MOBILE AND WIRELESS COMMUNICATIONS

IFIP - The International Federation for Information Processing

IPIP was founded in 1960 under the auspices of UNESCO, following the First World Computer Congress held in Paris the previous year. An umbrella organization for societies working in information processing, IFIP's aim is two-fold: to support information processing within its member countries and to encourage technology transfer to developing nations. As its mission statement clearly states,

IPIP's mission is to be the leading, truly international, apolitical organization which encourages and assists in the development, exploitation and application of information technology for the benefit of all people.

IFIP is a non-profitmaking organization, run almost solely by 2500 volunteers. It operates through a number of technical committees, which organize events and publications. IFIP's events range from an international congress to local seminars, but the most important are:

• The IPIP World Computer Congress, held every second year; • open conferences; • working conferences.

The flagship event is the IFIP World Computer Congress, at which both invited and contributed papers are presented. Contributed papers are rigorously refereed and the rejection rate is high.

As with the Congress, participation in the open conferences is open to all and papers may be invited or submitted. Again, submitted papers are stringently refereed.

The working conferences are structured differently. They are usually run by a working group and attendance is small and by invitation only. Their purpose is to create an atmosphere conducive to innovation and development. Refereeing is less rigorous and papers are subjected to extensive group discussion.

Publications arising from IFIP events vary. The papers presented at the IFIP World Computer Congress and at open conferences are published as conference proceedings, while the results of the working conferences are often published as collections of selected and edited papers.

Any national society whose primary activity is in information may apply to become a full member of IFIP, although full membership is restricted to one society per country. Full members are entitled to vote at the annual General Assembly, National societies preferring a less committed involvement may apply for associate or corresponding membership. Associate members enjoy the same benefits as full members, but without voting rights. Corresponding members are not represented in IFIP bodies. Affiliated membership is open to non-national societies, and individual and honorary membership schemes are also offered.

MOBILE AND WIRELESS COMMUNICATIONS

IFIP TC6/ WG6.8 Working Conference on Personal Wireless Communications (PWC'2002) October 23-25, 2002, Singapore

Edited by

Cambyse Guy Omidyar Institute for Communications Research National University of Singapore Singapore

SPRINGER SCIENCE+BUSINESS MEDIA, LLC

Library of Congress Cataloging-in-Publication Data

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

Mobile and Wireless Communications Edited by Cambyse Guy Omidyar

ISBN 978-1-4757-1033-5 ISBN 978-0-387-35618-1 (eBook) DOI 10.1007/978-0-387-35618-1

Copyright © 2003 by Springer Science+Business Media New York Originally published by Kluwer Academic Publishers in 2003 Softcover reprint of the hardcover 1 st edition 2003

All rights reserved. No part of this work may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photo­copying, microfilming, recording, or otherwise, without written permission from the Publisher (Springer Science+Business Media, LLC), with the exception of any material supplied specifically for the purpose of being entered and executed on a computer system, for exclusive use by the purchaser of the work.

Printed on acid-free paper.

Contents

Preface .......................................................................................................... xi

Part 1 Power Control I MIMO Receiver Algorithm

Second-Order Statistics of Closed-Loop Power Controlled Signals in Multi-Path Rayleigh Fading Channels Hafez Hadinejad-Mahram and Xiaolong Jiang ............................................. 3

Performance Comparison of Multiple-Transmit Multiple-Receive V -BLAST Algorithms Hufei Zhu, Zhongding Lei and Francois Chin ............................................. 11

Part 2 Ad-Hoc Networking

Market-based Network Formation for an Ad Hoc, P2P Wireless Network Yasunori Yamamoto and Junseok Hwang .................................................... 21

An Efficient Proactive Routing Method for Mobile Ad-hoc Networks Using Peer-to-Peer and Cellular Communication System Hiroaki Morino, Tadao Saito and Mitsuo Nohara ...................................... 29

A Mobile Multicast Framework for CDMA-based Ad Hoc Networks Hsu-Yang Kung and Su-Man Chen .............................................................. 37

VI

Multipath Routing in Ad Hoc Wireless Networks with Directional Antenna Somprakash Bandyopadhyay, Siuli Roy, Tetsuro Veda and

Contents

Kazuo Hasuike ............................................................................................. 45

A Reactive Service Composition Architecture for Pervasive Computing Environments Dipanjan Chakraborty, Filip Perich, Anupam Joshi, Timothy Finin and Yelena Yesha ................................................................. 53

Part 3 Personal Wireless Communications

Bluetooth PAN and External IP Networks Tore E. J¢nvik, Paal Engelstad and Do van Thanh ..................................... 63

DTV for Personalized Mobile Access and Unified Home Control Jianlin Guo, Fernando Matsubara, Johnas Cukier and Haosong Kong ............................................................................................. 71

A Novel Internet Radio Service for Personal Communications; The Private Channel Service Kensuke Arakawa, Yasushi Ichikawa and Yuko Murayama ........................ 79

Tools for On-Door Communications on WWW Keishi Suzumura, Hiromi Gondo and Yuko Murayama .............................. 87

Part 4 Buffer Control/Receiver

Buffer Control Using Adaptive MQAM for Wireless Channels Anh Tuan Hoang, and Mehul Motani .......................................................... 97

A Low Complexity Iterative Receiver Based on Successive Cancellation for MIMO Holger Claussen, Hamid Reza Karimi and Bernard Mulgrew .................. 105

Part 5 Satellites/High Altitude Platforms Station

Dedicated Bandwidth Approach for Channel Allocation in a Multi­Service Up/Down Link of a Low Earth Orbit Satellite Constellation RimaAbi Fadel and Samir Tohme ................................................. ............ 115

Mobile and Wireless Communications

Softer Handover Schemes for High Altitude Platform Station (HAPS)UMTS

vii

Woo Lip Lim, Yu Chiann Foo and Rahim TaJazolli ................................... 123

Part 6 Quality of Service QoS

Adaptive QoS and Handover Issues in Wireless Multimedia Networks Using a Dynamic Adaptive Architecture: DYNAA Rola Naja and Samir Tohme ...................................................................... 133

Dynamic QoS Guarantee with Repeater in Power Controlled WCDMA Urban Environment Mohammad N. Patwary, Predrag Rapajic and Ian Oppermann ............... 141

Part 7 UMTSIWireless LANs

Very Tight Coupling of Wireless LANs and UMTS Networks: A Technical Challenge and an Opportunity for Mobile Operators Manfred Litzenburger, Hajo Bakker, Stephen Kaminski and Klaus Keil .................................................................................................. 151

Dynamic UMTS Simulator for Congestion Studies and Evaluation of Resource Management Techniques Sami Nousiainen, KrzysztoJ Kordybach, Paul Kemppi and Veli-Pekka Kroger ..................................................................................... 159

Capacity and Coverage Increase with Repeaters in UMTS Mohammad N. Patwary, Predrag Rapajic and Ian Oppermann ............... 167

Pre-Authenticated Fast Handoff in a Public Wireless LAN Based on IEEE 802.1x Model Sangheon Pack and Yanghee Choi ............................................................ 175

Service Integration MUltiple Access (SIMA) A Protocol for Supporting Voice & Data in Wireless LANs Apichan Kanjanavapastit and Hassan Mehrpour ...................................... 183

Spatial Variation of Digital Television Signal in an Indoor Environment Ong lin Teong, Yan Hong and Shanmugam Ganeshkumar ....................... 191

viii Contents

Part 7.1 Security

Development of a Strong Stream Ciphering Technique Using Non-Linear Fuzzy Logic Selector Ahmed M. AI-Naamany and Afaq Ahmad .................................................. 199

Part 8 Multiple Access Techniques

Wireless MAC Scheme for Service Differentiation A Distributed Protocol Abdulla Firag and Harsha Sirisena ........................................................... 209

Packet Acquisition Evaluation of Slotted Spread ALOHA Data Networks Waseem librail and Ranjith Liyana-Pathirana ......................................... 217

Part 9 Code Division Multiple Access COMA

On Erlang Capacity of CDMA Systems Samad S. Kolahi ................................................ ......................................... 227

Power and Spreading Gain Allocation in CDMA Data Networks for Services with a Relative Priority Kwang-Seop lung, Sun-Mog Hong and Eun-Young Park ......................... 233

Adaptive Closed-Loop Power Control Using an MMSE Receiver in DS-CDMA Systems Lian Zhao and Jon W Mark ....................................................................... 241

CORDIC Based QRD-RLS Adaptive Equalizer for CDMA Systems Tim Zhong Mingqian, As Madhukumar and Francois Chin ...................... 249

Resource Allocation Using Dynamic Spreading Gain Control for Wideband CDMA Networks Supporting Multimedia Traffic Hailong Huang and Francois Chin ........................................................... 257

Mobile and Wireless Communications

Part 9.1 Turbo Code

On the Fixed-Point Implementation of Turbo Code in 3GSystem

ix

Sun Minying and Tan Wee Tiong ................................................. .............. 267

Part 10 Mobility

Cellular Positioning by Database Comparison and Hidden Markov Models Trond Nypan and Oddvar Hallingstad ...................................................... 277

Architectural Considerations for Personal Mobility in the Wireless Internet Mazen Malek Shiaa and Finn Arve Aagesen ............................................. 285

A Development of Flexible Access Control System for Advanced ITS Networking Mitsuo Nohara, Sheng-Wei Cai, Hitoshi Inoue, Yoshiro Okamoto and Tadao Saito ......................................................................................... 293

Ubiquitous Access to Personalised Services Tore E. 1;mvik, Anne Marie Hartvigsen and Do van Thanh ...................... 301

Preface

The Personal Wireless Communications (PWC) 2002 Conference was held in the beautiful garden city, Singapore. PWC has established itself as one of the IFIP TC6 conferences in the Mobile and Wireless Communications field. This year we received over 60 submissions, for a single-track conference with a keynote and three invited speeches.

Mobile communications and the Internet have created a major breakthrough in new telecom services around the world. It is expected that the number of global mobile connections will exceed the number of fixed connections in certain countries. The data service breakthrough enabled by the Internet will create new possibilities for mobile and wireless services.

We are living in a world of creation. Voice over cellular and data over Internet was created. When demand arose for mobility, we added data capability to handsets and voice to Internet. However, not all creations are useful; only a handful of people will use them.

In the area of telecommunications, we created Wide, Metropolitan, Local, Home, Vehicular and Desktop areas. The transport systems use wire-lines and wireless media. In the past, we created satellites and have done little with deep space beyond our reach. We are creative and we will create.

In all, the Personal Wireless Communications 2002 conference is what it is all about, to report on our findings and discuss our experiences and lessons learned from one another.

The PWC'2002 conference is a forum for tutorials, discussions and presentations of the new developments in mobile and wireless research and industry. The conference is arranged with one day for tutorials and two days for presentation. Topics presented in PWC 2002 are in the areas of Ad-Hoc Networking, Power Control, Personal Communications, Satellite, QoS,

xii Preface

UMTS and Wireless LANs, Handoffs, Security and Mobility, COMA and Physical Layer including modulation, coding and methods of communication functions including multiple access, error control, flow control and routing.

The Personal Wireless Communications PWC'2002 conference belongs to an IFIP workshop and conference series arranged by IFIP TC6 Working Group 6.8. Previous PWC events were held in Prague, Tokyo, Frankfurt, Copenhagen, Gdansk, and Lappeenranta in Finland. The PWC'2002 event was held in Singapore from October 23 to 25, 2002.

We did our best to bring you an outstanding PWC'2002 Technical Program and hope you will enjoy it.

Professor Kin Mun Lye Conference Chair

Organising Committee

Dr. Guy Omidyar Technical Program Chair

Kin Mun LYE, ICR, Singapore (Conference Chair) Guy OMIDY AR, ICR, Singapore (Tech Program Chair, WG 6.8 Co-chair) Koujuch LlOU, ICR, Singapore Michael ClllA, ICR, Singapore Jackson LAM, ICR, Singapore Beng CHEAH, ICR, Singapore

IFIP TC6 Working Group 6.S

Arup ACHARYA, USA Sathish CHANDRAN, Malaysia Marco CONTI, Italy Franco DAVOLI, Italy Silvia GIORDANO, Switzerland Veikko HARA, Finland Takeshi HATTORI, Japan Sonia HEEMSTRA de GROOT,

Netherlands Villy Baek IVERSEN, Denmark Ousmane KONE, France Pascal LORENZ, France Gerald MAGUIRE Jr., Sweden OlIi MARTIKAINEN, Finland

Ignacious NIEMEGEERS, Netherlands

Guy OMIDY AR, Singapore Guy PUJOLLE, France Debashis SAHA, India Tadao SAITO, Japan Jan SLAVIK (WG 6.8 Chair),

Czech Republic Otto SPANIOL, Germany Samir TOHME, France Andras G.VALKO, Hungary Adam WOLlSZ, Germany JozefWOZNIAK, Poland

POWER CONTROL / MIMO RECEIVER ALGORITHM

Second-Order Statistics of Closed-Loop Power Controlled Signals in Multi-Path Rayleigh Fading Channels

Hafez Hadinejad-Mahram and Xiaolong Jiang Institute of Communications Engineering Aachen University of Technology hafezGient.rwth-aachen.de

Abstract The second-order statistics of power controlled signals in multi-path Rayleigh fading channels are considered. These statistics can be used e.g. in design and evaluation of channel estimation and channel coding schemes. The simulation results presented here provide a comprehensive comparison between the figures obtained with and without power con­trol. The measures considered include the autocorrelation function and the power spectrum of the received signal, the autocovariance function of the signal power, and finally the mean and variance of the received signal power. The results are based on single-user link level simulations with a fixed-step power control scheme.

Keywords: Closed-loop power control, signal statistics

1. Introduction Power control (PC) is a key ingredient of direct-sequence code division

multiple access (DS-CDMA) systems [Gilhousen et al., 1991]. Its major task is to prevent the so called near-far effect, i.e. the situation in which the strong signal of one or several users overwhelms the signals of the other users resulting in a significant degradation of the system performance. Power imbalance among the users of a DS-CDMA system is due to several phenomena including path loss, shadowing (slow or long­term fading), and multi-path (fast or short-term) fading. The former two exhibit reciprocity in the forward and reverse links and thus can be combated by means of open-loop PC [Lee and Steele, 1996, Tam and Lau, 1999]. For example in the uplink, each mobile will measure

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

4 HaJez Hadinejad-Madram and Xiaolong Jiang

the averaged power from the base station over a long period of time and adjust its transmitted power inversely proportional to the averaged received power. The multi-path fading, on the other hand, varies very fast and is, in general, different for the forward and reverse link due to the large distance between the carrier frequencies of the two links and the frequency-selectivity of the channel. Thus, some form of closed-loop (adaptive) PC (CLPC) is needed to mitigate the effect of multi-path fading. However, it is intuitive and has also been shown by several researchers [Ariyavisitakul and Chang, 1993, Choclm.Jingam et al., 1998, Sim et al., 1999] that CLPC can only be effective if it is fast enough to track the variations of the channel.

The conventional fixed-step CLPC algorithm [Lee and Steele, 1996] is quite simple: the receiver measures the averaged received signal power or the averaged signal-to-interference power ratio (SIR) over the current PC period and compares it with a preset target value. A command is then sent to the transmitter over a feedback channel requesting it to reduce or increase its transmit power depending on the outcome of the comparison. Both fixed-step and adaptive step-size algorithms have been considered in the literature [Lee and Steele, 1996, Park and Nam, 1999]. Besides the conventional approach several predictive approaches have been proposed [Lau and Tam, 2001a, Lau and Tam, 200lb, Freris et al., 2001].

Most of the recent publications dealing with the statistics of the sig­nal and/or interference in DS-CDMA systems employing fast CLPC aim at estimating the system capacity based either on SIR cumulative distribution functions [Ariyavisitakul and Chang, 1993, Ariyavisitakul, 1994, Hashem and Sousa, 1999] or on the uncoded channel bit error rate (BER) [Chocka.Jingam et al., 1998]. In [Chockalingam et aI., 1998], in ad­dition to the BER calculation, the autocovariance function of the power in dB is given which quantifies the ability of the CLPC to compensate for the time-varying channel. The given power correlation statistics are, however, restricted to the case of a flat fading channel. Pirinen [pirinen, 2001] investigates, among others, the impact of mobility and CLPC on the autocorrelation function of the received signal in flat fading. How­ever, the simulation results seem to be partly flawed.

In this paper, we study the second-order statistics of power controlled signals in multi-path Rayleigh fading channels. The importance of these statistics lies with their application in the design and evaluation of chan­nel estimation and channel coding schemes. Extensive simulation results provide a comprehensive comparison between the figures obtained with and without power control parameterized by mobile speed and power control step size. The measures of interest are the autocorrelation func-

Second-Order Statistics of Closed-Loop Power ... 5

Figure 1. Closed-loop power control model

tion and the power spectral density (PSD) of the received signal, the auto covariance function of the signal power, and finally the mean and variance of the received signal power. The results are based on single­user link level simulations with a fixed-step CLPC scheme. Furthermore, path loss and shadowing are not considered in our model. The fixed pa­rameters like carrier frequency, symbol (or chip) rate, and power control command rate are chosen so as to match those of WCDMA.

2. System Model

The CLPC model used in this study is shown in Fig. 1. The trans­mitted signal traverses the fading channel and arrives at the receiver. There, the received power is estimated and based on this estimate a power control decision is made which is then sent back to the transmit­ter. The transmitter can now adjust its transmission power by executing the power control command, i.e. by reducing or increasing its transmis­sion power by a fixed factor ll. 2 (dividing or multiplying the transmitted signal by ll.). This procedure recurs every Tp seconds, where Tp is the power control command rate. Since the power command bit is assumed to be unprotected it is sensible to allow for a certain percentage of feed­back channel errors which is also included in the model.

The fading channel is modeled as a multi-path Rayleigh fading channel including the case of a single propagation path (flat fading). Each tap of the channel is assumed to have the classical (Jakes) Doppler spectrum [Jakes, 1974]. We assume that we can perfectly separate the multi-path components, i.e. there is no loss in diversity due to the correlation among the diversity branches. In practical systems this loss is present because of the imperfect autocorrelation properties of the waveforms resulting in the multi-path components to be correlated after demodulation. We believe, that the simplified setting of the ideal diversity system - which

6 Hafez Hadinejad-Madram and Xiaolong Jiang

has been adopted by many researchers before - captures the essence of the material and provides useful insight into the main characteristics of a power controlled DS-CDMA channel.

Power estimation is accomplished by an integrate and dump device which averages the instantaneous received power over the duration of a PC period Tp. The power command decision is made by comparing this power estimate with a preset threshold.

In the following we describe the statistics considered here in detail. First, we define the normalized sample autocorrelation function of the received signal of k-th tap as

where N is the number of samples and TA:(n) is the n-th sample of the k-th tap. The PSD estimate of the received signal is calculated for each tap by applying a 1024-point FFT to the above sample autocorrelation function [Oppenheim and Schafer, 1989]. The length of the autocorre­lation function is 199 (Le. 2·100-1) samples where the sampling rate is chosen to be the Nyquist rate, i.e. 2fD with fD being the maximum Doppler frequency. Finally, the normalized sample autocovariance func­tion of the (total) received signal power in dB is defined as

N N E R(n) R(n + m) - kl E R(n)12

C(m) = n=1 N N n=1

E IR(n)12 - 11 E R(n)12 n=1 n=1

where R(n) is the total received power in dB at time sample n. Table 1 summarizes the fixed simulation parameters.

