3 %. The blocking probability for a conventional Erlang B system with this same traffic load is 9.5% (from (3». Thus, we have lowered the blocking probability by more than a factor of three by adapting the data rate and transmit power to the traffic load.

If the cellular service provider has a blocking probability objective, a variable rate system can improve the system capacity. For example, we have plotted the offered traffic supported at a

~ m

:i"

co Cl ~.

<: gj iil I In

crea

sing

dat

a u

sers

Blo

ck

•

Leg

end

Voi

ce u

sers

D

ata

user

s V

oice

use

rs

Dat

a us

ers

~

8-

Sta

te N

o. ~ St

ate

No.

Full

Rat

e or

Hig

her

Fra

ctio

nal R

ates

(S

hade

d)

Fig

ure

I -

Sta

te d

iagr

am f

or m

= 4 s

yste

m w

ith

R =

(m -

kv)/

kd a

nd r

min

= lI

2

Blo

ck

00

o

81

blocking rate of 2% for our m = 4 variable rate system in Figure 2. Note that when half of the traffic is data, the capacity of the system is increased from 1.1 Erlangs to 1.8 Erlangs, an

improvement of 60%.

3.0-r----------------------------, "" ~ 2.5 '" .Q

"'" ~ 2.0

'" ;S .~

1.5 ~

~ <:)

~ 1.0

" ~ 0.5 ~ ~

0.0 0.00 0.25 0.50 0.75 1.00

Proportion of data users

Figure 2 - Offered traffic supported at 2% blocking (m = 4, 1I1lv = 100 s, 1IIld = 10 s)

5.0 Simulation Results

Analytical results are more difficult to achieve for larger systems. To gain results for a more realistic server size, we simulated an m = 35 channel system with a minimum rate of rmin = 113. The simulations used a mean call duration for voice of I/llv = 200 s, a full rate of 9.6 kbps, and a mean packet size for data of 96 kbits (l/Ild = 10 s). Two performance measures were evaluated: blocking probability and message wait time.

5.1 Blockin& probability.

In Figure 3, we have plotted.the blocking probability for two offered traffic loads before and after incorporating adaptive data rate communications. These two traffic loads are 33.5 and 29.7 Erlangs, corresponding to conventional system blocking probabilities of 10 and 5 per cent, respectively. The abscissa of Figure 3 is the fraction of total offered traffic attributed to data users.

For example, if the proportion of data users is 0.3, the fraction of offered traffic (Alll) comprising data users is 30% and the fraction comprising voice users is 70%.

82

10-1

, , I

I;J...... I I i I I ........

1 I I I I ! f--..-;.

10-2

....... i ;r- I I

10-3

r---;;.... ........ ! I- ----- 33.5 Erlangs, constant rate I ~ ~ ~

~

r= --0- 33.5 Erlangs, adaptive rate

l= "-I- -.k-- 29.7 Erlangs, constant rate

"-I-I ,''''-I- -----tJ- 29.7 Erlnngs, adaptive rate

I 1 I

10-4 ! I I

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

Proportion of Data Users

Figure 3 - Probability of Blocking with Adaptive Rate Modem in CDMA System (Simulation, m = 35, lIJlv = 200 s, lIJld = 10 s)

10%

5%

2%

1%

Note that the blocking probability decreases as the proportion of data users increases. This result satisfies our intuition, since a higher proportion of data users means we have more flexibility and "headroom." The reduction in blocking probability can be quite dramatic, especially when the offered traffic is low or the proportion of data users is high. For example, at an offered traffic of 29.7 Erlangs (5% blocking before adaptive rate), and 80% data users, the blocking probability is reduced by a factor of 50.2

5.2 Message Wait Time.

Lower blocking does not come free. Because data users must lower their bit rate to avoid blocking, a congested adaptive rate system will have a net increase in message wait time over the equivalent conventional system. In Figure 4, we have plotted the mean message wait time for two traffic levels, 33.5 and 29.7 Erlangs, corresponding to conventional system blocking of 10 and 5 per cent, respectively.

2The reader may notice that our model allows a maximum of 35-3 or 105 channels but the 1S-95 system is limited to 61 Walsh codes for traffic (plus one each for pilot, sync, and paging). This problem does not exist on systems that use a larger, non-orthogonal, code set like the Gold codes.

83

The reference mean wait time for the conventional system is 10 seconds (96k bits at 9.6 kbps). Note that a congested system with 33.5 Erlangs of offered traffic has a mean message wait time of between 6 and 20 seconds, depending on the proportion of data users. For lower traffic loads and/or lower proportions of data users, the mean wait time can actually be less than the reference

mean wait time.

22.0

20.0 --0-- 33.5 Erlangs

i 18.0

8 16.0 ~

--l!r-- 29.7 Erlangs

.s

.~ 14.0

.1::: 12.0

~ :: 10.0 <:l

~ 8.0

+---~;C-----_"""7":;" __ Wait time/or constant rate system

6.0

4.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Proportion of Data Users

Figure 4 - Mean Message Wait Time for Adaptive Rate Mobile Data Terminal in CDMA System

(m = 35, 1/llv = 200 s, l/Ild = 10 s)

Thus, there is an inherent tradeoff between wait time and blocking. When the proportion of data users is low « 25%), even a congested system with 33.5 Erlangs will have a net reduction in message wait time. But the blocking probability will be highest. As the proportion of data users increases, the message wait time increases while the blocking probability decreases.

5.3 Net Capacity Improvements.

From Figure 3, we see that an adaptive data rate system can reduce the blocking probability of a power-controlled CDMA system with 33.5 Erlangs offered traffic from 10% to 2% or less, depending on the proportion of data users. Consulting (3), or an Erlang B table, we find that 2% bloc.king on a conventional system corresponds to 26.4 Erlangs offered traffic. Thus, if the operator's standard is 2% blocking, we have increased the capacity of the cellular system from 26.4 Erlangs to 33.5 Erlangs, a gain of 27%.

84

6.0 Conclusions

By adapting the data rate and power of mobile data terminals, we can reduce blocking probability on a power-controlled CDMA system. The achievable reductions depend on the offered traffic level and the proportion of mean data users on the system. When the offered traffic is 33.5 Erlangs and the proportion of data users is 20%, blocking probability is reduced by a factor of 6. Message wait time is now a random variable and congested systems may cause delays to data users of up to a factor of two relative to conventional CDMA systems. If the cellular operator has a maximum blocking probability standard of 2%, this approach can increase system capacity by 27% on a typical 1.25 MHz CDMA system.

7.0 Acknowledgement

Ken Budka of AT&T Bell Laboratories provided several helpful suggestions on solving the analytical problem posed in Section 4.

8.0 References

[1] A.M. Viterbi and AJ. Viterbi, "Erlang capacity of a power controlled CDMA system," IEEE JSAC, Aug '93

[2] TIA IS-95, "Mobile station-base station compatibility standard for dual-mode wideband spread spectrum cellular system," Telecommunications Industry Association, July 1993.

[3] D.R. Cox and W.L. Smith, Queues, London: Methuen & Co., 1961.

[4] S.M. Ross, Introduction to Probability Models, New York: Academic Press, 1985.

[5] A.J. Viterbi, CDMA Principles of Spread Spectrum Communication, Reading, MA: Addison­Wesley, 1995.

9 A Simulation Study of Interference and SIR in Integrated

Voice/Data Wireless DS-CDMA Networks

Kaushik Das Salvatore D. Morgera

Information Networks and Systems Laboratory

Department of Electrical Engineering, McGill University

3480 University Street, Montreal, Quebec, Canada H3A 2A7

E-mail: {kaushik.sal}@brahms.insl.mcgill.ca

Abstract

The objective of this paper is to characterize the multiple access interference in a DS-CDMA integrated voice/data wireless network. The simulation model is designed for stream-based ser­vices and, consequently, accommodates voice calls and data calls of the long file transfer type. Our simulation results characterize a short term variation of uplink interference power. A key quantity that we investigate is the distribution of the cumulative instantaneous interfer­ence power. Another measure that we study is the fluctuation of the signal-to-interference ratio (SIR) at the base station. We also evaluate the SIR statistics in terms of cumulative distribution functions. Studies of this type are important for capacity enhancement in DS­CDMA wireless networks as they can lead to multiple access interference models that can be effectively employed in bandwidth management systems and in interference estimator-canceller algorithms.

1 Introduction

An integrated wireless network will support different services such as voice, facsimile, and asyn­

chronous data. The general aim is to transport these services in a seamless and efficient manner.

The critical issue for wireless networks is the multiple access technique which allows many unco­

ordinated users to share a common bandwidth. Code division multiple access (CDMA), a spread

spectrum signaling technique, is currently being proposed and implemented for such networks. In CDMA, each user transmits signals through a channel by employing an assigned spreading code

sequence and occupies the same entire allocated frequency band. We focus our attention on the

use of direct sequence CDMA (DS-CDMA) for an integrated multiservice wireless network. In

DS-CDMA, a carrier is phase modulated by a baseband information sequence and a much higher

rate pseudorandom (PN) spreading code sequence. The ratio of the spreading code rate to the

service rate is an important parameter and is known as the processing gain.

The primary purpose for using CDMA is to achieve high system capacity, defined as the number of simultaneous users per cell. CDMA wireless network capacity is limited by multiple

access interference. The crosscorrelation between any two arbitrary PN sequences contributes to

the level of multiple access interference. Nonzero crosscorrelation results because PN sequences

J. H. Reed et al. (eds.), Wireless Personal Communications

© Springer Science+Business Media New York 1997

86

chosen are typically not orthogonal due to bandwidth constraints. The total multiple access

interference power received is determined by all simultaneous users having nonzero crosscorrelation

functions. Also, the near-far problem is a major concern that arises in such a network and is due

to the fact that the power levels of received signals are very dissimilar. The usual remedy is to use power control; however, ideal power control is difficult to achieve in practice.

The objective of this paper is to characterize the multiple access interference in a DS-CDMA

integrated voice/data wireless network. The multiple access interference generated in a CDMA

system is a wideband, nonstationary stochastic process. The randomness of the interference comes

from three stochastic sources, namely, radio propagation, traffic variation, and mobile distribution. Due to the specific nature of the multiplexing method used for CDMA networks, the long term

average interference level depends to a large extent on certain general characteristics associated

with the entire user population; whereas, the short term, instantaneous interference level depends

on the temporal variation of the carried traffic, the spatial variation in portable terminal position,

and the temporal variation in voice activity. The fluctuation of the interference also depends

on certain CDMA system implementation details, e.g., the approach taken to mitigate fading, shadowing, and path loss through power control.

In this paper, we present the results of a detailed and precise simulation study and assess

the impact, of traffic parameters, voice activity, and spatial distribution of portables on uplink

(mobile to base) interference. The simulation model is designed for stream-based services and,

consequently, accommodates voice calls and data calls of the long file transfer type. We believe

that studies of this type are important for capacity enhancement in DS-CDMA wireless net­

works as they can lead to multiple access interference models that can be effectively employed in

bandwidth management systems and in interference estimator-canceller algorithms.

This paper is organized as follows. In Section 2, we present the system description and different

models used in our study. The simulation parameters and details are described in Section 3. The

results are discussed in Section 4. Finally, Section 5 contains our conclusions and remarks.

2 System Description and Models

2.1 DS-CDMA Wireless Network Model

We consider a DS-CDMA wireless network consisting of large number of mobile users. The whole

service area is divided into cells and each cell is served by a base station. The users communicate

via a radio link with one or multiple base stations interconnected to a mobile telephony switching office (MTSO) [1]. The same radio channel is reused in every cell. The following outlines the

general features of the system model for a DS-CDMA wireless network.

1. The cells are drawn with traditional hexagonal geometry as shown in Fig.l(a). The base

station and mobile user antennas have omnidirectional beam patterns.

® 'in e R BS

(a) (b)

MS: Mobile Station

BS: Base Station

Figure 1: (a) Layout of hexagonal cells (b) Enlarged view of a single cell

87

2. Separate frequency bands are used for the uplink (mobile to base station) and the downlink

(base station to mobile) channel. This ensures that the base stations experience interference

only from the mobile stations and the mobile stations experience interference only from the

base stations.

3. The downlink waveform is designed to carry a pilot signal for synchronization and power

control purposes. Power control is discussed in the sequel.

2.2 Spatial Distribution of Mobiles

The users are randomly located in the simulated system area. The location of mobiles is denoted

by (r,O) • The probability density functions of the polar coordinates are

(1)

p(O) = 211"' 0 ~ 0 < 211" (2)

where R is the radius of the circle that circumscribes the hexagonal cell. The parameters are shown in Fig.l(b).

2.3 Radio Propagation Model

The radio propagation is largely influenced by three nearly independent factors: path loss, shad­

owing and multipath fading [2]. Path loss is the deterministic attenuation of the signal power. It

is due to rm , the distance between the mobile unit and the fixed station and is proportional to

r;;''', where 1) is the path loss component. Typical values of path loss component 1) are 2, 3 and

4. Shadowing is the random variation of the signal due to local topographic conditions, antenna height, and environmental conditions. It is modeled by a log-normal random variable. Usually,

88

-pi e

-TI 1- e

-TI e

Figure 2: Two State Markov Model for Voice Activity

the standard deviation (TdB of log-normal shadowing is expressed in dB instead of natural units

and typical values range from 5-12 dB. Multipath fading is the result of signal ;efiections from

boundaries within the mobile environment. It causes the instantaneous envelope of the received

signal to be Rayleigh distributed. With these assumptions, the received power from a mobile at distance rm can be expressed as in [3]

(3)

where U is a Rayleigh distributed random variable with unit power, ee accounts for the shadowing

(~is Gaussian with zero mean and variance (T2), Kr;;." is the path loss, and PT is the transmitted

power. The variance (T is related to (TdB by the relationship (T = (0.110g.,1O)(TdB. Here, we assume

that the effects of Rayleigh fading are averaged, and, thus U is taken to be unity.

2.4 Service and Traffic Models

In this study, we consider voice and stream-based data services. For stream-based services, a

call is continuously transmitted over a relatively long duration. Though voice is a stream-based

service, it is modeled as being intermittent with a certain duty factor cr, in which conversational

speech is characterized by periods of activity called talkspurts and periods of silence [4]. This is

a two state Markov model as shown in Fig.2. The talkspurt and silence periods are independent

and exponentially distributed with means of 1/(3=1.0 second and 1/;=1.35 second for a voice

activity factor of cr= 1/(1 + 1.3)=0.425 or 42.5% [5]. Mobile terminals incorporate a voice-activity

monitoring technique to exploit this burst source model. This implies that the transmitter is not

active or is transmitting at reduced power (and rate) during silent periods in human speech [1,

6].

Stream-based data services include data services for which a large amount of digitized infor­mation is transmitted over a relatively long duration. The applications are facsimile and long file

transfer. These data services require continuous transmission on a constant power basis during

the entire service duration.

89

Figure 3: Power control schematic and interference geometry

In the integrated network, traffic is generated by voice and stream-based data users. A large population of users is assumed in each cell and a small number of this population needs service

at any time. The appearance and disappearance of the users are taken to follow a Poisson

distribution. An idle voice user generates new voice calls at the rate of Av (calls/sec/user) and an

idle data user generates new messages at the rate of A,d (calls/sec/user), both with exponential

inter-arrival time. Another important parameter for voice and data calls is the call holding time or

service duration, denoted by Tv and T,d, respectively. Holding time is a random quantity having

a negative exponential probability density function. The quantities AvT" and A,dT,d denote the

offered average traffic measured in Erlangs for individual voice and data users, respectively [7].

2.5 Power Control Model

Power control is essential in any DS-CDMA wireless network to mitigate the near-far problem.

This problem arises because the transmitters in a cell are located at different geographical dis­

tances from the receiver and the received powers at the base station are unequal. The power

control model we have chosen is taken from [1] and is very effective in combating path loss and

shadowing. In this model, the individual mobile adjusts its own transmission power in accordance

with the received downlink pilot signal level. Also, because of shadowing, when a mobile initiates

a call, it does not necessarily mean that the call will be served by that cell. The home cell for

the call will be chosen as the one with the smallest radio propagation attenuation to the mobile.

In fact, the mobile measures the power levels of pilot signals from all nearby base stations and is

connected to the base station that require minimum transmission power. Thus, the base station index k with which the mobile establishes radio link is chosen by the following equation

(4)

where rm is the distance of the mobile from the mth cell site and em, m = 0, 1, 2,. ,are independent Gaussian random variables with variance u. Note that the power control algorithm

90

Table 1: Simulation Parameters Used in the System Model

Item Symbol, unit Value Cell radius R, meter 200 Total number of cells N, - 37

Propagation path loss coefficient 1/, - 3

standard deviation of shadowing O'dB, dB 8

Average length of a voice call T., seconds 180

Average length of a data call T.dl seconds 100

Arrival rate of new voice calls .x., calls/sec/user 0.002, 0.004

Arrival rate of new data calls .x.d, calls/sec/user 0.004. 0.008

assumes reciprocity between uplink and downlink channels from propagation loss point-of-view.

In this study, the power control is implemented by scanning the forward link signal from seven

closest base stations as shown in Fig. 3.

3 Simulation Parameters and Schemes

The results of our study have been obtained by computer simulations. Table 1 summarizes the

simulation parameters used in this paper.

The outputs of the simulation are interference and SIR. We compute the interference power

received at the central base station. The uplink SIR is determined for an active mobile located

within the central hexagonal cell, and the mobile is assumed to be connected to the central base

station. The reason for computing the interference power and SIR at the central base station comes

from the assumption that the uplink performance of any mobile in the central cell approaches that

of any cell in an infinite network. This implies that a computation of this type will be applicable

to any mobile located in the system area.

The users are randomly located in the simulated system area with a uniform density of Mv voice users and M.d stream-based data users per base station. The uplink interference consist

of interference from mobiles within the same cell and, in addition, interference due to mobiles in

surrounding cells. They are called intracell, lint~ .. , and intercell, lintel" interferences, respectively.

The total interference I can be written as I = lint~ .. + lint... + Nth, where Nth is the background noise. Since we consider an interference limited case Nth is set to zero.

The simulation program generates call arrivals and departures for individual user in the time domain, Le., the simulation of calls from initiation to termination. It randomly selects a location for a call and computes the transmitted power assuming that the power control is in effect. The

received power at the central base station is then computed by considering the path loss and

.. u

... ~ j"r r-• I'·' I" ..

'.1

(~

" .. "00

[ ,. L. ~

•

r-r-r--r-r-

r-

n a ~ M M m m ~ m m ~ -(»

91

Figure 4: Simulation specifics: (a) Received power from a mobile at home base station ft)r a simulated

voice call (b) Histogram for simulated radius due to cell geometry.

shadowing. Fig. 3 shows the geometry for computing the power received at one base station (cell

#1) from a mobile station at distance d away in another cell (cell #2).

The duration of time simulated ranges up to 16000 seconds. Simulation clock resolution is

equal to 10 milliseconds. There are two simulation programs: simulation progmm-l and simulation progmm-2. The latter program employs a voice activity monitoring technique for voice users, whereas the former program does not. Other parameters are shown in Table 1. The traffic is

varied by changing the number of users per cell. The results were taken after the system has

reached steady state.

The simulation of voice and data calls follows our service model. Considering the effect of

voice monitoring, the power from each voice call becomes intermittent, i.e. Pv during talkspurt

and dpPv during silence period, where 0 < dp < 1. The reduced power is necessary in order to

maintain link synchronization. Here, the value of dp is set to 0.125. The transmission rate will

also be reduced during the silence period of a voice call in a DS-CDMA network. It is also found

that, in a call of 100 seconds in length, there are approximately 39-47 talkspurts which conforms

to the number given [5]. A typical simulated voice call is shown in Fig.4(a). The stream data

users, on the other hand, require continuous transmission with power P.d during the entire service

duration. The power levels mentioned here refer to the power at the receiver input point of the

base station in the home cell.

The coordinates of the mobile are randomly picked using (1) and (2). The physical layout of

the hexagonal geometry prevents the radius from being picked exactly as in (1). The resultant histogram is shown in Fig.4(b).

92

~r---~-----r----~----r---~-----r----~

18

~~--~~~--~1~~--~1~~--~~~--~~~--~~~~~~~ TIme In seconds

Figure 5: Temporal variation of interference power from simulation program-l as a function of voice

and data users. (a): {Mv=lO. M.d=5}. (b):{Mv=50. M.d=20} and (c): {M,,=lOO. M.d=40}.

4 Interference and SIR - Results

Instantaneous interference power is computed as a function of {Mv, M .. t} to assess the impact

of traffic parameters, voice activity, and spatial distribution of mobiles. The signal powers from

each interferer are summed at the central base station to determine short-term interference power.

Sets of users representing the different services have been chosen such that they categorize two

classes: one set {M" = 10, M.d = 5} represents a small number of users and the other two

sets {Mv = 50, M.d = 20} and {Mv = 100, M.d = 40} represent a large number of users. The

total instantaneous interference power is plotted at a resolution of 100 milliseconds from the two

simulation programs mentioned above. The reason for investigating two scenarios is that we are

also interested in characterizing the interference as a result of voice activity.

Fig.5 shows the instantaneous interference power from simulation program-I. It shows con­

siderable short-term variation of interference power. This variation is attributed to the sudden

arrival and departure of mobiles and the spatial distribution of mobiles. Fig.6 shows the total

interference power from simulation program-2. It shows that interference power fluctuates much

more rapidly due to the impact of voice activity. In either case, the interference is certainly a

non-stationary process.

An important observation is the variation in interference power. It is defined as the difference

between the highest and lowest received interference power. It is desirable that the variation be as

low as possible. The variation is best observed from Fig. 7 which plots the cumulative distribution

Ay = 0.002 Asd= 0.004

~O~---=~~--~'OOO~--~'~~---=~~--~2=~~--~~~--~~ Time in seconds

93

Figure 6: Temporal variation of interference power considering voice activity as a function of voice

and data users. (a): {Mv =10. Mod=5}. (b): {Mv=50. M.d=20} and (c): {Mv =100. M.d=40}.

function (cdf) of interference power. It is seen that the variation in interference power is relatively

large for small number of users. As the number of users in the network increases, the variation

decreases. This is indicated by the cdf's for {Mv = 50, M.d = 20} and {Mv = 100, M.d = 40} as

compared to the cdf for{Mv = 10, M.d = 5}.

Another measure that we study is the fluctuation of the signal-to-interference ratio (SIR) at

the base station. The SIR of the uplink received signal is important in determining the radio

link performance of any active user. Fig. 8 shows the short-term temporal variation of uplink SIR

at the base station during a time interval of three minutes. Results are plotted and compared

from two simulation programs. Dashed lines shows the results from simulation program-l whereas

the solid lines show the values of SIR considering voice activity monitoring, i.e., from simulation program-2. It is observed that voice activity monitoring technique improves the average SIR in

the mobile to base link, but it also introduces rapid variation of SIR over time. Since the average

length of a voice call is three minutes, it shows that the uplink SIR of an active talking mobile

varies many times during each call.

In order to examine the effects of individual arrival rates on the variation of SIR, we formulate

an experiment to simulate equal traffic in a single cell for a different number of users. For this,

we assume that the total arrival rates in a cell can be written as MvAv and M.dA.d respectively.

We have chosen Mv= 50, 100 and M.d= 20, 40 to represent a large number of users and the total

traffic simulated is 36 Erlangs/cell for voice users and 16 Erlangs/cell for data users, respectively. The cumulative distribution function of SIR is plotted in Fig.9. It is observed that the variation

94

:::::::::1::::::::r:::::::~::':::::;:'::::::::::::<IHZf/:,;.lt/J//· : ····T/······T·····rr······T········

···;··········v·······y····r·r·······r······· if ....... ; ....... )" ........ ; .... {-- .. ; ......... ; ........ .

: I: .:

0,) --,' ~ ; (c} ~ ;10-' ......... ; ............ ::::::~:::::::::;:::::::::~::::::~ ::~:::::::::~: ::,e:::: i:::::::::~::::::::. ~ ......... ; .... ::: :i:::::::::::::::::::::::::::::::::::::::::::::::::::: ;::::::::::::::::~::::::::: J :::::::::t:::::' .::t:::::::::t:::::::::t:::::::::~::::::'::::::::::::::~::r::::::L:::::::t:::::::: ... ::::: I: : I : :

......... ! .... ····~·········'·········:··::::·::I:::J.:::::::::::::::;:::::::L:::::::;:::::::::

.. ! ............. + .. ...j[ AV = '0.002 J ' :i:. A sd= 0.004

~ 10~L-__ ~ __ -L __ ~ ____ L-__ ~ __ -L __ ~~ __ ~ __ ~ __ ~

-2 0 2 6 8 10 12 14 16 18 Interterence ""- I (dB)

Figure 7: Cumulative distribution function of interference power as a function of voice and data users.

(a): {Mv =10, M.d=5}, (b): {Mv =50, M.d=20} and (c): {Mv =100, M.d=40}.

in SIR is nearly equal for two cdf's. This appears to substantiate the fact that variation of SIR

is due to the total traffic per cell rather than to the individual arrival rates of different services.

5 Conclusions

This paper has investigated the multiple access interference in an integrated voicel data network.

The simulation results demonstrate the temporal variation of uplink interference power. Different

factors such as arrivals and departures of cochannel portables, voice activity, and spatial distri­

bution of mobiles contribute to such temporal variation. The interference can be modeled as a

non-stationary process. It is observed that the variation in interference power is larger for a small

number of users per cell, as compared to a relatively large number of users per cell.

Temporal variation of interference has a considerable impact on radio link performance of

any active user. The SIR is found to fluctuate many times during a typical call duration. The

variation of SIR is seen to be dependent on total traffic per cell from different services.

There are other issues in a practical network which appear to further contribute to this non­stationarity. These are multi path fading; non-uniform traffic, known as hot-spot traffic; and non-ideal power control. The packet-based data services in a future generation wireless network

will also have considerable impact on interference in a DS-CDMA network [8]. Work is currently

underway to further understand the interference resulting from packet-based services.

Or---~--~--~---r---T--~----~--r---,

-2

-~O~--~=---~~~--~=---~~~~I~~~~I~~~I~~~~I~~~I~ Tlmeln ..... nds

95

Figure 8: Temporal variation of SIR (resolution-IOms) as a function of voice and data users. (a):

{M.=IO. M.d=5}. (b): {Mv =50. M.d=20} and (c): {M.=IOO. M.d=40}.

There are many important applications of this study. An important issue in DS-CDMA networks is the interference estimate that would form the input to a bandwidth management system (BMS), the output of which would be call admission and data congestion control policy decisions.

Another important applica.tion is in the area of interference cancelation strategies. Since a DS­

CDMA network is interference limited, interference cancelation methods will certainly improve the capacity. A direction seems to be toward interference estimation and cancelation algorithms,

where the challenge requires wideband, non-stationary interference estimation for subsequent cancelation.

Acknowledgments

This research was supported by a grant from the Canadian Institute for Telecommunications Re­search under the NCE program of the Government of Canada. The first author also acknowledges the financial support from the Canadian Commonwealth Scholarship Plan.

References

[1] K.S. Gilhousen, I.M. Jacobs, R. Padovani, A.J. Viterbi, L.A. Weaver, Jr., and C.E. Wheat­

ley III, "On the Capacity of a Cellular CDMA System", IEEE Transactions on Vehicular Technology, Vol. 40, pp. 303-312, May 1991.

96

l~r.-... -.. -... T •. -.. -... -.. , .• -... -.. -.. , ..• r .. -.. -... -.. r,.-... -.. -.. T.,.-.. -.. -... ',.-.~ ... ~.~~_~.~~ .. -.. r.-.. -... -.. T •. -... -.. -' ...

:::::::::~::::::: ::~:::::::::~:::::::.:~:::::: :::~ ::::::'. ~.: ;{:-::~::::::::: ::::::::~::::::::: : ::: ::: ::t::: ::::::~ ::::: ::::t:::::::: t: :::: :::~::: ': .:: ::~/:::: ::r:::: ::: ,-, ..... ! ........ .

: : : : : ~ : ........ ~ ...... ' .. . : ......... ~ ......... ~ ......... : .......... :- """;/~"""' ... : ................. ~ ........ .

......... , ......... , ......... , ......... , ....... , ..... ""/"., .......... , ........ . I m [' ___ j+/i m

[, AV = 0.004];···· .. · ..

i 10-' ............ ~~.~.O::.,=:::::::v::·::::=.:::::::±::~7::~:~;~:I:::::::: f •••••••• t •• : ••• :: •• : ••• : •• : ••••••• :.: •• : •• ::t.: ••• : •• t::.: ••• :i: ••••• : •• L •• : ••• i: •• :.:: •• l ::::: ::::l::: :::: ::t: ::: :::::t:::::::::: : ::::: it :::: ::: ::1::: :::: :::t::: :::::1:::: :::::t ::::::::

: I;::;: ......... ! ......•• ! .•••..•.• ,.... . ... " •. : ...••••• , .• : •••••••.•• : •••••.... ~ ......... ~ ••••...•. I: . : . .

....•........ 1 .•. ' ..• . r· r :

: 1 104L-__ ~ __ ~ __ ~~~ __ ~ __ ~ __ ~~ __ ~ __ ~~

-22 -21 -20 -19 -18 -17 -18 -15 -14 -13 -12 Signal 10 Intorterence Ratio (SIR). (dB)

Figure 9: Cumulative distribution function of SIR for equal traffic (Erlangs/cell) with two sets of

individual arrival rates for different services.

[2] W.C. Jakes, Jr., Ed., "Microwave Mobile Communications", New York: Wiley, 1974.

[3] M. Zorzi and R.R. Rao, "Capture and Retronsmission Control in Mobile Radio", IEEE

Journal on Selected Areas in Communications, VoL 12, pp. 1289·1298, October 1994.

[4] P.T. Brady, "A Statistical Analysis of On-Off Patterns in 16 conversations", Bell Systems

Technical Journal, VoL 47, pp. 73-91, January 1968.

[5] N. Amitay and S. Nanda, "Resource Auction Multiple Access (RAMA) for Statistical Multi­plexing of Speech in Wireless PCS", IEEE Transactions on Vehicular Technology, VoL 43,

pp. 584-596, August 1994.

[6] N. Guo and S.D. Morgera, "The Grode of Service for Integroted Voice/Data Wireless DS· CDMA Networks", Proceedings of ICC '94, pp. 1104-1110, May 1994.

[7] N.D. Wilson, R. Ganesh, K. Joseph and D. Raychaudhuri, "Packet CDMA Versus Dynamic TDMA for Multiple Access in an Integroted Voice/Data PCN", IEEE Journal on Selected

Areas in Communications, VoL 11, pp. 870-893, August 1993.

[8] N. Guo, S.D. Morgera and P. Mermelstein "Common Packet Data Channel (CPDC) for Integroted Wireless DS-CDMA Networks," to appear in IEEE Journal on Selected Areas in

Communications, VoL 14, May 1996.

10 Comparative Study of Adaptive Antenna Arrays

CDMA Communication Systems

Richard B. Ertel Dept. of Electrical Engineering Pennsylvania State University

Stephan V. Schell * Dept. of Electrical Engineering University of California, Davis

March 24, 1996

Abstract

. In

In this paper adaptive algorithms that may be used for spatial filtering or spatio-temporal

filtering in a direct sequence code division multiple access (DS-CDMA) communication system are

compared on the basis of signal-to-interference-and-noise-ratio (SINR) improvements, algorithm

convergence rates, capacity enhancement, and computational complexity. Algorithms that have

already been proposed as well as some new algorithms are presented for this application.

These algorithms fall into two categories: (l) conventional algorithms and (2) channel es­

timation algorithms. The conventional algorithms use knowledge of a desired reference signal

and measurements of the received signal to adapt the antenna weights. In contrast, the chan­

nel estimation algorithms use properties of the underlying code structure of DS-CDMA systems

to estimate the array response vectors. The estimated array response vectors are then used to

determine appropriate antenna weights, either directly or via Capon beamforming.

1 Introduction

Rapidly growing demands on wireless communication systems create the need for techniques that

increase system capacity. One method which has been shown to significantly enhance system

capacity is the use of adaptive antenna arrays in CDMA communication systems. [1].

While there has been a number of proposed adaptive algorithms for this application, the

absence of published comparisons of these techniques hinders practical engineering efforts such

as cellular system design. An objective of this paper is to remove this obstacle by presenting a

comparative study of these algorithms on the basis of signal-to-interference-and-noise-ratio (SINR)

*The work of the second author was supported in part by the National Science Foundation under grant MIP-9S-12732 and the U.S. Army Research Office under contract number DAAH 04-95-1-0328 (PI: W. A. Gardner).