Carrier frequency 20Hz Chip rate 3.84 Mchip/s PC command rate .J:. 1.5 KHz Spreading factor 256 PC loop delay Tp=~s Power threshold OdB

Table 1. Fixed simulation parameters

Second-Order Statistics of Closed-Loop Power ... 7

3. Numerical Results In this section we present and discuss our simulation results. First we

consider the case of a flat fading channel which already reveals the most important characteristics of power controlled signals. Then, we look at some results obtained for the multi-path case.

3.1. Flat Fading The normalized sample autocorrelation function for different step sizes

and different velocities was computed. Fig. 2(a) shows the results for 3 and 10 km/h and a power step size of 1 dB (if not otherwise stated, throughout the paper a step size of 1 dB is used). The results for other step sizes were quite similar. It is seen that the autocorrelation function slightly decreases with power control. This decrease becomes smaller for higher velocities and almost vanishes at velocities above 30 km/h (not shown in the figure) . Fig. 2(b) shows the autocorrelation function in a narrow interval around zero. The zigzag shape of the curve for 3 km/h can be explained as follows. When the channel changes very slowly a correct PC command can partly compensate the variations of the channel during the previous PC periods which leads to an instant increase of the correlation function at the transition between two PC periods. Note that the extent of this compensation depends on the Doppler rate, thus, this effect can be observed only at the very low velocities.

0.8

i 0.6

i 0.4

.; I 0.2

.. 0

1- PC on I - PCoH

-0.2 -0.1 0 0 .1 0.2 I[s)

(a)

PC on

0 .99 j ~ ,§ 0.98

1 .. 0.97

-0.006

(b)

.00 l[s)

0.006

Figure 2. Autocorrelation function of the received signal for different velocities.

8 Hafez Hadinejad-Madram and Xiaolong Jiang

-4 ......... ~ .... ............ .. -15 .----....--~-~-~_---. · . . · . .

-5 "I=~~I'''''''''''' -6 .. . ..... . .... . ~:: :: :: '1_---'-:, ~O::::g-~~..,.[ j': :::: :::L:::::':::::::'::: . "

- 7 ......... : ............... .. - 18 ......... : ......... : ......... : ........ . . : ... . ... .

-6 ......... : ................ . - 19 ....... ": ......... : ... " ............... ; ..... " . . . .

o -9 ......... ; ................. . ~

~ -20 """ ... ; .. " ..... ; ...... : ... . ... : .......

-10 .... · .. ·;· ...... ..

- 11 ......... ~ ...... ..

Il. : - - : · . . -21 ......... : ......... : .......... :-- ........ : ..... .

- 12 ......... : ....... ..

o 0.2 0.4 0 .6 0.8 0.2 0.4 0 .6 0.8 IIt_ IItD

(a) 3 km/h (b) 50 km/h

Figure 9. Power spectral density with and without PC.

Fig. 3(a) and 3(b) show the PSD for 3 and 50 km/h. In both cases the difference between the PSDs with and without PC is rather small as could be expected from the observation of the autocorrelation functions.

However, the differences for 3 km/h is larger than for 50 km/h. In

0.9

8 0.8

g 0.7

~ 0.6

i 0.5 .~

~ 0,4

iil 0.3

~ 0.2

~ 0.1

o~~~~/'U

-0.1

-0.1 -0.05 0 0 .05 0.1 t(s)

(a) 10 km/b

1- peoN 1 - peon

0.8

'1 _ 0 .6

!! .~ 0.4

~ iI 0.2

~ tf. o ...... ___ -r.v

-O.2L....o....,----~_~><.......__"_--'-_"----'-' -0.03 -0.02 -0.01 0.01 0.02 0.03

(b) 50 km/h

Figure 4. Power autocovariance function with and without PC.

the former case PC is capable of tracking the fading channel thus a noticeable difference in the spectrum, even though small, is expected.

Second-Order Statistics of Closed-Loop Power ... 9

Fig. 4( a) and 4(b) show the normalized autocovanance function of the received signal power in dB. Similar results were reported in [Chock­alingam et al., 1998] based on an analytical log-linear model. As can be seen from the comparison of the curves for 10 and 50 km/h, for low velocities, the difference between the curves with and without PC is larger. This is due to the random zigzag shape of the power controlled signal (in dB) at low velocities. The better the PC loop can track the channel variations the more random the curves will look and the less correlated the samples will be. Moreover, the main peak of the auto­covariance function is observed to be broader for lower speeds, whereas the sidelobes decrease faster.

step size OdB 0.5 dB IdB 2dB Velocity mean var mean var mean var mean var

3km/h 0.996 0.992 1.027 0.063 1.016 0.037 1.039 0.085 10km/h 0.997 0.993 1.192 0.727 1.080 0.241 1.063 0.171 5Okm/h 0.997 0.993 1.040 2.123 1.399 2.194 1.352 1.797

Table 8. Mean and variance of the received power for different parameters.

Table 2 shows the mean and variance of the received signal power pa­rameterized by mobile speed and PC step size. Obviously, the optimum step size, i.e. the one which yields a mean close to 1 and a variance close to 0 are different for different velocities.

3.2. Frequency-Selective Fading Similar simulations as with the flat fading case were conducted for

multi-path channels with 2 and 4 taps with different tap weight settings. The results were quite similar to those discussed in the previous section. The PSDs obtained for each tap had a shape like those shown in Fig. 3. However, the anyway small difference between the curves with and without PC was even smaller. As for the power autocovariance function, it could be observed that its sidelobes go to zero faster in the multi-path case and otherwise the shape of the curves are quite similar to those obtained with flat fading.

4. Conclusions We conducted extensive simulations to asses the changes of the cor­

relation properties of power controlled signals as compared to the case without power control. We looked at several measures including the autocorrelation function, the power spectral density, and the autoco-

10 Hafez Hadinejad-Madram and Xiaolong Jiang

variance function of the total received power. Our simulation results show that the changes of the correlation properties of the received signal are rather slight and can probabily be neglected for most applications even for low mobility.

References [Ariyavisitakul, 1994] Ariyavisitakul, S. (1994). Signal and interference statistics of

a CDMA system with feedback power control - part II. IEEE 1hlns. Comm., 42(2/3/4):597-605.

[Ariyavisitakul and Chang, 1993] Ariyavisitakul, S. and Chang, L. (1993). Signal and interference statistics of a CDMA system with feedback power control. IEEE 1hlns. Comm., 41(11):1626-1634.

[Chockalingam et aI., 1998] Chockalingam, A., Dietrich, P., Milstein, L., and Rao, R. (1998). Performance of closed-loop power control in DS-CDMA cellular systems. IEEE 7rans. Vehicular Technology, 47(3):774-788.

[Freris et aI., 2001] Freris, N., Jeans, T., and Taaghol, P. (2001). Adaptive SIR esti­mation in DS-CDMA cellular systems using Kalman filtering. Electronic Letters, 37(5):315-317.

[Gilhousen et aI., 1991] Gilhousen, K., Jacobs, I., Padovani, R., Viterbi, A., and Weaver, L. (1991). On the capacity of a cellular CDMA system. IEEE 1hlns. Vehicular Technology, 40:303-312.

[Hashem and Sousa, 1999] Hashem, B. and Sousa, E. (1999). Reverse link capacity and interference statistics of a fixed-step power-controlled DS/CDMA system under slow multipath fading. IEEE 1hlns. Comm., 47(12):1905-1912.

[Jakes, 1974] Jakes, W. (1974). Microwave Mobile Communication. Wiley, New York. [Lau and Tam, 2001a] Lau, F. and Tam, W. (2001a). Novel SIR-estimation-based

power control in a CDMA mobile radio system under multipath environment. IEEE 1hlns. Vehicular Technology, 50(1):314-320.

[Lau and Tam, 2001b] Lau, F. and Tam, W. (20mb}. Predictive closed-loop power control in CDMA mobile systems. Electronic Letters, 37(1):52-54.

[Lee and Steele, 1996] Lee, C.-C. and Steele, R. (1996). Closed-loop power control in CDMA systems. lEE Proc. Comm., 143(4):231-239.

[Oppenheim and Schafer, 1989] Oppenheim, A. and Schafer, R. (1989). Discrete­Time Signal Processing. Prentice-Hall, Englewood Cliffs, NJ.

[Park and Nam, 1999] Park, S. and Nam, H. (1999). DS/CDMA closed-loop power control with adaptive algorithm. Electronic Letters, 35(17):1425-1427.

[Pirinen, 2001] Pirinen, P. (2001). Impact of mobility and closed-loop power control to received signal statistics in Rayleigh fading channels. In Proc. IEEE Vehicular Technology Con/. VTC'Ol-Spring, pages 2859-2863, Rhodes, Greece.

[Sim et a!., 1999] Sim, M., Gunawan, E., Soong, B.-H., and Soh, C.-B. (1999). Per­formance study of close-loop power control algorithms for a cellular CDMA system. IEEE 7rans. Vehicular Technology, 48(3):911-921.

[Tam and Lau, 1999] Tam, W. and Lau, F. (1999). Analysis of power control and its imperfections in CDMA cellular systems. IEEE 7rans. Vehicular Technology, 48:1706-1717.

Performance Comparison of Multiple-Transmit Multiple-Receive V -BLAST algorithms

Hufei Zhu, Zhongding Lei and Francois Chin Institute for Communications Research 20 Science Park Road, Singapore 117674 Email: engp1735@nus.edu.sg.{leizd.chinfrancois}@icr.a-star.edu.sg

Abstract BLAST, a MIMO wireless communication systems can achieve very high spectral efficiency in rich multipath environment through exploiting the extra space dimension. A simplified version of BLAST known as V-BLAST has been proposed and implemented. In this paper we compare the performance of various V-BLAST algorithms through computer simulations. The results show that the nonlinear detection schemes with interference cancellation (Ie) have much better performance than linear detection schemes and Minimum Mean Squared Error schemes (MMSE) have better performance than Zero Forcing (ZF) schemes. The efficient square-root algorithm with Ie shows very attractive property that having low complexity while keeping the good performance compared with conventional IC scheme. In our simulation, little improvement is observed for the Pre-match filtering plus IC scheme.

Key Words V-BLAST, ZF, MMSE, Ie, an Efficient square-root algorithm, Pre-MF

1. Introduction

MUltiple-input multiple-output (MIMO) wireless communication systems can achieve very high spectral efficiency in rich multipath environment through exploiting the extra space dimension [1][2]. BLAST (Bell Labs Layered Space-Time) architecture [2][3] is such a system realizing very high data rates without additional power or bandwidth consumption. The diagonal BLAST or D-BLAST proposed by Foschini[4], utilizes multiple antenna at both transmitter and receiver and an elegant diagonally-layered coding structure in which code blocks are dispersed

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

12 Rulei Zhu, Zhongding Lei and Francois Chin

across diagonals in space-time. In an independent Rayleigh scattering environment, this processing structure leads to theoretical rates which increases linearly with the number of antennas (assuming equal numbers of transmit and receive antennas) with these rates approaching 90% of Shannon capacity. However, the D-BLAST suffers from low efficiency for short package transmission and it requires advanced encoding techniques and means to avoid catastrophic error propagation, which make it inappropriate for initial implementation.

In addressing these problems, a simplified version of BLAST known as vertical BLAST or V -BLAST has been proposed and implemented in real time in laboratory [3][5]. It is demonstrated that the V -BLAST is cost­effective and highly spectrally efficient, achieving as much as up to 60% of the capacity achievable by D-BLAST. Therefore V -BLAST has drawn lots of attention recently and much effort has been made to achieve better performance [6][7] or lower complexity [8][9]. In this paper, we will compare the performance of various V -BLAST algorithms through computer simulations over Rayleigh fading channels.

This paper is organized as follows. The V -BLAST system is overviewed in section II followed by the description of various V -BLAST detection algorithms in section m. The simulation results are shown in section N and concluded in section V.

2. V·BLAST system overview

The V -BLAST diagram is shown in Fig. 1. This system consists of M transmitters and N receivers where ~. At the transmitter end, a single data stream is de-multiplexed into M streams, and each sub-stream is then encoded into symbols and fed to its respective transmitter.

The wireless channel is assumed to be rich-scattering and flat-fading. The fading between each antenna pair of transmitter and receiver is assumed independent.

At the receiver end, each receiving antenna receives the signals from all

M transmitter antennas. Letting a = [a .. a2 ,. .. ,aM f ('T' denotes transpose operation) denotes the vector of transmit symbols from M antenna, then the received signal can be represented as

r=H·a+v (1) where H is the N-by-M complex channel matrix with statistically

independent entries and v is the complex Gaussian noise vector with zero d . 2

mean an vanance trv .

Simulation and Peiformance Comparison on V-BLAST algorithms 13

"il c:I

Vector c

TX ..

encoder -B

'f V-Blast RX data Signal

data ... Processing -5 it!

Figure 1: The block diagram of V -BLAST system

3. DETECTION ALGORITHMS

3.1 Linear Detection

Based on the formula (1), a linear detection is simply to mUltiply the received signal vector r with a linear transform matrix G, i.e. the estimated signal vector may be represented as

Ii = G . r = G . H . a + G . v (2) This linear processing is also known as "nulling". Because the effect of

the linear processing for each sub-stream is to keep the desired sub-stream signal while suppress or null the other sub-stream signals at the same time.

The linear detection algorithms differ from each other by the selection of G, which is derived based on different criterion. The most common criteria for nulling are Zero Forcing (ZF) and Minimum Mean Squared Error (MMSE), for which the corresponding linear transform matrix are

G = H+ (3)

(4)

respectively, where superscript '+', 'H', and '-1' represent matrix pseudo inverse, Hermitian, and inverse operation respectively. As can be seen, the linear detection obtain estimates of all M signals at the same time.

14 Hufei Zhu, Zhongding Lei and Francois Chin

3.2 Nonlinear Detection

3.2.1 Interference Cancellation (IC)

Actually, the nonlinear detection schemes [1] with interference cancellation (Ie) are borrowed from the multiuser detection context. This kind of detection methods does not detect the M signals at one run. Instead, it starts with linear detection of only one sub-stream by means of nulling with ZF or MMSE, i.e. multiply r with a row vector of G instead of the matrix G. This sub-stream was selected because it was the best one amongst the rows, in the sense that it gets highest post-detection signal-to-noise ratio (post-SNR). Then the effect of the detected signal is subtracted out from the received signal vector, resulting in a modified received vector with less "interferers". This process proceeds until all the signals are detected.

Let the ordered set q == {kp k2." ··,kM } (5)

be a permutation of the integers 1,2 .. ·· ,M specifying the order in which components of the transmitted symbol vector a are extracted. The full detection algorithm (with ZF nulling) can be described compactly as a recursive procedure, including determination of the optimal ordering for selection of the best row, as follows:

a) initialization: i ~ 1 (6a) G 1 = H+ (6b)

kl = arg~nll(GI) jlr (6c) J

b) recursion

W k j = [(Gj)k; f T Yk j = wkj ·rj

ilk; = Q(Ykj )

ri+l = rj - ilk; . (H) kj

G j +1 =H~. . kj+1 = arg minll(G i+l ) j r

jelkt···kj )

i~i+l

(6d)

(6e)

(6t)

(6g)

(6h)

(6i)

(6j)

where (G;) j denotes the jth column of G i' H k; stands for the matrix

obtained by zeroing columns kl , k 2 • ••• , k j of H, 11.11 is the length of the

Simulation and Performance Comparison on V-BLAST algorithms 15

vector and Q(.) denotes the quantization (slicing) operation appropriate to the constellation in use.

For nonlinear IC detection with MMSE, the steps are the same except Gp(i = 1, ... ,M) are calculated based on formula (4) instead of (3) above.

3.2.2 An efficient square-root algorithm for IC (SRIC)

An efficient square-root algorithm for nonlinear IC detection has been reported in [9]. This algorithm is to find an efficient way to compute the pseudo inverse in steps (6b)/(6i) to ease computation burden. Actually it reuses the intermediate computation results of previous iterations for the current iteration to avoid re-computing the pseudo-inverse (or QR decomposition) for each deflated subchannel matrix.

3.2.3 Pre-match mtering (pre-MF) with IC

The performance of IC detector for VBLAST varies with the channel matrix D. It may be improved through changing the distribution of the channel matrix. Pre-MF with IC scheme has been proposed based on this idea [10]. Before applying IC algorithm, it pre-process the channel matrix with a matching filter and fed the resulting cross-correlation matrix to the conventional BLAST IC detector for BPSK modulated signal. The signal after matched filter can be represented as

Y=HH ·r=HH ·H·a+HH ·v=R·a+v (7) where R = H H H is the cross-correlation cahnnel matrix. Therefore the matrix R, replacing D, is used to go through the process of IC scheme. (from (6a)to (6k».

4. Simulation results

In this section, we evaluate various detection algorithms over Rayleigh fading channel in the presence of white Gaussian noise through simulations. The algorithms are linear ZF, linear MMSE, IC with ZF, SRIC with ZF, SRIC with MMSE, pre-MF IC with ZF.

We consider the systems with 4 transmitter and 4 receiver antennas. Figure 2 shows cumulative distribution functions of SNR of these algorithms when pre-detection SNR is 10 db, while figure 3 shows corresponding cumulative distribution functions of channel capacity of these algorithms. The average BER (bit error rate) curves vs pre-detection SNR are shown in figure 4.

16 Hufei Zhu, Zhongding Lei and Francois Chin

From these figures we can see consistently that nonlinear detection schemes with IC perform much better than linear detection schemes in terms of post-detection SNR. capacity. and BER. but at the cost of more computational load. MMSE detection schemes (either linear or nonlinear) have better performance than their ZF counterparts because of taking into consideration of the effect of noises. It is notable that the SRIC algorithms have almost no performance degradation compared with their counterparts while keeping computational load quite low. As to pre-MF IC scheme. little improvement is observed compared with conventional IC scheme. This may be due to the assumption we made in our simulations that most of decisions for the first data stream are correct to ease computation load.

s. Conclusion

We present six V -BLAST algorithms. Simulation results show that the nonlinear detection schemes with IC has much better performance than linear detection schemes and MMSE has better performance than ZF. There is no obvious difference observed between the performance of the efficient square-root algorithm with conventional IC while the former has attractive low computing complexity. And little performance improvement is observed for the Pre-MF with IC scheme compared with conventional IC.

Reference [1) P.W. Wolniansky, GJ. Foschini, G.D. Golden and R.A. Valenzuela, "V-BLAST: an

architecture for realizing very high data rates over the rich-scattering wireless channef', in Proceeding of URSI International Symposium on Signals, Systems, and Electronics

(ISSSE'98), pp. 295 -300,1998.

(2) GJ. Foschini and M.l Gans, "On limits of wireless communications in a fading environment when using mUltiple antennas", Wireless Personal Communications, vol. 6, no. 3, pp. 311-335, Mar. 1998.

(3) GJ. Foschini, "Layered space-time architecture for wireless communication in a fading environment when using multi-element antennas", Bell Labs Tech. J., pp.41-59, Autumn

1996. (4) G.G. Raleigh and J.M.Cioffi, "Spatio-temporal coding for wireless communication", IEEE

Trans. Commun., vol. 46, pp. 357-366, Mar. 1998. (5) G.D. Golden, GJ. Foschini, R. A. Valenzuela, and P.W. Wolniansky, "Detection

algorithm and Initial Laboratory Results using the V-BLAST Space-Time Communication Architecture", Electronic Letters, Vol. 35, No.1, Jan. 7,1999, pp. 14-15.