J. H. Reed et al. (eds.), Wireless Personal Communications

© Springer Science+Business Media New York 1997

98

improvements, algorithm convergence rates, capacity enhancement, and computational complexity,

primarily with the use of extensive simulations.

2 System Modeling

This work considers the adaptation of a uniform linear antenna array containing M elements,

with an element spacing denoted by d. The signal, set), is incident upon the array at an angle ()

measured from the array normal. Assuming that the small time delays across the antenna array

cause only a phase shift in the carrier frequency, the signal present at the ith element may be

expressed as Xj(t) = s(t)ei(i-l),p using complex envelope representation, where 'if; = 27r~ sin((}) and

A denotes signal wavelength. Expressing the input signal in vector form gives x(t) = s(t)a((}),

where a((}) = [1 eN ei2,p ... ei(M-l),pV is the antenna response vector. In a CDMA system,

the incident signal, set), will be a sum of the signals from the various transmitters present plus

background noise. With N transmitters, and Ln multipath components of the nth transmitter the

received signal vector may be expressed as

N Ln

x(t) = L La((}n,l)an,lei<l>n"un(t - Tn,!) + net) (1) n=ll=l

where (}n,[, Tn,/' an,l, and ¢>n,1 are the angle of arrival, delay time, path attenuation, and path

phase of the lth path of the nth transmitter respectively (2). The additive noise interference, n( t) is

modeled as independent Gaussian noise present at each element. The code-modulated information

signal of the nth user is given by un(t) = bn(t)cn(t), where bn(t) and cn(t) are the information and

code sequences respectively, samples of which are taken from the set {± 1}.

Both dominant path and multipath reception are considered. For dominant path reception a

standard memoryless spatial filter may be used. In this case, the arraY' output may be expressed

as y[k) = wH[k)x[k), where x[k) is the input signal vector, and w[k) = [wl[k) w2[k) ... wM[kW is

the antenna weight vector. For multipath reception, a 2D RAKE receiver is used (3). A 2D RAKE

receiver is a bank of B beamformers, each of which is adapted to a different multipath component.

The outputs of the beamformers are then combined using a standard RAKE receiver.

99

3 Conventional Array Techniques

3.1 Sampled Matrix Inversion

The SMI algorithm minimizes the mean squared error between the desired array output, d[kJ, and

the actual array output, y[kJ. Defining the sampled correlation matrices, Rxx = kXXH and K

rxd = k 2: x[kJd[k]' the optimum weight selection becomes w[kJ = R;,;rxd [4J. k=l

To use the SMI algorithm with the 2D RAKE receiver, the beamformer weights are found

using the SMI algorithm, only with a delayed reference signal. The RAKE receiver weights may then K

be estimated from the scaled cross-correlation matrix, Krxd = 2: x[kJd[k], which was calculated k=l

when determining the beamformer weights. The inverse matrix calculation of R;,; only needs to be

calculated once for alI of the beamformers of each user, since the matrix will be nearly independent

of a time shift of a few chip durations.

3.2 Recursive Least Squares Algorithm

In [5J, Wang and Cruz present the use of the Baird method of the recursive least squares (RLS)

algorithm. Defining the sampled matrices, X = [x(1),x(2),··· ,x[K]], and P[kJ = X[k]XH[k],

the iterative relations for calculating w[kJ after the kth sample become:

1 P(k + 1) = l.[P[kJ _ P[k]X'(k+1)XT (k+1)P[k]J • f3 f3+x' (k+l)P[k]x'(k+l)

2 (k ) - [kJ P'[k]x(k+l) [d(k ' H[kJ (k )J* . w + 1 - w + f3+x H (k+1)p'[k]x(k+l) + 1) - w x + 1 .

3.3 Least Mean Squares

In [6J the use of the LMS feedback loop for adaptive arrays in biphase-modulated spread-spectrum

systems is presented. The LMS algorithm attempts to minimize the squared error between the

array output and the desired reference signal [4J by estimating the gradient of the mean squared

error function with respect to w[kJ and iteratively adapting the weights in the direction of steepest

descent. Defining the error sequence e[kJ = d[kJ -y[k], the LMS weight equation becomes w[k+ 1J =

w[kJ-,8V, where V = -2e[kJx[kJ is the estimated gradient and,8 is a positive real scaling constant.

The extension of the LMS algorithm for 2D RAKE receiver adaptation is presented in [7J.

3.4 Decision Directed Least Mean Squares

The use of the decision directed least mean squares (DDLMS) algorithm is presented in [8]. With

DDLMS, the array output is demodulated and a binary decision is made. The decision is used

as the desired reference signal, d[p], at the bit rate, where d[P] is the the pth decision. The error

100

signal at the bit rate, e[p) is taken as the difference between the reference signal, and the despread

array output, y[P). The average despread signal present at each of the antenna elements is found as pGc

vo[P) = L: X[k)CI[k). The adaptive weight is then calculated as w[P+1) = w[p)+.8e[p)*vo[P). k=(p-l)Gc+1

4 Channel Estimation Methods

Channel estimation algorithms use knowledge of the" special code properties of spread-spectrum

signals to obtain estimates of the channel parameters. Consider the reception of the /th path of the

user indexed by n = 1. By setting the adaptive array weight vector equal to the estimated array

response vector, i(91,1), the beamformer will point in the desired direction yielding a corresponding

increase in SINR. For improved performance, Capon beamforming, which minimizes array output

variance while maintaining unity gain in the direction of the desired user may be used [2).

4.1 Channel Estimation by Code Filtering

Channel estimation algorithms are presented in [2) [3) [9) [10) [11) and [12). These techniques

are based upon a code filtering step in which the received signal vector, x( t), is correlated with

the local code sequence, c(t), over intervals of n, the bit duration. With n = 1 indexing the

desired user, the code filter output for the pth data bit of the sampled signals is defined as v~[P) = pGc+~-1

L: x(kTc)CI(kTc-I)Tc), where I) determines the amount of delay ofthe local code sequence k=(p-l)Gc+~ used in the code filter. Without loss of generality, consider the code filter output for I) = TI,1 = 0,

where TI,1 is the time delay of the direct path of the desired user.

Provided that the variance of the desired signal is sufficiently large, then the output of the

code filter may be approximated as

(2)

Various means of exploiting the approximation in Eq. (2) for estimating channel parameters are

covered in the following sections.

4.1.1 Auto-Correlation Matrix

The first method, referred to as the channel estimation by the autocorrelation matrix (CEAM),

involves calculating the code filter output over multiple bit periods. The code filter matrix is given

by V D( 1)1,1) = [V'll,l (1), vm,,(2),· . " V~", [DJ], whose pth column is the pth code filter output. The

auto-correlation matrix of the code filter matrix is formed, RVV(I)I,I,I)I,I) = VD(I)I,dVZ(I)I,I)'

101

Using the approximation (2) the autocorrelation matrix may be approximated by Rvv( 1]1,l> 1]1,1) ~

Dai,lG~a(91,1)aH (91,1) [2].

The antenna response vector, 80(91,1), is estimated as the dominate eigenvector of Ryy ( 1]1,1, 1]1,1),

and the path amplitude is estimated as the squared root of the dominant eigenvalue. When mul­

tipath combing is desired, each of the required path parameters may be calculated from the cor­

responding code filter autocorrelation matrix, RYY(1]l,l, 1]1,1), It is not possible to determine the

phase using Ryy( 1]1,1,1]1,1) since, when calculating the autocorrelation matrix the phase infor­

mation of the path, .p1,/, is lost. In [2], the use of decision directed methods or training signals is

suggested for determining the phase. An alternative technique is to estimate the phase of the beam­

former outputs by repeating the procedure used to determine the array response vector estimates

on the RAKE receiver weights as described in [7].

4.1.2 Shifted Auto-Correlation Matrix

A method similar to the CEAM algorithm referred to as the channel estimation by the shifted

auto-correlation matrix (CESAM) algorithm exploits the properties of the shifted auto-correlation

matrix RVV(1]l,/,1]l,l) = VD(1]l,I)VZ(1]l,l) [2]. It can be shown that the array response vector,

a( 91,2) may be estimated by the dominant eigenvector of Rvv (1]1,2,1]1,1)'

As with the CEAM algorithm, it is possible to estimate the path array response vectors from

the dominant eigenvectors of the matrix as 1] is varied. Additionally, with the CESAM algorithm

the relative path amplitudes and phases of the multipath components may be estimated directly

from the shifted autocorrelation matrix.

4.1.3 Average Code Filter Output

When a training signal, or decision directed methods are used, then the data sequence may be

modulated onto the code sequence used for the code filter. The corresponding array response

vector may then be estimated as the average of the code filtered output over many bit intervals [3].

In this work, the algorithm will be referred to as the channel estimation by average code filter

output (CEACF).

The average modified code filter output denoted v ~[PJ is given by

DpGc+n-1

v~[P] = L x(kTc)C1(kTc - 1]Tc)b1(kTc - 1]Tc). k=(p-1)Gc +n

(3)

Provided that the variance of the desired term is sufficiently large, than the output of the code

filter may be approximated as vo[P] ~ a(91,1)C'l,lej</>,,1 DGc [2).

102

4.1.4 Matrix of Detected Paths

Another method of using code filtering for channel estimation which was presented in [3] uses

the CEACF algorithm to form a matrix of the average code filter output for each of the detected

paths. This algorithm will be referred to as the channel estimation algorithm via matrix of detected

paths (CEMDP). Since all path estimates are determined together, this method is performing joint

adaptation of the 2D array weights.

The CEMDP attempts to minimize self-interference due to the desired users own multipath

signal by taking into account the cross-correlations of the code sequence with the the multipath

delayed code sequences. The method forms a matrix of the detected paths, VI = [V~", [P] V~I,2[P]

... V~I,LI [PJ)] ~ AIRel where Al = (al,lal,l ... al,L, al,L,) and ReI is the code covariance matrix

defined in [3]. The array response vectors are then calculated by Al = YIR~/ [2].

4.1.5 Channel Estimation by Pre/Post Code Filtering

The channel estimation by pre/post code filtering (CEPPCF) algorithm presented in [9] [10] [11]

and [12] requires both the pre-processing and post-processing array covariance matrices to obtain

estimates of the array response vector a( Bl,I), and the interference and noise covariance matrix,

Q. In [10] it is shown that the preprocessing array covariance matrix may be estimated as Rxx ~ Ln N Ln

a~,la(Bl,l)a*(Bl,l) + Q where, Q = L: aLa(Bl,da*(Bl,l) + L: L: a~,la(Bn,l)a*(Bn,d + (7~I is the 1=2 n=21=1

interference and noise covariance matrix. From Eq. (2) it can be shown that the post processing

(Le., after code filtering) covariance matrix will be given by Rvv = G~a~,la(Bl,l)a*(Bl,l) + GeQ.

U sing the expression for the preprocessing and post processing array covariance matrices, it

can be shown that the array response vectors may be determined from the dominate eigenvector

of the difference, ctRvv - Rxx = (Ge - l)ar,la(Bl,I)aH (Bl,l). Likewise, the interference and noise

covariance matrix may be found as Q = G~:I (Rxx - J~ Rvv). The adaptive array vector is then

calculated as w = Q-I a [9]. ~

5 Simulation Results and Algorithm Comparisons

5.1 Simulation Parameters

For each of the trials, each user was randomly assigned three multipath signals. Each of the

multipath signals was given a time delay that is an integer multiple of the chip duration, te' The

maximum time delays which were assigned were 7te' The direct path of each user was assigned a

path amplitude of one, while all other path amplitudes were selected at random from a uniform

103

distribution from 0.5 to 1. The angles of arrival were randomly distributed between -90 and +90

degrees from the array normal from a uniform distribution.

The code sequences used were maximum length sequences of degree 13. A spreading factor

of 64 chips per bit was used. White Gaussian noise was added to each of the channel signals; the

noise power was 10dB below the direct path signal level. The element spacing was set equal to

half of a wavelength. For the 2D array configurations only seven taps were used. For each channel

condition, 25 trials were averaged to determine the SINR for the given parameters.

To obtain the SINR performance comparisons, the SINR of the array output of each of the

algorithms was calculated for various ranges of users, antenna elements, and adaptation length.

The number of users was varied from one to 100 in increments of ten. With the given channel

conditions, the number of antenna elements was then varied from one to ten. The number of bits

used to process the adaptive weight vector was varied from one to 29. The SINR of the signals over

50 bits was used to calculate the SINR.

5.2 Simulation Results

5.2.1 SINR Comparison

The SINR performance of each of the algorithms was calculated as SINR = R Pcp, where Pc = yy- c

Irl2 K lK K A:d ' ryd = t L: y[k]d[k]*, Rdd = K L: d[k]d*[k], and Ryy = t L: d[k]d*[k]. In each of the k=l k=l k=l

figures, the algorithm is followed by 'ID' or '2D' to indicate whether a single array is used, or

whether a 2D RAKE receiver is being adapted. Likewise, the use of Capon beamforming is indicated

by attaching 'CB' to the end of the abbreviation.

Figure 1 shows the SINR performance of a four element antenna array using 10 data bits for

weight adaptation as a function of the number of users. The 'Input SINR' curve is the SINRob­

tained with an omni-directional antenna pattern. Clearly the greatest loss in performance occurs as

the number of users is increased from one to ten. Adding more users further decreases performance,

but not at the same rate. This is due to the fact that with only one user the only interference

to the user is its own two multipath components and the additive Gaussian white noise. With

so few interference signals, the antenna array will have enough degrees of freedom to null out the

self interference. Therefore, significant performance enhancements are possible. The performance

difference with and without the use of Capon beamforming is also shown in Fig. 1.

The SINR performance of the 2D RAKE receiver algorithms, CEAM, CESAM, CEACF,

CEMDP, CEPPCF, 2DLMS, and 2DSMI, is shown in Fig. 2. Generally, when compared to the lD

array algorithms, the 2D RAKE receiver algorithms perform about 2 dB better. The performance

104

difference amongst the algorithms increases in the 2D case. In Fig. 1 the difference between the

algorithms for more than 20 users is less than 2 dB, whereas, in Fig. 2, the difference is more than

3 dB.

Of the algorithms, the 2DSMI algorithm yields the highest SINR. Each of the channel es­

timation algorithms with Capon beamforming performs nearly identically, the difference in their

SINR performances being less than 1 dB. The CEACF2D algorithm is only about 1 dB to 1.5

dB below the CEACF2DCB when ten or more users are present. Only when one user is present,

is the difference significant, approximately 6 dB, as shown in Fig. 2. Hence, the use of Capon

beamforming is most beneficial when there are relatively few interfering signals present.

5.2.2 Convergence Rates

Plots of the SINR performance as a function of number of bits processed for weight adaptation for

the case of four antenna elements are considered. The convergence rates of the ID array algorithms

can be inferred from Fig. 3 for the cases of 20 users. The primary differences among the algorithms

is the final steady state performance, not the rate of convergence. All of the algorithms converged

to within half a dB of steady state performance after processing between four to six bits.

The SINR performance of the 2D RAKE receiver algorithms, as a function of the number

of bits used for weight adaptation, is shown in Fig. 4. The convergence rates of the 2D RAKE

receiver algorithms are slower than their ID counterparts. Each of the other algorithms requires

closer to five bits to be within half of a dB of steady state performance.

5.2.3 Capacity Enhancement

System capacity is the number of users that can be supported with the given bandwidth allocation,

while still performing at a pre-determined maximum bit error rate (BER). In [IJ, the bit error rate

as a function of SINR is approximated by Pb = Q( ..j3GcSIN Rinputl, where Q(-) is the standard

Q-function. Requiring a BER of less than 10-3 with a processing gain of Gc = 64 gives a required

SINR of .05 (-13 dB).

The SINR simulations, for the case of four antenna elements and a weight adaptation length

of ten bits with varying numbers of users, are used to obtain the corresponding system capacity.

The maximum number of users, which could be present while still providing an SINR of .05 at the

demodulator input, was used to determine the system capacity. Fig. 5 shows the capacity levels that

can be achieved with the various ID algorithms. For a relatively small number of antenna elements,

the capacities are nearly equal, with a difference of only about five users between the best and the

worst performing algorithms. The difference increases as the number of elements is increased,

105

having about a 15-user difference when ten antenna elements are used. The CEPPCF1DCB,

CEACF1DCB, CEAMIDCB, and the SMI/RLS algorithms support nearly an identical number

of users over the range of antenna elements. Likewise, the LMS, DDLMS, and the CEACFID

algorithms support nearly an identical number of users.

The capacity performance of the 2D RAKE receiver algorithms is shown in Fig. 6. Again,

the spread between the capacity of the best and worst performing algorithms increases with the

number of antenna elements. The 2DSMI algorithm consistently achieved the largest capacity.

5.2.4 Computational Complexity

The level of computational complexity involved in obtaining the adaptive weights determines the

required processing speed, complexity of the hardware, and ultimately the cost of the system. In

this work, the computational complexity is estimated by the number of flops which are required

to process the data to obtain the desired adaptive weight vector. The flop definitions given in [7]

were used to determine flop count totals for each of the algorithms being considered.

The 1D array algorithm flop counts are shown in Fig. 7. In the figure, SMI Case 1 shows the

flop counts assuming that the inverse matrix is calculated independently for each user. SMI Case

2 indicates the flop counts required if the inverse matrix is calculated only once for each of the

users. In practice, the computational complexity would be somewhere between the two cases. The

computational complexity of the RLS, CEAM1DCB, and the CEPPCF1DCB algorithms are nearly

identical and are the most computational. The CEACF1DCB and the SMI (Case 1) algorithms are

roughly 5 to 6 dB below RLS and are nearly identical to each other. The DDLMS is only slightly

more computational than the LMS algorithm. The least computational of all the algorithms is the

CEACFlD algorithm.

Of particular interest is that the SMI (Case 2) algorithm is the second least computational of

all the algorithms when ten or more users are present. This is because SMI (Case 2) performs the

most computational part of the adaptation only once for all users in the channel. In practice, the

obtainable computational complexity of the SMI algorithm would fall somewhere between Case 1

and Case 2.

The flop counts of the 2D RAKE receiver algorithms are shown in Fig. 8. As compared

to the ID algorithms, the corresponding 2D algorithms are roughly 10 dB more computational,

with the exceptions of the SMI, CEACF2DCB, and the LMS algorithms, which are only about

5 dB more computational when used for 2D RAKE receiver adaptation. The CEPPCF2DDB,

CEAM2DCB, and the CESAM2DCB algorithms are the most computational of the 2D RAKE

receiver algorithms. This is expected since each of these algorithms must perform SVD for each

106

tap of the RAKE receiver.

The flop counts for the CEMDP2DCB, CEACF2DCB, 2DSMI, and 2DLMS algorithms are

nearly identical, and are roughly lOdB below the SVD based channel estimation algorithms. The

least computational of the 2D RAKE receiver algorithms is the CEACF2D algorithm, which simply

correlates the input signal vector with the local code sequence.

6 Conclusions

Of the algorithms considered, the CEACFID algorithm without the use of Capon beamforming was

the least computationally costly. Even though it was one of the worst performing algorithms, the

span between the best and the worst is not that great. Hence, if a low computational complexity

algorithm is sought, the CEACFID should receive some consideration. For a slight increase in

performance and complexity the CEACF2D algorithm is another good alternative.

For moderately complex algorithms, the 2DSMI offers exceptional performance. If means

of exploiting the inverse matrix for all users could be developed, this would be a very promising

algorithm.

The charmel estimation algorithms, that use SVD as part of the weight adaptation process are

among the most computationally costly of the algorithms. The primary benefit these algorithms

offer is that blind weight adaptation is performed. However, coherent detection is not used, and

hence, system performance suffers slightly.

References

[1] J. C. Liberti and T. S. Rappaport, "Reverse channel performance improvements in CDMA

cellular communication systems employing adaptive antennas," IEEE Proc. Globecom, vol. VI,

pp. 42-47, 1993.

[2] B. H. Khalaj, A. Paulraj, and T. Kailath, "Antenna arrays for CDMA systems with multipath,"

MlLCOM, pp. 624-628; 1993.

[3] B. H. Khalaj, A. Paulraj, and T. Kailath, "2D RAKE receiver for CDMA cellular systems,"

IEEE Pmc. Globecom, pp. 400-404, 1994.

[4] R. A. Monzingo and T. W. Miller, Introduction to Adaptive Armys. New York: John Wiley &

Sons, 1980.

107

[5] Y. Wang and J. Cruz, "Adaptive antenna arrays for the reverse link of CDMA cellular com­

munication systems," Electronic Letters, vol. 30, pp. 1017-1018, June 1994.

[6] J. R. T. Compton, "An adaptive array in a spread-spectrum communication system," Pro­

cIEEE, vol. 66, no. 3, pp. 289-298, 1978.

[7] R. B. Ertel, "Spatial filtering with adaptive antenna array algorithms in DS-CDMA commu­

nications systems," Master's thesis, Pennsylvania State University, 1996.

[8] D. Gerlach, "Base station array receivers in cellular CDMA," IEEE Asilomar Conf., pp. 646-

650,1992.

[9] B. Suard, A. F. Naguib, G. Xu, and A. Pauiraj, "Perfcirmanceof CDMA mobile communication

systems using antenna arrays," Proc. IEEE Int. Conf. Acoust., Speech, Signfll Processing,

vol. 4, pp. 153-156,1993.

[10] A. F. Naguid, A. Paulraj, and T. Kailath, "Capacity improvement of base-station antenna

arrays cellular CDMA," IEEE Asilomar Conf., pp. 1437-1441,1993.

[11] A. F. Naguib and A. Paulraj, "Performance of CDMA cellular networks with base-station

antenna arrays," Lecture Notes in Computer Science, no. 783, pp. 87-100, 1994.

[12] A. F. Naguib and A. Pauiraj, "Effects of multipath and base-station antenna arrays on uplink

capacity of cellular CDMA," IEEE Proc. Globecom, pp. 395-399,1994.

Figure 1:.1D Array: SINR vs. Number of Users (M=4 Sensors in Array) ,.

,.

-,. , , , ,

~.~~~=---~=---=~---=~~,oo Numberof~.

- - ItIputSINR • LMS o DOLMB + SMIIRLS -CEAM1oca ..... CEACFtDCB

·-··CEACF1D • CEPPCF1DCS

Figure 2: 2D RAKE Receivers: SINR vs. Num­ber of Users (M=4 Sensors in Array) ,.

~.~-,~~~~~~~~--~~~,oo Nl.mI:MItofUHrS

-OEAM2OCB ·-··CEACF2D .... CEACF2DCB o CEMDP2DCB

CESAM2DCB • CEPPCF2DCB • 2DlM8 + 2OBM.

108

Figure 3: ID Array: SINR vs. Number of Bits (N =20 Users are Present)

-9.'

o

___ ._,_.-'X~~ )( X )( .... ."'--(;'~ )( ~ iii A 0

x eMS o CDLMS

f -11 ~ ~,...o"'O" 0 : x 0

+ SMUALS -CEAM1DCB .... , CEACF1DCB ._ .. CEACF1D ~ 0.'

f~11.5 ~-,x

]II CEPF'CF1DC

-12 I

-12.5

-"'0:----;C---'''0--~15~-~20 Number 01 bita

Figure 5: ID Array: Capacity vs. Number of Antenna Elements ~r--~----~---~

70

J.

50 .' -".

20

10

~:---~--'~-~'~-~-~10-Number of antenna elements

x eMS o DOLMS + SMI/RLS -CEAM1DCB ..... CEACF10CB '-'·CEACF10 ]II CEPPCF10CB

Figure 7: ID Arrays: Flop Count vs. Number of Users (M=4 Sensors in Array, D=lO Bits)

50

30

i 25 /

"'0:---:20Cc---40Cc----:50~---:8~0 --':'00 Nul'l'lbeJof UsefS

x eMS o DOLMS

- - RLS ... SMICase 1 • SMICase2

-CEAM1OCB , .. --- CEACF1DCa ·_·-CEACF1D * CEPPCF1DCB

Figure 4: 2D RAKE Receivers: SINR vs. Num­ber of Bits (N =20 Users are Present)

--6

+++++++++++

-8

-14

-16

-180L:.--~--1~0---'~5--~20 Numberolbils

-CEAM2DCB · -. - CEACF2D

CEACF2DCB o CEMDP2DCB

CESAM2DCB • CEPPCF20CB )( 2DLMS + 205MI

Figure 6: 2D RAKE Receivers: Capacity vs. Number of Antenna Elements

100

90

80

70

.....

30

20

10 ~~-~-~.--~'--~8-~10·

Number 01 antenna alemen1S

-CEAM2DCB -'. CEACFZD

CEACF20CB o CEMOP2DCB

CESAM2DCB • CEPPCF2DCB x 20LMS + 208MI

Figure 8: 2D RAKE Receivers: Flop Count vs. Number of Users (M=4 Sensors in Array, D=10 Bits)

70

65

50

R R -gS5 Q .. ,,~ .... ·f- ~. + I" .~ .. + <, _ :f-

l., ~ ...

35

, 300L--~20~-~40~--!-60~--::aoC-----::'00

Number of antenna elements

-CEAM2DCB -,. CEACF2D

CEACF2DCB o CEMDP20CB

CE$AM2DCB • CEPPCF20CB x 2DLMS + 205MI

11 Cochannel Receivers for CPM Signals Based Upon the Laurent

Representation

Peter A. Murphy & Gary E. Ford

CIPIC, Center for Image Processing and Integrated Computing University of California, Davis, CA 95616

email: murphy@ece.ucdavis.edu, ford@ece.ucdavis.edu Tel: (916)752-2387, Fax: (916)752-8894

Abstract

In this paper, we develop optimum and suboptimum receivers for jointly detecting two cochannel continu­ous phase modulated (CPM) signals. These receivers are based upon Laurent's representation of binary CPM as the sum of a finite number of pulse amplitude modulated signals. We also provide a review of the Laurent representation and its application to the design of optimum and suboptimum single-channel receivers.

1 Introduction

Cochannel interference, which occurs when two or more signals share the same spectral and temporal chan­nels, is a major obstacle to high quality speech and data transmission in mobile radio systems [1, 2]. In a recent paper (3), we investigated a single-sensor technique for jointly detecting two cochannel CPM signals. The result was the development of the joint maximum likelihood sequence estimation (JMLSE) receiver based upon the conventional representation of CPM [4]. It was shown that this receiver performed well under certain conditions even when the power separation between the signals was negligible. One of the major drawbacks with this receiver, however, is its high degree of complexity.

In an effort to simplify the receiver and provide new perspectives on the joint detector, we have investi­gated the use of the Laurent representation [5] for binary CPM in developing both optimum and suboptimum cochannel receivers. In (5), Laurent showed that a binary CPM signal can be exactly represented as the sum of J{ pulse amplitude modulated (PAM) signals. The Laurent representation is a generalization of the interpretation of MSK-type signals (i.e., binary CPM with modulation index 1/2) as the sum of time and phase shifted PAM signals [6, 7]. For example, it is well known that MSK can be represented as an offset-QPSK signal in which the pulse shape is a half-cycle sinusoid with period four times the bit duration. There are three main advantages of the Laurent representation. First, it provides for easy calculation of the autocorrelation and power spectrum of CPM signals. Second, it provides for easy inclusion of channel effects in the expression for an LTI-filtered CPM signal. Third, and more important to the discussion in this paper, it provides a straightforward method of accurately approximating CPM signals using only a subset of the J{ PAM components in the exact representation.

A number of researchers have found the Laurent representation useful [8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18]. In (8), Kaleh employed the representation to develop the optimum maximum likelihood sequence estimation (MLSE) receiver for a single CPM signal in additive white Gaussian noise. It was pointed out that this optimum receiver has the same complexity as the MLSE receiver developed based upon the conventional representation of CPM [4]. The author also developed suboptimum receivers by approximating the CPM signals using a subset of the J{ PAM components in the Laurent representation. In Section 3, we will rely heavily on much of Kaleh's work.

Many researchers have used only the primary PAM component of the Laurent representation to develop simplified receivers. Baier [9] showed that when MSK-type signals are approximated using the primary component, a derotation technique simplifies the receiver design. Luise and Mengali [10] illustrated that for MSK-type signals, the average matched filter (AMF) [4, 19] used as the receive filter in a linear MSK-type receiver, is simply the time-reversed version of the primary pulse in the Laurent representation. In [11], Kaleh

J. H. Reed et al. (eds.), Wireless Personal Communications

© Springer Science+Business Media New York 1997

110

developed a differential detector for MSK-type signals and showed that the structure of the receiver is similar to a differential detector for BPSK, except that unlike a conventional differential detector, which multiplies two signals one of which is delayed by the symbol period T, a longer delay MT provides a performance improvement over a delay T. Del Re et al. [12, 13] developed a simplified receiver for a filtered normal burst in the GSM system. In [16], the authors describe a simplified Viterbi processor for the demodulation of the GMSK bursts used in the GSM system. The authors approximated the differentially encoded GMSK signal using the primary PAM component in the Laurent representation. In [17] the authors develop a noncoherent receiver for the GMSK signals used in the Digital European Cordless Telecommunications (DECT) system. Once again, the receiver is based upon approximating the CPM signals using the primary pulse of the Laurent representation.

The approach taken in this paper to develop optimum and suboptimum ML receivers for jointly detecting two cochannel CPM signals is based upon the Laurent representation. The paper is somewhat tutorial in nature, simply because the development of the joint detectors requires the necessary background material provided in the early part of this paper. Furthermore, these early sections include examples of the material applied to MSK signals and the GMSK signals used the European GSM [20] and DCS1800 [21] systems.

The structure of the paper is as follows. In Section 2 we review the Laurent representation for binary CPM signaling. Following this, Section 3 provides a discussion of optimum and suboptimum ML receivers for CPM signals based upon the Laurent representation and its approximation. This section, although primarily a review, does include some new results on the performance of suboptimum receivers applied to GMSK signaling. In particular, we show that for the GMSK signals used in the GSM and DCS1800 systems, a reduction in complexity by a factor of four is attainable with only a slight degradation in receiver performance relative to the optimum receiver. Next, Section 4 describes both optimum and suboptimum receivers for jointly detecting cochannel CPM signals. Once again, we illustrate that use of the Laurent approximation reduces the receiver complexity by a factor of four relative to the optimum joint detector. Finally, the paper provides some concluding remarks in Section 5.

2 Laurent's Representation

A binary CPM signal can be expressed as [4]

set)

B(t)

Re[Sb(t,Q.)eih!o'] t 2: 0

(2E;ei [8(')Hl t> 0 Vr -t N - 1

21rh 10 ?; a,g( r - iT)dr t 2: 0

N-l

1rh L aiq(t - iT) t 2: 0, i=O

(I)

(2)

(3)

(4)

where Ie is the carrier frequency, <i> is the carrier phase, which for the remainder of this section is assumed to be zero, Eb is the energy per bit, T is the bit period, h is the modulation index which takes on rational values (i.e., h = 2k/p k,p integers), {ad are the transmitted bits taken from the set {-I, I} with equal probability, get) is termed the frequency pulse and is nonzero in the interval [0, LT], has area equal to 1/2 and is symmetric about LT/2., and q(t} is the integral of the frequency pulse, such that

q(t)

q(LT}

l g(r}dr

1/2.