Simulation and Performance Comparison on V-BLAST algorithms 17

[6] Li, X.; Huang, H.; Foschini, GJ.; Valenzuela, R.A., "Effects of iterative detection and decoding on the performance of BLAST', Global Telecommunications Conference, 2000.

GLOBECOM '00. IEEE, Volume: 2,2000 Page(s): 1061-1066 vol. 2.

[7] Won-Joon Choi; Negi, R.; Cioffi, J.M. "Combined ML and DFE decoding for the V­BLAST system", in Proceeding of IEEE International Conference on Communications,

Vol. 3, pp. 1243 -1248, 2000.

[8] Wong Kwan Wai; Chi-Ying Tsui; Cheng, R.S. "A low complexity architecture of the V­BLAST system", Wireless Communications and Networking Confernce, 2000. WCNC.

2000 IEEE, Volume: 1,2000 Page(s): 310-314 voU.

[9] Babak Hassibi "An efficient square-root algorithm for BLAST'. [10] S. Sfar, N. Boubaker, R.D. Murch, and K.B. Letaief, "Performance ofpacketized layered

space-time detection over wireless links", ISCC' 2001, pp592-596, 2001.

_ . .... z,:: I II III I

~1

0.'

~

-0. IC..,." I I I I III 0.1 - ::::;.: :::~e - - -I_ ...j _ 1_ I-I Iu ' __ L _'- J..L LU

........ RI: ... lF I I I I I/~ .c I I I IIIII

..Q. SRICIllfIlMWSE I I I I J I "~ I I I 1 I I III 0.1 - - - ..... -"i - 17' 1'""14 r - - t" - ~..,. T t"'1"t

: : : : :: ::: : : I ~ : ~ /~ ,: ::: : : ::: - - r - r TT rnn---,-, ,L,-r' /T! nr - - r - r T T ro1

I I 1IIIIIt I I f 1/f /' 111 1 I I IIII I _ _ L _L11lIJU _ __ ' -/J_' .. ' ".J' __ _ L _'_1 1 LU'

I I I I I I III " I } I bI III I I I I I II I I I 1 111111 t 1 7t 111111 I I II III I

0.1 __ L _L~~LUU __ ~ _ J ~~ ~~ UL __ L_L ~ ~LU

0"

... : : : : : :::: /: ~J: I~ : : : : : :: : : :: :

- - ~ - I- + + ",-I H _l- -t - ,. - I I t-! -t H t- - - r- - ~ + + t'",.; I I 1 1111 11/ I /.1 PIII I I I I I I I1 1 1 I I 11 l 1 1 1j I II • 111 11 I I 11 111 1

--~ -: ~ ; :-:~f. --~T ~ :"ir:: ~: --:- -: ~ : ::-: I 0.2

o, ~o"

I I I I 1' 11 1 -'Ii 1' 1 I I I I I I I I I I I I I - - .. - t I ~ f II ii - - r l - J t - e-I-Ilti ,- - - r -.- II I1II

I I)"" 1 I I 1 -I l ei I I I 1I 1 I 1 I I I I I II I

,.' --Figure 2: Cumulative distribution functions of 8NR

18 Hufei Zhu, Zhongding Lei and Francois Chin

Figure 3: Cumulative distribution functions of channel capacity

D. R

. 10

., I.

10

. ' 1=_11' ~ W:::"'-lF 1-...... .,

I--- .,.., .... SE 04 -+- aRIC'llrllhlF

o ~SAIIC.e""E

o· I

-...: 1--:::::::'" I-

...........

I. _dOl "

Figure 4: Average BER

-J-

IS

AD-HOC NETWORKING

Market-based Network Formation for an Ad Hoc, P2P Wireless Network

YasUllori Yamamoto· School of Information Studies, Syracuse University Ministry of Public Management, Home Affairs, Posts and Telecommunications 5495-1-502 Ooyaguchi, Saitama 996-0934 JAPAN yayamamo@syr.edu

Junseok Hwangt School of Information Studies, Syracuse University 4-291, Center for Science and Technology Syracuse, NY 13244-4100 USA jshwang@syr.edu

Abstract In an ad hoc, arbitrarily formed peer to peer (P2P) network system, each user can select one of the services offered by multiple neighbor stations to have an access to the network. Every user has a preference function to determine a service to use, which is described as a total willingness to pay based on the bandwidth and the duration of using a service. In our study, we developed a market-based model for a user to determine a service in order to maximize the user's surplus. Our model allows a station to connect another station without reconnect ion as long as it wants or possible. By employing our model, stations of providing a service can offer a competitive pricing based on durations of connections such as a discount for their users to use it longer. The pricing of this kind is also preferable for both users and providers. In addition, based on our model, we developed an algorithm for a station to determine a way of making a connection to the network. We simulated some scenarios of ad hoc P2P wireless networks by using the algorithm. We found that the formed network is more efficient than former ceaselessly reconnecting networks in terms of the connectivity to get a certain QoS.

Keywords: Market-based networks, P2P networking, Wireless Communications

·Phone:+81(48)874-7233, Fax:+l(702)921-3399 tPhone:+l(315)443-4473, Fax:+l(315)443-5806

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

22 Yasunori Yamamoto

1. Introduction Due to rapid development of the wireless technology, demands on

high-speed personal wireless data communications has increased (In­ternational Telecommunication Union, 2000). Once in the range of propagation, individual users can access the Internet without any ca­ble wherever they are, whether they move or not. In this situation, decentralized ad hoc P2P networks are more suitable than centralized static network systems. Those networks allow every station to become a network provider for other stations and act as a repeater, so that a user out of the range of a base station could access the network via another station. We assumed the following scenario. In an ad hoc, arbitrarily formed P2P network system, each user selects one of the services offered by multiple neighbor stations to have an access to the network. Every user has a preference function to determine a service to use, which is described as a total willingness to pay based on the bandwidth and the duration of using a service. In our study, we developed a market-based model for a user to determine a service in order to maximize the user's surplus. In our model, a station can connect with another without a reconnection as long as it wants or possible. This feature allows stations to provide a competitive pricing such as a discount plan for their users to connect with them longer. In order to realize it, users are previously informed of costs according to the duration of a connection by provid­ing stations. The pricing scheme of this kind brings benefits to both of users and providers, since many users prefer a fixed rate service to that of usage-based charging (Ohu, 1999), while providers can get an optimal pricing to make their profits higher. In spite of many studies of optimal pricing, few of them take consideration of this users' preference.

In addition, based on the market-based model, we developed an algo­rithm of determining a way of connecting with the network. We simu­lated some scenarios of ad hoc P2P wireless network formations by using the algorithm. As a result, we found that an ad hoc P2P wireless net­work can be formed autonomously by using the algorithm. The formed network showed better performance than ceaselessly reconnecting net­works in terms of the connectivity.

2. Related Studies Some studies proposed a bidding method to allocate resource (i.e., a

seller allocates resource to those who value it most) and used a game theory to show an optimal condition under which users and a seller maximize their utilities. Users bid every certain period of time to be allocated. A recent study (Lazar and Semret, 1998) introduced the Pro-

Market-based Network Formation/or an Ad Hoc, ... 23

gressive Second Price (PSP) mechanism, which is derived from Vickrey Auction (second price sealed bid) and generalizes it. Resource is allo­cated to bidders according to the order of the bidding prices until no more resource is left. This mechanism is unprecedented in that it ap­plies to a generic arbitrarily divisible and additive resource model such as spectrum allocation. Their approach does not assume any specific mapping of resource allocation to quality of service. However, bidders are defined as having an explicit monetary valuation of quantities of resource which auctioneer does not or cannot know a priori.

Another approach (Altmann et aI., 2002) used a Markov model. In the model, an administrator offers several types of services at different priority levels, and a user decides one of them to use based on his or her job's priority and offered prices. A price at a priority level does not change but an actual QoS at the level varies depending on circumstances. Accordingly, users can see a QoS and its price of a service when deciding. Since this approach is job oriented, however, users cannot estimate total costs for their connections.

3. Assumption

We consider two main types of network topology when we develop a pricing model: Star and P2P. Star topology is that there is an access point used by multiple users simultaneously. Current cellular system is in this topology. P2P topology is that there are multiple stations and a station can be an access point to the others. Needless to say, this topology includes Star topology. An advantage of this can be shown when there is a station out of the range of an access point, but another station within the range can become a proxy to that station.

Since we assume that any user can also be a network provider to another, we do not care about a network topology but focus on behaviors of users and providers. Users prefer fixed rate services while providers seek maximum profits. We seek a way of providing a condition under which both players can get satisfied. A provider offers several types of services in terms of duration of service and bandwidth (QoS). A user chooses a service offered by neighbor providers according to his or her preference.

We assume a CDMA system to which our model applies, where users are assigned orthogonal spreading codes. In this system, a transmitted power and a length of codes determine a QoS. As in a study (Liu et al., 2000), we assume all codes have same length, and therefore a QoS is determined by a transmitted power. Let #Li(O < #Li < 1), di(1 ~ di), and P; represent the channel gain of user i, the distance between i and

24 Yasunori Yamamoto

the transmitter, and the allocated power of service j for user i at the distance of the one unit length from the transmitter, respectively. It should be noted that we ignore such propagation factors as shadowing and multipath. Since received power can be determined by the channel gain and the distance from the transmitter di and the path loss expo­nent K., J.LidiK. Pj is the average received power at user i with service j. In addition, an average transmission rate at user i is determined by the re­ceived power, which is Sji = (,diK. Pj{(i oc 1-',). Assuming that a provider sends data to user i with an average transmission rate Sj at the point of the provider, once a QoS of service j and the distance d is determined, its transmitted power can be settled. In this situation, a provider offers a user its service plans which consist of pricing and discount schedules. Those pricings are based on bandwidths (i.e., Sj) and service durations. Then, the user decides to use one of them, or otherwise gives up.

4. Model Utility function. Utility function Ui which determines the surplus for a user i E {I", ',I} is:

(1)

while tWi is a user i's total willingness to pay, Cj is the cost to use a service j E 1", " N, and v E [0, T} is the duration of service. A service provider does not determine a service price depending on bandwidth but on output power to meet designated bandwidth at the transmitter.

Let b E [0,00) represent the bandwidth. Since we assume that both of user i's preferences (i.e., service duration and bandwidth) need to be satisfied when accepting an offer, total willingness to pay can be:

(2)

where a is i's budget, tWiv(V) E [0,1] and tWib(b) E [0,1] are the nor­malized i's total willingness to pay functions of service duration and of bandwidth, respectively. The reason why we introduce the factor of the service duration is that the individual users prefer fixed rate services, and in other words, pricing plans for a certain bandwidth or power can be described as the service duration and a discount rate at the duration time. Therefore,

f HV) Ui = tWi{V, (diK.Pj) - cj[kj(v), t q,j{r) dr}, (3)

where kj{v) E [0, I} is the discount rate of making a service j with a duration v, and q,j{t) is the price of making a service j connection for one unit of time at time t.

Market-based Network Formation/or an Ad Hoc, ... 25

Pricing Scenario. As in a study (Wang et al., 1996), the ser­vice duration is exponentially distributed with mean 1/Tj for any ser­vice j, where Tj is the rate of service happening. We also assume that a provider gives a pricing and a discount schedule represented as [4>1 (t), 4>2(t), ... , 4>N(t») and [kl (d), k2(d) , kN(d»), respectively. Assum­ing that the exponentially distributed service duration is not affected by 4>j(t), users starting a service j at time t are expected to pay an average price of:

1+00 [1(T-t) 1 pj(t) = t Tje-rj(T-t). t 4>j(v)dv· (1- kj(r - t)) dr. (4)

Let Pb(t) be a spot price determined by how scarce the spectrum resource is at time t. Pb(t) would become higher if, for example, the propaga­tion condition is worse or number of users wanting to use it increases. Since 4>j(t) can be uniquely determined by Pb(t) if the discount rates are given, we assume that providers manipulate [Pl(t),P2(t),··· ,PN(t») to maximize their utility.

5. Case Study

5.1. Station's Behavior (With vs. Without a Discount)

As an example of comparison, we took three service plans 4>i(i = 1,2, and 3) with and without a discount plan kll. In addition, we took a user's preference tWI for those services as an example, which can be described as a Gaussian function. In this situation, the user's surplus U(t) can be described as follows:

U(t) = tWI(t) - kl(t) lot 4>i(V) dv (i = 1,2,3). (5)

Therefore, in this example, the duration of time to make the user's surplus optimal can be found by solving the following equation:

dtwl(t) dr· dt = dtkl(t) 10 4>i(V) dv (~= 1,2,3). (6)

s.t.

(7)

Figure 1 shows how these three pricing plans require a user to pay ac­cording to the duration of connection time, and a user's preference.

1 kl (t) = 1 in case of no discount.

26 Yasunori Yamamoto

45.0 price

40.5 +++;~l WithoUldlJcouO xxx:~2

36.0 ··· :~3

31.5

27.0

22.5

18.0

13.5

9.0

4.5 duration

0.0 0 4 8 12 16 20 24 28 32 36 40

Figure 1. Comparison of services, and a user's preference

5.2. Network Formation

Algorithm. In order to make a network formation autonomic, we developed an algorithm for a station to determine a way of establishing a connection to another station or giving up. Table 1 shows the algorithm, which employs our model discussed in this study.

Table 1. Algorithm to make a connection

P = {},Q= U; For i = 1 To the end of existing stations nearby

Calculate the distance di; For j = 1 To the end of services which station i offers

Q = QU{ QoS of Service Sij}; P = Pu{Pricing Schedule of Service Sij};

End; End; Find a service Si* j* :surplus sp( Si* j*) > sp( Sij) (i* I=- i, j* I=- j); If Si* j* < 0 Then

Give up making a connection; Else Make a connection to get the service Si* j* ;

End;

Market-based Network Formation for an Ad Hoc, ... 27

Discrete vs. Continuative Connection. In order to evaluate our model, we simulated ad hoc P2P network formations in two situations. One set of simulations employed our algorithm which enables a station to have a continuative connection to another over a period of time. The other scenario also employed our algorithm but continuative connections were disabled. All conditions were identical except for the difference in the continuation. In our simulation, 360 stations were generated ac­cording to Poisson distribution. Life spans of them also followed the distribution. Collected data from each simulation include: a total times to complete2, average numbers of active3 and inactive4 stations, and an average depthS of formed networks. Figure 2 and 3 show the difference between continuative and discrete connections. Both figures show the

lS

U ~lO

4 g ~

1.I£ ~l:! •

I i i~ '0

u_ .. II '0 1: 2 ..

E ! i lO IJ§

z ,; < I

I ~

1I

~~--------~~~----~I 0.101 tOOl O.ooJ 1.0IIII tOOl 0.11 ' .011 lOl2 D.I1IJ OJI4 tOil

Reciprocal T alai T me

Figure 2. Discrete Connections

)l

4J .lO

4 l j !ll IJ~ (/)

i~ I i '0

u_ .. II '0 ! :J E i" InacIlve SCaIIons

:>

,; z < , . I

~ " • . I. ':r - 1I

~~----~----~~~--~. 0,001 tOOl lOll! tOOl 0.009 D.I1I till 0.012 tll.l tOl4 D.I1IS

RecIprocal T eta Time

Figure 9. Continuative Connections

Average Number of Active and Inactive Stations

average numbers of active and inactive stations to the reciprocal total times to complete a simulation. As for the discrete connections, the average numbers of active stations are from 10.6 to 22.1 while those of inactive ones are from 2.06 to 4.40. On the other hand, concerning the continuative connections, those of active ones are from 18.6 to 32.5 and inactive ones from 0.50 to 1.12. This result indicates that more stations can have connections when they are continuative than discrete. In other words, the connectivity of an ad hoc P2P network improves when con-

2from the time of the first station comes up to the time of the last one disappears 3station having a connection 4station not having a connection 5number of connections from the root to a station (e.g., a star topology network has 1.00 of its average depth.)

28 Yasunori Yamamoto

tinuative connections are allowed. It should be noted that the average depths of both situations are almost identical (1.73 and 1.84).

6. Conclusion

In our study, we introduce a discount factor, and therefore fixed rate services which many individual users' preference can be described in our pricing model. Each user is previously shown with service plans which comprise a pricing and a discount schedule by multiple service providers. After that, the user can determine a service plan and a service provider to use. Besides, we do not assume that each service provider previously knows the users' preferences, but instead assume that their durations of time to use networks are exponentially distributed. In this assumption, we get an optimal pricing model.

In addition, we developed an algorithm for each station to determine how to make a connection to another station based on the surplus for it. Using the algorithm, we made simulations and confirmed that an ad hoc P2P wireless network can be formed autonomously.

In addition, our simulations showed that connectivity of formed net­works improves by introducing continuative connections.

References Altmann, J., Daanen, H., Oliver, H., and Suarez, A. S.-B. (2002). How to Market­

Manage a QoS network. In IEEE InfoCom 2002, Conference on Computer Com­munications, New York, USA.

Chu, K. (1999). User reactions to flat rate options under time charges with differen­tiated quality of access: Preliminary results from index.

International Telecommunication Union (2000). Main results ofWRC-2oo0. Retrieved September 26, 2001, from: http://www.itu.int/brconf/wrc-20oo/docs/index.html.

Lazar, A. A. and Semret, N. (1998). Design and analysis of the progressive second price auction for network bandwidth sharing. The 8th International Symposium on Dynamic Games and Applications.

Liu, P., Honig, M. L., and Jordan, S. (2000). Forward-link CDMA resource allocation based on pricing. In IEEE Wireless Communications and Networking Conference, Chicago, IL.

Wang, Q., Sirbu, M. A., and Peha, J. M. (1996). Telecommunications and Inter­net Policy, chapter Pricing of ATM Network Services. Lawrence Erlbaum Assoc, Washington, DC.

An Efficient Proactive Routing Method for Mobile Ad Hoc Networks using Peer-to-Peer and Cellular Communication System

Hiroaki Morinot, Tadao 8aitot:j: and Mitsuo Nohara:j:

t Research and Development Initiative Chuo University :j:TOYOTA InfoTechnology Center Co., Ltd Address: 42-8 Hommuracho Shinjuku-ku Tokyo 162-8479, Japan Tel: +81-9-5968-9571 Fax: +81-9-5968-9515 morino@tamacc.chuo-u.ac.jp

Abstract Information systems of automobiles in the future will have several broad­band wireless access systems to the Internet. Among them, DSRC (Ded­icated Short Range Communications)[lj is attractive as broadband mo­bile access system, though coverage will be limited. To extend coverage in effect from viewpoint of users, system can be effective in which termi­nals share data received from the Internet with other terminals located out of coverage using inter-vehicle communication. In this case, rout­ing protocol of multi-hop inter-vehicle communication network should achieve high throughput with low control overhead when there are many terminals and large amount of data traffic are offered to the network. However, conventional routing protocols using only peer-to-peer com­munication are difficult to meet these requirements.