(5)

(6)

As Laurent [5] showed, the complex baseband signal Sb(t,.Q.) can be expressed as the sum of [{ = 2L- 1 PAM signals, i.e.,

(7)

over the interval t E [LT, NT], where

n L-1

ak,n L (Xi - L fr n -jf3k,j

i=o j=1

£-1 av,n - L (};n-jf3k,jl

j=1

£-1 ao,n_£ + L '-"n_j(l- f3k,j) + '-"n,

j=!

where the set {f3k,j} are used in the binary representation of the index k:

L-l

k = L 2j - 1f3k,j k E [0, I< - 1] j=1

and f3k,j E {O, I}. Finally, the functions of time Ck(t) are are given by

£-1 Ck(t) = SO(t) II Si+£~k,j(t) k E [0, I< - 1],

j=1

where (t) = sin(\[1(t + jT)) = (t + T)

S3 sinC"'h) So J,

{ </I(t) t E [0, LT)

\[1(t) = 7rO

h - </I(t - LT) t E [LT,2LT] else

and

</I(t) = 27rh l g(-r)d-r

Example 2.1 The Laurent Representation for MSK

111

(8)

(9)

(10)

(11)

(12)

In this example, we present the Laurent representation for MSK. Recall, for MSK L = 1, h = 1/2, and

g(t) = { ~, 0,

t E [O,T] otherwise

Therefore, we have I< = 2£-1 = 1, so that the Laurent representation of (7) includes only one function of time co(t). It can be shown that

t E [0,2T) otherwise

This provides the well known interpretation of MSK as offset-QPSK in which the pulse shape is a half-cycle sinusoid with period 4T, The complex, baseband representation for MSK is given by

(13)

Example 2.2 The Laurent Representation for GMSK In this example, we present the Laurent representation for the G MSK signals used in the European GSM and DCS1800 systems. In particular, the GMSK signal parameters are BT = 0.3, L = 3, and h = 1/2. We see that I< = 2£-1 = 4, so that we have four signals components co(t), Cl(t), C2(t), and C3(t), as shown in Figure 1. The phase variables {ak,n, k E [0, 3]} are given by

112

0.' 0 .•

0.4

0.2

~~::::i··; ..•.. : i o 0.5 1 1.5 2 2.5 3.5 4

tiT

Figure 1: Signal components of GMSK with L = 3, BT = 0.3, and h = 1/2

aO,n

a3,n

n-3

LUi + Cl'n-2 + O'n-1 + an i=O n-3

L at + Ctn -2 + an i=O n-3

L O!i + Un-l + an ;=0 n-3

Lai+ an i=O

The Laurent representation for this signal is given by

(14)

(15)

(16)

(17)

(18)

We should highlight that the ca(t) signal component in general provides most of the signal energy for all CPM signals. For the GMSK signals described above, computer simulations reveal that 99.7236% of the signal energy is contained in the cart) component, CI(t) contains 0.2763% of the signal energy, and the remaining signal energy is shared between the C2(t) and C3(t) components.

3 Single-channel Receivers Based Upon Laurent's Representation

In this section, we develop coherent optimum and suboptimum demodulators for a single CPM signal based upon the Laurent representation. We should highlight that much of this section is a review of theory developed by Kaleh [8]. However, the performance results presented in Examples 3.1 and 3.2 are new. Section 3.1 includes a description of the optimum maximum likelihood sequence estimation (MLSE) receiver for CPM signals corrupted by additive white Gaussian noise (AWGN). Section 3.2 includes a description of less complex, suboptimum receivers for these same received signals based upon approximating the signals using a subset of the I< PAM components in the Laurent representation.

3.1 Optimum ML Receiver for CPM Signals in AWGN

The signal to be processed in the coherent receiver is

r(t) = Sb(t,g:) + n(t), t E [LT. NT] (19)

113

where net) is a realization of a complex, zero mean, white Gaussian noise process. The ML receiver maximizes the log likelihood function [22]

l NT in[P,(t)la(r(t)lgJ] <X - Ir(t) - sb(t,.Q:)1 2 dt

- LT (20)

with respect to the estimated sequence .Q: = [aL, aL+1,"', aN-d, where we assume knowledge of [aD, aI, ... , aL-d. This receiver criterion can be simplified to choose the.Q: that maximizes the metric

l NT A(.Q:) = Re{r(t)s;(t,.Q:)}dt

LT (21)

where Re(-) denotes the real part of (-) and (.)' denotes the complex conjugate of (-). Substituting (7) into (21) yields

A(.Q:)

where

~ NT { [K-IN~I ]'} V~ iT Re ret) {; ~ ej~hilk.nCk(t - nT) dt

f:!-f. Re{,\(n)}, n=O

K-I A(n) = L rk,ne-j·Mk,n

k=D

where the {ak,n} are calculated using (8) with {ad replaced by {ad and

rk,n = [ret) * ei( -t)Jlt=nT = J r(t)ek(t - nT)dt

(22)

(23)

(24)

The samples {rk,n} form a set of sufficient statistics for the computation of A(.Q:). Furthermore, (24) shows that they can be acquired by sampling at times nT the output of]{ matched filters {ci( -t), k E [O,]{ - I]} fed by the received signal ret). From Section 2, we know that the pulses {Ck(t), k E [O,]{ - I]} are real­valued and therefore the conjugation of ek( -t) in (24) is not necessary. However, we include the conjugation operation, because it provides for an easy extension of the algorithm to handle the effects of a possibly complex-valued LTI channel impulse response. The calculation of >.(n) at time nT, requires knowledge of all possible {ak,n, k E [O,]{ - I]}, which in turn depend upon the phase state 7rhaD,n_L, the state vector [an_I, an -2,'''' an-L+1], and the symbol an. For h = 2ilp (i,p integers), 7rhaD,n_L takes on p discrete values {O, 27r Ip, .. " 27r(p - l)/p}. It is easily shown that at each time nT, the maximum likelihood receiver computes A(n) for all possible 2L sequences [an, an-I,"', an-L+IJ and all p possible phase states 7rhaD,n-L. Therefore, there are p2L metrics calculated per symbol interval. As one might expect, this is the same number calculated by the MSLE receiver based upon the conventional representation of CPM [4J. The computationally efficient Viterbi algorithm can be employed to choose the sequence .Q: that maximizes A(.Q:). Figure 2 shows the structure of the receiver. The constant J2Eb/T in (22) is not necessary and can be omitted. However, we include it here because the joint detector of Section 4 requires it.

~

~ ~ ~

ret) r:: II

"" ~ ~

:;;:

Figure 2: Optimum MLSE Receiver for binary CPM based upon Laurent's Representation

114

Example 3.1 The MLSE Receiver for MSK Signaling In this example, we briefly describe the MLSE receiver for MSK. Recall, that in Example 2.1, we developed the Laurent representation for MSK and explained that for this signal, I< = 1. The ML receiver chooses the sequence .i! that maximizes (22), where

A(n)

TO,n [ret) * c~( -t)]lt=nT = J r(t)c~(t - nT)dt

(25)

(26)

The samples {rO,n} are obtained by sampling at times nT the output of the matched filter cjj(-t) fed by the received signal ret). The calculation of A(n) at time nT requires knowledge of the phase state %ilo,n-I and the symbol <in. It is easily shown that %aO,n-1 takes on 4 possible values {O, 7r/2, 7r, 37r/2} and <in takes on 2 possible values. Therefore, during each symbol interval 8 = 2 ·4 metrics are calculated.

Example 3.2 The MLSE Receiver for GMSK Signaling In this example, we develop the optimum receiver for the GMSK signals described in Example 2.2. The Laurent representation for this signal consists of the four signal components co(t), CI (t), C2(t), and C3(t) shown in Figure 1. The phase variables {ak,n, k E [0, I< - In are given by equations (14)-(17). It can be shown that this receiver requires the calculation of 32 = p2L metrics per symbol interval. The receiver structure is shown in Figure 3. Figure 4 illustrates the performance of this receiver in terms of bit-error· rate (BER) relative to the more conventional MLSE receiver [4]. The BER curves are calculated via computer simulations. We can see that the performance of the Laurent-based MLSE receiver developed in this section is identical to the conventional MLSE receiver, which is no surprise since the Laurent representation is an exact representation and since both receivers attempt to maximize the same likelihood function. The plot also includes .the performance of the reduced complexity receiver described in Example 3.3.

~ " u

" ret) ct: ~

"" t ;:

Figure 3: Optimum MLSE Receiver for GMSK Signals

3.2 Suboptimum Receivers for CPM Signals in AWGN

Less complex, suboptimum receivers can be developed for CPM by approximating the signals using a subset of the I< PAM components in the Laurent representation (see (7». Though suboptimum, these receivers admit only a slight degradation in performance relative to the optimum receiver. The suboptimum receiver is developed by first approximating the CPM signal using the first k components, where k ::; I<, of the Laurent representation, i.e.,

(27)

where the {ak,n, k E [0, k -In are calculated using (8). When k = I<, the approximation in (27) is equal to the exact representation of (7), and the suboptimum receiver becomes the optimum ML receiver of Section 3.1.

115

·"·1· •••••••• ··•···•· •••• ·· Lr ••• ~~ •••• · ••• : ••• • •••• l

Figure 4: BER for conventional, Laurent, and suboptimum MLSE receivers

Following a procedure similar to that of Section 3.1, let the approximate received signal be given by

ret) = 8b(t,S!) + net), t E [LT, NT] (28)

where net) is a realization of a complex, zero mean, white Gaussian noise process. Once again, the ML receiver maximizes the log likelihood function

in[Pr(t) Ig(r(t) Ig)] <X _ (NT Ir(t) - 8b(t,gWdt JLT

(29)

with respect to the estimated sequence g = [aL, aL+I,·", aN-d, where we assume knowledge of lao, ai, ···,aL_I]. Of course, this receiver is not optimum because 8b(t,g:) is an approximation. This receiver criterion can be simplified to choose the g that maximizes the metric

Substituting (27) into (30) yields

where

~(n)

lNT A(Q:) = Re{r(t)s;(t,g)}dt

LT

_ [2E;N-I_ A(Q:) = V T L Rep(n)},

n==O

K-I '" r e-j'lrho.k,n ~ k,n

k=O

[ret) * ek( -t)]lt=nT = J r(t)ek(t - nT)dt

(30)

(31)

(32)

(33)

Once again, the samples {rk,n} can be obtained by sampling at times nT the output of the j( matched filters {ekC -t), k E [0, j( -I]} simultaneously fed by the received signal ret). The calculation of ~(n) at time nT requires the knowledge of all possible {lik.n, k E [0, j( - I]}. We should highlight that in this case, the phase state is given by "haO.n-L+'. The value of i is dependent upon the {J3k,j} of (9) used in the binary representation of k. It can be shown that at each time nT, this suboptimum receiver computes ~(n) for all possible 2L-' sequences [an, an_I,'··, an-L+1+;] and all p possible phase states "haO,n-L+'. Thus, there are p2L -' metrics calculated per symbol interval. Recall from Section 3.1, the optimum receiver requires the calculation of p2L metrics. Therefore, this receiver provides a reduction in complexity by a factor of 2'.

116

Moreover, the receiver requires {(. matched filters instead of the K matched filters required in the optimum receiver.

Example 3.3 Suboptimum Receiver for GMSK Signaling In this example, we develop a suboptimum receiver for the GMSK signals in Examples 2:2 and 3.2. We approximate the GMSK signals using only the primary component co(t) and develop the ML receiver b.ased on this approximation. As stated in Example 2.2, the portion of the GMSK signal due to the co(t) component contains over 99% of the signal energy, and is, therefore, a very good approximation. The G MSK signal is approximated as

(34)

Using the results presented earlier in this section, the suboptimum ML receiver maximizes the metric A@, given by (31) with

(35)

where ro,,, is the output of the matched filter co( -t) sampled at times nT whose input is the received signal r(t). Using the results from Example 2.2, we know

,,-1

ao,n = an + L: 0:; ;=0

(36)

The calculation of X( n) at time nT requires knowledge of the phase state tilo,,,-1 and a". It can be shown that this suboptimum receiver computes X(n) for the 2 possible values of a" and all 4 possible values of the phase state. Therefore, there are 8 metrics calculated per symbol interval. Recall, the optimum ML receiver for GMSK developed in Example 3.2, required the calculation of 32 metrics per symbol interval. Therefore, this simplifiep receiver reduces complexity by a factor of 4. Figure 4 illustrates the BER performance of the simplified receiver. The plot reveals that the suboptimum receiver admits only a slight degradation in performance relative to the optimum receiver. Furthermore, this example highlights that by approximating a CPM signal with modulation index 1/2 using the primary PAM component of the Laurent representation, the suboptimum receiver has the same complexity as the optimum MLSE receiver for MSK (see Example 3.1). That is, the receiver requires the calculation of only 8 metrics per symbol interval.

4 Cochannel Receivers Based Upon Laurent's Representation

In this section, we develop novel optimum and suboptimum demodulators for jointly detecting two cochannel CPM signals. The receivers are based upon the Laurent representation described in Section 2. Section 4.1 includes a description of the optimum joint maximum likelihood sequence estimation (JMLSE) receiver and Section 4.2 includes a description of less complex, suboptimum receivers for these same signals.

4.1 OptimumML Receiver for Cochannel CPM Signals in AWGN

The signal to be processed by the coherent receiver is the sum of two complex baseband CPM signals and complex white Gaussian noise. The received signal can be expressed as

where

r(t) = 8b,I(t,Q,I) + Sb,2(t,~) + n(t), t E [LT, NT]

8b,I(t,Q,I) J2~'1 d[8,(')+4>d t 2: 0

8b,2(t,Q,2) = J2~'2 e;[8,(')+4>.1 t 2: 0

(37)

(38)

(39)

We should highlight that even though the two CPM signals are similar in structure, they may have different modulation indices hi and h2, different energy levels Eb,1 and Eb,2, different carrier phases <PI and <P2,

117

different frequency pulses gl(t) and g2(t), each with its own length of temporal support 11 and 1 2, and of course the transmitted bit sequences frl and fr2 are distinct. However, we do assume the signals share the same bit period T and that both signals transmit an equal number of bits N . Finally, 1 is chosen as the maximum value of 11 and 1 2.

The joint maximum likelihood receiver maximizes the log likelihood function

with respect to the estimated sequences '&'1 = [&I,L, &1',L+1,"" ,h,N] and '&'2 = [&2,L, &2,L+I,''', &2,N], where we assume knowledge of [0:1,010'1,11"", ll'l,L-l] and [a2.o, 0:2,1,"', 0'2,L-1]. The above maximization can be simplified to choose the '&'1 and'&'2 that maximize the metric

( 40)

Viewing (40), we see that the first two terms are basically the classic correlation receivers for detecting frd fr2 when Sb,2(t,fr2)/Sb,l(t,frl) is not present, The third term tries to ensure that the two signal estimates are uncorrelated,

Using the notation introduced in Section 3,1, substitution of (7) into (40) yields

(2ii;;: N -I (2E;; N - 1

A('&'I''&'2) = V -+ L Re{.\l(n)} + V -+ L Re{.\2(n)}-n=O n=O

N-l n K 1-IK2 -1

~v'EblEb2 L L L L Re{ej(1rhla~~),nl-7fh2ak;,n2+¢1-4i2) x

nl=O n2=O k 1::::O k2=O

[NT CI,k,(t _ nIT)c;,k,(t _ n2T)}dt, iLT

where a~~~ and a~~~ are calculated using (8) with {ad replaced by {&I,i} and {&2,i}, respectively, and

(41)

(42)

(43)

The samples r~l) n and r~2) n are obtained by sampling at times nT the output of the I<I + I<2 matched filters {cj,k,(-t), kl E'[O,I{I-'in and {Ci,k,(-t), k2 E [0,I<2 -I]} fed by the received signal r(t), We now turn our attention to the last term in (41). From Section 2, we know

CI,O(t) f. 0, t E [0, (11 + 1)11

C2,0(t) f. 0, t E [0, (12 + I)T]

using this result we can express the last term in (41) as

where

(44)

(45)

(46)

(47)

The terms R~k:.:,') are a function of the I<I + I<2 PAM components and can be determined in advance and stored in the receiver.

118

Finally, the optimum receiver chooses the.ih and.ib that maximize the metric

(48)

It can be shown that during each symbol interval the receiver requires knowledge of the phase states

1rhla~~~_Ll and 1rh2a~~~_L2J the joint state [&l,n-lJ Ch,n-2J"', &1,n-L 1+l I a2,n-b a2,n-2,"', a2,n-L 2+d and the symbols &I,n and &2,n. Therefore, there are PIP22L,+L, metrics calculated per symbol interval. As one migth expect, this is the same number of metrics required per symbol interval for the JMLSE receiver developed based upon the conventional representation of CPM [3]. Figure 5 illustrates the structure of the optimum cochannel demodulator.

(11 ra,n

(11 1"l,n.

;..

" (I) " " T[(l-l,n. ~ " ...

r(t) ~ UI, li2 :s ;..

(2) " rO.f!. ~ r(2)

l.n

n(kbk2) 7

(~) k. e[O.K.-l) rI\:J-1.n k~ E [O.K2 -1]

i E [0.L21

Figure 5: Optimum Cochannel Receiver for binary CPM based upon Laurent's Representation

4.2 Suboptimum ML Receivers for Cochannel CPM Signals in AWGN

In this section, we develop a suboptimum coherent joint maximum likelihood sequence estimation (JMLSE) receiver for jointly detecting two cochannel CPM signals received in additive white Gaussian noise. The approach is similar to that of Section 3.2, except we approximate the two cochannel CPM signals using only the primary PAM component of each signal CI,O(t) and C2,O(t) (refer to (7)).

We let the approximate received be given by

(49)

where Sb,l(t,fh) and Sb,2(t,fl:2) are an approximation to the cochannel CPM signals, such that

(50)

(51)

119

It can be shown using results from Section 4.1 that the suboptimum ML receiver chooses the !ll and !l2 that maximize the metric

L, L Re {R\O,O)ej(~h,a~:~-'h,a~~~_;H'-<P')}] (52) i=O

We can show that during each symbol interval, this receiver requires knowledge of the phase states 1I'hla~I~_1 and 7rh2a~~~_L" the state vector ["'2,n-l, "'2,n-2,"', "'2,n-L,+d and the symbols "'I,n and "'2,n. Therefore, during each symbol interval the receiver requires the calculation of P1P22L,+1 metrics. This is a reduction in complexity by a factor of2L ,-1 relative to the optimum detectors of [3] and Section 4.1. Furthermore, as we saw in Section 3.2, this receiver requires only two matched filters ci,o( -t) and ci,o( -t) compared to the K I + K 2 required in the optimum receiver.

At the moment no performance results for the suboptimum receiver are available. However, in the future we plan to investigate its performance in detecting cochannel GMSK signals and compare it to the optimum receiver of Section 4.1.

5 Conclusion

In this paper, we employed the Laurent representation to develop optimum and suboptimum joint maximum likelihood sequence estimation (JMLSE) receivers for cochannel CPM signals. In Section 2 we reviewed the Laurent representation and showed that a binary CPM signal can be expressed as the sum of K PAM signals. We also provided examples of how the representation can be used to express MSK signals and the GMSK signals used in the European GSM and DCS1800 systems. Following this, Section 3 included a review of optimum and suboptimum single-channel maximum likelihood receivers for binary CPM signals based upon the Laurent representation. This section also included new performance results highlighting that approximating the GMSK signals in the GSM and DCS1800 systems using only the primary PAM component of the Laurent representation, provided a reduction in receiver complexity by a factor of four, while admitting only a slight degradation in performance. In Section 4 we used the Laurent representation to develop optimum and suboptimum joint maximum likelihood receivers for detecting cachannel CPM signals. We showed that the receiver complexity in the optimum case is equivalent to the complexity of the JMLSE receiver based upon the more conventional representation of CPM. Furthermore, we highlighted that, as in Section 3, the suboptimum receiver provides a factor of four reduction in receiver complexity compared to the optimum joint receiver. In the future, we plan to investigate the performance of the suboptimum joint receiver applied to GMSK signaling. We also plan to examine the nse of the Laurent representation to include the effects of channel distortion on single-channel and cochannel CPM signals.

References

[1] D. Cox, "Cochannel interference considerations in frequency reuse small-cover age-area radio systems," IEEE Trans. Comm., vol. COM-30, no. 1, pp. 135-142, Jan. 1982.

[2] S. Wang and S. Rappaport, "Signal-to-interference calculations for balanced channel assignment pat­terns in cellular communications systems," IEEE Trans. Comm., vol. 37, no. 10, pp. 1077-1087, Oct. 1989.

[3] P. A. Murphy and G. E. Ford, "Co channel demodulation for continuous phase modulated signals," in Proc. of the Twenty Eigth Asilomar Cont on Signals, Systs., and Computers, 1995.

[4] J. Anderson, T. Aulin, and C. Sundberg, Digital Phase Modulation. New York, NY: Plenum Press, 1986.

[5] P. A. Laurent, "Exact and approximate construction of digital phase modulations by superposition of amplitude modulated pulses (AMP)," IEEE Trans. Comm., vol. 34, no. 2, pp. 150 - 160, Feb. 1986.

120

[6) S. A. Gronemeyer and A. L. McBride, "MSK and offset QPSK modulation," IEEE Trans. Comm., vol. COM-24, no. 8, pp. 809-819, Aug. 1976.

[7) M. K. Simon, "A generalization of minimum-shift-keying (MSK)-type signaling based upon input data symbol pulse shaping," IEEE Trans. Comm., vol. COM-24, no. 8, pp. 845-856, Aug. 1976.

[8) G. K. Kaleh, "Simple coherent receivers for partial response continuous phase modulation," IEEE J. Select. Areas Comm., vol. 7, no. 9, pp. 1427 - 1436, Dec. 1989.

[9) A. Baier, "Derotation techniques in receivers for MSK-type CPM signals," in Proceedings of the Fifth European Signal Processing Conference (EUSIPCO-90), (Barcelona, Spain), pp. 1799-1802, Sept. 1990.

[10) M. Luise and U. Mengali, "A new interpretation of the average matched filter for MSK-type receivers," IEEE Trans. Comm., vol. 39, no. 1, pp. 14 - 16, Jan. 1991.

[11) G. K. Kaleh, "Differentially coherent detection of binary partial response continuous phase modulation with index 0.5," IEEE Trans. Comm., vol. 39, no. 9, pp. 1335 - 1340, Sept. 1991.

[12) E. D. Re, R. Fantacci, L. Pierucci, G. Castellini, and G. Benelli, "Viterbi rp"eiver for mobil·· radio communications: Issues and implementation remarks," in Proceedings of the Fift·, Tirrenia International Workshop - Coded Modulation and Bandwidth-Efficient Transmission, (Tirrenia, Italy), pp. 169 - 178, Sept. 1991.

[13) E. D. Re, G. Benelli, G. Castellini, R. Fantacci, L. Pierucci, and L. Pogliani, "Design of a digital mise receiver for mobile radio communications," in Proceedings of the IEEE Global Telecommunications Conference (GLOBECOM), (Phoenix, AZ), pp. 1469-1473, Dec. 1991.

[14) G. K. Kaleh, "Differential detection via the Viterbi algorithm for offset modulation and MSK-type signals," IEEE Trans. Vechic. Tech., vol. 41, no. 4, pp. 401 - 406, Nov. 1992.

[15) O. Andrisano and M. Chiani, "The first Nyquist criterion applied to coherent receiver design for gen­eralized MSK signals," IEEE Trans. Comm., vol. 42, no. 2/3/4, pp. 449 - 457, Feb/Mar/April 1994.

[16) G. Benelli, A. Garzelli, and F. Salvi, "Simplified Viterbi processors for the GSM pan-European cellular communication system," IEEE Trans. Vechic. Tech., vol. 43, no. 4, pp. 870 - 878, Nov. 1994.

[17) S. Safavi and 1. B. Lopes, "A non-coherent equaliser receiver structure for DECT-type systems," in Proceedings of the 44th Annual Conference of the IEEE Vehicular Technology Society, (Stockholm, Sweden), pp. 1084 - 1088, June 1994.

[18) A. N. D'Andrea, A. Ginesi, and U. Mengali, "Frequency detectors for CPM signals," IEEE Trans. Comm., vol. 43, no. 2/3/4, pp. 1828 - 1837, February/March/April 1995.

[19) M. S. El-Tanany and S. A. Mahmoud, "Mean-square error optimization of quadrature receivers for CPM with modulation index 1/2," IEEE J. Select. Areas Comm., vol. 5, no. 5, pp. 896 - 905, June 1987.

[20) M. Hodges, "The GSM radio interface," British Telecom Technology Journal, vol. 8, no. 1, pp. 31 - 43, Jan. 1990.

[21] P. Ramsdale, "Personal communications in the UK - implementation of PCN using DCS 1800," Internat. J. Wireless Info. Networks, vol. 1, no. 1, pp. 29 - 36, Jan. 1994.

[22] J. Proakis, Digital Communications. New York: McGraw-Hill, 1989.

12 Site Survey for Wireless Communications Evaluation for

New York Mercantile Exchange

ABSTRACT

Carl Panasik, Dale Cone, Tom Siep

Texas Instruments Inc.

PO Box 655474 MS 446

Dallas TX, 75265

cpanasik@ti.com

The New York Mercantile Exchange (NYMEX) asked Texas Instruments (Tf) to study and

propose the implementation of a system to support commodities traders in the exchange trading pits. The system includes a custom hand held computer, an integrated wireless

communications system, a wireless-to-wired network bridge, and a fault detection and

recovery scheme.

This paper summarizes the findings of a site survey on the trading floor at NYMEX both

during off hours ("empty room" but with existing equipment) and during trading

(hundreds of traders in constant motion present). The goal was to evaluate the system

sensitivity and impact of the impairments in terms of friendly and external interferers.

Using a spectrum analyzer and a tracking generator, we measured the real-time path loss

andfrequency response from the temporary access point to the hand-held computer.

Measurements of the indoor propagation path loss at NYMEX show good correlation

with theory. Due to the large number of metal structures and plasma displays there is

significant multi-path throughout the room and, hence, very little shadowing.

With a location 100 floors below the World Trade Center broadcast center, the NYMEX

commodities floor is a particularly interesting environment from an external interferer

standpoint (i.e. intermodulation suppression, adjacent channel rejection). In addition,

the excessive trader activity and user density in a hyper-active physical environment places several constraints on communication system design.

J. H. Reed et al. (eds.), Wireless Personal Communications

© Springer Science+Business Media New York 1997

122

1. INTRODUCTION

The New York Mercantile Exchange (NYMEX) asked TI to study and propose the implementation of a system to support commodities traders in the exchange trading pits. The system includes a custom hand held computer, an integrated wireless communications system, a wireless-to-wired network bridge, and a fault detection and recovery scheme. This paper summarizes the site survey findings.

Communication system design is but one part of the proposed development effort for the Wireless Local Area Network (WLAN) and Hand Held Computer (HHC) Systems to be used by NYMEX commodities traders for receiving orders and reporting trades. The system development also covered issues of data integrity (assurance that all data will be delivered in an accurate, timely, secure manner) and development of a hand held computer (LCD s<;reen, electronic ink, 8-hour operation).

The system translates Trader data into a wireless media, to a bridge (Access Point), to a wired media, which is connected to Application Servers which track and process the Traders transactions. The wireless portion of the communications system is Radio Frequency (RF) based and, for simplicity of implementation, operates in an unlicensed band. The Industrial, Scientific, and Medical (ISM) bands are the appropriate bands for this type of application. In this large, multi-user environment, frequency hop-based spread spectrum RF is the technology of choice. This application calls for a robust communication system. Frequency hopped spread spectrum enables as many as 40 simultaneous communication channels, whereas direct sequence is limited to 5.

Data communications for the NYMEX Trade Reporting and Order Processing Application must be able to handle extreme loads. The communication requirements are different and more demanding than typical installations. Most wired Local Area Networks (LANs) have from 10 to 100 users. These users often use the LAN to transfer large amounts of data, such as file transfers. The NYMEX WLAN is designed for up to 2000 users. Each user performs relatively few, small transactions over the medium but the transactions tend to be concentrated over a short period of time.

The required system capacity is 188 transactions per second (tps). The system throughput can be less than the maximum capacity, provided that the backlogged trades are cleared promptly after a burst of traffic. "Promptly" is defined as a period not to exceed the length of time equal to the burst time, provided no more transactions are entered by the Traders during the catch-up time.

A transaction is defined as a TRADE or ORDER with an acknowledgment (ACK). A TRADE is the report from the Trader on the floor about the purchase or sale of a specified number of contracts to or from an opposite Trader. An ORDER is the request for the sale or purchase of a specified number of contracts from a source outside the trading pit. ACKs are application-level ACKnowledgements of a TRADE or ORDER which assure that no transactions will be lost. The TRADE has 80 bytes of application data and the ACK has 20 bytes of application data. ORDERs and their ACKs are somewhat smaller, but for the purposes of this study were presumed to be the same size, 100 bytes per transaction. The wired portion of the communications network must be

able to handle the traffic for 2000 users through a limited number of wireless to wired system Access Points. A summary of the requirements is listed in Table I.

2. RADIO COMMUNICATION IMPAIRMENTS

2. 1 Externallnterference

Wireless data communication in the 2.4-GHz Industrial, Scientific and Medical band is performed in a cooperative manner. Although FCC rules (CFR 47, Part 15) limit the unlicensed ISM transmitter power to I watt, there are, licensed (grand-fathered) higher power transmitters presently in use.

Externalinterferenceis defined as an uncontrolled (non-system) radio source. The interferer can be out of band or in-band and could overcome receiver filtering or compress the input active devices and thereby impair communication. Being in the path of a microwave link or in the vicinity of a competing wireless LAN are examples of external interference. Another example is a nearby high-powered transmitter in another frequency band which creates in-band signals via undesired mixing in the non-linear aspects of the receiver front-end active components.

The region surrounding the World Trade Center(WTC) in New York City is probably the most active RF environment in the United States. The television and radio broadcast center for New York City is located at the top of an adjacent WTC building. Several IOOKW FM transmitters, IOOKW to IOOOKW television transmitters (VHF and UHF) and high-power paging transmitters are sited there.

Tests in Dallas confirmed that the signal range of useful communication is from -15 dBm to -80 dBm I. Co-channel signals that are more than 13 dB above the -80 dBm receiver sensitivity could be a detriment. The wireless LAN transceivers used in the site survey had intermediate frequencies at 350 and 38 MHz. Any external interferers at these frequencies are of particular interest.

Measurements of external interferers were made with a spectrum analyzer. Plots were made which display the frequency response of the signals detected. We searched for signals outside the 2.4-2.48-GHz ISM band which exceeded -60 dBm and for in-band signals above -90 dBm and found none.

1 Based on a survey of 2.4-GHz WLAN transceivers available.

123

124

2.2 Propagation Path

To measure the propagation path, we have developed a set of RF propagation tests. These tests are used to determine clear area path loss, effects of shadowing, and to measure multipath nulls (number and depth). Using a spectrum analyzer and a tracking generator (Figure 1) we can measure (in real time) the signal strength and the propagation path frequency response from the stationary access point to the mobile. We then moved into areas that we had identified as being contenders for multipath (many reflecting surfaces) and shadowing (no direct line-of-sight to access point). We attempted to find the region of poorest reception, making graphs of the frequency response.

Spectrum Analyzer

Tracking 100 feet of RG58

Generatoy , _____ C_a_b_le_{_3_1-_d_B_lo_s_S_) _--j

-6 dBm out / -37 dBm

2 RF Amplifiers, (each 30-dB gain)

Figure 1. Propagation Path Measurement Schematic

There is also some correlation between the radio's Received Signal Strength Indicator and the received signal power, but it has considerable variation. The measured nominal (average) path loss values were later correlated with receive signal strength measurements made with the notebook computers transceivers. The notebook computer radios are used to verify that communication is still adequate in areas of high impairments.

We have determined that attenuation by human bodies exceeds 15 dB, for 4-inch thickness of flesh. To effect this loss the entire radio (antennas and transceiver) must be covered, on all sides. This is the only impairment found, to date, that will stop the radio transmission over the entire frequency band. If any portion of the radio is not covered, communication with the access point is only slowed.

3. NYMEX SURVEY RESULTS

The results of the NYMEX trading floor site survey are very encouraging. We found that reliable data communications are possible at any location on the main floor. In fact, the system worked well as far away as the fire exit2!

2 See location 7, Figure 3.

125

As expected, many external interferers were present. We found two VHF channels at 67 and 77 MHz, the PM band at 88 to 108 MHz, the high-band VHF stations between 174 and 216 MHz and a little activity in UHF Land Mobile or IMTS Paging. The majority of activity is in the cellular bands at 850 to 950 MHz with one cellular service provider's down-link (to the mobiles) at 870-882, the other provider at 883-895. NYMEX already has a wireless LAN operating at 902-928 MHz. There was no significant activity at 38 and 350 MHz, the IF frequencies of our wireless transceivers J •

An example of the propagation measurement is shown in Figure 2. The access point was placed 13 feet above the trading floor. The data in Figure 2 was taken at a location on the edge of the Unleaded Gasoline pit (see Figure 3, location 4), under a metal platform similar to the one on which the 'access point' was located. Note the distinct 30-dB null at center characteristic of multiple propagation paths combining exactly out of phase.

ATTE:N lOdS RL Oo;;;lSm

~ r--.... \~

lOdS/

--...., ~ \I Y

START 2.40000GHz

I

IVIKR -S4.:3:3o;;;lSm 2 44000GH:z: .

r ~ ~ ~ J ~

STOP O2.4S000GHz VI3W 30kHz: SWF' 50. Oms

Figure 2. Propagation Path Measurement from Access Point to Location #4.