This paper presents a cellular-assisted proactive routing protocol for multi-hop inter-vehicle communication network. In the method, each terminal is equipped with both ad hoc mode wireless LAN and cellu­lar system capability. It sends its link state packets to a control node provided in a fixed network using cellular system, and the control node sends link state packets to all terminals using multicast function of cel­lular system to reduce number of packets transmitted. Simulation us­ing straight road model confirmed that number of transmitted control packets in the proposed method reduces compared with conventional on-demand routing protocol, and amount of reduction increases accord­ing to increase of demands for data connections arise in the network.

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

30 Hiroaki Morino, Tadao Saito and Mitsuo Nohara

Keywords: ITS, In-vehicle terminal, inter-vehicle communication, multi-hop net­work, cellular system, link state routing protocol

1. Introduction With the emergence of broadband wireless systems including IMT-

2000, IEEE 802.11 based wireless LAN, DSRC (Dedicated Short Range Communication), capacity of wireless access systems from automobiles to the Internet will increase from conventional tens of kilo bit/sec to mega bit/sec. Among them, hot-spotted network by wireless LAN or DSRC is attractive in that broad bandwidth is available with lower cost than other access systems, although coverage is limited. In order to ex­tend coverage of hot-spotted network in effect from viewpoint of users, a system can be effective in which terminals receive data from the Internet from these systems and they share received data with other terminals located out of coverage using inter-vehicle communication network. As a system for this purpose, authors examine file transfer system using DSRC and multi-hop wireless LAN based inter-vehicle communication network. In the system number of cars participating in multi-hop inter­vehicle communication network can reach to order of hundred. Trans­port protocol and routing protocol in the network is required to achieve high throughput with low control overhead. In addition, it should scale to the network with large number of terminals. However, conventional routing protocols of mobile multihop network using only peer-to-peer communication are difficult to meet these requirements.

This paper presents a novel proactive routing protocol called base­station-assisted link state routing protocol, which is suitable for large scale multi-hop inter-vehicle communication network. In this method, each terminal is equipped with wireless LAN and cellular system. Termi­nals send link state packets to the control node in a fixed network using cellular system, and the control node sends them to all terminals using multicast function of cellular system to reduce number of transmitted control packets.

The rest of the paper is organized as follows. In section 2, the concept of file transfer system using multi-hop inter-vehicle communication net­work is described, and requirements of routing protocol of the system are presented. In section 3, operation of the proposed method is described. Section 4 presents results of performance evaluation by simulation and finally section 5 concludes the paper.

An Efficient Proactive Routing Method/or ...

2. Background of research

2.1. File transfer systems using inter-vehicle communication

31

The concept of the file transfer system using OSRe and multi-hop inter-vehicle communication network examined in this paper is illus­trated in Figure 1. From OSRe, cars can receive files containing traffic, map, entertainment information and so on. In this case a car (a) is ap­proaching OSRe, while other car (b) is out of coverage of OSRe. When (a) receives some files via DSRC, it becomes a file sender terminal, and it broadcasts a packet to advertise index of received files to other terminal. If (b) wants advertised files, it issues request to (a) and receives the files from (a) by unicast transport connection. In the rest ofthis paper, (b) is called file receiver terminal. Since broadcast traffic of file advertisement increases according to increase of file sender terminals, the system has to be designed so that broadcast traffic does not affect performance of file transfer in case there are many file sender terminals.

2.2. Related works There have been many studies of mobile multi-hop network routing

protocols[2, 3, 4, 5, 6], and they are mainly discussed in MANET(Mobile Ad-hoc NETworks) Working Group of IETF (Internet Engineering Task Force). Among them, ODMRP(On-Demand Multicast Routing Proto­col) [4] can work as a routing protocol of the file transfer system described in Section 2.1. In OOMRP, a source terminal with data to send broad­casts JOIN QUERY packets by flooding periodically to other terminals to make multicast group. Terminals that want to join the group send JOIN REPLY packet to the source terminal, and routes between them are constructed. Though ODMRP is simple and efficient protocol, it is in principle suitable to the network with a few source terminals and many multicast receiver terminals. For example, in case that many source ter­minals can co-exist and each wants to send data to other terminal as described in section 2.1, overhead of ODMRP caused by broadcast of JOIN QUERY packets may not be acceptable.

3. Base-station-assisted link state routing protocol

To resolve problems conventional routing protocols described in Sec­tion 2.2, this paper presents base-stat ion-assisted link state routing (BALSR protocol.

32

DSRC t..lcon

Hiroaki Morino, Tadao Saito and Mitsuo Nohara

(1) CI' CI) downlolCl nlelvllDSRC

Figure 1. File transfer system using Figure 2. Overview of the proposed inter vehicle communication system

3.1. Protocol operation

It is based on link state routing protocol such as OSPF[7]. Figure 2 shows overview of the system. It is assumed that each mobile terminal is equipped with ad hoc mode wireless LAN and cellular system capability, and both of them support IP (Internet Protocol) . Packet types handled in the protocol are listed as follows.

• Hello packet Packet containing network address of the terminal itself. It is exchanged by wireless LAN.

• Link state packet Packet containing list of network address of neighbor terminals in wireless LAN. It is exchanged by cellular system.

• File advertisement packet This packet is generated if a terminal is a file sender terminal ( c.f. Section 2.1) and contains indexes of files the terminal has. It is exchanged by cellular system.

• Data packet Packet containing user data. It is exchanged by wireless LAN.

Next, configuration of cellular system and fixed network is described. Coverage of a base station in the cellular system is assumed about two kilometers radius, and it supports multicast channel from base station to terminals. Requirements for multicast function in the cellular system are described in detail in Section 3.2. In the fixed network, control nodes called reflector nodes are provided, and they relay control packets

An Efficient Proactive Routing Method/or ... 33

exchanged between terminals. For terminals within coverage of one base station, one reHector node is assigned. Terminals within coverage area of each base station are informed of network address of the reHector node in advance.

Operation of the protocol is as follows. Each terminal periodically senses neighbor terminals in wireless LAN by exchanging hello pack­ets, and it generates a link state packet and it sends the packet to the reflector node. When a terminal becomes a file sender terminal, a file ad­vertisement packet and a link state packet are sent to the reflector node. When the reHector node receives a file advertisement packet or a link state packet from a terminal, it immediately sends to all terminals using cellular system. In this way, each terminal receives link state packets of all other terminals from the reflector node, and determines routes to other terminals on the multi-hop inter-vehicle network using dijkstra's algorithm. Furthermore, a terminal knows other terminals that received files from DSRe by receiving file advertisement packets.

3.2. Required functions of cellular system The proposed method is designed to send link state packets of multi­

hop network using multicast function of cellular system. To make the proposed method feasible, cellular system is required to support mul­ticast function at data link layer of downlink channel. Among packet switching method in the present cellular systems, GPRS (General Packet Radio Service) supports some of control functions of downlink channel for multicast in PTM (Point-To-Multipoint) service [8]. This can be one of important functions to realize the proposed method. Control func­tion of PTM is presently only specified in the radio interface in [8], and design of upper layer protocols are need to be further investigated.

4. Performance evaluation In this section, performance of the proposed method is evaluated by

computer simulation. In the following part, a model of file transfer using inter vehicle communication is defined at first and results of evaluation are shown.

Evaluation model. The model is shown in Figure 3. In this model, road is assumed to have three lanes (a),(b),(c), and DSRC is only avail­able at a spot area of lane (a). Among terminals moving on lane (a), predetermined ratio of terminals receives files via DSRC. Each of ter­minals that received files become file sender terminal, and it selects the furthest terminal from itself in multi-hop network as a file receiver termi-

34 Hiroaki Morino, Tadao Saito and Mitsuo Nohara

nal. Connection duration time for which one file sender terminal sends packets to one file receiver terminal is assumed to be 20 second. During connection duration time, file sender terminal sends data packet with rate of 50 packet/sec. When connection duration time ended, the file sender terminal selects another file receiver terminal and begins to send­ing packets

In each lane, a terminal enters from the left hand side and moves to the right hand side at constant speed. Velocity of each terminal is determined when it enters the lane, and it varies between 1l.5m/s (about 40km/h) and 16 m/s (about 60km/h). Average time interval at which a terminal enters a lane is 3.7 second, and thus average distance between newly entered terminal and the precedent terminal is 50m.

Parameters of ad hoc mode wireless LAN and cellular system are shown in Table 1.

Routing protocol to be evaluated. Based on the evaluation model, two protocols of BALSR and ODMRP are evaluated. Param­eters of each protocol are as follows. In the evaluation of BALSR, time interval of hello packet and link state packet is 2.5 sec and 7.5 sec respec­tively. Link state packet is also sent every 2.5 sec in case that topology is changed within this time interval.

In the evaluation of ODMRP, each file sender terminal sends JOIN QUERY packet by broadcast, and every terminal receiving JOIN QUERY packet sends JOIN REPLY packet to the file sender terminal. Among these terminals the file sender terminal selects the furthest terminal as a file receiver terminal and begins to send data packets. When connec­tion duration time ends, a file sender terminal broadcasts JOIN QUERY packet again and selects another file receiver terminal in the same way. Though ODMRP is mainly designed for multicast communication, per­formance enhancement for unicast communication is also studied[9], and operation of ODMRP in the evaluation follows description in [9].

Results. Figure 4 shows ratio of data packets successfully delivered to file receiver terminals. Ratio of file sender terminals means number of file sender terminals over number of all terminals on the road. BALSR and ODMRP can achieve data packet delivery ratio of almost 100% regardless of number of file sender terminal. Next, figure 5 shows number of control packets transmitted by all terminals and a reflector node to construct and maintain routes in BALSR and ODMRP. In results of BALSR, control packets include hello packets and link state packets. In results of ODMRP, control packets include only JOIN QUERY packets. As shown in the figure, BALSR improves control overhead compared

An Efficient Proactive Routing Method/or ... 35

with ODMRP in case that ratio of file sender terminals is more than 15% of all terminals. Amount of reduction of control packet increases according to increase of file sender terminals, and it is 50% at maximum.

These results show that the proposed method achieves the same data packet delivery performance as ODMRP, and reduces control overhead. This feature is effective in case that larger amount of data traffic is offered to the network.

5. Conclusion In this paper, routing protocol for wireless LAN based inter-vehicle

communication network is proposed where cellular system is used to ex­change control packets. Results of performance evaluation in the straight road model show that the proposed method reduces number of trans­mitted control packets compared with conventional ODMRP when ratio of terminals with data to send is more than 15 % among all terminals. It is also shown that amount of reduction of control packets reaches 50% at maximum. This feature will be effective in the environment of inter-vehicle communication in the future to handle large number of data connections. For future work, operation parameters of reflector node will be further evaluated by simulation.

Acknowledgments The authors wish to thank Prof. Shigehiko Naoe (Chuo University)

and Mr. Tadao Mitsuda (TOYOTA InfoTechnology Center Co., Ltd) for their valuable comments.

References

[1] ARIB Standard STD-T75 Ver 1.0, "Dedicated short-range communication sys­tem" Sep 2001.

[2] Charles Perkins et al., "Ad hoc On-demand Distance Vector (AODV) Routing," Internet-Draft <draft-ietf-manet-aodv-10.txt> Jan 2002.

[3] David B. Johnson et al., "The Dynamic Source Routing Protocol for Mobile Ad Hoc Networks (DSR)", Internet-Draft <draft-ietf-manet-dsr-07.txt> Feb 2002.

[4] Sung-Ju Lee et aI., "On-Demand Multicast Routing Protocol", Proc. of IEEE WCNC '99. pp.1298-1302, Sep 1999.

[5] Mario Gerla et aI., "Fisheye State Routing Protocol (FSR) for Ad Hoc Net­works," Internet-Draft <draft-ietf-manet-fsr-03.txt> Jun 2002.

[6] Amir Qayyum et aI., "Multipoint relaying: An efficient technique for flooding in mobile wireless networks," INRIA Technical Report RR n03898, Mar 2000.

[7] J. Moy, "OSPF Version 2", RFC 2328 Apr 1998.

36 Hiroaki Morino, Tadao Saito and Mitsuo Nohara

[8] ETSI TS 101 350: "Digital cellular telecommunication system (Phase 2+); Gen­eral Packet Radio Service (GPRS); Overall description of the GPRS radio in­terface; Stage 2" , Feb 2002.

[9] Sang Ho Bae et al., "Unicast Performance Analysis of the ODMRP in a Mobile Ad hoc Network Testbed.," Proc. of IEEE ICCCN 2000. pp.148-153, Oct 2000.

bmN" ..... DBRe

Figure 9. Evaluation model of the file transfer system

Table 1. Parameters of wireless communication systems

Type of communication system TI-ansmission speed Radius of radio range

Ad hoc mode wireless LAN Cellular system

0.98 ...., .,g ~ e 0.96 -0- BALSR

D- ODMRP ~~ !I .~ 0.94 1---------1 ~~ L-___ -\

0.92 1-------------1

0.9 '--__ -'-___ L.-__ -'

o 0.1 0.2 0.3 Ratio of file sender tenninals

Figure .4. Performance of data packet delivery ratio

11Mbps 100kbps

..., - I)

r~ 8~] 'E.s~ '-:l!l ~I)~ E"'" <> ~ t;

60

50

40

30

20

10

0 0

150m 2000m

0.1 0 .2

Ratio of file sender terminals

Figure 5. Characteristics of number of transmitted control packets

0.3

A Mobile Multicast Framework for CDMA-based Ad Hoc Networks

Hsu-Y ang Kung and Su-Man Chen Department of Management Information Systems National Pingtung University of Science and Technology, Pingtung, Taiwan 91207

Abstract: 'The objective of this paper is (i) to construct a robust ad hoc mobile network using the Direct Sequence CDMA technology, (ii) to self-create the transmission schedules for downlink and uplink communications, and then (iii) to multicast data packets with the collision-free transmission and the hidden-terminal avoidance. In this paper, we proposed the Hierarchical Linked Cluster Multicast (HLCM) network architecture and the HLCM Network Formation algorithm based on the characteristic of the DS-CDMA communication to construct a robust ad hoc multicast network. To achieve the collision-free and hidden-terminal avoidance multicast communications, the Uplink Multicast Schedule Algorithm (UMSA) and the Downlink Multicast Schedule Algorithm (DMSA) are proposed to generate the intra-cluster and inter-cluster transmission schedules, respectively. Each mobile node multicasts data packets to the destination nodes according to the pre-assigned time slots of the transmission schedules. 'The corresponding simulation results show that the proposed algorithms and control schemes effectively solve the hidden­terminal problem and achieve the self-organization and self-operation of a CDMA-based ad hoc multicast network.

Key words: Ad Hoc Network, CDMA, Hierarchical Linked Cluster, Hidden Terminal Problem, Multicast Communications.

1. INTRODUCTION

As the rapid progress of broadband wireless/mobile network technology, the state-of-the-art mobile multimedia applications, such as the Mobile Video Conferencing and Mobile Telemedicine, are eagerly required and developed [1, 2, 11]. For such popular mobile multimedia applications, many of them are based on an ad hoc network. There are two essential technology issues, which are (i) the creation and operation of an ad hoc network among the mobile hosts/devices to quickly adapt the mobility and

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

38 Hsu-Yang Kung and Su-Man Chen

(ii) the provision of the multicast communications to effectively reduce the consumption of the wireless network bandwidth [3, 4, 6]. Ad hoc wireless networks consist of mobile hosts and dynamically create a communication network among them without using any infrastructure or administrative devices [14]. In contrast to conventional wireless networks, which require the provisions of the prerequisite network infrastructure and centralized administration devices for mobile communications, ad hoc wireless networks can adapt the dynamic network topology to the mobile hosts and are self-creation, self-organization, and self-administration with the collision-free and hidden-terminal avoidance [3, 7, 12, 13, 15]. The "hidden­terminal" problem in an ad hoc network is that two mobile nodes simultaneously transmit messages to the third node and the transmission collision occurs. The third node is the "hidden-terminal".

The Third-Generation (30) wireless network, which provides higher bandwidth and could achieve mobile multimedia applications, is the key revolution of the broadband wireless/mobile network [5, 8]. The air interface of 30 focuses on the Code Division Multiple Access (CDMA) technology, which supports a high transmission rate and has the characteristics, including the anti-jamming, high capacity, and security, is very suitable for multimedia communications in ad hoc networks [6, 9, 10]. For example, each mobile device of a Mobile Video Conferencing, which is a CDMA-based communication system, can dynamically join or leave the conference and achieve perceivable presentation qualities with feasible communication schemes. In such kinds of mobile multimedia applications, it is critically required to rapidly construct an ad hoc mobile network with a robust topology, and distribute the data packets with collision-free multicasting [14, 15, 16]. There are few researches on the self-organization and self-operation of a feasible CDMA-based ad hoc network with the provision of multicast communications. To achieve the characteristics of ad hoc multicast networks, the state-of-the-art organization and communication schemes on the CDMA air interface are urgently required. In this paper, we proposed (i) the Hierarchical Linked Cluster Multicast (HLCM) network architecture, which is a robust ad hoc mobile network using the Direct Sequence CDMA (DS-CDMA) technology, and (ii) the transmission schedules with self­creation for downlink and uplink multicast communications with the collision-free transmission and the hidden-terminal avoidance.

The organization of the rest of this paper is as follows. In Section 2, we introduce the HLCM network infrastructure. Section 3 describes the proposed construction and multicast schedule algorithm. Section 4 describes the simulation results of the proposed algorithms. Section 5 presents our conclusion.

A Mobile Multicast Framework/or CDMA-based ...

2. IDERARCmCAL LINKED CLUSTER MULTICAST NETWORK INFRASTRUCTURE

39

The proposed Hierarchical Linked Cluster Multicast (HLCM) network architecture is based on the characteristic of the DS-CDMA communication. As depicting in Figure 1, the whole COMA frequency spectrum is divided into M frequency sub-bands, each of which conducts a communication range and forms a specific network topology. Each sub-band is composed of two communication periods, which are (i) the epoch period and (ii) the data communication period. An epoch period is the time period of creating and organizing an ad hoc mobile network. During each sub-band, the epoch period repeats periodically to ensure an up-to-date linkage of the network. A data communication period is the time period that all mobile hosts multicast data packets according to the transmission schedules for the specific HLCM ad hoc network.

HFBAND DIVIDED

INTO M

SUBBANDS

(SUBBANDM)

lOMIh Tanc

_ Network reorganization

c::::J Data communication time

_ Subband of HF (2 to 30 MHz) band

Figure 1. The frequency, time, and code division structure for the organization and communication of the HLCM network.

Figure 2 depicts a HLCM network topology, which is composed of hierarchical linked clusters. Mobile nodes in a HLCM cluster are defmed as follows. (1) Cluster Head. Each cluster has only a cluster head, which is the based

station (BS) in a cluster. For example, the nodes H], H3, H4, and H9 are the cluster heads of clusters C], C3, C4, and C9, respectively.

(2) Gateway Node. It is an interconnection node, which is responsible for forwarding data packets between/among clusters. To construct a robust ad hoc network, three kinds of gateway nodes are specifically defined. (i) The Overlapping Gateway. It is located in the overlapping area between clusters. (ii) The Non-overlapping Gateway. H two clusters don't overlap and have at least a pair of nodes, each of which is located

40 Hsu-Yang Kung and Su-Man Chen

in the corresponding cluster and can communicate with each other. These two nodes are the non-overlapping gateways. (iii) The Duality Gateway. A node not only is an overlapping gateway but also is a non­overlapping gateway.

(3) Ordinary Node. An ordinary node is neither the cluster head nor the gateway.