Figure 3 is a map of the NYMEX trading floor showing the Access Point and measurement locations. The propagation path measurements (Table 2. Receive Signal Level, Number of Nulls) were made during the evening of 22 June 1995 with no traders on the floor. The data-transfer measurements (Percent Error, Bit Error Rate, RSSI) were made during trading on 23 June 1995. For the propagation path measurement, the access point was the transmitter. For the data-transfer measurement, the access point was another laptop with data transceiver. In either case, the access point was located at the

3 GEe Plessey DE6003 WaveRider.

126

center of the platform between Natural Gas I Propane and Heating Oil, approximately 13 feet above the trading floor.

Table 2 is a compilation of the propagation path measurements taken during the NYMEX visit. Measurements 1-8 were taken during non-trading time, while measurements at locations 9-12 were taken during trading. Some locations (1 and 2) were not accessible during trading, while several measurements (9, 10, 11) were performed for the purposes of discerning measurement repeatability. The measurements in Table 2 are defined as:

Receive Signal Level is the average value of loss across the entire 2.400 to 2.483-GHz band.

Number of Nulls are those deeper than 10 dB.

Percent Error is the ratio of the number of packets received that have failed the header CRC check divided by the number of correctly received packets. Note, the number of correctly received packets is less than or equal to the number of transmitted packets. The measurement system does not account for packets sent but not received due to· transmission impairments.

Bit Error Rate is computed by dividing the number of bad bits by the number of bits received. Bad bits are the number of data bits that have been corrupted. Each time a packet is sent, it is a known pattern. The received data is compared with the known pattern. The bits that fail this comparison are defined as bad.

Receive Signal Strength Indicator is a voltage related to the power of the received signal (RSL). The relationship is given by RSSI = 0.04*RSL + 4.1 where the units of RSSI are volts and the units of RSL (receive signal level) are dBm.

127

Table 2. Measurements of NYMEX Trading Floor

Location Receive Signal Number Percent Bit Error Rate RSSI tv)

2

Level (dBm) of Nulls Error (xl 0") -4.0 -48 -67 -45

o 3

9 8.3 ~~1~0 __ -r ____________ ~ ______ 4-__ -41.0~0~ __ ~ __ ~29~. __ __

11 0.60-0.72 14.-27 . ....... 1.2.... ....................... ..... ............. ......................•....... . ............................. : .................... 0,.2.1 .......... _ ... 8c.10 ........ .

B ~

°85

( Gasdi", Heating 01 ~ 4

~~....!~------

OtdeOI

f1\ErudeOI \.}J Cp~ons

10 feet

9

@ @

CD

Figure 3. Map of NYMEX Trading Floor with Locations from Table 2.

128

3.1 Interpretation of NYMEX measurements

We found that reliable communication can exist for the entire main floor of the exchange. We define a reliable communication channel to have the following performance:

• Receive Signal Level above -80 dBm, or RSSI larger than 1.0 volt

• Number of Nulls less than 7

• Bit Error rate smaller than 10 x 10.4

The only locations in which the data throughput would be slowed significantly are locations 7, 8, 10 and 11. These locations were far beyond the trading area served by an access point. This level of performance can be used to provide additional redundancy against an access point failing.

The measured received signal level (data points in Figure 4) is plotted as a function of the distance along the trading floor. The theoretical equation for propagation loss in line-of­sight environments

L = 36*log(4*1t*d/A) - 46.7 dB for d > 8.5 m (1)

is used to calculate theoretical received signal level and shows good agreement with the measured data points. Measurements made in Dallas with many people crowded around the mobile transceiver (held at side) show that an additional 20-dB loss is incurred for close-packing of human bodies. Therefore, the close-packed RSL line denotes the RSL as a function of distance in a very crowded environment.

The suppressed signal of the data point at 36 feet is from measurements at position #3, the Gas and Heating Oil Options Pit. This location is in the shadow of a bank of plasma display panels which effectively attenuate the direct, line of sight signal. Position #3 only receives reflected signals.

3.2 Determination of Effective Range

To determine the effective range we have included the effect of other (i.e. external, non­cooperative) 2.4-GHz local WLANs, whether they are in the same room or on an adjacent floor. The receivers can handle a co-channel to interferer (ell) power ratio of 13 dB 4

when the interferer is another FSK signal. The third line from the top of Figure 4 denotes this level as a function of distance along the floor from the access point. The minimum receive signal level for BER better than 10.5 is 15 dB above thermal noise. In the 2-MHz receiver noise bandwidth (with a lO-dB receiver noise figure) the noise floor is at -101 dBm and the minimum RSL is at -85 dBm. This sets the worst-case range limit with zero

4 GEe Plessey Application Note ANI54-L3, January 1994

fade margin (and active nearby FSK interferers) to be 60 feet along the floor from the access point.

-20

-30

-40

E -50 m ~ -60 -;; >

" -70 ..J

iU -80 " til en -90

-100

-'-.... . -- . . '" --.. ---- ---

-." ---~-. - - -_. ---" - '::~~*i-""";"",-,

Range

I • Measured Signal

---Theoretical

- - - - Oose-Packed RSL

•. - ...• Co-Olannel Interference

- - - . - Mnirrum Receive Signal

Receive Noise Roor

-110 +--___ -+--___ -+ ___ Limit -0.00 20.00 40.00 60.00 80.00

Distance On Roor (feet)

Figure 4. Signal Level vs. Distance Along Trading Floor From Access Point.

3.3 Observation of Trader Activity

129

Almost all wireless communication links to date are portable, but not overly mobile. A quasi-stationary environment has the advantage that the multipath nuIIs are slowly moving with respect to the packet length. The NYMEX application adds the additional aspects of very quick movements, a compact high-user-density region, close packing of bodies that act as blocks to the RF signal, and a rugged, hyper-active physical environment.

Motion· At a given frequency in the 2.4-GHz band, a wavelength is about 5" (12.5 cm). In normal activity, the motion of a trader's hand moves through about 15" in each direction, traversing about 3 wavelengths, or more importantly, 12 potential nulls. The traders can make this motion in about a half second, so the radio is passing through 24 nulls/second. To communicate 1000 bits requires 1.6 msec. At 24 nulls/second, there is a clear communication channel for 42 msec between nulls. Hence, we could transmit 26 messages before the hand-waving user hits the next null.

Close Packing of Bodies - Unlike optical wireless communications, a radio-based system does not require 'line of sight' between access point and the hand held. In the Dallas tests, we determined that if the 2.4-GHz transceiver 'sees air' the signal will get to the access point. If there is anyway for air to get to the hand held (via the floor or bouncing off a wall or the ceiling) the unit will communicate with the access point. The only situation that stops communication is if one were to hug the HHC. This is deemed unlikely.

130

4. CONCLUSION

We found the NYMEX environment to be as 'hostile' as that of our Dallas test room, which had many metallic reflectors, creating high levels of multi path fading and shadowing as was the case at NYMEX.

In the area of external interferers, our location in the shadow of several VHF and UHF TV, FM radio and paging transmitters increased the external interference significantly. But all out-of-band signal levels were below our transceiver's requirements for nominal Bit Error Rate.

Through our measurements of propagation path, and experiments with the laptop data transceivers, we found that reliable data communications are possible at any location on the main floor using frequency hopped spread spectrum.

13 Measurement ofRF Propagation into Concrete Structures

over the Frequency Range 100 MHZ to 3 GHz

by

Clayborne D. Taylor Samuel J. Gutierrez Steven 1. Langdon Kenneth 1. Murphy

William A. Walton, III

Phillips LaboratorylWSM 3550 Aberdeen Ave. SE

Kirtland AFB, NM 87117-5776

ABSTRACT

A concrete structure with few ports of entry for RF was illuminated over the frequency range of 100 A1Hz to 3 Ghz to characterize the propagation ofRF signals into the interior region of the structure. The construction of the bUilding tested is standard for a blast hardened facility. Outside walls of the structure are two-feet thick concrete with horizontal and vertical rebar separated by only eight inches. And the flat rooflceiling is three-feet thick concrete with a two-inch steel plate base. With only one door located at the rear of the building and one set of windows across the front of the building. RF penetration through Windows, .doors and walls can be isolated and characterized separately.

Illumination of the concrete structure is provided by using two log-periodic antennas driven from two wideband linear amplifiers for the frequency range 100 A1Hz to 1 Ghz. Above 1 Ghz two traveling­wave tube amplifiers were used to reach 3 Ghz. Swept frequency measurements of the RF attenuation into the building were performed by using automatic network analyzers with the amplifier/antenna configurations. Logarithmic spacing of the frequency samples was used in the frequency sweep process. In all cases the antennas were oriented to produce a vertically polarized electric field with a horizontally polarized magnetic field

The RF penetration is characterized by transfer fUnctions that are ratios of the penetrant field to the incident field. In order to remove the antenna-to-measurement point distance variable, both the penetrant field and the incident field were recorded at the same distance from the antenna. The incident field is measured by directing the antenna toward an open region and measuring the horizontal magnetic field component at the surface of a large metal plate resting on the ground. Considering radiation conditions obtain, the incident electric field is computed from the magnetic field measurement.

Measured data for the penetrant field are compared with theoretical predictions to verifY the results. Interior measurements indicate that the rebar attenuation of the electric field was not significant above 120 A1Hz while the rebar attenuation of the magnetic field was not important above 200 A1Hz. Near the window there is little attenuation at all frequencies. However, there is significant attenuation, more than 50 dB, through the concrete rooflceilingfor all frequencies.

J. H. Reed et al. (eds.), Wireless Personal Communications

© Springer Science+Business Media New York 1997

132

1. INTRODUCfION

The electromagnetic coupling through structures from the exterior to the interior depends upon a number off actors [1]. Radio frequency signals in and around structures may vary considerably from point to point, since the energy may arrive from different directions at different times due to multiple propagation paths. The vector addition of signals from multiple paths may add together to produce a deep null at one point or a sharp peak at another. A complete deterministic theoretical analysis of the RF penetration through structures is not tractable, which suggests the use of a geometric/statistical model that accounts for both specular reflections from the walls and ceilings, and random scattering from equipment and various objects within the environment [1]. One such model that is being used by many investigators is called SIRCIM (Simulation of Indoor Radio-Channel Impulse-response Models) [2].

There are a variety of approaches to the study ofRF penetration of structures. For some applications a theoretical approach may be the best. This permits the penetration evaluation of structures that are not accessible for measurement. And a theoretical analysis is expected to be far less expensive than direct measurement. On the other hand, the results from the theoretical approach will have some associated uncertainty and may be highly inaccurate under certain conditions. Critical parameters may not be known and may not be attainable for some structures. Probably the greatest weakness of relying on theoretical analysis is that often erroneous calculations are not readily apparent.

Another approach to determining the RF penetration of structures would be one that relies primarily on a data baSe of measurements. For some applications this approach would be best. It would provide confidence in the accuracy of the penetration data for those structures tested. And there would be little need to determine the electromagnetic characteristics of the structure which may be inaccessible. However, there are some significant disadvantages to this approach. First, the data base that is accumulated may not be sufficiently complete to include all the structures of interest. Second, data collection is generally far more expensive than performing calculations.

It is expected that a combination of theoretical modeling and measurement is the best approach for the structures and frequencies of interest in wireless communication. For simple structures, the approach of relying primarily on theoretical analysis would be adequate where there is little risk in being able to develop an analysis with acceptable accuracy. For more complex structures it is recommended that the RF penetration be obtained from a few well-understood site-specific models that are verified and supplemented by measured data.

Two of the goals of the presented study are tIle quantitations of RF penetration through windows and concrete walls over the frequency range 100 MHz to 3 GHz. Personal-Comunication System (PCS) frequency bands extend from 800 MHz to 2.2 Ghz. In most structures there are a number of propagation paths between the source and the receiver operating in the PCS frequency regime. Consequently a structure was selected that essentially isolates propagation paths and provides direct quantitation of the attenuation for specific propagation paths. This can be achieved by illuminating a structure with only a few ports of entry for RF and with significant wall attenuation.

Building 12500 located on the A-15A site at Eglin AFB, FL was selected for this study. The walls of the building are 2-feet thick poured-in-place concrete and the ceiling is 3-feet thick concrete with a one

133

quarter-inch steel plate spot-welded to parallel 6" X 12" steel I-beam reinforcement. As a result of its construction, the flat roof/ceiling configuration should provide good, if inadvertent, electromagnetic shielding. The construction of the building tested is standard for a blast hardened facility. Outside walls of the structure are reinforced with horizontal and vertical rebar with 8-inch spacing both horizontally and vertically. There is only one exterior door located at the rear of the building and one set of windows located across the front of the building. Consequently, RF penetration through windows, doors and walls can be isolated and characterized separately.

The RF penetration is characterized by transfer functions that are ratios of the penetrant field to the incident field. In order to remove the antenna-to-measurement-point distance variable, both the penetrant field and the incident field were recorded at the same distance from the antenna. The incident magnetic field is measured by directing the antenna toward an open region to a magnetic field sensor that is attached to a large metal plate resting on the ground. Accordingly the measured magnetic field component is 2 X the incident field component. Assuming radiation conditions obtain, the incident electric field is 1201t Ohms X the incident magnetic field.

2. ANALYSIS

2 1 Analysis of RF Penetration through a slab of material

The analysis of the RF transmission through a slab of building material, as illustrated in Figure 1, may be accomplished by using transmission line theory. Considering plane wave propagation the line voltage represents the electric field, the line current represents the magnetic field, and the characteristic impedance of the transmission line represents the intrinsic wave impedance of the medium supporting the plane wave. In

Air Incident

Side ..

z =-d z=o

Air Transmitted

Side

Figure 1. Illustration of the transmission ofRF through a slab of concrete

134

detennining the transmission through a slab oflossy material with thickness "d" the analogous transmission line problem is the determination of transmission through a section oflossy transmission line, length "d", inserted in a lossless transmission line. The voltage Vo(z) and current Io(z) on the incident side of the section, for z :s: d, may be expressed,

and

Io(z) = ~O(e -)ll"z -roe +)ll"z)

o

Here Zo is the characteristic impedance of the transmission line and Po is the propagation constant. For the plane wave propagation problem being considered one would use

and

The reflection coefficient ro is determined by the application of appropriate boundary conditions. Note that <oJ = 21tfis the radian frequency.

For the region of the lossy line, -d :s: z :s: 0, the voltage and current may be expressed,

and

135

In the foregoing the complex propagation constant y and complex characteristic transmission line impedance for a lossy transmission line may be expressed in terms of the distributed series impedance and shunt admittance. However, for the plane wave propagation problem,

and

Y=jW~ /l(e _/J )=a+j~ w

Z =jW/l 1 Y

The complex amplitude VI and reflection coefficient r l may be determined by the application of appropriate boundary conditions.

For the exit medium, z ;, 0, the voltage and current take on relatively simple forms.

and

/2(Z)= ;2 e -}P.z o

Note that the exit medium and the incident medium are considered to have the same properties.

and

r _ I-(Z/Zo)

1 1 +(Z/Z)

136

Defining the transmission coefficient for the slab to be T = V2 / V. , while using V,(O) = V,(O) and V.(-d) = Vkd), yields the expression for the transmission coefficient

Note that there are two mechanisms for reducing the transmission coefficient, power reflection and power absorption through conduction and/or hysteresis.

2 2 Constitutive Electrical Properties of Building Materials

In order to evaluate the transmission coefficient, which quantitates the RF transmission loss, the constitutive electrical properties of the material must be known. Although some data exists, the constitutive electrical parameters are not well documented for building materials. Castillo and Singaraju [3] provide data recorded at Malmstrom AFB, MT for the frequency range 100 kHz to 100 MHz for concrete and soil. More recently Sou, Landron and Feuerstein [4] provide data over the range 1700 MHz to 18000 MHz for concrete, brick and limestone.

In Table 1. the effective permittivity and conductivity of a few building materials are provided for a range of frequencies from data collected by Sou et al. [4]. For their study samples of stone and brick were cut to waveguide dimensions using different kinds of band saws with some sanding to reduce air gaps as much as possible. However, the concrete samples were prepared by solidifying liquid concrete inside a waveguide for the measurements. The limestone came from Virginia Tech's physical plant, the brick samples were cut from a standard common face solid brick, and the concrete material was Bondex's concrete patch (item #39005).

In a recent study of the constitutive electrical properties of concrete, Halabe et aI. [5] developed a predictive model for the complex dielectric permittivity which considers frequency, temperature, moisture content, chlorine content and the concrete mix constituents. Parameters considered appropriate for the A-15A building used in this study ofRF penetration are 10% porosity, aggregate dielectric constant of 5, temperature of 25 ·C, and salinity = 50 ppt. An important factor is the moisture saturation level, i.e., the extent to which the air voids are filled with water. For horizontal bridge decks it has been reported that the moisture saturation level during the summer months varies from 40 to 50 percent [6]. Consequently, the moisture saturation level of the concrete in the A-15A building is assumed to be 40% since the RF penetration measurements were made during the summer months. And for the analysis the models developed by Halabe et al. [5] are used to compute the complex dielectric constant for concrete as a function of frequency.

137

Table 1. The Effective Permittivity and Conductivity of Building Materials (4)

FREQ. CONCRETE BRICK LIMESTONE

MHz E. o(S/m) E. o(S/m) ED o(S/m)

1723 6.11 0.133 4.62 0.0174 7.36 0.0235

2000 5.64 0.153 4.50 0.0173 7.25 0.0213

2600 5.10 0.146 4.26 0.0197 7.06 0.0179

3230 4.40 0.00324 7.13 0.0232

8200 4.45 0.0225 6.73 -0.0259

12000 4.40 0.0502 29.6 0.434

18000 4.11 0.0364 6.83 0.183

Combining the data from the Halabe model with the results from Castillo and Singaraju (3) and with the results from Sou et aI. [4] provides the constitutive electrical parameters that are shown in Table 2.

Table 2. Constitutive Electrical Parameters Measured for Concrete

FreQuencv in MHz Dielectric Constant Conductivity in S/m

10 (18.6*) (0.0102*)

50 (13.0*) (0.0215*)

100 7.93 (I 1.1*) 0.0337 (0.0347*)

1000 6.07 0.0684

2000 5.87 (5.66**) 0.083 (0.142**)

3000 5.80 0.095

* This result is taken from Castillo and Singaraju [3]. ** This result is obtained by averaging the parameters measured by Sou et aI. [4] over the range 1800 MHz to 2100 MHz.

The results shown in Table 2. exhibit sufficient consistency considering that different concrete samples are used in the measurements. However the measured conductivity from Sou et aI. [4] depart somewhat from the other data. Perhaps the Bondex Concrete Patch material that is used by Sou et al. contained an additive which altered the electrical conductivity. It is a common practice to add an epoxy resin

138

to concrete patch material to improve its adhesion property.

2 3 Computed Transmission Losses for Building Materials

Using the constitutive electrical properties in Table I, results are computed for the transmission loss through 4" and 8" walls and shown in Table 3. These results generally indicate that the attenuation losses are not very significant for the frequency ranges and the wall thicknesses considered. Since reflection losses are included along with the material attenuation versus frequency, some oscillations in the transmission losses are expected when the ohmic losses are low.

Table 3. RF Attenuation through Building Materials

FREQ. CONCRETE PATCH BRICK LIMESTONE

MHz 4 inches 8 inches 4 inches 8 inches 4 inches 8 inches

1723 10.4dB 19.5dB 3.4dB 3.3dB 3.0dB 5.5dB

2000 12.3dB 22.9dB 1.9dB 3.9dB 4.2dB 5.ldB

2600 12.0dB 22.8dB 3.2dB 4.4dB 3.7dB 5.0dB

3230 2.2dB l.3dB 2.8dB 5.3dB

8200 3.ldB 5.2dB 2.2dB -_ .... _-12000 4.7dB 9.0dB 18.9dB 32.ldB

18000 4.ldB 7.3dB 13.5dB 25.2dB

For the measurements that are presented in a subsequent section, the structure had concrete walls that were two feet in thickness with steel reinforcement rods forming a mesh inside the concrete. Since the electrical and magnetic connectivity of the rods is not known, the initial consideration of RF attenuation will focus only on the power loss associated with the concrete material. Using the constitutive electrical parameters shown in Table 2, computations are made for the attenuation expected for the concrete walls and ceiling of Building 12500 at the A-15A site. Of course the total attenuation for the ceiling including the 1-beams and the steel plate should exceed the concrete losses by a significant amount and the total attenuation through the 2-foot walls containing the steel reinforcement rods may exceed the concrete losses by a significant amount at the lower frequencies near 100 MHz. Estimates of the attenuation from rebar are provided by Rohrbaugh for a variety of parameters [6]. Results for the concrete attenuation without rebar are provided in Table 4.

139

Table 4. RF Attenuation through Concrete Walls

Freauencv in MHz 2-Foot Wall Thickness 3-Foot Wall Thickness

100 13.9dB 19.4dB

200 16.7dB 23.9dB

400 21.0dB 30.5dB

600 24.2dB 35.3dB

1000 28.7dB 42.2dB

1400 31.9dB 47.0dB

1800 34.5dB 50.9dB

2000 35.6dB 52.6dB

2400 37.8dB 55.9dB

3000 41.1dB 60.8dB

MEASUREMENTS

The building selected for studying RF penetration into buildings is a blast-hardened structure at the Eglin Air Force Base A-15A site on Santa Rosa Island, FL. Low-level continuous-wave illumination was used for the frequency band 100 MHz to 3 Ghz with the illumination and interior measurement points illustrated in Figure 2. Two antennas and four amplifiers were required to cover the frequency band. The

I XIII_il,"I ... ~Iian.'t'''''

I I I

Figure 2. Diagram of Building 12500 at the A-15A Site on Elgin Air Force Base, FL

140

antennas, log-periodic linear arrays, were positioned as close to the building as possible yet keeping the structure within the main lobes of the antenna patterns. Both antennas were oriented so that the illuminating electric field was vertical and the illuminating magnetic field was horizontal.

In order to quantitate the attenuation the illumination field was measured by directing the antennas toward an open region and measuring the magnetic field at points corresponding in distances and orientations to the internal measurement points of the illuminated building. The incident field is measured by placing a 4' X 8' aluminum sheet on the ground and locating a magnetic field sensor near the center of the sheet. The magnetic field sensor responds to the induced surface current on the plate that is simply 2 X the incident magnetic field strength. Only the incident magnetic field is recorded. The incident electric field strength is obtained by using the properties of the far zone radiation fields, i.e. the ratio of the electric field strength to the magnetic field strength is 120n Ohms.

The measurements of the "illuminating" field are shown in figure 3. These measurements were performed to establish the uniformity of the field within the main lobe and the reproducibilty of the measurements. At low frequency there is good uniformity but above 1 GHz the uniformity degrades. These do not represent the field developed by the antenna only, since reflections from the ground are included in the measurements.

10"

Aim

Frequency(Hz)

Figure 3. Incident magnetic field measured at the surface of a 4' X 8' aluminum sheet placed on the ground surface.

For the propagation loss in a two-feet thick concrete wall, figure 4 shows the ratio of the principal component of the internal magnetic field (HJ to the incident field for the measurement point B and the source point X. Minimum attenuation seems to occur in the frequency range 120 MHz to 205 MHz. In this range the

dB

1~ __ ----__ --__ ~~~~~~ ________ ~ __________ ~

10· Freq1.Ienc:y(Hz)

10'·

Figure 4. The internal magnetic field (HJ at measurement point B relative to the incident magnetic field for illumination source point X.

10· Frequency(Hz)

10';:)

141

Figure 5. The internal electric field (Ey) at measurement point B relative to the incident electric field for illuminating source point X.

142

attenuation is less than 20 dB, which is consistent with the calculations shown in Table 4. Apparently below 120 MHz rebar attenuation is significant. Note that the maxima and minima in the data of figure 4 are consistent with internal standing wave patterns and external resonances. Corresponding results for the electric field are shown in figure 5. Both figures are consistent with one another and with calculations. Because of the locations of the source and the measurement points for figures 4 and 5, multipath contributions are expected to be negligible.

Propagation through the windows of Building 12500 is shown in figure 6. Here the ratio of the principle component of the internal electric field (Ey) to the incident electric field is shown for measurement point A and source point A. These data show little attenuation of the field propagating into the interior of the

-40

·50

~O'":·:---------~~~'"';O·:-------~----10'1tl Frequancy(Hz)

Figure 6. The internal electric field (Ey) at measurement point A relative to the incident electric field for illumination source point A.

building. Correponding results for the principle component of the magnetic field (HJ are shown in figure 7. For the same illumination, a second measurement point is considered, where the internal measurement is located some distance from the window. In figure 8 the ratio of the principle component of the internal electric field to the incident field is provided for measurement point B and source point A. At low frequency significant attenuation is seen that appears to decrease with increases in frequency, which is consistent with propagation through apertures.

Although the data are not shown, Building 12500 was illuminated from directly overhead and internal field measurements were performed. In all cases the field attenuation exceeded 50 dB.

10~----------------------~----------------------,

-40

_~OL·------_--~---~-----1~O·-------------------....J10;O Frequency(Hz)

Figure 7. The internal magnetic field (HJ at measurement point A relative to the incident magnetic field for illwnination source point A.

10

0

-10

dB -30

-40

-50

~o· 10· Frequeney(Hz)

I

I

143

Figure 8. The internal electric field (Ey) at measurement point B relative to the incident electric field for illwnination source point A.

144

CONCLUSIONS

By illuminating a blast hardened concrete structure with only one door and one set of windows it is possible to isolate and study the modes ofRF penetration from the exterior to the interior of concrete structures. For two-feet thick concrete walls, the measured attenuation above 200 MHz appears to increase with an increase in frequency. And for frequencies near and below 100 MHz, rebar attenuation becomes significant and seems to affect the magnetic field more than the electric field penetration. In the frequency range between 120 and 205 MHz the attenuation is often less than 20 dB. For four-inch and eight-inch thick walls the attenuation is much less (see Table 3).

Propagation through the windows seems to be primarily line-of-sight propagation with very little attenuation seen near the window. At distant points from the window the attenuation is about 20 dB in the lower frequencies and the attenuation appears to decrease with increase in frequency. Inside the structure standing wave conditions occur that result in substantial oscillations in the variation of the penetrant field with frequency.

For propagation through the structure concrete roof/ceiling significant attenuation occurs.

REFERENCES

l. T.S. Rappaport and S. Sandhu, "Radio-Wave Propagation for Emerging Wireless Personal­Communication Systems," IEEE Antennas and Propagation Magazine, Vol. 36, No.5, pp. 14-23, October 1994.

2. T.S. Rappaport, S. Seidel and K Takamizawa, "Statistical Channel Impulse Models for Factory and Open Plant Buildings Radio Communication System Design," IEEE Transactions on Communications, May 1991.

3. J.P.· Castillo and B.K Singaaju, "Effects of Wave Reflection on Objects Near a Plane Ground," ATHAMAS Memo 8, Phillips Laboratory, Kirtland AFB, NM 8717,1 May 1975.

4. C.K. Sou, O. Landron, and MJ. Feuerstein, "Characterization of Electromagnetic Properties of Building Materials for use in Site-Specific Propagation Prediction," MPRG Technical Report #92-12, VPI&SU, Blacksburg, VA, 27 June 1992.

5. U.B. Halabe, A. Sotoodehnia, KR Maser, and E.A. Kausel, "Modeling the Electrromagnetic properties of Concrete," ACI Materials Journal, Vol. 90, No.6, pp 552-563, November-December 1993.

6. J. Rohrbaugh, "Shielding Effewctiveness of Concrete Literature Search Results," Internal report, Philliups Laboratory, PLlWST, Kirtland AFB, NM, October 1995.

14 Interference from Wireless Devices into Network

Equipment Inside COs Between 450 and 914 MHz

1. Introduction

Mahbub Hoque

Steve Harbin

ZhaofengMa

The increasing need for wireless communication in or near telephone exchange facilities has raised

concern for sustained integrity of telecommunications switching equipment operation through

immunity to electromagnetic interference (EMI). In addition to portable communication systems

(PCS), several wireless devices such as cordless phones, handi-talkies, and other portable

transceivers may be used near telephone company Central Offices (COs), huts, and remote nodes

at customer sites. The intent of this work is to perform a realistic analysis by considering the

presence ·of equipment cabinets and other scattering objects inside a CO and to better quantify the

electromagnetic compatibility of wireless transmitting devices with switching equipment inside

telephone company facilities. Measurements have been performed to determine the immunity of a

few network switches in the presence of wireless devices.

An engineering tool called Electric-field Prediction inside a Central Office (EPCO) is being

developed to determine the distribution of potentially interfering electromagnetic fields at any point

inside an equipment room, considering multiple reflections from the metal cabinets, walls, and

ceiling when the layout of the room is known. Such interference has also been observed during

measurements. Finally, critical distances that should be maintained between portable handsets and

telecommunications equipment to avoid interference have been calculated.

2. Immunity of Telecommunications Equipment

The potential for interference to telecommunications equipment depends on the immunity of that

equipment. The Network Equipment-Building System (NEBS) immunity level for network systems,

given in GR-I089-CORE [I], is 2 VIm from 120 kHz to IO GHz, and IO Vim from 10 kHz to 24

kHz. It then decreases logarithmically to 2 Vim at 120 kHz. An additional objective for immunity

to radiated electric fields is 10 VIm from 10kHz to 10 GHz. These electric-field immunity criteria

J. H. Reed et al. (eds.), Wireless Personal Communications

© Springer Science+Business Media New York 1997

146

can be used to calculate a critical distance around a system from which wireless device use should

be excluded. This distance will then provide a guideline for the use of wireless devices in

equipment areas. Recently, great concern has arisen about interference from cellular and other

wireless devices to medical electronics. In response to reports of interference to electric

wheelchairs, pacemakers, hearing aids, and emergency and operating-room monitoring equipment,

several hospitals have banned cellular phones beyond their lobbies. Similarly, several phone

companies ban cellular phones from equipment spaces.

3. Scattering from System Cabinets and Other Objects Inside COs

Most COs contain digital switches and other large network systems shielded by metal cabinets.

During some operation and for any fault detection procedures, the cabinet doors are kept 'Jpen.

When the doors are open, this shielding is removed, and use of wireless devices has a greater

chance of causing interference if the radiated field exceeds the system's immunity level. In order to

perform a realistic analysis, field distributions were calculated for different cabinet distributions

inside a CO. Figure 3-1 shows a typical cabinet distribution in a CO, where switch cabinets are

located next to each other in parallel rows.

Central office wall

Central office wall

Figure 3-1. Typical Cabinet Layout Causing Multiple Reflections Inside CO

With this configuration, radiation from a wireless device at position PI will depend mostly on

reflections from the walls and ceiling. However, the field exposed to the system's circuit packs

when cabinet doors are open will be higher than that of the free-space condition. The present

analysis models the scattering and multiple reflection from such metallic structures inside a CO.

However, the most common location for the use of wireless devices is at position P2 with anyone

of the cabinet doors open. Here, the switch cards are exposed to a high electric tield due to

reflection from the cabinet door on the opposite side.

The effects of radiation from a portable handset depend on the radiated power and the distance

between the handset and equipment. The radiated power of wireless devices varies over a wide

range, from 0.5 mW to 5 watts.

To determine the radiated fields from a portable handset, the handset was modeled as a monopole

antenna and a rectangular metal box[2]. The handset body is considered to be a perfect conductor.

The handset antenna and body are modeled as a monopole antenna, 8 cm long, attached to a

rectangular, conducting box with dimensions 15 cm by 2.5 cm by 6 cm, representing the overall

mechanical structure of a typical handset. The handset and its field distribution were modeled using

the Finite Difference Time Domain (FOTD) technique[3]. This analysis is focused on operation at

914 MHz, but some predictions were made for 450 MHz. The handset model is used in conjunction

with scattering objects to predict the electric'field inside network equipment cabinets.

4. Effects of Emission from Wireless Devices

The electric field from wireless devices was calculated at a circuit pack located inside a network

system cabinet when the rest of the cabinet doors were closed. The resultant scattered field was

calculated by using the Uniform Geometrical Theory of Diffraction[ 4]. The electric field inside the

cabinet was calculated while a wireless transceiver was moved toward cabinet SW I along paths I

and 2. No multiple reflections from the cabinets along path 1 were observed. However, the received

field at SW1 will depend on the scattering and multiple reflection from the ceiling and walls. Such

scattering was considered during this analysis, where ceiling and walls were treated as lossy

dielectric materials. Figures 4-1 and 4-2 show the received field distribution at SWI when a

wireless handset was placed at different distances along path 2.

147

148

'C' Q)

10 Gi E 8 £l '0 6 ~ 'C

4 Ql ;;:::: u 2 ~ Q)

0 iii

---250 mW along path-2 in Figure 1

- - - - 250 mW under free-space condition

....... 100 mW along path-2 in Figure 1

_. -. -100 mW under free-space condition .. : ... -. .. ,

.--.":-.-..... ~.~,:

0 0.25 0.5

."".,,'"

0.75 1.25 1.5 1.75 2 2.25 2.5 2.75 3

Distance from the handset (meters)

Figure 4-1. Electric Field Inside Cabinet SWI at Different Distances Down an Aisle from a Handset Operating at 450 MHz

16

14 --250 mWaloDg path-2 In Figure I

- - --250 mWunder rree-space condition

12 ..•..•. 100 mWaloDg path-2 In Figurr 1

- - -100 mWuDderrree-space condition

10

E VIm ... .:.~

"'" \

<~"\~~'<""\.,

0.2 0.4 0.6 0.8 1.1 1.4 t.6 1.8 2.2 2.4 2.6 2.8

Distance rrom the haRdset (meters)

Figure 4-2. Electric Field Inside Cabinet SWI at Different Distances Down an Aisle from a Handset Operating at 914 MHz

Figures 4-1 and 4-2 show that the radiated electric field is reflected from the two sides of the

cabinet walls, experiencing a waveguide effect caused by multiple reflection. The presence of

walls and ceiling was also considered for the analysis. When the handset was 1.25 meters from the

receiving point (SW2), the received field was 20 to 30 percent higher than the field under free­

space conditions at the same distance. However, the received field can be as high as 6 Vim (at 450

MHz) when the handset is 1.25 meters from the cabinet. Under free-space conditions, the received

field is only 3 to 4 Vim at 1.25 meters from the handset. Similarly, the scattered field can be as

high as 5 to 6 VIm (at 914 MHz) at 1.2 meters from the cabinet, which is almost 30 to 40 percent

higher than the field under free-space conditions at the same distance.

5. Field Exposure Measurements in a Digital-Switch Room

Experiments were performed to estimate the immunity of a few digital switches - including

5ESS® I, DMSTM2_100, and others - to radiation from wireless devices like cellular phones,

cordless phones, and prototype pes devices. The purpose of these experiments was to obtain

performance data on network switches under realistic conditions of interference from wireless

equipment. Measurements were made with the cabinet doors open. Measurements were also made

to observe the immunity of the switch to radiation from prototype pes radio ports and handsets.

5.1 Cellular and Cordless Phone Emission Inside a Switch Room

To simulate a worst-possible scenario, measurements were made with the cabinet doors open and

with handsets very close to the network equipment and its circuitry, as close as 2 to 3 centimeters

in some cases. Load boxes were used to simulate traffic, and systems alarms were monitored.

Measurements were made with a Handie-talkie radiating 3 watts at 450 MHz and a Micro-talk

cellular phone radiating a maximum of 600 m W at 830 MHz. Measurements were repeated for all

the switch cabinets. No major disruption in switch performance was observed. However, a few

minor alarm messages were triggered by the exposure. Among the equipment affected were the

administrative module, a power supply module, and an opto-electronics transceiver module.

Similar measurements were performed with a low-power cordless phone radiating a maximum of

0.5 mW at 914 MHz. These measurements were made in two parts. In the first, the immunity of

base-set radiation was studied. For this, the base set was placed inside the switch room, close to the

system cabinet, and the cabinet doors were opened. Measurements were made of the electric field

incident to the cabinet. Measurements indicated that electric fields as high as I VIm could be

incident at electronic components inside the switch cabinet.

I 5ESS is a registered trademark of AT&T.

2 OMS is a trademark of Northern Telecom

149

150

5.2 Emission from a Digital Cellular Phone

Measurements were also taken inside the switch room in the SWBT Technology Resource Institute

at Austin, Texas. Here a Freedom Link (FDL) system (including a base station and a handset) was

tested close to different switches and SONET equipment. The base station and the handset were

allowed to radiate at a maximum power level of 10 mW. To simulate a worst-possible scenario,

measurements were made with cabinet doors open and with the handset very close to the network

equipment and its circuitry, as close as 2 to 3 centimeters in some cases. Switch performance and

systems alarm were monitored. No disruption in switch performance was observed. However, when

the handset was allowed to radiate 600 mW (under external mode), an error in pulse mask in the

opto-electronic module of an FLM 150 system was observed. The objective of these tests was to

determine the qualitative effects of exposure from Freedom Link devices.

Further tests were conducted with the FDL wireless device close to 5ESS and DMS-JOO switches.

The switches were connected with the loadrack to simulate a normal traffic pattern. The base

station and the handset were allowed to radiate at a maximum power level (specified for internal

mode) of 10 mW. To simulate a worst-possible scenario, measurements were made with cabinet

doors open and with the handset very close to the switch circuitry - as close as 2 to 3 centimeters

in some cases. Switch performance and systems alarm were monitored. No disruption in switch

performance was observed. Later, the handset was allowed to radiate 600 mW (under external

mode), and was used close to the 5ESS switches (with the cabinet door open). A few alarm

messages were observed in the processor control unit. In fact, interference caused by such exposure

affected the administrative module, a power supply module, and an opto-electronic transceiver; as

a result, permanent disruption of the 5ESS switch performance was observed.

Next, a more detailed test was performed to determine the interference effect with the handset

radiating 10 mW of power. The test was repeated with more than 10 channels over the operating

frequency band of the handset. No interference was observed. The objective of these tests was to

determine the qualitative effects of exposure from FDL devices.

6. Critical Distance from Ii Typical Network Equipment

The primary objective of this investigation was to quantify the effects of EMI caused by wireless

devices on telecommunications equipment located inside COs, which effects also depend on the

immunity of the equipment. To provide an indication of the effects of using typical handsets inside

COs, critical distances for such handsets were calculated at 450 MHz and 914 MHz for 2 Vim and

1 Vim immunity levels of the equipment, as shown in Figure 6-1 .

'" 7 ..

~ 6 E 5 .5 ~ 4 .; 3 .., "5 2 .,c "C U

0

0

........ 10 vim at 914 MHz -C- 10 vim at 450 MHz __ 2 vim at 914MHz -0-2 vim at 450 MHz __ I vim at 914MHz -lr- 1 vim at 450 MHz

50 100 150 200 250 300 350 400 450 500

Radiating po\\er in mW

Figure 6-1. Critical Distances Between Telecommunications Equipment in Two Cabinets Along Path 2 (Figure 3-1) and a Handset Operating at 450 and 914 MHz

The analysis assumed that the handset was moved along path 2. At 914 MHz and 100m W

radiating power, the critical distance is estimated to be 30 cm when the immunity level of the

switch is 10 Vim. This distance increases to 2.8 meters and 3.8 meters for 2 Vim and 1 Vim

immunity levels. A similar analysis at 450 MHz revealed that for the 10 Vim immunity level the

critical distance is still 30 cm. This distance increases to 2.2 meters and 4.2 meters for 2 V 1m and

I Vim immunity levels of the switch.

7. Conclusions

151

The object of this investigation was to quantify the effect of interference on telecommunications

equipment inside COs that can be caused by operation of wireless handsets. The analysis

considered the multiple scattering effect due to the presence of metal cabinets inside a CO. It was

observed that under such a realistic environment the telecommunications equipment can be exposed

to a field intensity 4 to 5 times greater than that it would be under free-space conditions. This

preliminary analysis was performed using an engineering tool called EPCO, which considered

mUltiple reflections from the cabinets, walls, and ceilings. It was observed that for a typical

handset radiating 250 mW of power, the maximum radiated field can be as great as 9 Vim at

152

65 cm from the handset, while at the same distance the electric field along the antenna axis is about

5 Vim.

Measurements were also perfonned with a wireless device close to several telecommunications

switches, including 5ESS, DMSIOO, and FLM150. A few alann messages were observed in the

processor control unit when the handset was operating with 600 mW of power. In fact,

interference caused by such exposure affected the administrative module, a power supply

module, and an opto-electronic transceiver; as a result, permanent disruption of the 5ESS switch

performance was observed. Similarly, an error in the pulse mask in the opto-electronic module of

an FLMI50 system was observed when the handset was allowed to radiate 600 mW of power.

Analysis revealed that wireless devices operating between 450 and 914 MHz and radiating 100

mW of power should not be located within 2 to 5 meters from switch circuitry when the switch has

an immunity level of 2 Vim and I Vim. This restriction can be relaxed to 20 to 30 cm when the

switch immunity level is 10 Vim. In the future, more detailed investigations will be perfonned to

determine such critical distances. Analyses will also be perfonned to determine the compatibility of

PCS devices inside COs.

References

[I] GR-I089-CORE, Electromagnetic Compatibility and Electrical Safety - Generic Criteria for

Network Telecommunications Equipment (a module ofLSSGR, FR-64 and TSGR, FR-440),

Issue I (Bellcore, November 1994).

[2] SR-3277, Emissionjrom Wireless Devices Inside a Central Office, Issue 1 (Bellcore,

December 1994).

[3] G. J. Burke and A. J. Poggio, Numerical Electromagnetic Code (NEC) - Method of Moments.

Naval Ocean Systems Center, Technical Document 116 (January 1981).

[4] D. A. McNamara, C. W. I. Pistorius, and J. A. G. Malherbe, Introduction to the Uniform

Geometrical Theory of Diffraction (Boston: Artech House, 1990).

15 Fading Phenomena in the Wireless Environment

by

Frank Bohn, u.s. Anny Communications-Electronics Command

Paul Major, U.S. Anny Communications-Electronics Command

Louis Taylor, Telos Corp.

ABSTRACT

The personal radio communication industry has spawned much research into propagation phenomena in the 1 to 3 GHz frequency spectrum, and has provided technological advances that create opportunities to capitalize on the use of these higher frequencies for military tactical communications and civilian wireless uses. However, the interests of the personal communication industry has focused upon propagation path configurations that can be considered "high-low, " that is, paths in which one end is close to the ground (an individual user) and the other end is at a higher elevation and utilizes an antenna support structure such as a tower or a building, aircraft or orbiting satellite. Tactical military communication and other emerging wireless applications, on the other hand, usually employ "low-low" path configurations-faths between individual users where the antenna heights at both ends of each link will be 1.5 meters above the ground or less.

The difference in geometry of a "high-low" and a "low-low" path is rather obvious, but what is not so obvious is the difference in propagation phenomena, especially fading characteristics, of the two types of paths. A "high-low" path in an urban area is usually characterized by Rayleigh propagation in which no direct line-ofsight propagation path exists and all of the energy from radio transmitter to receiver is by forward scatter and reflections. A "low-low" path between two users is usually via a direct line-of sight propagation path, but with insufficient terrain clearance to support Gaussian propagation; therefore, the path is characterized as Rician.

This paper reviews the Rayleigh, Gaussian and Rician propagation phenomena; describes the different factors that are associated with each of these three types of propagation; and discusses research and experimental work to enable more accurate prediction of Rician propagation loss, expected fading and bit-error rates under different environmental (terrain, foliage, weather, etc.) conditions.

J. H. Reed et al. (eds.), Wireless Personal Communications

© Springer Science+Business Media New York 1997

154

INTRODUCTION

The purpose of this paper is to assist in avoiding some of the potential problems that arise

as higher and higher frequencies are used for communications in situations where both ends of

the path may be near ground-level and in opticalline-of-sight (LOS) with each other, but with

insufficient terrain clearance to achieve true radio LOS, i.e., at least six-tenths of the first Fresnel

zone clearance. Such situations arise when centimeter wavelengths are used for short-range

communications such as wireless (personal radio) applications and military tactical situations.

One specific problem being addressed in this paper is a problem resulting from the

program to provide voice and digital communications for the individual soldier on patrol. Such

a concept has been with us for quite some time (Taylor 1992), but when it comes to

implementing the concept there arises the very practical problem of selecting a frequency band

to use for such a service and the means of providing the service in the selected frequency band.

Another is the forthcoming use of frequencies above 1.0 GHz for personal communications

which will certainly lead some users to attempt to pass digital communications over paths with

low-low configurations that include the one described above.

If one considers the use of hand held radios operating at a frequency of 1.5 GHz over a

path of one kilometer, the radius of the first Fresnel zone at the midpoint of the path is 7 meters.

If the path is flat and unobstructed and the radios are held at head height of approximately

1.5 meters above ground the clearance of the LOS will be only 0.2 of the first Fresnel zone.

Therefore the path loss will not be 96 dB (the free space loss over I kilometer at 1.5 GHz), but

will average 113 dB with noticeable variations (fades) as either the transmitter or the receiver (or

both) move. The frequency of occurrence and the depth of these fades is the main point of

discussion in this paper.

REVIEW OF LITERATURE

In civilian configurations of cellular and personal communication systems the users of

the communication systems are highly mobile with a relatively fixed density distribution, i.e.,

locations with a high demand for channels today usually will be the same locations that have a

high demand tomorrow. In addition, most of the areas with a high demand for communication

channels will have suitable fixed locations for mounting base stations with adequate access to

the public telephone network. In the military configuration the individual soldiers who will use

155

obstruction, walking or running; the center of action may be highly volatile so a fixed base

station is not practical nor desirable, and areas of high density one day may have no need for

communication a day or two later; and access to higher echelons will be via radio

communications due to the absence of a wire line infrastructure in the forward combat zone,

Users of civilian personal communications in a rural environment will experience

Gaussian propagation or Rayleigh propagation most of the time interspersed with small areas

where Rician propagation is predominant (Steele 1992, Chapt 1), Users of civilian personal

communications in an urban environment will usually have Rayleigh propagation provided by

reflections from buildings since they rarely will have a direct, uninterrupted LOS path to the base

station (Lee 1993, 3),

In the military environment the lack of a base station will deny sufficient antenna height

for Gaussian propagation, and the poor antenna gain and low transmitter power necessitated by

the use only of battery-powered portable radios will deny use of forward scatter (and Rayleigh

propagation) because of the lack of sufficient power margin to overcome the greater loss

associated >yith forward scatter (unless there happen to be large reflecting objects between the

transmitting and receiving sites), As a result, most of the tactical communications at the soldier

level usually will be via Rician propagation,

In Figure 1 these three modes of propagation are illustrated, It must be realized that there

exists a continuum of propagation modes between and beyond the three distinct modes

illustrated in the figure, Pure LOS, i.e" free space propagation in which the propagation loss

decreases smoothly inversely in proportion to the square of the distance only exists between

orbital and outer space vehicles. The propagation loss of such a mode of propagation is

expressed:

L=32,45+20LOG(f)+20LOG(D) where:

f is frequency in MHz, and

D is distance between antennas in kilometers

At the other extreme are earth-bound antennas separated by such a great distance that the

only mode of propagation between them is by energy refracted from ionized regions above the

surface of the earth, i.e., skywave propagation, which includes near vertical incidence (NVI)

156

between approximately 300 kHz and 30 MHz (Goodman 1992, 20)-a mode of propagation that

appears to be gaining in tactical usefulness (Braunberg 1995). Between these two extremes, and

in the frequency range between 20 or 30 MHz and 1.5 or 2 GHz are the frequencies considered

for tactical ground-to-ground mobile and portable communication that employ the propagation

modes illustrated in Figure 1. The effect of the transitioning between the three modes illustrated

in Figure I is shown in Figure 2. When less than 0.6 of the first Fresnel zone radius in clearance

exists between the direct LOS path and the nearest obstruction the propagation loss is greater

than the theoretical free-space propagation loss. Under this condition the propagation is

identified as Rician and the loss is expressed by the "Plane Earth Pathloss Formula" (Steele

1992, 156):

L=40LOG(D)-20LOG(hlh2) where:

hi and h2 are the heights of the two antennas above effective

earth surface expressed in meters and D is the distance

in meters between them.

It is to be noted that when 0.6 of the Fresnel zone radius clearance does not exist the

propagation loss is independent of frequency and is only dependent upon antenna heights (Jakes

1974,83). Figure 2 shows the dependence of propagation loss upon antenna height when range

is held constant. Rician propagation prevails when the antennas are sufficiently close together

and the earth topography is sufficiently level that the propagation can be assumed to be over a

smooth flat earth, and the path clearance is from 0 to approximately 0.6 of the first Fresnel zone.

Rician propagation is characterized by energy going directly from the transmitting antenna to the

receiving antenna canceled by some energy which is reflected from the ground between the two

antennas. (The reflected energy undergoes a 180-degree phase reversal upon reflection, hence

the cancellation). The average propagation loss for this mode of propagation is represented by

the equation immediately above. As the antenna height becomes sufficient that greater than 0.6

of the first Fresnel zone clearance is experienced the propagation becomes Gaussian, as

represented in Figure lA and in the region indicated in Figure 2. As the antenna height is

increased the reflected energy oscillates between in-phase and out-of-phase with the direct

energy due to changes in the length of the path traversed by the reflected energy. As shown in

157

propagation. The amplitude of the resulting fluctuations shown in Figure 2, and the number of

cycles until the loss essentially can be considered equal to the free space value, is a function of

the characteristics of the reflecting objects, and ground reflection, that support reflected energy

adding to and subtracting from (depending upon phase) the direct energy.

By the same token, Rician propagation loss actually experienced will vary about the

value given by the formula dependent upon the reflections supported by natural objects and

artifacts between the two antennas. Smooth terrain covered with short, tufted grass between the

two antennas will provide very little reflection and the propagation loss will be essentially that

predicted by the formula provided above (Lee 1982,91). Terrain covered by mineralized rocks

of about one-half wavelength in diameter will provide reflections that could be greater than the

direct energy and the received energy could vary from zero to twice that predicted by the

formula. The rate of variation in received energy will be a function of the movement of the two

antenna with respect to each other and is caused by the reflecting centers moving about in the

reflecting area.

If the earth bulge interferes with the direct beam between the two antennas due to

distance between the antennas, or topography, as illustrated in Figure lC, there will be no direct

energy transmitted between the two antennas and all energy will be conveyed via reflections and

the propagation mode is known as Rayleigh. Rayleigh propagation is also experienced when

objects such as buildings separate the transmitting and receiving antenna so the energy must

travel a circuitous path or paths. Path loss for Rayleigh propagation paths are empirical and very

difficult to predict. Bullington has performed seminal work in this area for urban and rural areas

(Bullington 1947) and Okumura has accomplished the same for suburban areas (Okumura 1968).

Bullington'S work was expressed with nomograms and was updated by Egli (Egli 1957) and later

put in a form useful for computer solution by Longley and Rice (Longley 1968), and more

recently has been put in a form that can automatically use digital map terrain information, by the

Joint Spectrum Center (ECAC 1985 and ECAC 1993)

158

Steele introduced a K-factor to differentiate between the probability density functions

(PDF) experienced in Gaussian, Rayleigh and Rician propagation modes. He defined K as:

K=(power in the dominant channel)l(power in the scattered paths).

When K approaches a value of 32 Gaussian propagation is experienced, when K approaches 0

Rayleigh propagation is experienced. Between these two values the propagation mode is Rician

(Steele 1992, 12). He goes on to show that when K is sufficiently high that the propagation is

Gaussian a signal-to-noise ratio (SNR) of only II dB is required to achieve a bit-error rate (BER)

of 10-3 with non-coherent frequency-shift keying (FSK), while with Rayleigh propagation, with

K =0, a SNR of 30 dB is required to achieve the same BER due to the deep fades associated with

Rayleigh fading.

THE TACTICAL ENVIRONMENT

It can be seen from the above that the soldier with a hand-held radio will usually not be

in a sufficiently. elevated (and exposed) position to support Gaussian propagation, and has

insufficient transmitter power and antenna gain to support Rayleigh propagation; therefore the

soldier usually will experience Rician propagation.

As stated above, the variation in received power with Rician propagation is a function of

the reflecting environment between the transmitting and receiving antennas. The reflection from

a plane surface that is many wavelengths in dimensions can be easily predicted as it is a function

of the angle of incidence and polarization of the energy and the conductivity and dielectric

constant of the material (Jakes 1974, 81). This information can be used for design and

procurement planning, but it is only of general utilitarian value for determining what will happen

in a tactical environment where irregular surfaces of a large variety and shape, size and electrical

characteristics are present. Also, it must be considered that energy at any frequency above about

30 MHz will be appreciably attenuated by vegetation with the attenuation increasing with

frequency, density and moisture content of the vegetation (Welch 1995, Weisberger 1982).

The propagation loss formulae presented above explain the decrease in signal strength

that are observed when the distance between a transmitter and a receiver are increased, but do

not explain the more rapid fluctuations in signal strength observed at a mobile station, which are

known as short term fading phenomena (Lee 1993,361). The amplitude of the fluctuations is a

159

function of the size, shape, orientation, dielectric constant and conductivity of any departure

from a perfectly· smooth earth surface between the transmitting site and the receiver. The

average rate at which these irregularities occur is a function of the size, shape, and orientation of

these same irregularities and the speed at which the propagation path is lengthening or

shortening.

The soldier in the field cannot be expected to analyze the factors that will affect the

ability to communicate, but will be very concerned if denied communication. What is needed

are some typical scenarios, e.g., desert, snow, ice, rough terrain, grassy terrain, mountains, urban

buildings, etc., with examples of each, and training of the soldier in being able to recognize the

typical scenarios and relating them to potential communication impairments and methods to

surmount the impairments. For example, if large reflecting objects are available, they can be

used to good advantage by proper positioning of the transmitter and receiver sites; rough terrain

can be used to obtain a height advantage to facilitate communication; mountainous terrain can

provide height and reflecting advantages; defiladed positions will not enable communication

over any di~tance unless a nearby reflecting object is available; and maximum communication

range over smooth terrain can easily be determined, etc. Empirical measurements of typical

scenarios, coupled with photographs, could be used to provide course material for such training.

EFFECTS OF FADING ON DATA AND VOICE COMMUNICATIONS

In the analog voice world there is an apparent graceful degradation in voice quality

accompanied with an increase in noise as the SNR degrades. Shaping the analog bandpass and

use of emphasis and de-emphasis are the main tools available to increase intelligibility at a given

SNR. The success of a transmission was measured by the time it took to successfully transmit a

message of a given size from originator to recipient.

The tools available for digital transmission still include bandpass shaping, and include

use of packets or frames, error detection and correction, and interleaving. In addition, the digital

signal world provides tools to increase the probability of the successful transmission of a signal

in an environment where the presence of noise is a problem, if one is willing to trade-off time for

160

Since a digital message must be received correctly in its entirety to be useful, dividing it

into packets and/or frames permits correction of, or re-sending of only a small part of the

message which contains one or more errors instead of being concerned with the entire message.

Errors tend to occur in clumps, but error detection and correction is facilitated by dispersion of

errors; therefore, interleaving (dividing adjacent parts of a frame into error-control blocks that

are transmitted out of sequence--i.e., not putting all of one's eggs in the same basket) spreads

the clumped errors and reduces the burden on the error detection and correction scheme used.

This in tum reduces the number of overhead bits that must accompany information bits to

support error detection and correction operations. The disadvantage of packetizing framing is

that each transmitted packet requires a fixed overhead consisting of the identification of the

packet, the originator, the recipient, etc.; therefore, the smaller the size of the packets the smaller

the ratio of information bits to overhead bits.

This brings us back to the physics of fading. Rayleigh fading is typified by aperiodic

deep fades of very short duration; Rician fading is more uniform (but more frequent) in

occurrence,. shallower and of longer duration; and Gaussian fading is very shallow, similar to

white noise and .has the highest frequency of occurrence. Packet lengths that would have a

relatively high probability of being transmitted between Rayleigh fades would have a very low

probability of being transmitted between Rician fades under Rician fading conditions of the

same average SNR, and practically no probability of being transmitted under Gaussian fading

conditions with the same SNR.

If a system is to operate under Rician fading conditions, i.e., a "low-low" path

configuration using hand-held units with minimum transmitter power output, a modulation

scheme, packet (or frame) length and error correction and detection scheme should be selected

by the system designers to optimize the operation of the system and maximize the range over

which they will operate. This should be accompanied with a program to train the users of the

units to recognize situations where they can and where they cannot be used effectively.

The references used in this paper present cite abundant data concerning reflection

characteristics of various terrain and structures pertaining to Rayleigh propagation, but there is a

dearth of data pertaining to reflection characteristics pertaining to Rician propagation. The

variation in amplitude of a received signal under Rician conditions is caused by the variations in

amplitude and phase of the reflected signals that combine with the direct signal. The variations

161

and are a function of the size and distribution of reflecting points (areas) in LOS to ends of the

link.

The authors are venturing to obtain data through literature, ongoing experimental work of

others, and planned measurements to determine the reflection characteristics indigenous to

different types of environments and terrain, and the effects of the characterized reflection

characteristics upon Rician and Gaussian fading characteristics. For example, the sinusoidal

undulations in the Gaussian propagation region shown in Figure 2 will vary in amplitude in

proportion to the magnitude of the reflected energy arriving at the receiver, and the fading

characteristics (depth, frequency of occurrence, amplitude and duration of fades) in the Rician

propagation region will be a function of the reflection characteristics of the terrain between the

transmitter and receiver.

CONCLUSION

The purpose In preparing this paper was to deal with recent studies for military

communication systems that used an analysis methodology derived from commercial practice

that did not address the reality of the military situation. While it may be desirable to educate

military mobile radio users in the propagation phenomena they may encounter, it also seems

necessary to better define the situation to the system designers and analysts who are not familiar

with military operations.

In addition, it is imperative that the need be defined for a new propagation program to

deal with the envisioned situations whereby the Army proposes to use mobile communications

systems in frequency bands never before utilized for such service. Older propagation models

were narrowband and analog oriented; the future is digital and wideband. While the system

developer may appreciate more data being transmitted in narrow bandwidths, there is also the

BER and multi path effects on range and interference tolerance to be considered.

While this paper has been primarily propagation loss oriented, the effects on digital

signals, packet transmission, etc., must be considered as related matters. In the growing density

of users seeking use of the spectrum, sharing frequencies by different signal types must also be

studied---but this is a topic for another paper.

162

t=ud~uu_luumuummm~:uuuuuuJ .------~----~--:-------·-·-~·---·-----·-·-·l-·~~_~·;"·;_~-;~~~~;~;~=~ .... ~--:~:~.-.-------------------.