3. SYSTEM IMPLEMENTATION

In this Section, we proposed three algorithms to achieve the self­construction of the HLCM network and the uplink/downlink multicast communications with collision-avoidance. To achieve the network self­organization and self-linkage, we proposed an HLCM Network Construction (HLCM-NC) algorithm, which operates in the epoch period, to rapidly construct a two-tier hierarchical ad hoc network. Figure 3 depicts the structure of the epoch period. The HLCM-NC algorithm divides each epoch into two frames and some time slots. Two frames F/ and F2 are further divided N time slots and N is the number of all nodes in the ad hoc network. Each node broadcasts the control message in its assigned time slot of the two frames according to the node 10. The transmission order in frame F/ is with the increased manner, i.e., from node NJ to node NN, and the

IRlJIIIJUlI2 IRlJI3 BO:HM IRlJII

.......... I I I I·· .. · .. ··· ......... · I I I ......... .

/~ I!!A>,£I IRM£2 ••..•.•.•.••.• r-+l

1--' ~ ~~ • Cluster Head l; Group G 1

H-l · .. · ....... 1 I ........... 1-+-1 !UI" !lilf !lilf !lilf 9.Or 9.Or sa 12 N-INN-I 1 I

Gateway Group Gz 0 Ordinary Node • Group G3

Figure 2. A HLCM network topology. Figure 3. The structure of the epoch period.

transmission order in frame F2 is with the decreased manner, i.e., from node NN to node N/.

The construction operations during frame F/ are as follows. (i) The node NH, which doesn't "hear" any node announce to be a cluster head

A Mobile Multicast Frameworkfor CDMA-based ... 41

before NH'S announcement, NH becomes a cluster head and broadcasts this message in its assigned time slot. (ii) A node Nit which "hears" the node NH announce to be a cluster head before its announcement, the node Ni becomes a member of the cluster CH and broadcasts its node ID. (iii) A node Ng,

which "hears" two or more than two nodes announce to be cluster heads before Ng's announcement, Ng becomes an overlapping gateway and broadcasts this message in its assigned time slot. (iv) A node NJO which "hears" a node NH announce to be a cluster head and the other node Ny announce to be a member of another cluster before Nx's announcement, N:x pairs for Ny and becomes the non-overlapping gateway. Nx broadcasts this message in its assigned time slot. (v) A node Nd, which already announced to be an overlapping gateway and then "hears" the other node Ny announce to be a member of another cluster, Nd becomes a duality gateway. (iv) If a node Hj joins a multicast group Gk, Nj announces its multicast group ID (Gk) in its broadcasting time slot.

After frame FJ of the epoch period, we have the following realization. (i) Each cluster head realizes its cluster members. (ii) Each overlapping gateway realizes its cluster head. (iii) Each non-overlapping gateway announces its cluster heads and the linked clusters. However, the cluster head doesn't agree yet. (iv) Each ordinary node realizes its cluster head. (v) Each cluster head realizes the multicast groups in its cluster.

The construction operations during frame F2 are as fellows. (i) Each cluster head announces its members, the multicast groups IDs, neighbouring cluster heads, and the gateways including the overlapping and non­overlapping gateways. (ii) Each overlapping gateway announces its cluster heads and the linked clusters. (iii) Each non-overlapping gateway announces its cluster head and the linked clusters after selecting by the cluster head. (iv) Each ordinary node announces its cluster head. (v) Each cluster head realizes the groups in the neighbouring clusters.

To achieve the collision-free and hidden-terminal avoidance multicast communication, we proposed the Uplink Multicast Schedule Algorithm (UMSA) and the Downlink Multicast Schedule Algorithm (DMSA), which construct the Intra-Cluster and the Inter-Cluster transmission schedules, respectively. To achieve message routing for building transmission schedules, there are four kinds of cluster heads are specified: the Start Head (SH), the Mediate Head (MH), the End Head (EH), and the Isolated Head (IH). Between/among the neighbouring clusters, the smallest (largest) ID cluster heads are the SHs (EHs), and the other cluster heads are MH. The IRs have no neighboring cluster head. Based on the transmission of SHs, MRs, and ERs, the multicast schedules for uplink and downlink communications are self-created.

Based on the UMSA and DMSA, each cluster head creates the uplink and downlink multicast schedules and broadcasts the schedules to all nodes during the epoch period. Using the uplink multicast schedule, each node of

42 Hsu-Yang Kung and Su-Man Chen

the cluster can transmit data packets to its head in the pre-assigned time slots with collision-avoidance during the data communication period. Using the downlink multicast schedule, each cluster head can multicast data packets to all members of a multicast group between clusters via gateway nodes. Based on the transmission schedules, the hidden-terminal problem is effectively solved.

4. SIMULATION RESULTS

,~"-(0) (b)

(c) (d)

'~' ".':' 14 1 .

1 2 17 iI 1 3

, J a

(e) (I)

Figure 4. An example of the connection topologies based on six different radio frequencies.

(0) (b)

Billl ~'vlllli¥l ESiili I M;:;""1' "I

(e) (I)

Figure 5. The corresponding HLCM network topologies based on Figure 4.

As depicting in Figure 4, the simulation models are composed of 20 mobile hosts, which form different connection topologies and have different mobility capabilities based on six different radio frequencies. Each mobile host is initially positioned randomly and free to move. By applying the proposed HLCM Network Construction algorithm, a robust ad hoc network is self-created and self -organized among mobile hosts. The corresponding HLCM networks based on different connection topologies are illustrated in Figure 5.

Tables 1 and 2 show the uplink and downlink multicast schedules corresponding to the examples of Figure 5 based on the UMSA and DMSA

A Mobile Multicast Framework/or CDMA-based .. . 43

schemes. In the uplink multicast schedule, the "*,, represents the start head, the "#" represents the end head, the "+" represents the isolated head, and the mediate node is represented with blank. Each cluster member is assigned unique time slots for transmitting data packets to its head to ensure collision­free and hidden-terminal avoidance. In the downlink multicast schedule, each cluster head assigns dedicated time slots for each of its gateways to avoid collision. The cycle length of the transmission schedule is the sum of the time slots that the related cluster heads arrange for a specific gateway.

Table 1. The uplink multicast schedule corresponding to Figure 5.

Table 2. The downlink multicast schedule corresponding to Figure 5.

--HJD£ LOr 1 l J .. , , , , , ICI II 11 IJ ~IIJIN -w - I J"!"., IOU W., 11,1 J .. J .,

1 '10 II W " n tt» , to 11 .. II 11 1'»

IODT

, 1 , 1\ 11 1) lJ n II .. ]. ., , 11 11 U --tolJDB. 1 1 ] .. " , , • to )II II 12 D ~ IS II

-'" -, 41'''''0 U un.. J • , lit "" U I? :a t 111 I' Hi IT \J '10 W J'" It:ll , 10

, , ., I , 11 l) U .. , , T I • 11 I)

tIl ''7'' nun 11,9 I 1 , 11 D '1 tl tt

.... -~1UIr 12: , .. j. • , • , 10 1I11lJ 1-1 U ..

-...,., ."II014Un.Jll..,,,,,un 1 ~"~M"~ ~"W~"WZ~" n ".,., ttU u .. J , ., I , 11 I' --r<OiaJ'.1J 11: , .. J , 1 •

_~ .""IOWU4,.to'4"4,,1O ww."ZW'lto"_U ,I"JD"""1D • 7 \I 11 U. 1$.1. 1 II II U ., I •• 1 , 1 10 II 17 II , 7 10 II 17 II , J \10 It --P'CJOISam 1 2: , .. , , f •• 10 1111 u .. lui"

-~~ . , .. ".,. to 1<1 lJ .. J , • 10 ... lJ .. J 1 U~~ZW~~ZWM~Z~N~Z

., .. , l' • liD II 11 U ".1" , , • 1011

.... -fCOIi SU1r I 1 , .. , • 1 I , 10 \I 11 u "lui"

-<') -, .. " 1 I 110 l' lJ 1t • , 7 I 10 W U It

Q M~~~"Z"~ZM~ZN"~K

110 "S' J • ,u u U 171& " J , J

5. CONCLUSION

-~: I

1«IDI )" 1011 '" IJI' II AlII ....... 11 I I I I I I I

1 J J 1 11 J, 1

... ~: 2

-w 1'CEII! 'tollun,. 1 ........ ,,11Jl I.,. ..... , 1 1 1: , J

NXJ& "'"UUnll,. 1 ....... 21) 11 J, 11

IC\dIr ..... 2. l' 1 1 1'11

_"': 1

JrOJB "110"'''" .. I 11 I 111

I ........ , J J , 1 1 2.

...-a .... ! 1

.. 1 1 1 1: ,2. J 1

f'oICt'IIII " , ..

-(lit

... ...

-14

... .. -cr:t.

.-

... , . • ~ ~ , , , , I I , , IO" I " 1 1 I · -. • , • " II

1 1 , , 10 , "

, ~ " 1 1 · 1

1 · 1 1 1 ... ,. I I

-. • 10 \I I I I 1 , 1 1 1 1 .. " .. 111

, , 1

" • , , 1 1 ... " , ,

" J · , . 1 1 , 1 , , , , , 1

In this paper, we proposed the Hierarchical Linked Cluster Multicast (HLCM) network architecture and the HLCM Network Construction algorithm based on the characteristic of the OS-COMA communication to construct a robust ad hoc multicast network. To achieve the collision-free and hidden-terminal avoidance multicast communications, the Uplink Multicast Schedule Algorithm (UMSA) and the Downlink Multicast Schedule Algorithm (DMSA) are proposed to generate the intra-cluster and inter­cluster transmission schedules, respectively. Each mobile node multicasts data packets to the destination nodes according to the pre-assigned time slots

44 Hsu-Yang Kung and Su-Man Chen

of the transmission schedules. The simulation results show that the proposed algorithms and control schemes effectively solve the hidden-terminal problem and achieve the self-organization and self-operation of a CDMA­based ad hoc multicast network.

References

[1] F. Adachi, M. Sawahashi, and H. Suda, ''Wideband OS-COMA for next-generation mobile communications systems," IEEE Communications Magazine, Vol. 36, No.9, pp. 56 -69, Sept. 1998.

[2] C. P. Blum, R. Dubois, R. Molva, and O. Schaller, "A development and runtime platform for teleconferencing applications," IEEE Journal on Selected Areas in Communications, Vol. 15, No.3, pp. 576-588, Apr. 1997.

[3] S. Chakrabarti and A. Mishra, "QoS Issues in Ad Hoc Wireless Networks", IEEE Communications Magazine, Vol. 39, Issue 2, pp. 142 -148, Feb. 2001

[4] Y. Chun, X.L. Ming, and S .. M. Lin, "Hierarchical On-demand Routing for Self­Organized Networks," Fifth Asia-Pacific Conference on Communications, Vol. I, pp. 128-133, 1999.

[5] C. Comaniciu, N. B. Mandayam. O. Famolari and P. Agrawal, "QoS Guarantees for Third Generation (3G) COMA Systems via Admission and Flow Control," Proceedings of IEEE VTC'OO- Fall, Sep. 2000, pp. 249-256.

[6] N. Dimitriou, R. TafazoUi, and G. Sfikas, "Quality of service for multimedia COMA," IEEE Communications Magazine, Vol. 38, No.7, pp. 88-94, July 2000.

[7] I. Gupta, "Minimal COMA Recoding Strategies in Power-controlled Ad-hoc Wireless Networks", Proc. 1st International Workshop on Parallel and Distributed Computing Issues in Wireless Networks and Mobile Computing, April, 2001.

[8] O. Gurbuz and H. Owen, "Oynamic Resource Scheduling Schemes for W-COMA Systems," IEEE Communications Magazine, Vol. 38, No. 10, Oct. 2000, pp. 80-84.

[9] Y. Han and H. G. Bahle, "COMA Technology: Present Status and Future Prospects," Asia Pacific Microwave Conference, pp. 165-168, 1997

[10) P. M. Mistry, ''Third Generation Cellular (3G): W-COMA & TO-COMA," Wesconl98, pp. 227-231,1998.

[11] M. W. Oliphant, ''The Mobile Phone Meets the Internet," IEEE Spectrum, Vol. 36, No. 8, pp. 20-28, Aug. 1999.

[12] G. Pei, M. Gerla, M., X. Hong, and C. C. Chiang, "A Wireless Hierarchical Routing Protocol with Group Mobility," Wireless Communications and Networking Conference (WCNC), Vol. 3, pp. 1538 -1542, 1999.

(13) C. E. Perkins, "Mobile Networking in the Internet," Mobile Networks and Applications, Vol. 3, 1998, pp 319-334.

[14] R. Ramanathan and M. Steenstrup, "Hierarchically-organized, multibop mobile wireless networks for quality-of-service support," Mobile Networks and Applications, Vol. 3, No. I, June 1998, pp. 101-119.

[15] J. Ryu, S. Song, and O. H. Cho, "A Power-Saving Multicast Routing Scheme in 2-tier Hierarchical Mobile Ad-Hoc Networks," Vehicular Technology Conference(VTC), Vol. 4, pp. 1974 -1978,2000.

[16] M. O. Sunay, S. Tekinay, and S. Z. Ozer, "Efficient allocation of radio resources for COMA based wireless packet data systems," Global Telecommunications Conference (GLOBECOM), Vol. 1b, pp. 638 -643, 1999.

Multipath Routing in Ad Hoc Wireless Networks with Directional Antenna

Somprakash Bandyopadhyay·, Siuli Roy·, Tetsuro Ueda", Kazuo Hasuike" • MIS Group, Indian Institute of Management Calcuna, Jom, Calcuna 700104, India e-mail: somprakash@iimcal.ac.in ph: +91-33-467-8300 fax:+91-33-467-8307 A ATR Adaptive Communications Research Laboratories, Kyoto 619-0288, Japan

Abstract: Multipath routing protocols are distinguished from single-path routing by the fact that they look for and use several routes from a source to destination. Several routing schemes have been proposed in the context of mobile ad hoc networks that uses multiple paths simultaneously by splitting the information among the multitude of paths. However. the effect of route coupling in this environment can severely limit the gain offered by multipath routing strategies. Route coupling is a phenomenon of wireless medium and occurs when multiple routes are located physically close enough to interfere with each other during data communication. In this paper, we investigate the effect of directional antenna on multipath routing. We have shown that the effect of route coupling across multiple paths with directional antenna is much less compared to that with omni-directional antenna. As a result, the routing performance using multiple paths improves substantially with directional antenna compared to that with omni-directional antenna.

Key words: Ad hoc networks. multipath routing, route coupling. directional antenna.

1. INTRODUCTION

Ad hoc wireless networks [1,2] are envisioned as infrastructure-less networks where each node is a mobile router, equipped with a wireless transceiver. Recently, there is a growing interest in ad hoc networks and its applications. Usually, the user terminals in ad hoc wireless networks use omni-directional antenna. However, it has been shown that the use of directional antenna can largely reduce radio interference, thereby improving

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

46 Somprakash Bandyopadhyay, Siuli Roy et al.

the utilization of wireless medium and consequently the network throughput [3, 4]. To achieve this, a Wireless Ad Hoc Community Network (W ACNet) testbed has been developed at ATR where the user terminals are equipped with small, low-cost directional antenna, known as ESP AR (Electronically Steerable Passive Array Radiator) antenna [3]. ESPAR antennas have a much higher gain than their omni-directional counterparts; so their use significantly reduce the RF power necessary to transmit packets. They can suppress co-channel interference and can therefore enlarge the capacity in terms of node-density (more terminals per unit area) in the network. In our earlier work, we have developed the MAC and routing protocol using ESPAR antenna [5]. The objective of this paper is to illustrate the advantages of multipath routing with directional antenna in the context of WACNet.

The routing schemes for ad hoc networks usually employ single-path routing which might not ensure desired end-to-end delay. However, once a set of paths between source s and destination d is discovered, in some cases, it is possible to improve end-to-end delay by splitting the total volume of data into separate blocks and sending them via selected multiple paths from s to d, which would eventually reduce congestion and end-to-end delay [6]. Utilization of multiple paths to provide improved performance, as compared to a single path communication, has been explored in the past in the context of wired networks [7,8].

The application of multipath techniques in mobile ad hoc networks seems natural, as mUltipath routing allows to diminish the effect of unreliable wireless links and the constantly changing topology [9]. The On-Demand Multipath routing scheme is presented in [10) as a multipath extension of Dynamic Source Routing (DSR), in which alternate routes are maintained, so that they can be utilized when the primary one fails. It has been shown that the frequency of searching for new routes is much lower if a node keeps multiple paths to the destination. However, the performance improvement of mUltipath routing on the network load balancing has not been studied extensively. M. R. Perlman et al.[11] demonstrates that the multipath routing can balance network loads in their recent paper. However, their work is based on multiple channel networks, which are contention free but may not be available in most cases. The Split Multipath Routing (SMR), proposed in [12], focuses on building and maintaining maximally disjoint multiple paths.

However, it has also been shown that deployment of mUltiple paths does not necessarily result in a lower end-to-end delay. In [11], the effect of Alternate Path Routing (APR) in mobile ad hoc networks has been explored. It was argued that the network topology and channel characteristics (e.g., route coupling) can severely limit the gain offered by APR strategies.

Multipath Routing in Ad Hoc Wireless ... 47

Suppose, only two sources Sl and S2 are trying to communicate data to dl

and d2 respectively. Let us assume that we select two node-disjoint paths for communication: Sl Xl Yl dl and S2 X2 Y2 d2• Since the paths are node-disjoint, the end-to-end delay in each case should be independent of each other. However, if Xl and X2 and/or Yl and Y2 are neighbours of each other, then two communications can not happen simultaneously (because RTS I ers exchange during data communication will allow either Xl or X2 to transmit data packet at a time, and so on.) .So, the end-to-end delay between any source and destination does not depend only on the congestion characteristics of the nodes in that path. Pattern of communication in the neighbourhood region will also contribute to this delay.

This is a phenomenon known as route coupling. Route coupling occurs when two routes are located physically close enough to interfere with each other during data communication. As a result, the nodes in those two routes are constantly contending for access to the medium they share and can end up performing worse than a single path protocol. Thus, node-disjoint routes are not at all a sufficient condition for improved performance in this context.

In this paper, we propose a notion of zone-disjoint routes in wireless medium where paths are said to be zone-disjoint when data communication over one path will not interfere with data communication in other path. However, getting zone-disjoint or even partially zone-disjoint routes in ad hoc network with omni-directional antenna is difficult, since the coverage area of each node is high and the MAC has to take care of hidden terminal problems as well. One way to reduce the coverage area of a node is to use directional antenna. In this paper, we investigate the effect of directional antenna on multipath routing. We have experimented with zone-disjoint paths and compared their effectiveness with respect to node disjoint paths. We also show that the probability of getting zone disjoint paths are much higher with directional antenna as compared to that with omni- directional antenna. As a result, the routing performance using multiple paths improves substantially with directional antenna compared to that with omni-directional antenna.

2. ZONE-DISJOINT ROUTES

The effect of route coupling has been measured in [13] using a correlation factor 11. The correlation factor 11 of two node-disjoint paths is dermed as the number of the links connecting the two paths. The total correlation factor of a set of multiple paths is defined as the sum of the correlation factor of each pair of paths [13].