L = 32.45 + 20*LOG(t) + 20*LOG(D) dB, fin MHz, D=d1 + d2 in meters

Fig. 1 A, Gaussian Propagation

L=20*LOG(D) - 20*LOG(h l * h2), hi and h2 in meters

Fig. 1B, Rician Propagation

Fig. 1 C, Rayleigh Propagation

PROPAGATION MODE VERSUS FRESNEL CLEARANCE

FIGURE 1

2~.I------~~~~~~~~--~~~~~~~~~~~--~~~~~--~~~--~~~----

__ ~ __ ~

~~~ __ ~

.

200

~""'l.,.-,; ...

",--.,

..... "

.,.-j;'.

,., .. " .

. ,,, .. ,

.,,,

~;'i

',·:

"+::

,.H:,·,·L·'+'·;"',·.,'f;".'··:·;;:"·;·:';;;j'.'+:'·';,.":'·";'i,<.'·';'·""·':'··"·"';':'"

•.•. , .

. ;",

.•.. ,:

... , .

.. ",

! 1~ I IL

10

0

~ R

ayle

igh

Reg

ion

Gau

ssia

n R

egio

n

a I

1 1

1 1

'I 1

1 l' 1

1

1 I

I 1

1 II

i 'I

I I

I 1 II

I '1

'1

I I

I 1

1

I 1

I I'

I I

I ,

I I

I 1

I 1

I " II"

, f

11

I

I I

..a:

C'!

0 C'

! 'III

: fC

! CIC

! q

q 0

0 0

0 '"

..a

:: ~

~

N

C'!

..a::

~

~

N

N

N

N

..,

"! ..,

Fre

snel

Cle

aran

ce (u

nits

of

Fres

nel z

on

es)

Pro

paga

tion

Loss

Ver

sus

Fres

nel C

lear

ance

Fig

ure

2.

"': ..,

U! ..,

"! ..,

...

"! ...

..a::

"!

IX!

... ...

... C'

f "f

;

'" '"

.......

0\

Vol

164

REFERENCES:

Braunberg, Andrew. 1995. "High Frequency Radio Enjoys Renaissance." Signal. March: 65.

Bullington, K. 1947. "Radio Propagation a Frequencies above 30 MC." Proceedings of the IRE.

October: 1122-1136.

ECAC. 1985. "Terrain-Integrated Rough-Earth Model (TIREM)." Annapolis, MD: lIT

Research Institute for Electromagnetic Compatibility Analysis Center.

ECAC. 1993. "TIREMISEM Handbook" ECAC-HDBK-93-076U. Annapolis, MD: ITT

Research Institute for Electromagnetic Compatibility Analysis Center.

Egli, J. 1957. "Radio Propagation Above 40 MC Over Irregular Terrain." Proceedings of the

IRE. October: 1383-1391

Goodman, John M. 1992. HF Communication, Science & Technology. New York: Van

Nostrand Reinhold.

Jakes, William C. 1974. Microwave Mobile Communications. New York: John Wiley & Sons.

Lee, William C. 1982. Mobile Communications Engineering. New York: McGraw-Hill.

Lee, William C. Y. 1993. Mobile Communications Design Fundamentals. New York: John

Wiley & Sons.

Longley, A. G. and P. L. Rice. 1968. "Prediction of Tropospheric Radio Transmission Loss Over

Irregular Terrain." ESSA Research Laboratories ERL79-ITS67. Washington, D.C.: US

Goverrunent Printing Office.

Okumura, Y. and E. Ohmori, T. Kawano, and K. Fukuda. 1968 "Field Strength and its

Variability in VHF and UHF Land Mobile Service." Review of the Electrical

Communication Laboratorv. Sept.-Oct.: 825.

Steele, R. 1992. Introduction to Digital Cellular Radio. London, UK: Pentech Press Ltd.

Taylor, L. L. 1992. "Whither Tactical Communications." presented at MILCOM '92.

Weisberger, Mark A. 1982. "An Initial Critical Summary of Models for Predicting the

Attenuation of Radio Waves by Trees." DoD ECAC, ESD-TR-81-101. U.S.

Goverrunent Printing Office.

WeIch, C., Catherine Lemak and Lee Corrington. 1995. "A Model for Estimating

Electromagnetic Wave Attenuation in a Forest (EW AF) Environment." presented at

Advisory Group for Aerospace Research and Development "SPP Symposium 'Digital

165

Communication Systems: Propagation Effects, Technical Solutions, Systems Design.'"

Athens, Greece, 18-21 September.

16 Transmitter Timing Optimization in Digital Simulcast Networks

Rade Petrovic

Center for Wireless Communications

University of Mississippi

Abstract In this paper we analyze effects of differential delay of simulcast signals on an FSK receiver. Laboratory

and field tests indicate that a significant deterioration in performance occurs when participating signals

have similar amplitudes. FM clicks are observed at the symbol edges. Several improvement techniques

are considered. and the emphasis is put on the transmitter timing optimization. A new algorithm is

proposed for calculating the average differential delay over the whole multitransmitter simulcast area

weighed by a user population density distribution. Effect of the signal transmission delays is investigated.

and significant gains are found when the transmitters have widely different antenna heights. The proposed

technique can be used to optimize the transmitter network layout.

1 Introduction

Simulcast technique is widely used in RF communications in order to increase coverage area,

reduce coverage holes due to shadowing effect, and improve building penetration [1,2]. In

particular, it is applied in paging systems, and in the recently deployed Narrowband Personal

Communication Service (NPCS) system. Simulcast means that a signal is simultaneously

broadcasted by multiple transmitters into mutually overlapped areas. These multiple signals

arrive at the receiver with a random phase, and produce interference effects similar to multipath,

i.e. fading and phase noise [3-11]. This interference is especially damaging if the simulcast

signals have nearly equal power at the receiver and the difference in propagation delays is a

significant fraction of the symbol interval. Such situations occur, for example, when one

transmitter has a higher antenna and/or less shadowing than another. In paging systems, the

differential delay of one quarter of the symbol interval is considered acceptable [12].

NPCS system proposed by Mtel [13], and deployed in 1995, has the simulcast interference

problems enhanced in comparison to a typical paging system, due to a shorter symbol interval

J. H. Reed et al. (eds.), Wireless Personal Communications

© Springer Science+Business Media New York 1997

168

(higher bit rate) and a higher transmitter power (better coverage). The NPCS system can

compensate its disadvantages by a better transmitter timing optimization, which is the topic of

this paper.

2 Effects of the Simulcast Interference

In this section we will consider effects of the simulcast interference on Frequency Shift Keying

(FSK) signals. This modulation technique is widely used in paging systems and in the NPCS.

In particular, we will report our findings based on the laboratory and field tests using 4x4 FSK,

designed for Mtel's NPCS. In this modulation technique four subcarriers are spaced 10 kHz

apart, and each is modulated by four level FSK. The peak frequency deviation is 2,400 Hz, and

the signaling rate is 3,200 baud. The total bit rate in the channel is 25,600 bit/so

Due to the difference in propagation delay from different transmitters participating in the

simulcast, .or due to their synchronization errors, at each boundary between symbols there is an

uncertainty interval, where the receiver may see different symbols coming from different

transmitters. During this uncertainty intervals the simulcast signals may come with opposite

phases, which results in the phase shift acceleration [11], and FM clicks similar to those

described originally in [14, IS].

Those FM clicks have been observed both in field and laboratory tests. Their amplitude increases

as the difference of powers of participating simulcast signals decreases. Their width depends on

the bandwidth of the receiver; for wide bandwidth receivers the clicks are confined to the

uncertainty intervals, but for narrowband receivers FM clicks are broadened and cover the whole

symbol interval. Wide bandwidth receivers have disadvantage of excess noise, and in 4x4 FSK

narrow bandwidth is needed in order to reject the adjacent subchannels. Therefore we need to

optimize the receiver bandwidth.

We have investigated the effects of simulcast interference in laboratory using RF channel

simulator (HP 11759B), which can simulate the differential delay and the multipath fading. We

have measured the block-error-rate (BIER) as a function of the differential delay. The block

169

consists of 32, bit interleaved, BCH(32,21) codewords. Fig. 1 shows results for a static

simulcast (no multipath fading), of two strong signals (-80 dBm each, within 0.1 dB) and with

slight transmitter frequency offset (10 Hz). This offset emulates the real world situation and

eliminates some artifacts of frequency locked oscillators. The results are obtained for two

receiver bandwidths, 6 kHz and 8 kHz, and with and without FM click clipping. The clipping

is done after discriminator and before the data filter in the receiver based on digital signal

processing. In order to accommodate the possible local oscillator variations the clipping is done

at relatively high level that corresponds to 2,800 Hz deviation.

The results show that the receiver performance is significantly damaged by the differential

propagation delay, and both the increase of the receiver bandwidth and the introduction of the

clipping have positive effects. However, the equisignal conditions established in the lab are rare

in the field. So, we introduced independent Rayleigh fading for each simulcast signal,

corresponding to 2.5 Hz Doppler shift, while other conditions are unchanged. The results are

shown in Fig. 2. We can see that the BIER has improved significantly in comparison to Fig. I.

This can be explained by the decrease of FM click amplitudes due to the increased discrepancy

of the simulcast signal powers in the Rayleigh fading.

However, the effects of the simulcast interference are not negligible even in the case of

optimized receiver design. They depend on the environment, differences in antenna heights, type

of fading (Rayleigh or Ricean), position of obstacles, reflection surfaces, or scatterers. Some

improvements have been achieved in the field tests by equalizing the antenna heights or

downtilting their radiation patterns, which impose the penalty of reducing the coverage range.

The best results are obtained by tuning start of transmission moments for all transmitters in a

simulcast area.

3 Transmitter Timing Optimization

First, it should be recognized that it is impossible to completely equalize delays throughout the

area, even for two transmitters. Typically, the objective is to match the arrival time of simulcast

signals at an equisignal point. However, due to the shadowing effects, the equisignal points are

170

Fig. 1 Static Simulcast equal signals -80 dBm, 10Hz offset

0.9'---'--'---'--~--~-----'---'--~1---'

o . 8 -+-----,--------j-----+--r- I

0.7 j-----+----/'_i_ I--i----i-----+--+.--+--- ! ..

! I l,,/ Ql ! i A '0 0.6 !--Il-:T , ~ 0.5 --!--+~f--i----+----,/----l--~~~--l

! 04 -!I-I/I' j---j--~:~ ... f ... "J"-:"7::':-:~.-:':-~-+J.-~:;--:--'·~1·+-/-H-11-8:· O.O+---r-~--~---!---+---r--+--~---!----!

0.6

0.5

.! 0.4 o ... ... o t 0.3 Ql

.::£ ()

:§ 0.2

0.1

0.0

o 10 20 30 40 50 60 70 80 90 100

I.

differential delay [us]

Fig. 2 Rayleigh Faded Simulcast equal signals -80 dBm, 10Hz offset

JliI'

/ /;

/ I~/

--y /

/ /1/ /

~ --~C~~p ~""""'r ---_." '--------.--;.--------- . o 10 20 30 40 50 60 70 80 90 100

differential delay [us]

.-.\3 ...

8 kHz & clip. -I--

8 kHz, no clip.

6 kHz & clip. -.-6 kHz, no clip.

. .. \3-"

8 kHz & clip. -I--

8. kHz, no clip.

6 kHz & clip. -.-6 kHz, no clip.

171

arranged in an irregular contour and no perfect equalization is possible even on this line.

Further, this technique does not provide information about the effects of the simulcast

interference throughout the overlap areas of multiple transmitters. Different simulcast overlap

areas have different sizes, different importance, two or more transmitters can participate in the

overlap, etc. We feel that only a computer simulation can perform an optimization for a

complex outlay of transmitters with a meaningful information of the overall system performance.

The objective of the computer simulation should be the minimum differential delay in the whole

simulcast area, not only portions of it or characteristic points. Therefore we should define an

average differential delay for the whole area, 7.v ' It is calculated as an average of the

differential delays over a mesh of points in the coverage area. The algorithm for the calculation

can be described through following steps:

1. - Generate a mesh of m points in a simulcast area. The density of points

corresponds to a prediction of user density. Alternatively, a uniform distribution

of mesh points is chosen but a weight to each point is assigned according to the

user density prediction.

2. - Calculate signal strengths of all relevant transmitters (with power above a

threshold), Pi' i = 1, 2, ... n, for each mesh point. In this phase it is done by

Hata's model for path loss [16], that does not include topological details. In a

later phase a software that uses the digital terrain data will be included.

3. - Calculate the signal arrival time, ti , for signal from i-th transmitter at each point.

It is calculated based on a direct path. If there are large reflecting surfaces we

can consider them as independent transmitters.

4.- Calculate the average arrival time of the signal at each point [1]:

172

n

tav=LPitilLPi i-l i",l

5. - Calculate the differential delay at each point:

n n

'j= L Pi(ti -tavl 2 ILPi' j=1,2,00.,m i=1 1=1

6.- Calculate the average differential delay for the whole mesh of m points:

m

'av= L a j , jim j=l

(1)

(2)

(3)

where aj are weighing coefficients that are chosen according to the area

importance. For example more weight can be assigned to downtown areas than

to suburban areas.

It is important to note that a single parameter is used to characterize the whole simulcast area,

so that random errors in the signal power calculations (for example, due to the shadowing

process) will be smoothed by the averaging process. Therefore we expect that in most cases the

method based on Hata's model for the signal power calculations is adequate.

Fig. 3 shows the differential delay calculated for a coverage area of the size 60 kIn by 40 kIn

with two transmitters with equal powers, one with the antenna height 200 m, having coordinates

(10 kIn, 0), and the other with the antenna height 100 m, located at (-10 kIn, 0). Both

transmitters operate simultaneously, and the zero differential delay occurs along the x = 0 line.

On the side of the higher antenna the differential delay is relatively low, and drops to zero, as

the coverage of the lower antenna ends. However, on the side of lower antenna the differential

delay is much larger, especially for x < -10 kIn. The higher antenna provides less path loss and

overshoots the lower antenna. This effect was verified in our field tests.

Fig 4 shows the distribution of the differential delay under the same conditions, except that the

lower antenna signal is delayed by 30 JLS after the higher antenna signal. Naturally, the

173

60

tau [us]

-20

Fig. 3 Differential delay variations over the coverage area; simultaneous broadcast.

tau [us]

Fig. 4 Differential delay variations over the coverage area; 30 J.lS delay at the lower antenna.

174

differential delay behind the lower antenna is reduced by about 30 /LS. However, the differential

delay between transmitters and behind the higher antenna are increased. However, the average

propagation delay is still lower.

Fig. 5 shows how the average differential delay changes as a function of the signal delay at the

transmitter with lower antenna (Txl). The averaging is done on 61 by 41 points square mesh

with equal weights. The conditions are the same as for Fig. 3 and 4, except that the height of

the antenna at (10 lan, 0) is varied from 100 m to 200 m in 10 m steps to produce a family of

curves. For a small difference in antenna heights only a small gain could be obtained, but for

a large difference the gain is significant. The minimum of the curves is broad, so that a small

timing error of the order of 5 /LS is insignificant. The timing of NPCS transmitters in Mtel

network is controlled by the Global Positioning System with precision of 1 /Ls, so the above

conditions can be easily met.

Fig. 5

>. m

Q3

20

-c 15 m

:.;::; c: ~ ~ 10 -c Q)

~ 5 ~ m o o

Effect of antenna height h1 =1 OOm, h2=100m-200m

hl .. JOOm

20 40 60 Tx1 delay [us]

175

The optimization of transmitter timings for a multi-transmitter simulcast network is done through

an iterative process, where in each step we calculate new average differential delay. When the

optimization is done for a given layout of transmitters we can investigate effects of changing the

layout. New layout would produce a different minimum average oelay which should be a

criterion for the transmitter network design (together with the coverage prediction and the price

for deployment).

4 Conclusion

Simulcast interference has adverse effects on the system performance especially in the case of

similar simulcast signal amplitudes anda large differential delay. We have illustrated the effects

in the case of 4x4 FSK modulation used in the NPCS network. The effects can be mitigated by

receiver optimization, by antenna height equalization or by antenna radiation pattern downtilting.

All those techniques have some adverse effects on the overall coverage area. The only technique

that brings. no such penalty is the transmitter timing optimization.

We have proposed here a computer simulation technique that calculates average differential delay

throughout the simulcast overlap area, weighed by signal power ratios and user population

density. Calculation is done over a computer generated mesh points within a predefined coverage

boundaries. Signal strength is calculated using the Hata model. Optimization of transmitter

timings is done through an iterative process, with the objective of minimizing the average

differential delay. Repeated calculations with modifications in the transmitter layout and/or

modifications in antenna radiation patterns and/or heights can be used for the network design'

optimization.

References

[1] G. C. Hess, "Land-Mobile Radio System Engineering", Artech House, Inc., 1993.

[2] G. D. Gray, "The Simulcasting Technique: An Approach to Total-Area Radio Coverage," IEEE

Transaction on Vehicular Technology, Vol. VT-28, No.2, May 1979, pp. 117-125.

[3] T. Hattori, K. Hirade, and F. Adachi, "Theoretical Studies of a Simulcast Digital Radio

176

Paging System Using a Carrier Frequency Offset Strategy", IEEE Transaction Vehicular

Technology, Vol. VT-29, pp 87-95, Feb. 1980.

[4) T. Hattori and S. Ogose, "A New Modulation Scheme for Multitransmitter Simulcast

Digital Mobile Radio Communication", IEEE Transaction Vehicular Technology, Vol.

VT-29, pp. 260-270, May 1980.

[5) J. Ade, "Some Aspects of the Theory of Simulcast", 32nd IEEE Vehicular Technology

Conference, San Diego, CA, May 1982, pp.133-163.

[6) T. Hattori, K. Kaneko, and T. Nagatsu, "Multitransmitter Simulcast Digital Signal

Transmission System Using Waveform Offset Strategy", Rev. Elec. Commun. Lab, vol.

30, no. 2, pp. 299-307, 1982.

[7) F. Muratore, V. Palestini, and F. Pattini, "4-PAM.FM with Noncoherent Detection for a Pan­

European Paging System", IEEE Trans. on Comm., Vol. 41, no. 11, pp. 1618-24, Nov. 1993.

[8) R. Petrovic, W. Roehr, and D. Cameron, "Multicarrier Permutation Modulation for Narrowband

PCS", 3rd Virginia Tech, Symposium on Wireless Personal Communications, June 1993.

[9) R. Petrovic, W. Roehr, and D. Cameron, "Muiticarrier Modulation for Narrowband PCS", IEEE

Transaction on Vehicular Technology, Vol, 43, No 4. November 1994.

[10) R. Petrovic, W. Roehr, D. Cameron, K. Tano, "Performance Analysis of Digital FM in

Simulcast Environment", 4th Virginia Tech. Symposium on Wireless Personal Communications,

June 1994.

[11) R. Petrovic, W. Roehr, D. Cameron, "Simulcast in Narrowband PCS", Proceeding of IEEE

MTT-S Symposium on Technologies for Wireless Applications, February 1995, pp 65-69.

[12) PacTel Paging Telesis Technologies Laboratory, "Advanced Architecture Paging Experimental

License", Presentation to the Telocator High Speed Committee, May 29, 1992.

[13) Mobile Telecommunications Technologies Corporation, "Petition for Rulemaking to Allocate 150

kHz in the 930-931 MHz Band and to Establish Rules and Policies for Nationwide Wireless

Network (NWN) Service", Before the FCC, Nov. 12, 1991.

[14) S.O. Rice, "Noise in FM Receivers", Chapter 25, pp. 375-424, in Proceedings,

Symposium on Time Series Analysis, M. Rosenblatt (ed.) John Wiley & Sons, Inc.,

N.Y., 1963.

[15) J.E. MalO and J. Salz, "Theory of Error Rates for Digital FM", BSTJ, vol. 45, pp.

1511-35, Nov. 1966.

[16) M. Hata, "Empirical Formula for Propagation Loss in Land Mobile Radio Services," IEEE Tr.

on Veh. Tech., Vol. VT-29, No.3, August 1980, pp. 317-325.

17 The Prediction of the Performance of Diversity Systems

by Modal Analysis and FDTD

1 Abstract

G.Dolmans Electromagnetics Division

Eindhoven University of Technology, the Netherlands e-mail: G.Dolmans@ele.tue.nl

Two theoretical full-wave models of propagation inside indoor environments are developed using modal analysis and the finite difference time domain technique. Propagation and system analysis are combined, resulting in aprediction of bit error rates, signal-to-noise ratios and coverage inside buildings. A comparison is made between the developed three-dimensional deterministic models and a stochastic model based on Rayleighfading characteristics.

2 Overview of European Radiowave Propagation Research

In Europe, many cooperations between universities, industry, and research institutions exist for obtaining insight into the characteristics of the mobile radio channel for the VHF and UHF frequency bands [1]. An example of such a cooperation is the European Cooperation in the Field of Scientific and Technical Research (COST). As a result of several COST projects, deterministic and stochastic prediction models have been devised for the 900 and 1800 MHz bands. Deterministic models are often used for electromagnetic field prediction and the stochastic modeling is normally used for system simulation. It is important that both re­search areas will merge to a common one [1]. One research activity that combines the electromagnetic field approach and the system simulation analysis has been started by a cooperation between Philips Research Laboratories and the Eindhoven University of Technology.

The goal of this cooperation is to obtain a detailed knowledge of radio wave propagation inside indoor environments. Knowing the field distribution inside a building, and describing the interaction between antennas and the electromagnetic fields precisely, smart antennas can be designed which are optimized for indoor use. Due to the fact that portable terminals are in close proximity to the human body or to other objects, the antenna and object must be considered together. This means that the antenna must be designed in such a way that the antenna and the nearby object functions properly in a rapidly changing propagation environment.

3 Review of Propagation and System Analysis

In the following section, two deterministic near-field propagation models will be described. In section 3.2, the interaction between radio waves and the receiving mobile system will be analyzed (system properties of a mobile receiver).

3.1 Propagation Models

A communication system devised for the UHF or VHF frequency band is subjected to multipath fading. Multipath fading can be described by envelope spatial fading (narrowband fading), frequency selective fading (time delay spread) and time selective fading (Doppler spread). Only the spatial narrowband fading will be considered in this paper. For indoor communication, it is a valid assumption to neglect the Doppler spread due to the low speed of the mobile receiver.

J. H. Reed et al. (eds.), Wireless Personal Communications

© Springer Science+Business Media New York 1997

178

Although the model can be used to predict time delay spread, the sensitivity of the receiver to delay spread will be presented in a future publication. We will focus on propagation models that are capable of predicting full wave solutions inside indoor environments, including near-field effects.

Most deterministic models use ray-tracing to describe wave propagation inside bUildings. The high­frequency approximation of the ray-tracing technique limits the application of the model to large objects and canonical geometries with known diffraction coefficients. Furthermore, the ray series is weakly con­vergent inside highly reflective environments, indicating that other techniques may be better suited to predict the electromagnetic fields. Recently, several two-dimensional full-wave propagation models are proposed for solving Maxwell's equations for indoor propagation problems. In these approaches, the two­dimensional propagation problem is solved by using the Finite Difference Time Domain Technique [2], the Method of Moments [3], and modal analysis [4]. A two-dimensional solution can be used for power level predictions, however, it is our aim to investigate three-dimensional models that incorporate all fading effects.

The first three-dimensional propagation model that will be proposed is based on an eigenfunction expansion of the electromagnetic fields inside a shielded room with two dielectric side-walls. This model can be used to predict wave propagation inside highly reflective environments. Furthermore, it can be used to evaluate the accuracy and efficiency of numerical approaches. .

Wave propagation inside a room containing objects will be analyzed with a numerical technique based on a space-time discretization of Maxwell's equations. The finite difference technique (FDTD) has been changed in such a way that the two propagation models can easily be compared to each other.

It is not sufficient to consider wave propagation only, for example, the received signals depend on the polarizations of the antennas. Usually, the signal received by the antenna system is obtained by multiplying each ray with the gain pattern. However, this calculation is based on far-field approximations, which may not be valid for a portable receiver close to an object or to a human body. The receiving properties of the antenna system in a rapidly varying electromagnetic field will be described rigorously in this paper by using the Lorentz reciprocity theorem.

3.2 System Properties of a Mobile Receiver

Using the Lorentz theorem, the induced voltage on the antenna can be calculated, which will depend on the incident electromagnetic fields and the current distribution on the receiving antenna. The signal power is obtained by calCulating the induced voltages, whereafter the signal-to-noise ratios are available at each location inside the building. In literature, closed-form expressions are available for calculating the proba­bility of error in terms of the average signal-to-noise power at the input of the receiver. The probability of error will depend on the dimensions of the room, the dielectric properties of the walls, the transmitting and receiving antennas and the chosen modulation scheme.

Diversity techniques can reduce the spatial fading in such a way that acceptable transmission of information is obtained. Three implementations of a space diversity receiver will be compared to each other: selection diversity, maximum-ratio combining and equal-gain combining. The coverage inside the room for a specific probability of error will be obtained using the deterministic propagation models.

Stochastic techniques can be employed to analyze the performance of receivers in a multipath environment. Statistical models are based on making an assumption of the spatial and temporal field distribution inside an indoor environment. The value of the probability distribution can also be obtained by measuring a large set of channel responses. The parameters used in the statistical models should be chosen carefully in order to reproduce most of the propagation conditions that can occur inside an indoor environment.

179

Because it is our goal to study the influence of the layout of several buildings on the field pattern, and the coupling of the antenna to nearby objects, we will focus on deterministic models. However, in the last section a comparison will be made between a statistical model and the developed deterministic full-wave models.

4 Indoor Radiowave Propagation: Full Wave Description

4.1 Modal Analysis

Recently, a rigorous modal solution is described in [4], using a two-dimensional waveguide model. This method has the ability to evaluate the effects of thickness of the walls, material properties and frequency on the electromagnetic field pattern. Although two-dimensional propagation models can be used for power level predictions, a three-dimensional model will be better able to predict the variations of the electromagnetic fields on a wavelength scale. Dolmans et al. [5] described the electromagnetic fields inside a room with perfectly conducting walls using modal techniques. This technique will now be extended to a perfectly conducting room with two lossy dielecttic side-walls. The geometry is shown in figure 1.

Z

s c

V, S, S,

~ r,=(x"y"z,)

o x ~'-----+-----r-----------------+----~

a d, d, 0

Figure I: Side-view of room

The dimensionE of the room in the X-, y- and z- directions are given by a, band c, respectively. The boundary S is made of perfectly electric conducting material. The volumes Vi and V3 consist of lossy dielecttic material. The thicknesses of the dielectric walls are denoted by d] and a - d2 , respectively. In each volume, the permittivity is given by E( T) = < - j E:' , for i=1 ,2,3. A point-source transmitter is located at the coordinates (X., Ys, zs).

The electtic and magnetic fields are expressed in terms of the vector potentials by using

~ [~I ~] Ei(T) = -jw Ai(T) + k; \7\7.Ai(T) , ~ I ~

Hi(T) = -\7 x Ai(T), JLi

(I)

where Ei(T) and Hi(T) are the electtic and magnetic fields, respectively. The wave number in volume i is denoted by ki and the angular frequency is given by w.

Two components of the vector potential are needed to determine the electromagnetic fields of point sources directed into the y- and z- directions. An x-directed point source needs only one component. In this case, the vector potential ,( can be written as Aixex.

180

Using the boundary conditions of the vector potential Aix on the surfaces 5, 51 and 52, and using the Helmholtz equations results in

Aix = L L fmn(r', r-;). (2) m=ln=l

Substitution of equation (2) in (1) yields an eigenfunction expansion of the electromagnetic fields inside the room. The expressions for the electromagnetic fields are described in [6] and will not be repeated here. The infinite double series will be truncated for computation on a computer. An error analysis has revealed that the truncation error does not depend on the length of the room (which is denoted by the symbol a) [6]. This means that propagation in large rooms can be calculated very efficiently with the presented double series.

Up to now, electromagnetic fields inside a room caused by an electric point-source bave been analyzed. The results can be used for calculating the fields caused by arbitrary current distributions on the transmitting antenna. For example, the x-component of the electric field due to an x-directed half-wave dipole transmitter is given by

l x ,+>./4 Eix = Ef: COS(k2(X, - x'))dx',

x,->.j4 (3)

where Ef: is the electric field caused by a point source and A is the wavelength of the transmitted signal. The electromagnetic fields caused by a half-wave dipole transmitter can also be described in terms of a double series.

We will now investigate the scattering and absorption of electromagnetic fields at the dielectric walls. The configuration is presented in table 1.

Table 1: Parameters room with two dielectric walls

I dim I source lobs plane I dielectric walls I frequency

a=5m x, =0.752 m 0<x<5m E~ = 4.5, E~ = 0.13 f =2GHz b=4m y, = 0.202 m 0<y<4m d1 = 40 em, a - d2 = 40 cm c=3 m z, =0.4m z =2.26 m

The magnitude of the x-component of the electric field (radiated power is equal to 200 m W) will be visualized in a large observation plane (33 A x 27 A). It is clear from figure 2 that the electric field pattern is very irregular inside the observation plane. At some points inside the room, the electric field almost vanishes. A receiving antenna positioned in a fade will not function properly, therefore, diversity systems will be analyzed in section 6 to reduce the rapid signal fluctuations inside the room. The effect of the dielectric wall on the electric field pattern is demonstrated in figure 3.

181

width[mj

: Pill -10 -3 12 19 26

Ex (d8V/ml

Figure 2: lEx I in a room with two dielectric walls

Ex [dBV/m)

Figure 3: IExl near a dielectric side-wall

The modal expansion method can be used to analyze the electromagnetic fields inside two perfectly con­ducting rooms divided bya dielectric wall. The electric field in this configuration is shown in figure 4. The shape of the dielectric waIl is clearly visible in this figure. Up to now, wave propagation in an empty room is described. These results will be used later on for system performance calculations. Furthermore, the results will be used for comparison with stochastic models. The field distribution inside a room containing objects will be investigated in the next section by using a finite difference algorithm.

182

0.0 0.6 1.2 1.8 2.4 3.0

widtb{mJ

-10 -0 19 29 38

Ex [dBV/m]

Figure 4: IExl inside two rooms divided by a dielectric wall

4.2 Finite Difference Time Domain Technique

The Finite Difference Time Domain technique is a computational method that calculates the temporal evo­lution of electromagnetic fields within a region of space by stepping through time. Yee [7] has proposed a staggered grid for the placement of the electric and magnetic fields in space and time. A combination of first and second order accurate boundary conditions at the truncation of the mesh will be used for implementation of the method on a computer.

The FDTD method is a flexible numerical tool that can handle combinations of metallic and lossy dielectric structures. Using this method, the influence of the human head on the electromagnetic field pattern inside an indoor environment will be studied. The head will be modeled by a sphere of radius 9 cm, the hand is modeled as being 10 em wide and 2 cm thick wrapped around the lower part of a telephone hand-set. The hand and the head are modeled by a dielectricum with a real part of the permittivity equal to 50 and losses equal to 1.2 Sim [8]. The telephone hand-set is modeled by taking the tangential electric fields equal to zero (perfectly conducting). The combination of hand-set and human head is placed inside a small room. The walls of this roorri are 10 cm thick and the dielectric properties of the walls are given in table 1.

Usually, the excitation in the FDTD method is modeled by a voltage source, which enforces the electric fields to a particular value at one cell of the FDTD mesh. In the previous modal model, current distributions on the source were used to obtain the electromagnetic fields inside a room. In order to be able to compare the two propagation models, a current point source will be used in the FDTD method. The temporal behavior of the current source is modeled by a raised cosine pulse with a frequency of 900 MHz. This vertically polarized current source is located just above the perfectly conducting hand-set.

The dominant z-component of the electric field will be calculated in a vertical plane crossing the handset, the head and the dielectric walls of the room. The cell size is I cm x Icm x Icm (33 cells per wavelength) and the distance to the absorbing boundary is one wavelength. The total number of cells is 200 x 200 x 200 = 8 million cells. The antenna is positioned at the coordinates (1.3 m, 1.3 m). The magnitude of the z-component of the electric field is presented in figure 5. The dielectric walls are clearly recognizable, furthermore, it is evident that the dielectric head positioned in the middle of the room has a large influence on the electric field pattern.

183

0.0 0.5 1.0 1.5 2.0

height [m]

-10 -4 14 20

Ez [d8V/m]

Figure 5: Effect on thejield pattern of the presence of a person inside a small room

5 Signal Description at Receiver Including Near-Field Effects

The reception of electromagnetic fields by the receiving antenna is detennined by the shape and the direction of the antenna. In order to design an optimal receiver, it is necessary to model the interaction between the electromagnetic fields and the type of receiving antenna precisely. This will be done by deriving a Thevenin representation of the receiving antenna using the Lorentz reciprocity theorem.

The antenna operates in two different states: the receiving state and the transmitting state. In the transmitting state, the electromagnetic fields are given by (ET , fiT). In the receiving state, the electromagnetic fields are incident upon the receiving antenna. These electromagnetic fields are represented by ("E R, fi R). The fields in the receiving state can be divided into the incident field (E I , fiI), which is the field inside the room without the receiving antenna present, and the scattered field (Es, fiS). In the tenninal domain of the receiving antenna, a low-frequency approximation of Maxwell's equations will be used. Applying Lorentz's reciprocity theorem to the part of the antenna that is susceptible to electromagnetic radiation and to the tenninal domain results in