48 Somprakash Bandyopadhyay, Siuli Roy et al.

In this paper, we propose a notion of zone-disjoint routes in wireless medium where paths are said to be zone-disjoint when data communication over one path will not interfere with data communication in other path. In other words, if there is no link (11 =0) between two node-disjoint paths, we say the two node-disjoint paths are zone-disjoint. Otherwise, the two node­disjoint paths are 11 - related. This is shown in figure 1 (from [13]).

• b "

'<I><I><r>D d

Figure 1. Two node-disjoint path with 11 = 7 (taken from [13]).

It has been shown that larger the correlation factor, the larger will be the average end-to-end delay for both paths. This is because two paths with larger correlation factor have more chances to interfere with each other's transmission due to the broadcast feature of radio propagation. In addition, larger the correlation factor, the larger will be the difference of end-to-end delay along mUltiple paths [13]. Based on this study, it can be concluded that the success of multipath routing in ad hoc networks heavily dependent on the correlation factor among multiple routes.

However, it is difficult to get multiple zone-disjoint routes using omni­directional antenna. With directional antenna, it is possible to de-couple multiple routes, thereby reducing the correlation factor among multiple routes. For example, if each of the nodes in figure 1 uses directional antenna towards its target node only, then the communication between S-a-b-c-D will not affect the communication between S-d-e-f-D.

Even if we get multiple zone-disjoint routes using omni-directional antenna, the best-case packet arrival rate at the destination node will be 1 packet at every 2*lp, where tp is the average delay per hop per packet of a traffic stream on the path p. The best-case assumption is, single traffic stream in the network from S to D with error-free transmission of packets. In contrast, if we use directional antenna, best-case packet arrival rate at destination will be one packet at every tp. Table 1 and 2 illustrate this point. Let us refer to figure 1 and assume that each node is equipped with omni­directional antenna. Let us further assume that the two paths shown are zone­disjoint i.e. 11 = O. Let us denote tp as a time-tick, i.e. at each time-tick, a packet is getting transmitted from one node to other. Consider table 1. S is sending a data-packet PI to node a at time-tick To and node a is sending data­packet PI to node b in the next time tick i.e. TI. With omni-directional

Multipath Routing in Ad Hoc Wireless ... 49

antenna, S has to sit idle during T 10 because S has received RTS from node a. So, S can only transmit its second packet P2 to node d (fIrst node of the second path) at time-tick T2• Similarly, when c is sending data to D, b will also be affected and node a cannot send data to b during that time to avoid collision at b. The packet transition is shown in Table 1 and destination D will receive packets in alternate time-tick. Even if we increase the number of zone-disjoint paths, the situation will not improve with omni-directional antenna.

Table 1 Packet Arrival Rate at D with Omni-directional Antenna

S a b c d e f D To PI>a TI PI>b T2 P2>d PIX T3 PI>D P2>e PI T4 P~a P2>f Ts P~b P2>D P2

T6 P..>d P~

T7 ... ... ... . .. ... ... ... ...

However, with directional antenna, when node a is transmitting a packet to node b, S can transmit a packet to node d simultaneously. Thus, as shown in Table 2, destination D will receive a packet at every time-tick with two zone-disjoint paths using directional antenna. It is to be noted here that two zone-disjoint paths with directional antenna is sufficient to achieve this best-case scenario.

Table 2 Packet Arrival Rate at D with Directional Antenna

S a b c d e f D To PI>a TI P2>d PI>b T2 P~a PIX P2>e

T3 P..>d P3>b PI>D P2>f PI T4 P,>a P~ P~ P2>D P2 T, P6>d P,>b P~D P.c>f P3 T6 P7>a P,X P~a P.c>D P4

T7 ... .. , ... ... ... ... ... ...

We have done a simulation study in order to establish that it is much easier to get a set of two zone-disjoint paths with directional antenna than that with omni-directional antenna. In this study, nodes were randomly placed into an area 1000 x1500 at a certain density. A source and destination

50 Somprakash Bandyopadhyay, Siuli Roy et al.

were randomly selected such that they are multi-hop away from each other. First, we have assumed that all the nodes are equipped with directional antenna with fixed transmission range. Between the selected source and destination, two 4-hop, zone-disjoint routes were found out. If two 4-hop, zone-disjoint routes were not available for that source-destination pair, another source-destination pair was selected. Then we have assumed that each node is having omni-directional antenna and computed the correlation factor 'Ilomni among those two routes that are zone-disjoint with directional antenna. This experiment was repeated for 25 source destination pair. As discussed, in each case, 'Ildir is zero and we compute 'IlOJmi • Then, the average 'Ilomoi were found out. Then, we change the node density and repeat this experiment. The results are shown in figure 2. As the number of nodes in the system increases, average 'Ilomoi increases. However, 'Ildir is zero in all the cases. This indicates that it is possible to get zone-disjoint paths with directional antenna at different node densities but those zone-disjoint paths with directional antenna will have high correlation factors, if we use omni­directional antenna.

-s ~ 7.5 .- B ~ ~ 7 +------------.".--i 9";i /"" ~ § 6.5 +------------::= ./"'"----1

<=I .::1 ".." .g ~ 6 -f--:-:----.-'--=:..,..::;;.....-----1 .!!:8 ~ ~ .~I 5.5 ~ u 0 5 +--~-_r_----r---r__-___i

40 50 60 70

Number of Nodes

Figure 2. Average Correlation factor llormi at different number of nodes when T\dlr =0.

3. A MECHANISM FOR MULTIPATH ROUTING USING DIRECTIONAL ANTENNA

In order to make the directional routing effective, a node should know how to set its transmission direction effectively to transmit a packet to its neighbors. So, each node periodically collects its neighborhood information and forms an Angle-SINR Table (AST) [3]. G\m(t) is the strength of radio

Multipath Routing in Ad Hoc Wireless ... 51

connection from node n to node m at an angle u with respect to n and as perceived by m at any point of time t. AST of node n specifies the strength of radio connection of its neighbors with respect to n at a particular direction. Mfinity of node m with respect to node n, aW u,m(t), is a number associated with a link IW D,m at time t , such that a W D,m(t) = Max [Gu D,m(t), 0< u <360]. In other words, the transmission angle w with respect to n maximizes the strength of radio connection from n to m, as perceived by m at any point of time. Based on this, a Neighborhood-Link-State Table (NLST) at each node is formed

We have designed a modified link-state protocol to make the nodes in the network topology aware [5]. Our primary aim is to collect all topology­related information from each node in the network and distribute them periodically (as updates) to only one of its neighbor, without flooding the network with topology-update packets. A node maintains a complete (but approximate) topology map of the entire network, called Global Link State Table(GLST). It not only depicts the connectivity between any two nodes but also the strength of connection or affinity value of the connection. In order to use the directional antenna, a node propagates its perception of the topology-information to only one of its neighbors at a periodic interval. Selection of target neighbor to propagate topology-map based on a criterion termed as least-visited-neighbor-first. Bach node monitor a metric called recency of its neighbors to decide which of them has received least number of update messages. The neighboring node that has received least number of update messages so far will be the target node for updating.

Whenever, a source S wants to communicate with a destination D, it computes multiple node-disjoint routes from S to D. From these mUltiple routes, it computes zone-disjoint routes. However, if zone-disjoint mUltiple routes do not exist between S and D, it selects single route. The assumption here is that, unless correlation factor is zero, performance improvement through multipath routing cannot be guaranteed. However, due to mobility and slow link-information percolation, it may not be possible to maintain zero correlation factor among multiple routes. To improve performance under mobility, the source node periodically computes multiple zone-disjoint paths and adaptively modifies its routing decision.

4. CONCLUSION

In order to make effective use of multipath routing protocols in the mobile ad hoc network environment, it is imperative that we consider the effects of route coupling, especially in single channel networks. However, high degree of route coupling among multiple routes between any source and

52 Somprakash Bandyopadhyay, Siuli Roy et al.

destination pair is inevitable, if we use omni-directional antenna. This paper has analysed the problem and proposed a mechanism to alleviate the problem of route coupling using directional antenna. As a result, the routing performance using multiple paths improves substantially with directional antenna compared to that with omni-directional antenna.

REFERENCES

1. E. M. Royer and C-K Toh, "A Review of Current Routing Protocols for Ad hoc Wireless Networks", lEEE Personal Communication, April 1999, pp. 46-55.

2. J. Brach, D. A. Maltz, D. B. Johnson, Y. C. Hu, and J. Jetcheva, "A Performance Comparison of Multi-Hop Wireless Ad Hoc Network Routing Protocols,' Proc. ACMIIEEE Mobile Comput. and Network., Dallas, TX, Oct. 1998.

3. S. Bandyopadhyay, K. Hasuike, S. Horisawa, S. Tawara, "An Adaptive MAC Protocol for Wireless Ad Hoc Community Network (W ACNet) Using Electronically Steerable Passive Array Radiator Antenna" , Proc of the OLOBECOM 2001, November 25-29,2001, San Antonio, Texas, USA

4. Y.-B. Ko, V. Shankarkumar and N. H. Vaidya, "Medium access control protocols using directional antennas in ad hoc networks," Proc. Of the IEEE lNFOCOM 2000, March 2000.

5. S. Bandyopadhyay, K. Hasuike, S. Horisawa, S. Tawara, "An Adaptive MAC and Directional Routing Protocol for Ad Hoc Wireless Network Using Directional ESPAR Antenna" Proc of the ACM Symposium on Mobile Ad Hoc Networking & Computing 2001 (MOBIHOC 2001), Long Beach, California, USA, 4-5 October 2001

6. N.S.V.Rao and S.O. Batsell, QoS Routing via Multiple Paths Using Bandwidth Reservation, in Proc. of the lEEE lNFOCOM98, 1998.

7. S. Bahk and W. El-Zarki, Dynamic Multi-path Routing and how it Compares with other Dynamic Routing Algorithms for High Speed Wide-area Networks, in Proc. of the ACM SIGCOM, 1992

8. S. Murthy and lJ. Oarcia-Luna-Aceves. Congestion-oriented Shortest Multi-path Routing, in Proc. of the IEEE lNFOCOM96, 1996.

9. Aristotelis Tsirigos Zygmunt J. Haas, Siamak S. Tabrizi , Multi-path Routing in mobile ad hoc networks or how to route in the presence of frequent topology changes, MILCOM 2001.

10. A. Nasipuri and S.R. Das, "On-Demand Multi-path Routing for Mobile Ad Hoc Networks," Proceedings oflEEE ICCCN'99, Boston, MA, Oct. 1999,

11. M. R. Pearlman, Z. 1. Haas, P. Sholander, and S. S. Tabrizi, On the Impact of Alternate Path Routing for Load Balancing in Mobile Ad Hoc Networks, MobiHOC 2000, p. 150,3-10.

12. S.J. Lee and M. Oerla, Split Multi-path Routing with Maximally Disjoint Paths in Ad Hoc Networks, ICC 2001.

13. Kui Wu and Janelle Harms, On-Demand Multipath Routing for Mobile Ad Hoc Networks EPMCC 2001, Vienna, 20th - 22nd February 2001

PERSONAL WIRELESS COMMUNICATIONS

A Reactive Service Composition Architecture for Pervasive Computing Environments

Dipanjan Chakraborty, Filip Perich, Anupam Joshi, Timothy Finin, Yelena Yesha Department of Computer Science and Electrical Engineering. University of Maryland, Balti­more County

dchakr1 @cs.umbc.eclu,fperlc1 @cs.umbc.edu,joshl@cs.umbc.edu,finln @cs.umbc.edu,yeyesha@cs.umbc.ed

Abstract Development of customized services by integrating and executing existing ones (refered to as service composition) has received a lot of attention in the last few yem with respect to wired, infrastrutcure based web-services. With the advancement in the wireless technology and pervasive computing, we eovison that in the near future, we will have such information or services embedded in various wireless devices in our vicinity. However, wired infrastructure-based service discovery and composition architectures do not take into consideration factors arising from the possible mobility of the service providers. In this paper, we present Anamika: a distributed, de-centralized and fault-tolerant design architecture for reactive service composition in pervasive environments.

1. Introduction Service Composition can be defined as the process of creating customized

services from existing services by a process of dynamic discovery, integration and execution of those services in a planned order to satisfy a request from a client. Research in the area of service discovery [1, 15, 3, 8, 18]and service composition [6, 17, 12, 16,4, 14]has focused on trying to leverage the wide array of e-services available over the network to provide customized services to e-customers, for example planning a business trip for a person. There has been a sharp increase in these types of wired infrastructure-based services in the last few years. Existing service composition systems [14, 6, 12]broadly address the problems associated with composing various services that are available over the fixed network infrastructure. They primarily rely on a centralized composition engine to carry out the discovery, integration and composition of web-based e-services.

However, with computing today becoming increasingly pervasive, we envis­age that in the near future, mobile and embedded devices will also be capable

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

54 Dipanjan Chakraborty, Filip Perich et al

of providing customized information, services and computation platforms to peers in their vicinity. People will need the cooperation of services available in their resource-rich vicinity to satisfy their information needs.

Service Composition systems for the pervasive computing environments need a different design approach than those developed for wired services. This is because many of the assumptions of standard composition architectures of wired services are no longer valid in such dynamic environments. Service Composition architectures in wired infrastructure assumes the existence of a centralized composition entity that carries out the discovery, integration and execution of services distributed over the web. They also need a tighter coupling with the varying network layer protocols.

We have designed a distributed architecture to perform service composition in pervasive computing environments. Central to our system is the concept of a distributed broker that can execute at any node in the environment. An individual broker handles each composite service request, thus making the design of the system immune to central point of failure. A broker may be selected based on various parameters such as resource capability, geometric topology of the nodes and proximity of the node to the services that are required to compose a particular request. Current prototype of our system has been implemented over Bluetooth.

Service discovery and composition is an important and active area of research [5, lO]and has been studied widely in the context of web-services. Most of the research in realizing service composition systems for web-based services have a centralized architecture for service integration and execution management. We are aware of systems like eFlow [6], CMI [17}, Ninja Service Composition Architecture [12], Sheng's framework [4]on declarative web service compo­sition based on state charts that broadly address various problems related to service composition in the context of wired services. Due to lack of space, we are unable to present details of these systems.

2. System Architecture and Design Principles In this section, we describe a genera1layered architecture that enables service

composition in pervasive computing environments. Our architecture introduces two distributed reactive techniques to carry out service composition in purely ad-hoc environments. Our composition architecture primarily deals with the discovery, integration and execution of the components of a composite request. Figure 1 depicts the different layers and modules in the architecture. In the rest of the paper, we shall refer to a client as a device from where the service composition request originates. A broker is a device that coordinates the different components to calculate the result.

A Reactive Service Composition Architecture ... 55

Application Layer

~ervlce Compo.ltlon Service Exeoutlon Layer Layer

I a.:::.::::;'.:~ .. ".~:. I I Paull '".oovery ModUI_1

J .~~_.rv oe ::~ .oov·rya~A~V'·r Cllluetooth SOP .:.,u,.uon ...... ... I

Network Layer (RFCOM/PPP/TCPIIP. CDPD. CDMA etc)

•

Figure 1. General Architecture for ad-hoc service composition

2.1 System Components Network Layer:. The Network Layer forms the lowest layer in the ar­chitecture and encapsulates networking protocols that provide wireless/ad-hoc connectivity to peer devices in the vicinity. We assume the existence of a suitable network layer that provides us with connectivity to the neighboring devices.

Service Discovery Layer:. The service discovery layer is required for the proper functioning of the composition platform. There is a direct dependence of the success of the composition techniques on the underlying service dis­covery mechanisms. This layer encompasses the protocol used to discover the different services that are available in the vicinity of a mobile device. Our design of the service discovery mechanism is primarily based on the principles of Peer-to-peer service discovery, Dynamic caching of neighboring service de­scriptions, Semantic description based service matching, service request rout­ing and propagation control. We do not employ central lookup-server based service discovery and maintenance. Each device has a Service Manager where the local services register their information. Service Managers advertise their services to neighboring nodes and these advertisements are cached. Services are described using a semantically rich language DAML+OIL [13]which is used in service matching also. On cache miss, the service request is forwarded to neighboring nodes. We describe our protocol in detail in [7].

Service Composition Layer:. This layer is responsible for carrying out the process of managing the discovery and integration of services to yield a

56 Dipanjan Chakraborty, Filip Perich et al

composite service. The process model of the composite service is supplied as input to this layer. In our current implementation, we have used the 'compos­iteprocess' definition of DAML-S to describe a process model. We describe our two reactive techniques of composition in detail in next two subsections.

Service Execution Layer:. The Service Execution Layer is responsible for carrying out the execution of the different services. Prior to this, the service composition layer provides a feasible order in which the services can be composed and also provides location and invocation information of the service(s). This layer has a module called the "Fault Recovery Module", which is responsible to guard against node failures and service unavailability. We explain the fault-tolerant techniques in the next two subsections. The Service Execution Layer and the Service Composition Layer are tightly coupled with each other due to their dependence on each other.

Application Layer:. The application layer embodies any software layer that utilizes our service composition platform. The application layer encompasses different GUI facilities to display the result of a composed service and provides the functionality to initiate a request for a composite service.

2.2 Dynamic Broker Selection Technique This approach centers on a procedure of dynamically selecting a device to

be a broker for a single request in the environment. In the following section, we describe three distinct features of the Dynamic Broker Selection Technique.

Broker Arbitration and Delegation:. When a request for service composi­tion arrives at the service composition module in a mobile device it finalizes a platform that is going to carry out the composition and monitor the execution. Once the platform has been chosen, the device is informed of its responsibility. The mobile device acting as the broker is responsible for the whole composi­tion process for a certain request. The selection of the broker platform may be dependent on several parameters: power of the platform (battery power left), number of services in the immediate vicinity, stability of the platform, etc. The brokerage arbitration might make the originator of the request to be the broker for that particular composition.

Each request thus may be assigned a separate broker. This makes the architecture immune to central point of failure and the judicious choice of brokerage platform has the potential of distributing the load appropriately within the different devices. This avoids the problem of swamping the central composition entity by numerous requests.

A Reactive Service Composition Architecture ... 57

Service Integration and Execution:. The assigned broker's first job is to discover the services from its vicinity. The broker progressively increases its search "radius", a number of devices that it can reach by asking other devices in its radio range to forward service request, to discover all of the different services necessary for the composition. The broker returns failure when it fails to discover all of the required services. Service discovery and integration is followed by service execution. The information obtained during the service discovery (service address, port, invocation protocol) is utilized to execute the services.

Fault Recovery:. Faults in ad-hoc environment may occur due to a service failure, due to a sudden unavailability of the selected broker platform, or due to network partition. The standard solution to this problem is to make the requester to initiate a new request for every composite service. This is very inefficient and not applicable in our environment due the relatively high occurrence probability of the above failures. The fault -tolerance module in the architecture employs check pointing to guard against such faults. The broker for a particular request sends back checkpoints and the state of the request to the client of the request after a subtask is complete. The client keeps a cache of this partial result obtained so far. If the broker platform fails, the source node detects the unavailability of updates. The source of the request then reconstructs the query that is still left unsolved by the broker. This request is now treated as a different service composition request in the environment. Thus, If a node currently acting as the broker of a request fails, then using the same principle the architecture adapts itself to select other brokers dynamically.

2.3 Distributed Brokering Technique

The key idea in this approach is to distribute the brokering of a particular request to different entities in the system by determining their 'suitability' to execute a part of the composite request.

Broker Arbitration:. This module performs almost the same set of actions as described in the previous section. However, the key difference is that it only tries to determine the broker for the first few services (say SI to Si) in the whole composition. This layer tries to utilize the resources available in the immediate vicinity instead of looking for the resources required to execute the whole composition. Thus, a single broker only executes a part of the whole composite process (based on the resources that it currently has available to it).

Service Integration and Execution:. The broker is responsible for com­posing the services SI to Sn. The broker decides on a service search "radius". The composition is carried out among services discovered within this radius.

58 Dipanjan Chakraborty, Filip Perich et al

~~ I I I

r------ -------------

, i~ I I~ 1 ________________________ ..L _____ •

Figure 2. System Components in Anamika Reactive Service Composition Architecture

Suppose a broker determines that it has services S 1 to Si available in its vicin­ity(within radius r). It goes ahead and carries out the partial integration and execution. It then informs the requester (source node) about the 'current state' of the execution. Secondly, it uses the 'Broker Arbitration' Module to select another broker which has the ability to carry out a subset or whole of the re­maining composition. In this manner, the composition hops from one node to another till the final result is obtained. Then the current broker returns the final answer of the composition to the client.

3. Implementation and Experiments In our initial implementation of the design architecture, we have developed

a reactive service composition system called Anamika. The individual com­ponents of the Anamika system existing in participating mobile devices are described in Figure 2. Current prototype of the architecture has been im­plemented over Bluetooth [18]. Composition knowledge is described using DAML-S[UJin terms of subset of individual services that might be able to satisfy a composite request. Service discovery is done in a peer-to-peer manner using semantic description of services using DAML-S and our light-weight rea­soning engine present on participating devices. Anamika implements Dynamic Broker Selection mechanism and decides the best platform to carry out the com­position based on a combination of the processor power of the platform and number of services that the platform has. The Network Manager implements the API required for higher layers to reliably communicate to neighboring peers over Bluetooth. We implemented the networking level API over RFCOMM [18]protocol over ffiM's BlueDrekar transport driver [2]for Linux kemeI2.4.2-2. Service Discovery Manager provides the functionality to local Composition Manager to discover services in Bluetooth peers. The principles of our Service Discovery Manager is described in detail in [7]. The Service Composition Manager is the principal component that is responsible for service composition

A Reactive Service Composition Architecture ... 59

and management. Peer Service Composition Managers collaborate with each other to implement different techniques of service composition. The Compo­sition Knowledge base has been modeled in DAML-S. Due to lack of space, we are unable to provide a greater detail of the implementation. However, it is available in our technical report [9].

Experiments:. We carried out various experiments to validate the proper functioning of the Anamika Reactive Service Composition system and test dif­ferent features of the Dynamic Broker Selection Technique. In our experimental setup, services reside on different laptops. These services are registered to the local composition managers residing on the machines. Bluetooth modules pro­vided by Ericsson and mM's BlueDrekar software stack [2]and transport driver are used by the Network Manager to communicate between devices. Some of the laptops also have 802.11 b connectivity to the Internet. Every Anamika sys­tem displays a graphic user interface for clients to access composite services formed by the services available in the vicinity. There is no central "Broker service" in the system and all the participating systems are liable to be bro­kers. We carried out experiments to test the proper functioning of the different mechanisms for service composition and under different states of the ad-hoc environment. However, due to space limitations, we are unable to provide details of the experiments.

4. Conclusions

In this paper, we have introduced a novel design approach for service com­position in pervasive computing environments. Our architecture for service discovery and composition is distributed, decentralized and fault-tolerant to service and network unavailability. Service Discovery is done in a peer-to-peer mode rather than a centralized mode, and service descriptions are cached for scalability. We introduce two reactive techniques, Dynamic Brokerage Selec­tion mechanism and Distributed Brokerage technique to accomplish service composition in dynamic environments. Our approach enables any device par­ticipating in the composition to act as the broker, making the design immune to single point of failure. We use a source-monitored fault-tolerance mecha­nism using checkpoints and rollbacks to the last completed service. We have also presented the Anamika system, an initial implementation of our design architecture. Anamika has been implemented over Bluetooth using RFCOMM as the network layer. Our future work includes implementing the "Distributed Brokerage" mechanism in Anamika, perform assessment of the different mech­anisms with respect to factors like mobility of the environment, availability rate of the services.

60 Dipanjan Chakraborty, Filip Perich et al

References [1] The Salutation Consortium Inc 1999. Salutation architecture specification (part 1), version

2.1 edition. World Wide Web, http://www. salutation.org. [2] IBM alphaworks. BlueDrekar protocol driver. World Wide Web, http://www.

alphaworks.ibm.com/tech/bluedrekar. [3] Ken Arnold, Ann Wollrath. Bryan O'Sullivan, Robert Scheifter, and Jim Waldo. The Jini

specification. Addison-Wesley, Reading, MA, USA, 1999.

[4] B. Benatallah, M. Dumas, Q. Sheng, and A. Ngu. Declarative composition and peer-to­peer provisioning of dynamic web services. In 18th International Conference on Data Engineering., February 2002.

[5] F. Casati, D. Georgakopoulos, and M. Shan editors. Special issue on e-services. VLDB Journal,2001.

[6] F. Casati, S. Dnicki, L. Jin, V. Krishnamoorthy, and M. Shan. Adaptive and dynamic service composition in eflow. Technical Report, HPL-200039, Software Technology Laboratory, Palo Alto, CA, march 2000.

[7] Dipanjan Chakraborty, Anupam Joshi, Tim Finin, and Yelena Yesha. GSD: A novel group­based service discovery protocol for MANETS. In 4th IEEE Coriference on Mobile and Wireless Communications Networks (MWCN 2002). Stockholm. Sweden, September 2002.

[8] Dipanjan Chakraborty, Filip Perich, Sasikanth Avancha, and Anupam Joshi. Dreggie: A smart service discovery technique for e-commerce applications. In 20th Symposium on Reliable Distributed Systems, october 2001.

[9] Dipanjan Chakraborty, Filip Perich, Anupam Joshi, Timothy Finin, and Yelena Yesha. A service composition architecture for pervasive computing environments. Technical report, University of Maryland Baltimore County. USA, March 2002. TR-CS-02-02.

[10] G. Weikwn. Editor. Special issue on infrastructure for advanced e-services. IEEE Data Engineering Bulletin, 24(1), March 2001.

[11] DARPA Agent Markup Language for Services. World Wide Web, http://www.ai . sri.com/daml/services/daml-s.pdf.

[12] R.H. Katz, Eric. A. Brewer, andZ.M. Mao. Fault-tolerant, scalable, wide-areaintemet ser­vice composition. Technical Report. UCB/CSD-I-1129. CS Division. EEeS Department. UC. Berkeley, January 2001.

(13) DARPA Agent Markup Language and Ontology Inference Layer. http://www . daml. org/2001/03/daml+oil.daml.

[14] David Mennie and Bernard Pagurek. An architecture to support dynamic composition of service components. Systems and Computer Engineering. Carleton University, Canada.

[15] UPnP White Paper. World Wide Web, http://upnp.org/resources . htm. [16) Chaitanya Pullela, Liang Xu, Dipanjan Chakraborty, and Anupam Joshi. Component

based architecture for mobile information access. In Workshop in conjunction with International Conference on Parallel Processing (ICPP)., August 2000.

[17] H. Schuster, D. Georgakopoulos, A. Cichocki, and D. Baker. Modeling and compos­ing service-based and reference process-based multi-enterprise processes. In Proc. Inti. Conference on Advanced Information Systems Engineering, Sweden., June 2000.

[18] Bluetooth Specification. World Wide Web, http://www.bluetooth.com/ developerlspecification/Bluetooth_ll_Specifica%tioBook.pdf.

BLUETOOTH PAN and external IP networks

1 Tore E. ]j1jnvik, 2Paal Engelstad & 2 Do van Thanh 1. Unik - University of Oslo - Norway - tlf: +47 90199176 - torejoen@iji.uio.no 2. Telenor R&D - Norway - tif: +4790977 10 2-{Paal.Engelstad, thanh-van.doJ@telenor.com

Abstract: This paper discusses how ad-hoc Personal Area Network (PAN) based on Bluetooth technology may be connected to external networks. We assume that the Bluetooth network (piconet) is formed by the automatic SAPIFO procedure [4], that one or more piconet devices have access to external networks, and that IPv4 is used for external communication.

Key words: Bluetooth, Ad-hoc Networking, PAN Personal Area Network, BNEP Bluetooth Network Encapsulation Protocol, IP.

1. MOTIVATION

Bluetooth was initially designed as an efficient cable replacement technology primarily for handheld devices. With the forecasted abundance of Bluetooth-enabled devices, it is, however, reasonable to assume that Bluetooth will evolve from being a cable replacement to become a network infrastructure connecting multiple devices together into a piconet. Indeed, all the devices belonging to one person can form a PAN (Personal Area Network) using Bluetooth. After the piconet is formed, devices that have external access to the Internet may provide Internet access to other piconet devices that are not directly connected to an external network. Bluetooth has specified a PAN profile for IP over Bluetooth, which uses BNEP (Bluetooth Network Encapsulation Protocol) to emulate an Ethernet segments between master and slave. If the master has an additional Ethernet connection to an external network. it uses the NAP (Network Access Point) role to interconnect the Ethernet segments and form a piconet. If the master has no

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

64 Tore E. Nmvik, Paal Engelstad and Do van Thanh

external Ethernet connections, on the other hand, it uses the GN (Group ad­hoc Network) role. The Ethernet segments are connected using functions from 802.ld. This paper outlines a new role, which accommodates more than one network access points by combining NAP and GN. The new role also allows a slave to serve as an access point.

2. OVERVIEW OVER BLUETOOTH TECHNOLOGY

Bluetooth is a wireless technology communicating in the 2,45 GHz ISM band and is based on a frequency hopping spread spectrum. Bluetooth has a Master/Slave architecture where one master can control up to 7 active slaves. Each Bluetooth transceiver is allocated a unique 48-bit Bluetooth Device Address (BD_ADDR) based on the IEEE 802 Standard

2.1 BNEP

NetworiUng Applications U 0 4 a 12 16 20 2. 28 31

MPoJ:= lEI I TCPAJOP U OestRiIi:nAdcDss ~ 0-2)

I IP JJ DesInaIm AdIml (8)ta ~ 1$Qm~

U~ iB)iI 01 '

I BNEP

LI SaItIl ~(8)1Os 1-4)

t2CAP

I ~ ~~I ~ProIDcdT)I)Il I~~ (8)'11 5) O'PayIJad _

.I Bluetooth Baseband U • • II Bluetooth Radio 1J

Figure lao The networking reference stack for the PAN profile. with the Bluetooth radio (as OSI layer 1) Baseband and L2CAP as OSI layer 2) and BNEP (as the network adaptation between Bluetooth layer 2 and the IPI Figure lb. BNEP with an Ethernet packet payload sent usingL2CAP

The Bluetooth Network Encapsulation Protocol, BNEP, emulates an Ethernet on a broadcast network segment, hiding the underlying master­slave based piconet topology. BNEP runs over L2CAP, as illustrated in figure la. BNEP reuses the Ethernet packet format commonly used for local area networking technology. The 48 bits Bluetooth addresses are used as IEEE source and destination addresses. The format of the BNEP header is shown in Figure lb. The BNEP header may be extended with one or more extension headers that allows for additional capabilities. BNEP also defines

BLUEI'OOTH PAN and external IP networks 65

connection control messages. Before completion of the BNEP connection setup, the initiator has to indicate the roles of both end-points. For bandwidth saving purposes, protocol and multicast filter commands have been defined to indicate which protocol types and multicast addresses a device wants to receive. All these control messages have to be confIrmed before the new confIguration applies. BNEP accommodates IP communication by transporting IP packets between two Ethernet-based link-layer end-points on an IP segment. It encapsulates the IP packets in BNEP headers, letting the source and destination addresses reflect the Bluetooth end-points and setting the 6-bit Networking Protocol Type fIeld to code for an IP packet in the payload. BNEP fInally encapsulates the BNEP packet in an L2CAP header and sends it over the L2CAP connection.

2.2 PAN profile

The PAN Profile[2] identifIes two configurations of a Bluetooth PAN: The Network Access Point (NAP) confIguration is used when the master is connected to an external network, and the Group Ad-hoc Node (ON) confIguration is used when no devices have a network connection. In both cases, the Bluetooth device that uses the NAP service or ON service is a PAN User (P ANU). The NAP and ON forward BNEP packets between P ANUs according to the BNEP protocol, which implements parts of the IEEE 802.1D standard.

2.3 Piconet formation

SAPIFO [4] represent a suggested procedure for automatic piconet formation. This procedure is based on the assumptions of the existence of at least one possible piconet among the available Bluetooth enabled devices that will participate in the PAN. All devices will start a procedure to detect Bluetooth enabled devices within radio range and their Bluetooth address (BD-ADDR). SAPIFO presupposes the use of a Dedicated Access Code (DIAC) in the Inquiry phase that is reserved for certain computing class of devices. It is therefore not necessary to set up a L2CAP connection and use the SDP protocol to search for devices with computing capabilities.

When all devices have detected all other devices within radio range, they will inform their neighbours about the detected devices. After this is fInished, all devices will have a table of all devices and their possible connections in the future piconet. This table will be basis for the distributed procedure to select possible Master candidates for the piconet. All devices with the highest number of detected devices will be candidates. H more than one is device is candidate, the one with highest BD-ADDR will be selected

66 Tore E. 1¢mvik, Paal Engelstad and Do van Thanh

master. The selected master will now page the other devices and form the piconet. If there is more than one master candidate, the others can be used as backup master(s). SAPIFO also contains procedures for piconet maintenance taking care for devices entering or leaving. A consequence of the maintenance procedure is that a new master can be selected.

3. USING IPFOR INTERNET ACCESS

Slave 1

/' ~Ma&a ~ave3

\

Slave 2 !\ , \~~

•

Figure 2. The new role proposed in this paper allows one or more slaves to serve as access points to external networks

The new role presented in this paper and illustrated in Fig 5 assumes a mobile piconet that is able to connect to different access technologies including WLAN, GSM GPRS or GSM HSCD. IP is a technology that allows such inter-technology communication, and the ubiquitous IPv4 protocol is therefore assumed [5]. Since the network may be mobile, access points may gain and lose Internet connectivity in a non-deterministic manner as the Bluetooth network moves, and the network must reconfigure itself automatically to changing Internet connectivity conditions. Due to the dynamic nature of our scenario, one or more access points may pop up on slaves as well as on the piconet master, which is the situation that the new role must cope with (Fig. 5). In comparison, access points on slaves cannot be fully utilized as a resource if the existing Bluetooth NAP profile is being used. NAP only allow the master to connect to one external Ethernet-like network by means of BNEP-based bridging.

The most limiting factor for IPv4 is the scarcity of IP addresses. ISPs are often reluctant to allocate global IP addresses to roaming nodes, which often have limited access privileges. This means that an access point is likely to

BLUETOOTH PAN and external IP networks 67

receive one external IP-address at most - probably a private IP-address (i.e. the ISP is implementing a NAT solution) or a global IP address at best. The wayan access point acquires the external IP-address - assuming automatic address allocation - is dependent on the link-layer technology used for the external access. If the external network is Ethernet based [11], it will likely use DHCP [9], or it may receive the address from a Mobile IP foreign agent through an ICMP Router Advertisement [10]. On PPP-enabled links, however, the access point will likely be authorized an IP-address after successful PPP authentication [14]. On 2G and 3G cellular networks, other techniques may apply. Different nodes in the Bluetooth piconet must share the external IP-address that the access point acquires. The easiest way to accommodate this is to allocate private IP addresses [7] or IPv4 link-local addresses [8] to hosts and routers on the piconet, and use Network Address and Port Translation (NAPT) for Internet Access [13]. A NAPT is a router that replaces a private or link-local IP source address and port number for outgoing IP packets with a global IP address and a unique port number before forwarding them towards the Internet. It performs the reverse translation with the destination addresses of incoming packets before forwarding them into the internal network.

4. THE PROPOSED IP SOLUTION

The new role proposed in this paper attempts to locate all essential state information centrally on the master. The piconet is then more easily maintained, and there is far less fate sharing, i.e. the piconet does not depend on a slave being present in addition to the master. The back-up masters assigned during the SAPIFO procedure may also easily replicate the state information directly from the master, and services take over network based on existing state information without disruption.

As a result of this design choice, the new role mandates that: • The master serves as a network router, which intercepts and forwards IP

packets, and maintains IP state information about the slaves on the piconet. Slaves, on the other hand, can be IP hosts.

• A slave acquires a private or link-local IP address for its own from the master. Thus, the master implements DHCP-server for allocation of private IP addresses, and answer DHCP request from slaves [8].

• The master serves as the default gateway of hosts on the piconet. Since the master has full control with all local IP-addresses, it answers

ARP requests directly without broadcasting the request to other slaves. The master also answers requests from slaves trying to claim a link-local IP address [8], and ensures that all link-local addresses are unique.

68 Tore E. Jrpnvik, Paal Engelstad and Do van Thanh

One problem with using NAPTs and multiple access points is that all outgoing packet of a stateful session (e.g. a TCP session) must pass through the same address translator where the state information about the address translation is stored. Otherwise, the IP-packet will be assigned a new global source IP-address and the session will break. Furthermore, since Internet Service Providers (ISPs) are probable to implement ingress filtering, the packet should be sent over the access point corresponding to the IP-address that the private IP-address is translated into. The new role therefore mandates that the master assigns an access point to each communicating host, and that this information is stored on the master. When the master­router receives an IP-packet bound for an external network, it checks the table to fmd which access point to forward the packet to. If the master has a connection to an external network, it may naturally serve as an access point for some of the slaves.

Some access points may implement a NATP-router. It these cases, the master forwards packets unaltered to the access point, which in turn translates the packet. A host residing on the access point itself uses the private IP-address as source address, and outgoing packets destined for the Internet may be sent directly from the access point, without being sent via the master. The packets must however be passed through the NAPT module to ensure correct and consistent address and port translation of all hosts residing on the piconet. However, it is anticipated that some access points may not be able to implement NAPT or serve as a router. In these cases the master should do the translation on behalf of the access points. All incoming packets carrying an IP payload from the external network, is sent directly to the master by copying the IP-payload into a BNEP packet. All outgoing IP­packets are sent from the master, and the access point uses copies the payloads from a BNEP packet into a header corresponding to access technology for external access.

The master needs a method to acquire the external IP address from the access point, while the access point acquires a private IP address from the master as described above. In this mode of operation, the access point must send all packets originated from itself to the master, using its private address as a source address.

Packets destined for the access point is only accepted from the master, and the IP-header must carry its private IP-address as a destination address. The access point needs a method to distinguish IP-packets that are to be blindly forwarded to the access network from those destined for the access point itself. We propose to introduce a BNEP extension header type for this purpose: All packets carrying the specific extension header type will be blindly forwarded to the external network. Some additional filtering rules may be specified to optimize the solution.

BLUETOOTH PAN and externallP networks 69

5. THE PROPOSED BLUETOOTH SOLUTION

It should be clear that the proposed IP solution requires support from the underlying Bluetooth technology. We propose that the piconet is formed automatically using the SAPIFO procedure. The procedure allows a master to be elected, while other candidate masters serve as back-up masters of the piconet. It also mandates how the piconet is set up.