ER = v R - ziR = J Isa (EI X h,T - if' X fiI) . ridS, (4)

where the impedance z of the antenna is given by the quotient of the port voltage vT and the port current iT in the transmitting state. The surface Sa is chosen to coincide with the conductors of the antenna. The transmitted fields are nonnalized to the input current, resulting in the nonnalized fields (if', h7).

184

It will be assumed that the antenna system is terminated with a perfectly matched load. The equivalent Thevenin representation is given in figure 6. The receiving antenna is represented by an ideal source ERin series with the antenna impedance and the load of the antenna.

+

z· vI:

Figure 6: Thevenin representation of the receiving antenna

The time-average power that will be delivered to the load is given by

P _ ~R { R(.R)*} _ IERI2 _ II Is.((ii x -;;7). if + (ii x tT) .jjf)dSI2 L - 2 e vL ~L - 8Re(z) - 8Re(z) (5)

It will be assumed that the thermal noise can be described by a Gaussian process. The noise power at the load of the antenna can be written as

Pn = kTB, (6)

where k is Boltzmann's constant, T is the temperature of the resistor and B is the bandwidth of the noise. We can now define the local signal-to-noise ratio as the local time-average signal power divided by the time average noise power

S So PL I I Is. ((ii X liT) . if + (ii x tT) . jjI)dS12 N = IOlog No = 10 log Pn = IOi0g 8kTBRe(z) (7)

In most modern mobile communication links, digital transmission techniques are used to transmit and process the information. As an example, we will study the bit error rate for a noncoherent FSK (frequency shift keying) modulated signal. The bit error rate of a receiver located in a room is given by

1 -So 1 -I I Is. ((ii x -;;7) . if + (ii x tT) . JJI)dSI2 Pe = zexp 2No = zexp 16kTBRe(z) (8)

The bit error rate depends on the dimensions of the room, the dielectric properties of the walls, the directions and locations of the source and receiver, the impedance of the antenna and the bandwidth of the transmitted signal. In the next figure, the relationship between the bit error rate and the signal-to-noise ratio will be. compared for a fading (modal analysis) and a non-fading environment. The room configuration is given in table 1. The receiving antenna is modeled by a half wave dipole antenna. The received bit error rates are calculated at 6400 points inside the room. Due to the rapid spatial variations of the received voltages inside an indoor environment, a very high signal-to-noise ratio is needed in order to get an acceptable quality of the signal at the receiver. It is well known that diversity receivers can reduce the signal fluctuations inside fading environments. The effect of space diversity will be studied in the next section using the full-wave prediction models.

l.Oe - 01

~ l.Oe - 02

~ ..... l.Oe - 03 o

& LOe - 04 ~ .g LOe - 05 i:l..

""" no fading

""""""""""~~~"" , , ,

LOe - 06 L--....l...._....l.... ___ "--__ -L __ --I..I

10.0 20.0 30.0 40.0 signal-to-noise ratio SIN [dB 1

50.0

Figure 7: Bit error rates infading and non-fading environments

185

6 Performance Analysis of Diversity Receivers in Indoor Communication Links

An effective way for reducing the effect of fading is to use multiple receiving antennas. If the distance between the antennas is made large enough to ensure independent fading, the receiver is called a space diversity receiver. The received voltages on the two antennas can be combined in many different ways. We will distinguish between

• Selection diversity. The output of one antenna at a time will be used. Using selection diversity, the received voltage is equal to the maximum value of the voltages of the two antennas. The received voltage can be written as: Vout = max{lvtI, Ivzl}.

• Maximum ratio combining. Using this optimal technique, the received signals are added, each weighted with the complex conjugate of the original signals. The output voltage can be written as:

Vout = Vlvtlz + Ivzlz.

• Equal gain combining. When using this combining method, the gains of the maximum ratio combiner are all set to a constant value of unity. The detection of the signals can be obtained by using a simple phase-locked summing circuit. The output of the equal gain combiner is given by: Vout = IVII + Ivzl.

The induced voltages on a receiver without and with diversity (maximum ratio combining) will be compared to each other. The configuration of the room is given in table 1. The received voltages are normalized to the port voltage of an x-directed point-source transmitter. The amplitude of the received voltage of a single half-wave dipole antenna (without diversity) is. shown in figure 8. This voltage is calculated in a section of a vertical plane inside the room. The amplitude of the received signal of a maximum ratio combiner is shown in figure 9. The total noise power of an equal gain combiner or a maximum ratio combiner exceeds the noise power of a single antenna receiver. The differences in noise power have been accounted for in figure 9.

186

1.0 1.0

1.4 1.4

1.8

! " 2.2

1.8

! , " 2.2

2.6 2.6 " 3.0

1.5 1.6 1.7 1.8 1.9 2.0 1.5 1.6 1.7 1.8 1.9 2.0

z[m] z[m]

-55 -53 -52 -50 -49 47 ·55 -53 -52 ·50 -49 -47

-~ -~ Figure 8: Voltage of single antenna Figure 9: Voltage of max. ratio combiner

It is clear from the figures that the number of deep fades inside the room is decreased by using diversity reception.

The most powerful method for comparing the various receivers is given by a full link-budget calculation. The coverage in the room will be calculated for various values of transmitted power (point source) for a maximum bit error rate equal to 0.001. The coverage is calculated by evaluating the received signal-to-noise ratios at 6400 locations inside the room. The coverage (number of locations inside the room with bit error rates below 0.001) versus the transmitted power is presented in figure 10.

~ " on e ~ 0 ()

100.0 99.8 99.6 99.4 99.2 99.0 98.8 98.6 98.4

",,".;;J";.;;C­l''',~/

,I,' :/ / I' ,/

/:' ,"" I, : I' :

" :' f ,' :' I, : I,:

//; . : I, :

no diversity max. ratio equal gain

selection diversity 98.2 ,.' . : 98.0 '-'----'--'-----'-'----'----'-----'

-25.0 -20.0 -15.0 -10.0 -5.0 0.0

transmitted power [dB]

Figure 10: Coverage inside the room versus the transmitted power

Notice that the transmitted power using diversity inside indoor environments can be 10 dB lower compared to telecommunication links without diversity.

7 Comparison between Full Wave Solutions and Statistical Methods

In this section, the relationship between bit error rate and signal-to-noise ratio will be examined by comparing the full wave deterministic models with stochastic techniques. For frequency-nonselective, slowly fading channels, the signals in the multipath environment are often described by a Rayleigh distribution.

187

This stochastic distribution will be used for evaluating the bit error rate versus the signal-to-noise ratio for fading environments with and witbout diversity reception. Approximately, the error probabilities for non-coherent FSK modulation are given in [9]. The approximations for Rayleigh fading environments are valid for signal-to-noise ratios larger than 5 dB.

I S f d' P = -exp-- non- a mg e 2 2N' ,

Pe = 1/ (S / N), fading, without diversity,

Pe = 1/(S/Nf, fading, witb diversity.

(9)

(10)

(II)

Using tbe deterministic full wave model, the electric fields have been calculated at 6400 locations inside the room with the eigenfunction technique. At each location, tbe received voltage, the signal-to-noise ratio and tbe bit error rate have been calculated by using formulas (4), (7) and (8) for various values of transmitted power. The results for the stochastic and deterministic model are given in figure 11.

~ 1.0e + 00

.... 1.0e - 01 g 1.0e - 02 0)

4-< 0 1.0e - 03

~ 1.0e - 04 .. 1.0e - 05 .!:J e c.. 1.0e - 06

10.0 20.0 30.0 40.0 50.0

signal-to-noise ratio SIN [dB]

Figure 11: Bit error rate versus signal-to-noise ratio, curve 1: no fading, curve 2: Rayleigh fading with diversity, curve 3: modal analysis with diversity, curve 4: modal analysis without diversity, curve 5: Rayleigh fading without diversity

The predictions made by the two different techniques are very similar, except for the prediction of tbe performance of the diversity receiver. Notice tbat the actual spatial distribution of the received signals depends on the room dimensions and the used transmitting and receiving antenna, which is predicted by the deterministic full wave model.

8 Conclusions

Wave propagation and system performance inside indoor environments are studied with two full-wave models. It is shown that the necessary transmitted power for a diversity system is 10 dB lower compared to receivers without diversity. Furthermore, it is shown that the predictions made by tbe deterministic models are not always the same as the outcomes of stochastic techniques. In future, more comparisons will be made between the developed models and 'established' prediction tools such as ray-tracing.

188

9 Acknowledgements

This research was supported by Philips Research Laboratories. The author wishes to thank Dr. M. Jeuken and Jr. L. Ley ten for the helpful discussions.

10 References

[I] Fleury, B.H. and P.E. Leuthold, RADIOWAVE PROPAGATION IN MOBILE COMMUNICATIONS: "An Overview of European Research" IEEE Communications Magazine, Feb. 1996, p. 70-81.

[2] Lauer, A. et aI., FDTD SIMULATIONS OF INDOOR PROPAGATION In: Proc. IEEE Vehic. Tech. Conf., 1994, p. 883-886.

[3] de Backer, B. et aI., AN INTEGRAL EQUATION APPROACH TO THE PREDICTION OF WAVE PROPAGATION IN AN INDOOR ENVIRONMENT In: Proc. IEEE 3rd Symp. on Commun. and Vehic. Techn. in the Benelux, 1995, p. 28-33.

[4] Whitman, G.M. et aI., A THEORETICAL MODEL FOR RADIO SIGNAL ATTENUATION INSIDE BUILDINGS IEEE Trans. on Vehic. Techn., Vol. 44, No.3, 1995, p. 621-629.

[5] Dolmans, G. and L. Ley ten, APPLICATION OF GREEN'S FUNCTION TECHNIQUE TO MODEL THE RECEIVED VOLTAGE OF AN ANTENNA IN A MULTI-PATH ENVIRONMENT In: Proc. ICAP'95, Eindhoven, The Netherlands, 1995, p. 186-189.

[6] Dolmans, G., DIVERSITY SYSTEMS FOR MOBILE COMMUNICATION IN A LARGE ROOM Eindhoven University of Technology, EUT Report 96-E-297, 1996.

[7] Yee, K.S., NUMERICAL SOLUTION OF INITIAL BOUNDARY VALUE PROBLEMS INVOLVING MAXWELL'S EQUATIONS IN ISOTROPIC MEDIA IEEE Trans. on Ant. and Prop., Vol. 14, No.3, 1966, p. 302-307.

[8] Toftgard, J., et aI., EFFECTS ON PORTABLE ANTENNAS OF THE PRESENCE OF A PERSON IEEE Trans. on Ant. and Prop., Vol. 41, No.6, 1993, p. 739-746.

[9] Proakis, J.G., DIGITAL COMMUNICATIONS New York: McGraw-Hill, 1995.

18 The Spectrum Management Process: How it is Working to Support the New

Wireless Services

ABSTRACT

David 1. Cohen

Information and Telecommunication Studies

Graduate School of Management and Technology

University of Maryland University College

College Park, Maryland 20742

This paper, intended for wireless researchers and developers is an introduction to the

ongoing spectrum management activity for wireless services. An overview of the spectrum

management process is presented and a summary is provided of the specific actions the spectrum

management community has completed during the past several years to support the new wireless

services. The paper addressees these subject areas: Organization of u.s. Spectrum Management,

spectrum allocation and assignment, wireless allocation policy, spectrum engineering and spectrum

economics.

1 INTRODUCTION

Spectrum management, principally a governmental activity, has the mandate to assure

effective, efficient and prudent use of the electromagnetic spectrum and to insure electromagnetic

compatibility arnong radio equipments (i.e. transmitters, receivers). Spectrum management is an

interdisciplinary activity involving technical, economic and governmental policy and regulatory

considerations.

The growth in the wireless services has highlighted the importance of spectrum as a natural

resource and, indeed, the Congress has held hearings to learn about "spectrum". Spectrum managers

have been required to act expeditiously to identify additional spectrum (i.e .. spectrum allocations)

J. H. Reed et al. (eds.), Wireless Personal Communications

© Springer Science+Business Media New York 1997

190

for the new wireless services and also develop the appropriate supporting regulatory provisions

needed to implement the new allocations. Especially noteworthy, spectrum managers devised

many new innovations. These new spectrum management innovations will be highlighted in this

paper.

2 ORGANIZATION OF THE UNITED STATES SPECTRUM MANAGEMENT

Historically, the need for spectrum management first arose in 1912 when it became

necessary to establish frequency registrations for equipments to deal with interference problems

involving important and critical radio transmissions. Two decades later, the Communications Act

of 1934 gave to the Federal Communications Commission (FCC) a broad regulatory mandate for

regulating wireline and wireless communication equipment and inherently authority for spectrum

management. However, the same Act gave to the President the authority to assign frequencies to

all Federal Government owned and operated radio systems. (Note: The term "radio" is a generic

regulatory term for all RF communication equipments). As a result of this action governmental

spectrum management is for the United States a dual organizational structure. The National

Telecommunications and Information Administration (NTIA) in the Commerce Department acts

on behalf of the President to manage the Federal Government's use of the radio spectrum and the

FCC manages all other uses.

Two sets of Governmental regulations implement effective spectrum management. The

FCC Regulations for Spectrum Management are contained in Title 47 of the Code of Federal

Regulations (CFR) and regulations and procedures which apply for Federal Government spectrum

management are found in the NTIA Manual of Regulations and Procedures for Federal Radio

Frequency Management [1].

3 SPECTRUM ALLOCATION AND ASSIGNMENT

A key principle of both national and international spectrum "regulation" is that spectrum is

organized into block of continuous frequencies for particular radio services (e.g. mobile, fixed etc.)

191

with somewhat uniform controlling technical standards. These blocks of spectrum are termed

"allocations". It is emphasized that spectrum allocations are made to radio services and not to

technical descriptions such as equipment types. modulation method, or message protocol (e.g.

continuous voice, data). There are also statuses among allocations. The highest allocation status

is Primary (and an additional but closely related allocation status termed Permitted). A lesser class

of service is Secondary. Secondary allocation services are on a non-interference basis to Primary

and Permitted services. Some spectrum is allocated exclusively for non-Federal Government use

and some solely for Federal Government use. Most spectrum is allocated for shared Government

/ non-Government use.

The block allocation scheme has many advantages [2]: assist in planning by apportioning

spectrum for future uses, sets out reasonable easy coordination procedures to avoid interference,

provides for stability and design certainty for equipment manufacturers and reserVes spectrum for

socially desirable but otherwise uneconomic uses, and additionally the block allocations are well

known nationally and internationally.

Due to the rapid advances in wireless use and technology some are seeking increased

flexibility in spectrum use. It has been suggested to advance this flexibility the block allocation is

not needed and that users be allowed to use the spectrum as they wish as long as "there will be some

minimal rules necessary such as maximum allowable power and power densities." [3] The issue of

block vs. non-block spectrum allocation is a fundamental concern. Further technical and economic

study is needed. The questions to be asked are: How much is needed in these minimal rules? Is the

block allocation method (with appropriate service sharing) the best practical method to assure

electromagnetic compatibility among equipments and insure economic viability? The FCC recently

[11] established an allocation with greater flexibility by allocating the band 4660-4685 MHz to a

General Wireless Communication Service (GWCS) for both Fixed and Mobile Services. Example

applications for this new General Wireless service allocation are dispatch services, point to point

microwave, aeronautical audio/visual service, wireless local loop services and terrestrial fixed and

mobile auxiliary broadcast services.

The FCC, grants to individual users/organizations a license which gives the right for a

specific user or station to use a specific frequency( or frequencies). This is a frequency assignment.

192

The Federal Government sector through the NTIA, "authorizes" frequency assignments in a process

similar to licensing.

4. WIRELESS ALLOCATION POLICY

The FCC when contemplating new spectrum allocations for non-government applications

first issues Notices of Inquiry to get industry and user comments about what is needed. This is

followed by Notices of Proposed Rulemakings and Report and Orders. This process (working

jointly with the NTIA) revises both the U.S. spectrum allocations and the other supporting

regulations for the allocations. Internationally allocations are decided by World

Radiocommunication Conferences (WRC). A recent WRC was held in 1995 and the next WRC will

be held in 1997.

Changing spectrum allocations takes time and there are both national and international

considerations. Many studies, [4], [5], [6], [7] have projected future wireless growth to aide future

spectrum planning and allocation actions.

The FCC began in the early 1990's the process to find spectrum for the new emerging

wireless technologies. Deciding the amount of added spectrum needed for the new wireless

services is a difficult task. The Congressional Office of Technology Assessment (OTA) in a study

[8] of the role of wireless in the National Information Infrastructure (NIl) called attention to the

difficUltly in charting the growth of wireless by stating "the level of uncertainty is much higher and

more pervasive than usual: all aspects of the wireless industry- technologies, markets and rules are

changing almost constantly." Additionally, the OTA states that "Rapid technology advances,

unfocused user needs .... make predicting the future of wireless technologies and services highly

speculative.

Traditionally, spectrum allocations for new services is accomplished either by having the

new service move higher in frequency where spectrum is available or share with an existing service.

Based upon propagation and equipment characteristics it was decided, by the FCC, for the initial

allocations for PCS not to go higher in frequency and, instead, to utilize spectrum around 2 GHz

currently occupied by the Fixed Microwave service [9]. The specific allocations are shown in the

first block of Table 1.

193

Table I. "New" Spectrum Allocations for Wireless Services

Block A 1850-1865/1930-1945 MHz

BlockB 1870-1885/1950-1965 MHz

BlockC 1895-1910/1975-1990 MHz

BlockD 1865-1870/1945-1950 MHz

BlockE 1885-189011965-1970 MHz

BlockF 1890-1895/1970-1975 MHz

Unlicensed (asynchronous) 1910-1920 MHz

Unlicensed (isochronous) 1920-1930 MHz

Unlicensed 2390- 2400 MHz (shared with amateur)

asynchronous

General Wireless Communication Service 4660-4685 MHz

Unlicensed (Millimeter wave) 59-64 Hz

Generally, when new services are added to an existing allocation the new service must devise

and implement sharing 1 coordination procedures to avoid interference to the existing service

equipments. A coordination procedure [9] was implemented for the 2 GHz deployment but

additionally, in a new and innovative action, when sharing is not feasible a new policy emerged that

requires the existing fixed microwave equipment to be moved out of the band with costs to be borne

by the new PCS service users.

Another innovation, Congress has involved itself in spectrum allocation. The Omnibus

Budget Act of 1993 required the Department of Commerce ( i.e. NTIA) to identify an aggregate of

at least 200 MHz of Federal Government spectrum below 5 GHz to be reallocated by the FCC to

promote the development of new telecommunication technologies, products and services. The

NTIA identified 235 MHZ [10]. Fifty megahertz of this spectrum has been reallocated by the FCC

for wireless applications [11]. The bands 2390-2400 MHz and 2402-2417 MHz are to be used for

unlicensedPCS operations in accordance with Part 15. The band 4660-4685 MHz is allo~ated for

a new General Wireless Communication Service (GWCS) for both Fixed and Mobile operations [4].

194

Concerning the additional 185 MHz of Federal Government spectrum to be reallocated the

FCC will establish additional rulemakings [12]. The FCC has promised in their reallocation

proceedings to consider among a variety of services public safety use. Additionally, moving higher

in frequency to the millimeter wave part of the spectrum, the FCC has allocated the band 59-64

GHz for use by general unlicensed devices. Noting the severe propagation losses in this

"millimeter wave" band one use for this allocation may be for short distance computer to computer

communications. Table 1 summarizes the new wireless allocations enacted to date.

5 SPECTRUM ENGINEERING

Spectrum engineering is one of the disciplines in spectrum management. The role of

spectrum engineering is to prevent interference (i.e., electromagnetic compatibility) and to insure,

subject to other concerns that the spectrum resource is. used efficiently. Spectrum engineering

employs various interference prediction models and field and laboratory measurements to

accomplish its goals.

Based on interference studies the FCC and NTIA develop and adopt various regulations

which specify technical criteria for equipments which limit intra -service and inter-service

interference. Internationally, the ITU-R (formerly the CCIR) Study Groups recommends similar

criteria (i.e. termed ITU-R Recommendations). These ITU-R recommended technical criteria are

often times later included or referred to in the International Radio Regulations.

Note that interoperability is not one of the concerns of spectrum engineering. Rules and

standards found in the FCC Rules for PCS [9] are only to prevent interference. The FCC

acknowledges the importance of interoperability but believes it will emerge without the FCC

intervention.

5.1 Prevention of Interference

A typical interference study uses analytic or computer models to calculate wanted and

195

interference signal levels for an environment of transmitters and receivers. The interference study

is a parametric analysis to determine the technical criteria which avoid interference. Spectrum

engineers, among other things, can limit transmitter power, transmitter power densities, transmitter

emission spectra, receiver selectivity and antenna directivity. Also frequency separations (i.e. Off

Frequency Rejection) and spatial separations (i.e. increased propagation loss) can be specified.

Sometimes a "worst case" spectrum analysis is appropriate. In this case, possibly, free space

propagation or even anomalous ducting propagation is assumed to be the coupling mode for the

interference.

A special particular type of interference study is called a "prior coordination procedure" or

simply prior coordination. The requirement to accomplish the studyl coordination is mandated by

the Radio Regulations. The need for a prior coordination is to identify possible interference

sources when a new system (in service X) is to share the spectrum with existing users (in service Y).

For example, the PCS deployment must protect existing microwave users and therefore prior to

deployment the FCC requires PCS developers to coordinate with the existing microwave users.

The FCC PCS Coordination procedure is described in [9].

A critical parameter for interference analyses is the interference criteria. Interference criteria

are thresholds for determining when an interference condition may exist. Often the interference

criteria, called a "protection ratio" is a ratio of signal power S, to interference power, I. Table 2

lists several interference criteria which are used for wireless service applications.

196

Table 2 WIRELESS SERVICE INTERFERENCE CRITERIA

Wireless Service Interference Criteria

Analog AMPS Cellular Ref. [13] S/I >18 dB

PCS Sharing with Fixed Microwave Ref [9] lIN < -6 dB (analog)

10-5 to 10-6 change in bit error (digital)

Unlicensed PCS Ref. [9] Monitor Speech Window

(asynchronous)

Monitor Time/Speech Window (isochronous)

The first block in Table 2 reflects a specific communication performance ( e.g. a voice

articulation measure). The lIN (I, interference power, N, noise power) is utilized when the effect

of interference is to be minimized and thus the interference must be below the noise level and

contribute only a small increase to the noise level.

5.2 Unlicensed Sharing

Currently, unlicensed low power PCS devices operate in several ISM bands (902-928 MHz,

2400-2483.5 MHz, 5.725-5.875 GHz). The FCC has adopted power limits and other technical

criteria for these bands for wireless communication equipment but these wireless equipments must

share the spectrum with a host of other non-communication equipment such as microwave ovens.

No operational spectrum sharing constraints such as use time, separation distance are included in the

FCC Rules and Regulations for unlicensed wireless operations in the ISM bands.

Mindful of expanding demand for unlicensed PCS devices such as wireless LAN s, wireless

PBXs, and PDAs the FCC when allocating spectrum for PCS allocated specific bands for unlicensed

operations. See Table I under the headings "unlicensed". Based upon spectrum engineering studies

the FCC adopted several innovation technical and operation restrictions to promote the success of

the service [9]. These rules are identified in Table 3.

Table 3. Innovations for Unlicensed Spectrum Use

Dedicated spectrum for unlicensed services

Separate subbands for asynchronous and isochronous traffic

Spectrum Etiquette:

(A) Listen before transmit spectrum access

(B) Limited transmitter power

(C) Limited time duration for transmissions

197

The restrictions / operational constraints listed in Table 3 are really an experiment [14 ].

The operational success of these rules will not be fully known until deployment.

5.3 Efficient Use of the Spectrum

Ano~her goal of spectrum engineering is to promote efficient use of the spectrum. The

spectrum management community [15] defines spectrum efficiency as

M Spectrum Efficiency = -

U

where M is the amount of information transferred. U is the amount of spectrum utilized. U is the

product ofB (frequency bandwidth), S (geometric space, usually area) and T (time). Note that

in this definition of spectrum utilized U, is not just frequency but has components of frequency,

geometric space (geography) and time. The literature contains many examples of calculations

comparing spectrum efficiencies for wireless applications for the various over the air interface

methods.

198

6 SPECTRUM ECONOMICS

More organizations I users and the Federal sector (Congress and the Executive Branch) are

becoming increasing aware of the economic importance of the spectrum natural resource. This has

affected the manner in which the FCC has apportioned spectrum licenses for wireless. The earliest

used "competitive decision" method called "comparative hearing" uses a judging scheme to award

licenses. When the cellular licenses were to be awarded there were, as expected, numerous

applicants and the FCC instituted a lottery scheme as a fair way to avoid the time consuming process

of comparative hearings. Noting the economic value of the spectrum when the PCS licenses were

to be awarded the Congress gave the FCC the authority to utilize competitive bidding (i.e. auctions)

to apportion the licenses with the added conditions that the proceeds from the auctions be placed in

the Treasury of the United States. The initial PCS auctions for Blocks A and B (See Table 1) were

immensely successful and yielded 7.7 billion dollars [16].

7 CONCLUSION

The expansion of new wireless services and technologies is being supported by actions of

the spectrum management community. Probably the most important action is the allocation of

additional spectrum for wireless services. Several new spectrum management innovate actions

have been implemented to support the wireless expansion. These new innovations include: (1)

Buyouts of existing fixed microwave operations, (2) Congressional mandated Federal Government

Reallocation (3) Competitive Bidding (Auctions) and (4) Dedicated spectrum and technical sharing

criteria for unlicensed communication.

199

8 REFERENCES

[I] National Telecommunications and Infonnation Administration, Manual of Regulations

and Procedures for Federal Radio Frequency Radio Management. US Government

Office. Published periodically (e.g. January 1995) with periodic supplements.

[2] National Telecommunications and Infonnation Administration, U.S. Spectrum Management

Policy: Agenda for the Future, NTIA Special Publication 91-23.

[3] Noam, E. M., "Taking the next step beyond spectrum auctions: open spectrum access",

IEEE Communications Magazine, pp 66-73, December 1995

[4] National Telecommunications and Infonnation Administration, Land Mobile Spectrum

Planning Options, NTIA Report, October 1995

[5] National Telecommunications and Infonnation Administration, National Spectrum

Req';lirements, Projections, and Trends, NTIA Special Publication 94-31.

[6] Personal Communications Industry Association, "PCIA 1995 PCS Technologies Market

Demand Forecast Update 1994-2005. Washington D.C. Jan 1995.

[7] Davison, A. and L. Marturano, "Impact of digital techniques on future LM spectrum

requirements", IEEE Vehicular Technology Society News, May 1993.

[8] U.S. Congress, Office of Technology Assessment, "Wireless Technologies and the

National Infonnation Infrastructure", OTA-ITC-622 (Washington D.C., Government

Printing Office, July 1995).

[9] FCC, 47 CFR Parts 2,15 and 24, Memorandum Opinion and Order GEN Doc No. 90-34,

FCC 94-144, July 1994.

[10] National Telecommunications and Infonnation Administration, Spectrum Reallocation

Final Report, NTIA Special Publication 95-32.

[11] FCC, First Report and Order and Second Notice of Proposed Rulemaking, Allocation of

Spectrum Below 5 GHz from Federal Government Use, ET Docket 94-32, February 1995.

[12] FCC, "FCC approves reallocation Plan for 185 megahertz of spectrum released from

Government use, Establishes Rule Making Schedule", Report No. GN 96-4, March

1996.

200

[13] Rappaport, T.S., Wireless Communications - Principles and Practices, Prentice Hall

PTR,1996.

[14] Steer, D. G. "Coexistence and access etiquette in the United States unlicensed PCS band,

IEEE Personal Communications Magazine Fourth Quarter 1994 pp.36-43.

[15] ITU, ITU-R Recommendations, 1994 SM Series Volume Spectrum Management

Techniques Recommendation ITU-R SM 1046.

[16] Flanagan, P., Personal Communications Services: The Long Road Ahead,

Telecommunications Magazine, pp. 23-28 .February 1996.

19 Analysis of an Eigenstructure Technique

for DSSS Synchronization * Nitin R. Mangalvedhe and Jeffrey H. Reed

Mobile and Portable Radio Research Group (MPRG) Bradley Dept. of Electrical Engineering

Virginia Polytechnic Institute and State University

Abstract

A new technique for synchronization of direct sequence spread spectrum (DSSS) sig­nals is presented. This technique exploits the eigenstructure of a frequency-channelized DSSS signal to directly estimate the underlying spreading code. The synchronous es­timate of the spreading code steadily improves with the increase in the collected data. In this paper, the algorithm used for estimating the spreading code is deri~eri. It is shown that under infinite time-average assumptions, a perfect code estimate can be obtained when the signal is received in arbitrary levels of white background noise. The only requirements are that the spreading code must truly multiply the message signal, the code must have a constant modulus, and the message and code repetition rates must be practically inco=ensurate. It is shown that the technique is insensitive to frequ~ncy offsets on the received signal that are integer multiples of the code repetition rate. The technique is also shown to exploit multi path.

1 Introduction

Code synchronization constitutes an important aspect of a direct sequence spread spectrum (DSSS) receiver. It is a process by which the receiver synchronizes to the pseudonoise (PN) spreading code of the received signal. The first stage of synchronization, termed PN acquisition, involves a coarse estimate of the PN code within a fraction of a chip.

Conventional acquisition techniques [1, 2, 3] exploit the knowledge of the internal struc­ture of the spreading code to achieve synchronization. They exhibit good acquisition perfor­mance in low noise environments but are not suitable for environments with high levels of noise and interference. Factors such as Doppler shift and multipath complicate the problem further.

DSSS packet radio and military systems often require frequent, fast and robust syn­chronization. Blind estimation of the spreading code without the a priori knowledge of its structure, timing or the Doppler-shift is useful in achieving these objectives. A blind technique directly estimates the delayed and Doppler-shifted code at the receiver.

The eigenstructure technique presented here is a blind technique based on the dominant mode despr:eader, first proposed in [4], and estimates the spreading code by exploiting its constant modulus property and the eigenstructure of the frequency-channelized DSSS signal. Under infinite time-average assumptions, the technique provides a perfect estimate of the

"This work was sponsored by Virginia's Center for Innovative Technology and the Research Center for Excellence project from the Federal Highway Administration.

J. H. Reed et al. (eds.), Wireless Personal Communications

© Springer Science+Business Media New York 1997

202

spreading code in the presence of arbitrary levels of temporally-white background noise and for arbitrary codes satisfying the constant modulus property.

This paper presents the development of the algorithm used in the eigenstructure tech­nique. It is shown that the technique is insensitive to frequency offsets in the signal that are greater than half the code repetition rate. Any small residual signal frequency offset can be removed at the despread signal-to-noise ratio (SNR). It is also shown that when the signal is received in a multi path environment, the technique provides a multipath estimate of the code, which may be used to enhance the demoduiated signal. Simulation results for different channels are presented in [5, 6] .

. This paper is organized as follows. In section 2, the algorithm is derived for a signal in a single user environment. In section 3, the algorithm is analyzed for a signal received in a multipath environment. Conclusions are presented in section 4.

2 The Eigenstructure Technique

Linear periodically time varying (LPTV) processing techniques have been developed for the direct extraction of the message signal from the received DSSS signal by employing an algo­rithm that adapts to the channel to remove the spreading code as part of the demodulation process. The eigenstructure technique for synchronization is based on these techniques.

2.1 Processor Development

The baseband transmitted DSSS signal is modeled as

s(t) = c(t)d(t), (1)

where c(t) is the spreading code and d(t) is the data (information) signal. The message is recovered at the receiver by

d(t) = Ld a [c*(t)x(t)], (2)

where x(t) is the received signal and Ld is a linear lowpass filter operator with frequency response Ld(J) that covers the passband of d(t). The spreading code c(t) is periodic with a repetition rate Ir = liT" such that c(t + Tr) = c(t). Therefore, it can be replaced by its Fourier series expansion:

where

c(t) = L C(k)ei21rkfrt , k

(3)

Tr /2

C(k) = ~ J c(t)e-i27rkfrt dt. (4) Tr

-Tr /2

Substituting this Fourier series representation (FSR) of c(t) into (1) and (2), the spreading and despreading processes may then be expressed as:

where

d(t) s(t) BW ~ Ie

Figure 1: Fourier series representation of the DSSS spreader.

d(t)

L C(k)[d(t)ei21rkfrt] k

Cd 0 { [~C(k)ei21rkfrt] * x(t) }

L C*(k)xdt),

203

(5)

(6)

(7)

The processors implementing these operations (known as the FSRs of the DSSS spreading and despreading operations) are as shown in Figures 1 and 2, respectively. The presence of the frequency-shift operations in the two processors suggest that they are linear periodically time-variant operations. The nonblind time-dependent processor discussed in [7] and [8] has the general form

d(t) = L 1ik 0 [x(t)e-i21rkfrt], k

(8)

where {1id are a set of LTI filters. The FSR despreader here may be modeled by this processor but with the constraint

(9)

The adjustment of the weights of the FIR filter bank implementing the LTI filters to minimize the error between the despread signal d(t) and the original signal d(t) (as done in [8]) requires the knowledge of the message signal.

x(t) d(t)

e-j27r(k+l)f. t

__ Xr~<.!~e~::~ 5~~~r:e!i~~r ____ _

Figure 2: Fourier series representation of the DSSS despreader.

x(t) d(n) rate jd

Figure 3: Blind despreader, frequency-domain representation.

205

x(t)

BW ~fc

Figure 4: Blind despreader, time-domain representation.

If the filters are constrained by (9), then Figure 2 leads to the structure of Figure 3. Performing the frequency shift operation on the received signal results in the vector signal x(t) = [Xk(t)]k which is linearly combined to obtain the despread signal

d(t) L W*(k)Xk(t) k

wHx(t), (10)

where w = [w(k)] and (-)H denotes the conjugate-transpose (Hermitian transpose) opera­tion. Figure 4 illustrates an implementation of the processor that resembles a conventional synchronizer. In this structure, the processor weights are used to estimate the spreading sequence, which is then used for despreading. This spreading sequence is synchronous with the spreading waveform component in the received signal and hence eliminates the synchro­nization process carried out in conventional receivers. The algorithm for the computation of wcan be the same as in Figure 3. The despreading is carried out in the time-domain by

d(t) = Cd 0 [C*(t)x(t)],

where c(t) is obtained from w by the Fourier relationship

c(t) = L w(k)ei21rkJ,t. k

2.2 Algorithm Development The frequency-channelized DSSS signal s(t) = [Sk(t)]k is given by

Sk(t) = Ld 0 [s(t)e- j2,,-kf,t]

(11)

(12)

206

= Cd 0 [L C(l)d(t)e-j2,,(k-ll/rt], I

using (1) and (3). Letting k -l = m, we have l = k - m and (13) may be rewritten as

L C(k - m){Cd 0 [d(t)e- j2"mfrtj} m

L C(k - m)dm(t). m

Then

set) [Sk(t)]k LCmdm(t) m

Cd(t)

(13)

(14)

(15)

where d(t) = [dk(t)]k is the frequency-channelized message signal and C is the matrix of Fourier coefficients of the spreading code. That is,

C [C(k - m)km [ ... C-l Co Cl ... J,

Cm [C(k - m)]k

C(-l-m) C(m) (16)

C(1- m)

Similarly, the delayed, attenuated and Doppler-shifted DSSS signal set) = aei2"ll.ts(t - r) (where a is the attenuation, r is the delay and ~ is the Doppler shift) has a frequency­channelized representation

set) Cd(t) (17) C [C(k - m)]k,m (18)

d(t) = [Cd 0 (£l(t)e- j2"k frt )]k, (19)

where

£l(t) aei2".olrtd(t - r)

C(k) C(k - ko)e-j2,,(k-kol!rr

c(t) c(t - r)ei2"kofrt

and where ~ = (ko + eo)/" and ko has an integer value such that leol ~ 1/2. The received data signal x(t) therefore has the frequency-channelized representation

x(t) = Cd(t) + i(t) (20)

207

in the general single signal environment where the frequency-channelizer receives a single attenuated, delayed and Doppler-shifted DSSS signal in background noise i(t), and i(t) is the frequency-channelized noise-component,

x(t) = aei27r~ts(t - T) + i(t).

The autocorrelation matrix of x(t) is

where (-)T denotes time-averaging over the reception interval [0, T).

Rxx ([Cd(t) + i(t)][Cd(t) + i(t)]Hh C(d(t)dH (t))TCH + C(d(t)iH (t))T +(i(t)dH (t))TC H + (i(t)iH (t))r.

(21)

(22)

If d(t) and i(t) are statistically independent, then as the averaging time grows to infinity, the autocorrelation converges to

• -. - H Rxx = CRJ.J.C + Rii· (23)

Using the theory of spectral correlation [9, 10], the identity

Uk(t) 11. 0 [u(t)ei27rklrt]

=> Ru ... , ~ (uk(t)uj(t)h 00

J 8".",(J) df, (24) -00

S ..... ,(J) I H(JWS:" (!1)

where S~,,(J) is the cyclic spectrum of u(t), Q being the set of cyclic frequencies:

Q (k -l)fr

!1 = f + (k ; l) fr.

Thus (24) becomes

Ru.", = j IH(JWS~~-I)lr (f + (k; l) fr) df -00

(25)

Equation (25) holds for arbitrary finite power waveform u(t) and LTI filter 11., where S~,,(J) has zero value at every nonzero value of Q if u(t) is stationary, and nonzero value at discrete nonzero values of Q if u(t) exhibits second-order cyclostationarity at those values of Q.

If the background noise i(t) is temporally stationary and white, then the cyclic spectrum of i(t) is given by

S~(J) = { ~o, otherwise. (26)

208

Then it follows from (25) that

where

-00

k = I otherwise

(27)

(28)

for any LPF frequency response Ld(f), and Ok is the Kronecker delta function. Similarly, if d(t) is stationary with power spectrum Sdd(f) , then the cyclic spectrum of d(t) is given by

and (25) can be used to show that

Rdkd.Ok-1 00

00=0 otherwise,

J ILd(fWSdd(f - colr + klr) dl -00

(29)

(30)

(31)

for any LPF frequency response Ld(f). Then substituting (27) and (30) into (23) yields the autocorrelation matrix

(32)

Equations (30)-(31) also hold if d(t) is nonstationary but does not possess second-order cyclostationarity at any multiple of Ir. In particular, if d(t) is a PSK signal with data rate Id, power Rd, pulse shaping Pd(f) and temporally white message sequence d(n), then the cyclic spectrum of d(t) is given by [10]

S'Jd(f)

S~d(f)

lal2 S~d(f + colr )e-i27rar

{ Rd Pd(f + ~/d) Pd,(f - Ud), 0,

a = kid otherwise,

(33)

and (25) can be used to show that Rdkdl is given by (30)-(31) if the message and code-repeat rates are incommensurate, such that Id/lr is irrational. This condition can also be shown to approximately hold if Id and Ir are practically incommensurate such that Id/lr = p/q where p and q are large relatively-prime integers. In this case, Rdkdl reduces to

{ lal2ej211'if,r Rd I ILd(f - colr - klrWPd(f)P*(f - iplr) dl,

0,

::::; [lal2 Rd _LILd(f - col. - klr WlPd(fW dl] Ok-I

I = k + iP'(34) otherwise

(35)

209

if q!d (P!r) is larger than the effective bandwidth of Pd(f) , and (32) holds for this signal as well.

If c(t) has a constant modulus, then Parseval's relation can be used to further simplify (32) by noting that the Fourier coefficients of the spreading code satisfy the shift-orthonormality property:

L C(k)C*(k -I) k

Tr /2

L 2- J c(t)e-j21rkfrt dt C*(k -I) k Tr

-Tr /2

T r /2

2- J e-j27rlfrtdt TT

-Tr /2

8l

Applying (36) to the Fourier coefficient matrix C yields

(36)

(37)

Equations (36)-(37) hold for any spreading waveform as long as it has a constant modulus over the entire code period. Thus (37) also applies to the frequency-shifted Fourier coefficient matrix C.

Then it follows from (32) that the Fourier coefficients of c(t) can be computed from the eigenstructlire of Rxx. Since C is a unitary matrix satisfying the orthonormality property (37), according to the theory of matrices, the eigenvectors of the autocorrelation matrix are equal to C multiplied by the eigenvectors of diag{RJkd.}, which are equal to the set of unit vectors {ek}' Thus the eigenvectors ofRxx are equal to {Ced = {Ck}. Hence,

- -H (R"nI + Cdiag{ ReM.}C )Cl

R"ncl + Cdiag{ReMk}CHcl

(R"n + Rd1d,)Cl

R"n + Rdldl

(38)

(39) (40)

Furthermore, if d(t) is a PSK waveform and Ld(f) is set equal to Pd(f + CO!T) (or Ld(f) is set equal to Pd(f) and CO!T « Id), then (38) is maximized for I = 0 and the ko-shifted Fourier expansion of the code in the received signal can be estimated from the dominant mode (eigenvector associated with the maximum eigenvalue) of Rxx. The dominant mode can also be used to directly estimate d(t) from x(t) using (10), or it can be used to estimate c( t), and despread the data signal using (2). For this reason, this technique is referred to here as the dominant mode despreading (DMDS) algorithm.

The basic DMDS algorithm may be summarized as follows.

1. Frequency-channelize the received data signal x(t) to form the vector signal x(t), using the formula

x(t) = [.cd 0 (x(t)e- j21rkfrt )]k,

where Ir is the code-repetition rate and .cd is an LTI lowpass filter that is matched to the pulse shape of the message signal.

210

2. Compute the autocorrelation matrix of x(t) using the formula

Rxx = (x(t)x H (t))T

where OT denotes the time-averaging operation over the collection interval [0, t), and (.)H denotes the conjugate-transpose operation.

3. Compute the dominant mode of Rxx by solving for the maximum eigenvalue Amax and the associated eigenvector W max of the eigenequation

Rxxw= Aw.

4. Compute the processor output signal using either of the formulas

J(t) = w:/.axx(t) frequency domain despreader,

J(t) = Cd 0 [c*(t)x(t)] }. . c(t) = L,kWmax(k)ej27rkfrt tlme-domam despreader,

where Cd is an LTI lowpass filter (not necessarily equal to Cd) that is matched in some way to the pulse shape on the message signal.

The technique provides a delayed and Doppler-shifted version of the message signal, since the alg~rithm estimates c(t) (or C(k)) rather than c(t). This requires keeping track of the timing of the collected data in order to correctly use the estimated code. In addition, the DMDS algorithm automatically down-converts J(t) to within ±lr/2 carrier cycles by subsuming the component of the carrier offset with periodicity Ir into the spreading code. Thus, the estimation of c(t) is insensitive to carrier offsets beyond ±lr/2 carrier cycles and it may still be necessary to perform a fine tuning carrier-recovery procedure to fully recover the message signal, however, this can be carried out at the despread SNR.

3 Application of the DMDS Algorithm in a Multi­path Environment

Multipath propagation is inevitable in the wireless environment, so it is important to see how the algorithm performs in this environment. A two path multipath model is used here:

x(t) = set) + 9 set - 7) + i(t), (41)

where 9 is the complex scaling factor of the multipath component and 7 is the delay. Then the frequency-channelized signal is given by

Cd 0 [x(t)e-j27rkfrt] Cd 0 [(c(t)d(t) + gc(t - T)d(t - T) + i(t))e-j27rkfrt] L C(k - m)dk(t) + L C(k - m)dk(t) + ik(t), m m

(42)

where

Then,

C(k) C(k)e-j27rk j,T

dk(t) Ld 0 [g d(t - r)e-j27rkf,t].

x(t)

C d(t)

[Xk(t)]k Cd(t) + Cd(t) + itt) [C(k - m)km

[dk(t)k

Then the autocorrelation matrix of x(t) is

Rxx (x(t)xH (t))T = (Cd(t)dH (t)C H + Cd(t)dH (t)C H

+Cd(t)dH (t)C H + Cd(t)dH (t)C H + i(t)iH (t))y

assuming that d(t) and itt) are temporally independent.

211

(43)

(44)

Rxx = CRddCH + CRda.CH + CRa.dCH + CRa.a.CH + Rii (45)

averaging time grows to infinity.

Sd,d)!) Uk(t) Vl(t)

Su,v/ (f)

-00

ILd(f) 12 SUkV/ (f),

d(t)e- j27rkf,t

9 d(t - r)e-j27r1f,t

F{ (Uk (t + ~) vt (t - ~))y}v

( 46)

(47)

(48)

(49)

where F{·}v is used to denote Fourier transformation with v as the variable of integration.

Su,v/(f) = F{(d (t +~) e-j2rrkf,(H'i)

x [gd (t - r -~) e-j27rlf,(t-T-'i)r ej2rrlf,T)r}v.

Letting t - ~ = tl and ~ + ~ = ~, it can be shown [11] that

S"kV/(f) = g*F{(d (tl +~) e-j27rkf,(tl+~) x [d (tl -~) e-j27rlf,(tl-~)r)T}Vej27rlf'T

g* S~d(fl)ej27rlf'T, (50)

212

where

a = (k -l)fr

f -f (k+l)j 1 - + 2 r'

Thus (46) becomes

RMI g*ej21rlfrr lILd(JWS~~-I)/' (f + (k; l) fr) df -00

g*d21r1 /,r Rdkdl

where use has been made of (25). Since

Then

where

RdkdktJk-l

a=O otherwise,

g*ej21r1/,r RdkdktJk-l.

diag{Rd• d.}

g*diag{ ej21rk /,r Rd•dk }

g*Rdd EH ,

E ~ diag{e-j21rk/,r}

In a similar fashion, it can be shown that

and

Using (54), (56) and (57) in (45),

Rxx CRddCH + g* C~d EH CH + 9 CE RddCH + Igl2 CE Rdd EH CH + Rii

(51)

(52)

(53)

(54)

(55)

(56)

(57)

= (C + 9 CE)Rdd(C + 9 CE)H + Rjj. (58)

The columns of (C + gCE) are {cm + gCm e-j21rm/,r},

Cm [C(k - m)]k, cm [C(k - m)]k.

In this case, the shift-orthonormality property is only approximately satisfied. The dom­inant mode corresponds to m = O. The eigenvector of Rxx associated with the maximum

213

eigenvalue is approximately equal to Co + g Co and the (approximate) estimated code is obtained from its inverse Fourier transform,

c(t) :::::: F-1{co+gco} c(t) + g c(t - r).

The demodulated signal is given by (4),

d(t) :::::: £d 0 [(c*(t) + gc*(t - r))(c(t)d(t) + gc(t - r)d(t - r) +i(t))] £d 0 [c*(t)c(t)d(t)] + £d 0 [gc*(t)c(t - r)d(t - r)] + £d 0 [c*(t)i(t)] +£d 0 [gO c*(t - r)c(t)d(t)] + £d 0 [lg12 c*(t - r)c*(t - r)d(t - r)]

(59)

+£d 0 [gO c*(t - r)i(t)]. (60)

If the spreading code has good autocorrelation properties [1], then for r > Te, the second and third terms are negligible. Assuming that the spreading code and noise are statistically inde­pendent, the third and the sixth terms approach zero. Furthermore, Ic(tW = Ic(t- rW = l. Hence (60) becomes .

d(t) :::::: d(t) + Igl2 d(t - r). (61)

Thus, the estimated data also contains a delayed and attenuated component. In fact, the attenuation of the delayed component is greater than that in the received signal as indicated by the factor Ig12. When this factor is not negligible, equalization of the data can be carried out. A point- worth noting is that equalization needs to be performed only at the despread SNR and the data rate since the eigenstructure technique automatically synchronizes with the multi path component also.

4 Conclusions The eigenstructure technique for synchronization of DSSS signals is presented. The algorithm is developed and it is shown that under infinite time-average assumptions, a perfect estimate of the spreading code can be obtained for arbitrary levels of white background noise. The only requirements are that the code have a constant modulus and that the data and code repeat rates be practically incommensurate. The insensitivity of the technique to arbitrary frequency and time offsets is also proven.

The technique is analyzed when applied to a signal received in a static multipath envi­ronment. It is shown the the technique provides a multipath estimate of the spreading code, which can be exploited by employing an equalizer on the despread data at the despread SNR and at the data rate. Extensive simulation results are presented in [5, 6].

References

[1] R. E. Ziemer and R. L. Peterson, Digital Communications and Spread Spectrum Systems. New York: Macmillan Publishing Company, 1985.

[2] M. K. Simon, J. K. Omura, R. A. Scholtz, and B. K. Levitt, Spread Spectrum Commu­nications Handbook. New York: McGraw Hill, revised ed., 1994.

214

[3] S. S. Rappaport and D. M. Grieco, "Spread-spectrum signal acquisition: Methods and technology," IEEE Communications Magazine, vo!. 22, pp. 6-21, June 1984.

[4] B. G. Agee, "Blind despreading of PNSS signals using the constant modulus algorithm." Tech. Report no. AGI-88-09, AGI engineering Consulting, Woodland, CA, September 1988.

[5] N. R. Mangalvedhe and J. H. Reed, "An eigenstructure technique for soft synchroniza­tion of DSSS signals," in Proceedings of ICASSP, 1996.

[6] N. R. Mangalvedhe and J. H. Reed, "Evaluation of a soft synchronization technique for DSSS signals." To appear in the J. Se!. Areas Commun., Special Issue on CDMA Networks III, October 1996.

[7] C. D. Greene, J. H. Reed, and T. C. Hsia, "An optimal receiver using a time-dependent adaptive filter," in Proceedings of IEEE MIL COM, 1989.

[8] J. H. Reed, C. D. Greene, and T. C. Hsia, "Demodulation of a direct-sequence spread spectrum signal using an optimal time-dependent filter," in Proceedings of IEEE MIL­COM,1989.

[9] W. A. Gardner, ed., Cyciostationarity in Communications and Signal Processing. New York: ~EEE Press, 1994.

[10] W. A. Gardner, Statistical Spectral Analysis: A Nonprobabilistic Theory. Englewood Cliffs, NJ: Prentice Hall, 1988.

[11] N. R. Mangalvedhe, "An Eigenstructure Technique for Direct Sequence Spread Spec­trum Synchronization," Master's thesis, Virginia Polytechnic Institute and State Uni­versity, 1995.

20 Future Wireless Challenges for Integrated Circuits

Abstract

Daniel E. Fague Analog Devices, Inc. 804 Woburn Street

Wilmington, MA 01887

The popularity of mobile communications has created a large demand for both end user equipment and the integrated circuits inside them. While early cellular and cordless phones had relatively low complexity chips with discrete circuits implementing many functions, modern phones have highly integrated baseband sections and increasingly integrated· radio sections. The next challenge for the industry is to successfully bring mobile communications to the consumer market. Semiconductor technology is a key enabler to meet that goal. The future of integrated circuits will include highly integrated radio solutions, including integration of key filters. Baseband sections will have communications-specific DSP solutions and highly integrated audio sections. These advances will enable very small handsets to be built at a lower cost than ever before.

1. Introduction

The rapid growth of personal communications services and networks (PGS/PGN) has created a large mobile communications market. Hundreds of service providers and original equipment manufacturers (OEMs) are scrambling to build enough infrastructure and mobile equipment to satisfy the demand of consumers world-wide for communication any time, any place, in any form. This demand for communications comes not only from the traditional, mostly business-oriented, users of mobile communications, but also from a new, growing group of recreational or occasional users. All users of mobile devices demand smaller, lighter equipment with longer times between battery charges, and ever­increasing feature sets. But, the new users of wireless devices bring an additional requirement that business users traditionally have foregone: low price. Service providers and eqUipment providers have also determined that low price is a requirement to allow the personal communications market to continue to grow at a rapid rate.

While the early mobile telephone units had largely discrete designs with very low levels of semiconductor integration, the demand for lower cost phones has driven semiconductor manufacturers to design circuits with higher and higher levels of integration. The result is that a phone built for today's marketplace will most likely have a highly integrated data processing section, encompassing one or two application specific digital signal processors (DSPs), a microcontroller, and a data conversion chip, and a medium integration radio portion, usually including several integrated circuits (lGs) and some supporting discrete circuitry. The challenge that looms ahead for semiconductor companies is how to continue to reduce the cost of their circuits so that their customers may continue to reduce the price of the mobile equipment.

This paper will address some possible avenues for the future of integrated circuit technology. It will focus on the advantages of integration, and how semiconductor companies may be able to leverage those advantages to bring value to OEMs and, ultimately, the wireless consumer. The second section will discuss the state of the art of integration, giving some examples of highly integrated solutions. The third section will begin discussions on the efforts to integrate pieces of the telephone circuitry that until recently have been the sole dominion of passive components - the filter. The fourth section will continue the discussion with a review of sampling receivers. Finally, the fifth section will draw some conclusions.

2. Integration Trends

The mobile telephones that exist today present to the observer a wide variety of solutions. Older analog cellular and cordless phones with relative low complexity have several baseband processing IGs, and primarily discrete radio front ends (see Figure 1). The early models of digital cellular phones are

J. H. Reed et al. (eds.), Wireless Personal Communications

© Springer Science+Business Media New York 1997

216

much the same, showing more complex baseband solutions, but still low integration levels. The late models of both analog and digital cellular and cordless phones, however, show the change that has occurred in the industry. These phones show high levels of integration [1-3], sometimes single chip radios [4,5] and only a couple of baseband processing chips

RF

000000 0 OeD Oooe o 0 0 00 00 o 0 00

~ DO 0 o 0 0 .. ········uuO oDu o ooOeo o ODD

oeooo O 0 0

DO 800d DO BB

DOD

............... ~ .. ~~~ ..

Qo 1990 1994 1998

Figure 1 - Progression of integration of mobile phones

As an example, the baseband section of a Global System for Mobiles (GSM) digital cellular telephone is shown in Figure 2. It can be seen that this solution consists of three ICs: an algorithm signal processor (ASP), which primarily implements the speech encoder/decoder and the equalizer, a microprocessor chip that does system control and channel encoding/decoding, and a baseband interface chip that has both audio and RF data conversion functions. In this configuration, the various ICs are chosen for their specific functionality, and are pieced together to form a complete solution. A companion radio solution is shown in Figure 3, where it can be seen that most of the non-filter portions of the radio have been integrated onto two ICs. Thus, the baseband chipset and the radio chipset form a 5 chip solution for the cellular phone. Commodity items such as memory chips, crystals, and regulators, and specialty items such as power amplifiers and filters are combined with the ICs to form the total bill of materials (BOM) for the phone.

RF

Display

Figure 2 - Block diagram of a three chip baseband section for a digital cellular phone.

A key enabler to mobile telephones is the DSP engine used. The DSP must have capabilities to perform communications-specific tasks such as voice compression/expansion and data encoding/decoding, as well as channelizing and demodulating the signal. In most cases, a general purpose DSP can handle the job, but may not be the most efficient implementation. As requirements for implementations of specific voice or data codecs become known, a need is seen for application specific DSPs. In anticipation of this, DSP vendors have begun to license their cores to third party vendors or customers who will produce these application-specific DSPs. Future years will see more communication­specific DSPs produced, perhaps with baseband and voiceband interfaces included.

["'­

Figure 3 - A two chip solution for digital cellular radios.

217

Typically, the type of processes available to a semiconductor vendor will dictate, or at least influence, the partitioning of a solution. A vendor with a good sub-micron CMOS process will typically integrate as much as possible onto its baseband solutions. Manufacturers with Gallium Ar3enide (GaAs) processes will target the power amplifier and RF front end components. Vendors with high frequency bipolar processes will provide solutions for the radio front end as well as intermediate frequency ICs. They may also compete with GaAs suppliers in the power amplifier market. Companies with combined bipolar and CMOS (BiCMOS) processes will typically offer highly integrated solutions, since high frequency bipolar circuits can be combined with digital logic easily in BiCMOS designs. It is the semiconductor company's responsibility to determine the optimal partitioning for a given design and available processes. A company with a combination of any or all of the above processes may produce several different solutions for the same standard to address different customer segments.

3. New Horizons

As semiconductor vendors examine the state of the art for integration of cellular or cordless phones, they can quickly come to the conclusion that the traditional portion of the phone allocated to semiconductor technology ranges from 30% to 40% of the total BaM of the phone. Specialized semiconductors, such as power amplifiers, LCD drivers, and memory chips, account for an additional 20% of the BaM. The balance of the BaM includes items such as printed circuit board, battery, display, and plastics that must be included and can never be integrated (until the time that wristwatch phones become a reality!). The BaM also includes items such as crystals and filters. It is these last two items that have particular appeal to semiconductor companies, because high levels of integration could eliminate the need for some filters, orthe filters themselves may be integrated. It is the integration of the ·passive" SAW filters that has some interesting consequences.

There are primarily two approaches to integrating SAW filters that can be used. The first approach is to build on-chip spiral inductors and enhance their a using transconductance amplifiers acting as negative resistance cells [5). The typical a of 2 or 3 of an on-chip spiral inductor is enhanced to as much as 100, allowing measured fractional bandwidths of 1% at 200 MHz with very modest current consumptions (3 mA at 3V supply voltage). The drawbacks of a-enhanced filters are that the dynamic range of the filter is typically limited by the amount of enhancement done on the inductor, and the filters must be designed to allow for tuning to overcome variations in gm due to process and temperature. Research is progressing in these areas, however, and today dynamic ranges of 50-70 dB for out-of-band signals are quoted.

The second, more recent approach to integrating SAW filters uses a technology currently utilized in the automotive market to deploy automobile air bag passive restraint systems. The device used in these new air bag deployment systems are tiny, micromachined accelerometers. These accelerometers have tiny, moving parts that vibrate in response to large G-/orces, causing capacitances to change and inducing a voltage change in a monitoring circuit that can then activate the air bag deployment circuitry. The intriguing part of the accelerometer is that it is a mechanical device integrated on silicon technology.

218

It could be presumed that the structures used to sense G-forces can also be used as resonators for oscillators or filters.

In fact, such utilizations have been fabricated, with satisfactory results [7,8]. Phase noise of oscillators using micromachined electro-mechanical devices (MEMs) is comparable to conventional crystal oscillators [7]. Filters have also been demonstrated [8], though at frequencies much below normal surface acoustic wave (SAW) filter frequencies. The promise of replacing non-silicon mechanical resonators with silicon resonators is fueling the research in this area. The challenge for this technology is the lithography requirements. Very fine lithography (below 0.3 micron) is required if the 10's or 100's of MHz range of frequencies are to be reached with MEMs technology. Electron beam lithography is a possible solution, but its cost must be reduced for the technology to become mainstream.

4. Digital Domain

The discussions about integrating SAW filters wouldn't be complete without mentioning another approach - sampling the received signal at a suitable intermediate frequency (IF) and filtering the channel with digital filters. As far-fetched as this idea sounded a few years ago, this method is being used today in cellular base stations [9]. The principle is simple enough: downconvert the desired carrier frequency (or group of frequencies) to a suitable IF, normally around 40-50 MHz, sample the carrier(s) with a high dynamic range, high resolution analog-to-digital converter (ADC), and use digital signal processing and digital filtering to capture the desired signal. This approach has clear advantages: the channel filter choice is never a permanent choice, since a quick software change can put new tap coefficients into the filter, the radio itself can be reconfigured with software, allowing a single radio to serve, for example, the analog cellular market or the digital cellular market, with software downloadable from a remote site. This allows service providers maximum flexibility in configuring their coverage areas.

(a) (b)

Figure 6 - Architectures for IF sampling receivers. (a) Narrowband (GSM, DCS-1800, PCS-1900) (b) Wideband (AMPS, IS-54, PDC)

Figure 6 shows two examples of IF sampling receivers used in base stations. The wideband IF sampling receiver can greatly reduce cost of a base station when one RF channel carries one traffic channel, as is the case in analog cellular (AMPS, TACS, ETACS) today. In this case, one ADC and one SAW replace one SAW and one ADC per channel. The more channels a single ADC can handle, the more cost effective the wide band IF sampling receiver becomes. The narrowband IF sampling receiver is used when a single RF channel carries multiple voice traffic channels. Because each RF channel already carries multiple voice channels, the base station is most cost effective by simply replacing the analog downconverter circuitry with digital circuitry.

A key enabler to software radios is the baseband interface. The ADC's used today for software radios have spurious free dynamic ranges (SFDR) of 70-80 dB, and resolutions of 14 bits. With this level of performance, these ADCs can be combined with an RF block downconversion stage, a digital channelizer IC, and a DSP demodulator to implement a complete radio receiver. Such systems exist today, and will only extend higher in frequency in the future. Today, this technology is limited to base stations due to the high current consumption required by the channelizer, the ADC, and the DSP. However, as semiconductor technology advances, the software radio may become a reality for handsets as current consumption is reduced.

219

5. Conclusions

The mobile communications market has rapidly grown from addressing a niche for the business community to being poised for the mass consumer market. As OEM's have struggled to provide enough mobile communications eqUipment to their customers, semiconductor companies have also struggled to meet the demand for wireless IC's. The next few years will see the mobile communications market transition to a consumer market. This will put increased pressure on semiconductor companies to provide their customers with more competitive solutions than ever before. This will result in communications ICs that include:

- Customized or application-specific DSP functions using standard, licensed cores, - Integrated channel selection filters at high IF's, - Integrated resonator circuits that eliminate the need for external crystals. - Digital receivers with current consumption low enough to warrant handset usage,

In short, as the demand for lower cost mobile communications products continues to grow, semiconductor manufacturers will continue to integrate more of the phone in an effort to provide more value to end equipment manufacturers. This will help to achieve the goal of personal communications services: communications at any time, in any place, with anyone.

Acknowledgements

The author would like to thank Tom Gratzek, Esben Randers and the members of the Wireless Communications Group in the Communications Division for their thoughtful input to this article.

REFERENCES [IJ I. Koullias, J. Havens. I. Post. and P. Bronner, "A 900 MHz Transceiver Chip Set lor Dual-Mode Cellular Radio Mobile

Terminals", 1993 IEEE Intematlonal Solid· State Circuits Conference, February 24-26, 1993, San Francisco, pp. 140-141.

[2J B. Bierde, J. Lipowski, J. Petranovlch, and S. Gilbert, "An Intermediate Frequency Modulator Using Direct Digital Synthesis Techniques lor Japanese Personal Handy Phone (PHP) and Digital European Cordless Telecommunications (DECTj", Proceedings of the 44th IEEE Vehicular Technology Conference, Stockholm, Sweden, June 8-10, 1994, pp. 467-471.

[3J B. Wuppermann, S. Atkinson, B. Fox, B. Jansen, and G. Jusul, 'A 2.7 V Two Chip Set Transceiver lor DECT', Proceedings of the Fourth Intema#onal Symposium on Personal, Indoor, and Mobile Radio Communlca#ons, Yokohama. Japan, September 8-11.1993, pp. 407-411.

[4J K. Hess and D. Fague, "Performance Evaluation 01 a Single Chip Radio Transceiver". Proceedings of the 4Sth IEEE Vehicular Technology Conference, Atlanta. Georgia. April 28-May I, 1996. pp. 1048-1051.

[5J C. Takahashi, et. al., "A 1.9 GHz Si Direct Conversion Receiver IC for apSK Modulation Systems", Proceedings of the 1995 IEEE ISSCC. Feb. 1995. San FranciSCO, CA, pp. 138-139.352.

[6J W. Kuhn, F. Stephenson, and A. Elshabini-Rlad. "A 200 MHz CMOS a-Enhanced LC Bandpass Filter". submitted to IEEE Joumal of Solid State Circuits lor publication.

[7J C. T.-C. Nguyen and R.T. Howe, "CMOS Micromechanlcal Resonator Oscillator". Technical Digest, IEEE Intemaffonsl Electron Devices Meeffng, Washington. D.C .• December 5-8.1993. pp. 199-202.

[8J C. T.-C. Nguyen. "Micromechanical Resonators lor Oscillators and Filters", Proceedings of the 1995 IEEE Intemadonsl Ultrasonics Symposium, Seattle, WA. November7-10, 1995.

[9J B. Brannon, "Using Wide Dynamic Range Converters lor Wide Band Radios", RF Design. vol. 18. no. 5, May 1995, pp. 50-55, 63-65.

[10J R. Baines. "The DSP Bottleneck",IEEE Communications Magazine. vol. 33. no.5, May 1995. pp. 46-54.

INDEX

A Accelerometers 217 Adaptive algorithms 97 Adaptive antenna arrays 97 Adaptive sampling 45, 52, 53 Algorithm Signal Processor (ASP) 216 ALOHA 65 Analog Interface Circuit 60 Antenna response vector 98, 100, 10 I, 102 Attenuation loss 136

B Beamformer 98 BER 156 BiCMOS 217 Bit error probability 67 Bit error rate (BER) 104, 156 Blocking probability 75, 78 Boomer sites 25 BPSK 33 Braunberg, A. 154 Bullington, K. 155

C Capacity 97, 98, 104, 105, 108 CAPITAL Project 12 Capon beamforming 100 Carrier to Interference Ratio 21, 22 CDF ofinterference power 93,94 CDF of SIR 93,96 CDMA interference model 92 CDMA Mobile System 61 CDMASIR 93 Cellular Digital Packet Data (CDPD) 23 Cellular radio 75 Central Office (CO) 145 Channel estimation 100, 10 I, 102 Co-channel 109, 116 Co-channel interference 29 Code Division Multiple Access 57,75,76,83,

84,97,98 Code filtering 100 CODECS 216 Computational complexity 97, 98, 105, 106, 108 Concrete 136 Concrete Buildings 139 Constitutive Electrical Parameters

Concrete 13 7 Brick 137 Limestone 13 7 Concrete patch 13 8

Convergence rates 97, 98, 104, 108 Core sites 25

CPM 109 Critical distance 145 CSMA 65 CSMA/CA 41

D DCSl800 109 Decision directed 99, 107, 108 Diversity 51,52,177,178,180,184-187 DMSTM-IOO 149 DSP processors 216

E Egli,J.155 Electric-field prediction inside a Central Office

(EPCO) 145 Electromagnetic interference (EMI) 145 Electronic Serial Number (ESN) I, 2, 4, 5, 6,

7,8 Emission 147 Equal gain combiner 185 Error floor 45-56 Error region 50-52

F Fading 158 Finite Difference Time Domain (FDTD) 147, 177,

178,182 5ESS® 149 FLMI50 150 Floating point operations (flops) 105, 106, 108 FM clicks 168 Forward Control Channel (FOCC) 3,4,6,7,

8,9 Freedom Link (FDL) ISO Frequency Shift Keying (FSK) 168 Fresnel zone 154 Full-wave prediction 184 Fully Adaptive Smart Antenna 19

G Gallium Arsenide (GaAs) 217 Gaussian 153 Geolocation 12, 13 Goodman, J.M. 154 GSM 109,216

H Hand-held computers 121,122 Hopping channels 28

Immunity 145 Improved Gaussian approximation 67

222

Incident Field measurement 140 Indoor propagation 178 Interference 195 Interference cancellation 72 Intersymbol interference 45,46,48 Intersystem Signal System 41 (lS-41) 4, 5 IS-95 75 ISM 33,122

J Jakes, W.C. 154156 Joint detection 109, 116 Joint Spectrum Center (ECAC) 155 JMLSE 109

L LAN 33,122 Laurent 109, 110 Least mean squares (LMS) 99, 105, 107, 108 Lee, W.C.Y. 154,155,157 Line-of-sight 153 Line Spectral Pairs 58 Linear antenna arrays 98 Linear Predictive Coefficients 58 Longley, A.G. 155 Lorentz theorem 178

M Maximum ratio combiner 185 Mega Instruction Per Second 60 Message wait time 82, 83 Military communications 153, 159 Mobile communications 215 Mobile Identity Number (MIN) 1,2, 3,4, 5,

6,7,8 Modal analysis 177, 178, 184, 187 Modem Mode 7, 8 Mobile Switching Center (MSC) 2,4, 5, 6 MSK 109 Multipath 98, 102 Muitipath Statistics 17 Multiple-access interference 63

N Narrowband packet throughput 65 Narrowband PCS 167 N-delay channel 53-55 Network equipment 145 NYMEX 121-130

o Oakmura, Y. 155

P

Packet radio 63 PCMCIA 33 Personal communications 153 Phase shift keying 45,46 pi/4DQPSK 45,46,51,52 Portable communication system (PCS) 145 Power control 75, 76 Power-Delay-Angle Gram 14, 15 Pulse Code Modulation 60

Q Qualcomm Code Excited Linear Prediction 57

R RAKE receiver 98 Rayleigh 153,169,177,187 Rebar 138 Recursive least squares (RLS) 99, 105, 107, 108 Reflection loss 136 Regular Pulse Excited-Long Term Prediction 57 Reverse Control Channel 3, 4, 7 Rice, P.L. 155 Rician 153 Roamer 1,3,4

S Sampled matrix inversion (SMI) 99, 103, 105,

106, 107, 108 SAW filters 215 Selection diversity 178, 185 Signal-to-interference-and-noise-ration (SINR)

97, 103, 104, 107, 108 Simulation of a COMA system 90 Simulcast 167 Software Radio 11,13,19 SONET 150 Spatial filter 98 Spectrum allocation 190 Spectrum efficiency 197 Spectrum engineering 194 Spectrum management 189 Spread spectrum 63, 100 Standalone Mode 7, 8 Steele, R. 154, 157 Stream-based data services 88 System Identity (SID) 4

T Tactical military communications 153 Taylor, L. L. 154 Thevenin 183, 184 20 RAKE receiver 98, 99, 103-108 2.4-GHz 123, 125, 128, 129 Two-delay channel 46-50

U Unlicensed 193

V Vector Sum Excited Linear Prediction 57 Visitor Location Register 6 Viterbi 1\3 Voice call simulation 91

W Weisberger, M.A. 157 Welch, C. 157 Wideband packet throughput 68 Windows 144 Wireless 33 Wireless communications 97,121,122,129 WLAN 122, 128

223