After the L2CAP connections have been established, the master uses SOP to check the capabilities of different slaves, and determines which slaves are willing to and capable of servicing as access point. A service class for Internet Access Points with a predefmed UUID[ 1] should be defined for this purpose.

The master reads the mode of operation and status of the access point from appropriately defined SOP attributes. It fmds whether the access point is capable of dynamically allocating of external IP-addresses and port­numbers to Bluetooth hosts (i.e. by implementing NAPT on the access point) or the allocation should be managed by the master. This mandates the behavior of the access point and how the master forward packets over the access point, as outline above.

If the master allocates external IP-addresses (e.g. by performing NATP­translation) on behalf the access point, it may fmd the external IP address or prefix of the access point through an SOP attribute. It should be noted that the number external IP-address managed by the master may exceed the number of Bluetooth hosts requiring Internet access, especially if some access points have acquired an IP-prefix from the external network. In this case, the master may not need to do port translation. Each Bluetooth host may even be assigned an external IP-address directly through OHCP, instead of using private address, and not even address translation will be required.

Another SOP attribute that gives the status of the external access should also be defined. Hence, the master may use SOP periodically to check if an access point have gained or lost Internet access over the external network, or if the external IP-address or prefix of the access point has changed.

Back-up masters should be ready to take over the network if the master goes down. A potential back-up master should therefore periodically download the state information cached at the master. SNMP [12] is an example of an IP-based protocol that may be used for this purpose. However, a Bluetooth specific solution is preferable, since the back-up mechanism should not mandate that the Bluetooth nodes run IP - some devices may not even be configured with an IP stack.

The already existing Bluetooth protocol, SOP, might be used for replicating state information. Backup-masters would poll the SDP-server on the master periodically and download (i.e. pull) status information from the

70 Tore E. lr/mvik, Paal Engelstad and Do van Thanh

master. Alternatively, new Bluetooth protocol functions may be defined for this purpose. The master may, for example, use some new LMP commands to upload (i.e. push) state information onto the backup-masters. Another approach would be to introduce an entirely new protocol over L2CAP with a new reserved Protocol Service Multiplexor (PSM). The proposed solution will be shown in more details in a subsequent paper.

6. CONCLUSION

This paper proposes a new Bluetooth role, which allows ad-hoc Personal Area Network (PAN) based on Bluetooth technology to connect to external networks. Unlike for the NAP and ON roles, a slave may provide access to the Internet and multiple Internet access points may be used simultaneously. The proposal assumes that the Bluetooth network (piconet) is formed by the automatic SAPIFO procedure [4], and that IPv4 is used for external communication. The new role will be detailed in a subsequent paper.

REFERENCES

[1] Specification of the Bluetooth System http://www.bluetooth.comldev/specifications.asp [2] Personal Area Networking Profile

http://www.bluetooth.comlpdflP AN_Profile_0_95a.pdf [3] [Bluetooth Network Encapsulation Protocol (BNEP) Specification

http://www.bluetooth.comlpdfIBluetooth_II_Specifications_Book.pdf [4] Tore J~nvik, and Do Van Thanh "Ad-hoc formation of Bluetooth Piconet for data

communication" [5] 3Gwireless and Beyond. San Francisco May 2002 [6] Internet Protocol http://www.ietf.orglrfclrfc791.txt [7] IP Mobility Support for IPv4 http://www.ietf.orglrfclrfc3220.txt [8] Address Allocation for Private Intemets http://www.ietf.orglrfclrfc1918.txt [9] Dynamic Configuration of IPv4 Link-Local Addresses http://files.zeroconf.orgldraft-ietf-

zeroconf-ipv4-linklocal.txt [10] Dynamic Host Configuration Protocol http://www.ietf.orglrfclrfc2131.txt [11] 802.1d http://www.ieee802.orgillpages/802.1D.htmI [12] Ethernet http://standards.ieee.orgigetieee802/ [131 A Simple Network Management Protocol (SNMP) http://www.ietf.orglrfclrfc1157.txt [14] Traditional IP Network Address Translator (Traditional NAT)

http://www.ietf.orglrfclrfc3022.txt [15] The Point-to-Point Protocol (PPP) http://www.ietf.orglrfclrfc1661.html

DTV for Personalized Mobile Access and Unified Home Control

Jianlin Guo, Fernando Matsubara, Johnas Cukier, Haosong Kong Mitsubishi Electric Research Labs, 558 Central Avenue, Murray Hill, NJ 07974, USA

Abstract: Today, all devices are trying to connect and communicate each other. There is an increasing trend for self-configuring networks that allow devices to easily and automatically join and leave networks and learn about other connected devices. Digital television (DTV) is one of the most significant developments in recent digital technologies. It will become a natural center of the digital home because of its rich functionality and simplicity of use. In this paper, we present a home networking system to show the newly developed technologies that allow DTV and mobile phone to deliver complete new experiences to the end users.

In order to achieve a total and seamless connectivity, a mobile DTV (MDTV)­centric home network architecture has been proposed and implemented. First, a MDTV link protocol is designed and implemented in the home network system to connect DTV and mobile phone. Once connected, the mobile phone provides the wireless connectivity to Internet and also acts as an identifier for the personal data access, DTV is used as a displayer. Then, Universal Plug and Play (UPnP) protocol and Simple Control Protocol (SCP) are implemented to guarantee a total connectivity of all home devices. The enhanced DTV provides features of service discovery and ubiquitous home devices control, and web browsing. In the proposed home network system, the DTV is the central control point situated at home, users can use remote control at home or mobile phone away from home to access or control home devices. Our experiments have showed that the proposed home network system provides a simple and unified home device control and access. The MDTV architecture alone accomplishes an instant personalized mobile access anytime, anywhere.

Key words: DTV, mobile phone, personalized mobile access, UPnP, SCP, home networking, unified home control.

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

72 Jianlin Guo, Fernando Matsubara, Johnas Cukier, Haosong Kong

1. INTRODUCTION

In today's homes, TV is only for watching. PC is used to access the Internet for email, Web browsing, and other applications. Users can't watch TV and access Internet simultaneously because TV and PC are usually in different rooms and TV can't access Internet. The role of the TV has been changing since the arrival of the DTV. TV has its advantages over PC for home networking. It is where people spend most of their time, and most importantly, it presents a simple interface to all users. It is natural to predict that TV will be the center of the networked home. The goal of this paper is to present new technologies that allow DTV and mobile phone to deliver new values to end users in their networked home by fully utilizing the advanced features of the DTV.

DTV brings to TV much more capabilities than ever before. One of the most attractive ones is the Internet. The high quality DTV display can be utilized to navigate Internet, view pictures, read messages, etc. However, the question is how to easily connect DTV to the Internet without any complicated wires and devices, and more importantly, how to get instant access wherever and whenever by the most convenient means at hand. Today's TV and mobile phone work independently. Can TV and mobile phone complement each other by using TV's large screen, high quality display capability and mobile phone's mobility? This paper presents the MDTV architecture (Fig.I) that allows DTV and mobile phone working together to provide an easy-to-use, anytime, anywhere instant Internet access solution with no new wires.

Connecting home devices together is hard to many families, especially devices in different categories, for example, electric devices such as television and audio/video devices, and power line devices, such as lighting and air conditioner. Many of home devices are connected today to form isolated "micronetworks" by using different technologies [1]. USB is used to connect printers, MP3 players, and digital cameras to PCs. IEEE 1394 is used to connect video cameras to PCs. Ethernet connects PCs to broadband modems. A bewildering variety of analog and digital interfaces are used to interconnect DVD players, DVRs, DTVs, and digital set-top boxes with other audio and video components. How to interconnect these isolated "micronetworks" raises challenge to home networking. It is believed that a combination of technologies will be used in many homes and UPnP is a technology to bring these isolated "micronetworks" together. In this paper, we present a MDTV -centric home networking system (Fig. 3) to deliver a total and seamless connectivity. This system achieves a simple and

DTV for Personalized Mobile Access ... 73

ubiquitous home networking solution. DTV is the center of the proposed home networking system. The enhanced DTV provides features of service discovery, unified home device access and control, and web browsing. Users can use DTV remote control to access and control home devices, subscribe on-demand multimedia services such as movies and MPEG video, and access personalized information such as messages and pictures. The content provider can multicast the multimedia services to multiple homes, multiple home devices such as TV, PC, laptop, etc. When away from home, users can use their mobile phones to access and control their home devices. The MDTV -centric home networking system can easily be extended to include more devices that use different technologies.

2. A MOBILE DTV ARCmTECTURE FOR PERSONALIZED INSTANT INTERNET ACCESS

The MDTV architecture (Fig.I) consists of the MDTV server, the MDTV, and optionally Internet Service Provider (ISP). The MDTV is comprised of a wireless mobile device, a DTV, and a device adapter that connects the two. Both MDTV server and ISP are the content providers. The MDTV server, where user profiles are stored, is to provide user's personal information in a highly customized form according to user profile. The wireless mobile device communicates with the MDTV server via both a base station and the Internet. The adapter includes a MDTV Link protocol and an upstream protocol stack configured to generate IP data. The user uses DTV as user control interface as long as the DTV is connected to the wireless mobile device via the adapter. The MDTV architecture supports multiple upstream and downstream communications paths, not necessarily involving the same technology. User's personalized service requests are transmitted in the form of IP packets to the MDTV server over the already set-up wireless channel. In response to the IP data, the MDTV server unicasts the required service data to the DTV via the same wireless downstream or an alternate high-speed downstream communication path. For general services such as web browsing, the MDTV architecture automatically adapts to high-speed communication path if such path available.

In our implementation, the mobile phone is used as the wireless mobile device. A MDTV Link protocol is designed and implemented in DTV and the device adapter. The serial technology is used in current version of implementation and Bluetooth will be used in future version. In order to provide an easy-to-use solution, the physical interconnection between DTV

74 lianlin Guo, Fernando Matsubara, lohnas Cukier, Haosong Kong

and mobile phone is as easy as plugging the mobile phone into a DTV docking station. The MDTV supports intelligent programming functionality. When user plugs hislher mobile phone into the docking station an IP connection is established between the DTV and the Internet through the cellular wireless network. Once connected, DTV turns into an Internet TV. With a web browser being integrated into DTV, users can watch TV program and navigate Internet or access personal information simultaneously. The mobile phone is charged while plugged-in. When un­plugged, the mobile phone and DTV can be operated independently. User's personal information is automatically deleted from DTV. The proposed MDTV architecture provide end user with a mechanism to access hislher personal content anytime, anywhere in the home or hotel wherever a MDTV available.

DTV

MD1V SelVer ISP

Figurel. MDTV Architecture OvelView

To access personal information, content, or to activate other channels of services such as pay-per-view channels, the mobile phone is used as user identifier to conveniently enable user profile. The user then accesses personalized information and content through a customized interface. User can perform such access anytime, anywhere as long as a MDTV exists.

The user profile totally depends on user's preferences. For instance, some users might want to have personal pictures in their profiles and others might want to have personal messages in their profiles. The profile can be updated any time. When displayed on DTV, the MDTV content, such as messages and pictures, overlaps TV video. Users can choose either transparent or opaque background for the MDTV content. For example, user might choose opaque background for web browsing and transparent background for text

DTV for Personalized Mobile Access ... 75

messages. When MDTV server receives an instant message such as email for connected user a notification is sent to DTV immediately. DTV then notifies user by displaying an icon on screen or playing a special sound. User may download message and read it or ignore the notification. Fig.2 shows a message being displayed on DTV.

Figure2. MDTV Snapshot

The MDTV architecture enables end user to access to personalized information anytime, anywhere. It provides end-to-end connectivity between DTV and Internet services. User personal information associated with the mobile phone. No explicit dialogue is required in order to connect to the desired services. No requirement to boot the device or for installation of software by the user. No new wires are needed.

3. A MOBILE DTV-CENTRIC HOME ETWORKING SYSTEM

Many of the forces that lead to the creation of business networks, such as distributed data availability, cooperation, and the optimized use of resources, are becoming the requirements for the home. The office products have been modified for the home networking market [3]. Unfortunately, the home presents some novel and unusual challenges that have not been primary concerns in networking deployments until now [2]. For example, low cost, easy of installation, and extensibility must be considered. Most importantly, home networks are for less knowledgeable users. Users want simple, personalized, and integrated experiences. To be successful, the home networking deployments must satisfy user's requirements and rely on consumer products [3].

76 lianlin Guo, Fernando Matsubara, lohnas Cukier, Haosong Kong

The self-configuring networks that allow devices to easily and automatically join and leave networks and learn about other connected devices will greatly benefit end users. One of such technologies is UPnP. UPnP is an architecture for pervasive peer-to-peer network connectivity of intelligent devices. It is designed to bring zero-configuration, automatic device discovery, dynamic join or leave, standard-based connectivity to ad­hoc or unmanaged networks in the home or other spaces. UPnP is a distributed, open networking architecture that leverages TCPIIP and web technologies to enable seamless proximity networking in addition to control and data transfer among networked devices. SCP, architecturally an extension of UPnP, is a PHY -independent device control protocol that allows the manufacturers to make small and intelligent devices for limited bandwidth networks such as Power Line Carrier (PLC) networks. It is a lightweight protocol to provide peer-to-peer, event-based control of devices in the home. SCP aims at providing an inexpensive solution due to its simple architecture, small memory footprint, and using the existing infrastructure such as electric wiring.

TV is the center of home entertainment. It presents a simple and familiar interface to all users. TV -centric home networks have unique advantages and great potential in home networking market. Based on MDTV architecture, UPnP and SCP technologies, this paper has proposed and implemented a MDTV -centric home networking system (Fig.3) to provide a mechanism for the unified access to, and control of, home devices.

ISP Figure3. MDTV -centric Home Networking System Overview

The proposed home networking system consists of UPnP domain, 1394 domain, PLC domain, and wireless domain. In the implementation, the

DTV for Personalized Mobile Access ... 77

UPnP domain is comprised of a DTV, a security camera, a multimedia server and a PC. DTV is a UPnP device and a UPnP control point. PC can optionally serve as another control point. The SCP technology is used to network the power line devices that include a lighting and an air-conditioner. IEEE 1394 is used to connect DTV to a DVD player. DTV is a central control point of all devices. DTV bridges PLC domain and 1394 domain to UPnP domain. It initiates and runs a proxy device in the UPnP network for every device in the PLC network and 1394 network. The proxy device in the UPnP network represents its counterpart in PLC network or 1394 network. The proxy devices participate in UPnP activities. The bridge, that is DTV, translates UPnP commands and queries received by the proxy device into the corresponding messages in the PLC network or 1394 network, which are in turn passed to the target device via SCP or 1394. Thus, all devices in the PLC network and 1394 network are visible in the UPnP network, and can be controlled by DTV or other UPnP control point. The mobile phone is only device in wireless domain. The MDTV Link protocol is used to connect mobile phone to DTV. Once mobile phone joins system, it provides home network system with the wireless Internet connectivity. If there is already a high-speed connection, for example via a broadband modem, between home networking system and Internet this wireless link will only be used as an up stream link for personal data access and on-demand service. The high-speed connection will be used for the downstream path. User simply uses DTV remote control to access and control home devices, subscribe on-demand multimedia service, access personalized information and navigate Internet at home. User may also access or control home devices while away from home. Using mobile phone, user can send command to the MDTV server. The DTV (set-top box which is always live) pulls command from MDTV server periodically. If the command exists DTV then executes command and sends the result back to the mobile phone. In this case, home networking system needs a connection to the Internet or smart phone system. The proposed home networking system achieves a simple and ubiquitous home device control and access solution. Fig.4 shows devices that are currently connected to the home networking system.

The MDTV server stores user profile, customizes user's personal data, and provides on-demand multimedia services such as standard video, MPEG video, and HDTV video. The MDTV server supports multicast functionality. This allows multiple homes to share the resources without wasting network bandwidth. User may request multimedia services by using mobile phone as an identifier. The MDTV server may intelligently multicast multimedia content across Internet to multiple homes, multiple home devices such as

78 lianlin Guo, Fernando Matsubara, lohnas Cukier, Haosong Kong

DTV, PC or laptop, etc. The rate adaptation capability and flexible multicast functionality have been integrated to handle network and user

Figure4. MDTV -Centric Home Networking System Snapshot

4. CONCLUSION

The technologies have begun migrating to the homes from the offices. It presents tremendous opportunities for the home networking deployments. The MDTV -centric home networking system is designed to bring user a new mechanism to access and control home devices in the home, in the car, in the office, and virtually everywhere. It provides a total and seamless connectivity, and a unified access to, and control of, devices regardless of the technology used. It also presents a simple and familiar interface to all users. Our implementation demonstrates the effectiveness of a unified control over different types of technologies. The technologies presented in this paper will greatly increase the market opportunity for both home networking and DTV itself, will also provide more opportunities to service providers, and most importantly will benefit consumers. We will continue to improve our technologies by extending the device scope, enhancing security, and utilizing more advanced technologies to deliver high values to the consumers.

5. REFERENCES

[1) Sandy Teger and David J. Waks, "End-User Perspectives on Home Networking", IEEE Comm. Magazine, April 2002 vol. 40 No.4

[2) Dave Marples and Stan Moyer, "In-Home Networking", IEEE Comm. Magazine, April 2002 vol. 40 No.4

[3) Bill Rose, "Consumer Requirements for Home Networks", Proceedings of 2002 IEEE 4th International Workshop on Networked Appliances, January 2002

A Novel Internet Radio Service for Personal Communications; The Private Channel Service

Kensuke Arakawa Faculty of Software and Information Science Iwate Prefectural University araken@comm.soft.iwate-pu.ac.jp

Yasushi Ichikawa Graduate School of Software and Information Science Iwate Prefectural University ichikawa(!!comm.soft.iwate-pu.ac.jp

Yuko Murayama Graduate School of Software and Information Science Iwate Prefectural University murayama@iwate-pu.ac.jp

Abstract The World-wide Web(WWW) works now as the infrastructure over

the Internet for multimedia applications. Internet Radio is one of those applications, and provides the users with streaming services over the network regardless of such geographic restrictions as the traditional ra­dio broadcast service systems have. Its growth is explosive. We have started operating an Internet Radio station with streaming music since April 2000. We are planning to provide users with private channels, so that users can listen to their favorite music. This novel type of service is only possible on the Internet, but not on the traditional radio systems. This paper reports our idea on a private channel, a novel service on the Internet Radio, and its design as well as our implementation of a prototype system.

Keywords: Internet radio, personal communications, music delivery service, Private channel

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

80 Kensuke Arakawa, Yasushi Ichikawa and Yuko Murayama

1. Introduction It broadcasts over the Internet regardless of such geographic restric­

tions as the traditional radio broadcast services have. There are more than 5000 Internet radio stations. One of these stations: http://www.woljJ.m.com.

deals with the various types of streaming such as MP3 streaming and Windows Media streaming. This site provide users with a shopping function as well so that users can purchase CDs of their favorite music.

The existing stations provide their services in such a way that a radio station provide their listeners with programs on one or more channels. It is very much like the way that the traditional radio broadcast services do. In the case that a station provides more than one channel, the ser­vice is called "multi-channel," and for example, some stations broadcast different types of music per channel.

We propose a novel type of service, what we call a private channel on which a station provides each user with her Ihis favorite program. Since the station unicasts streaming data to each user site, it is possible for a station to deliver different streaming data to each user site. The idea is that a user creates a program of channel in terms of the list of music for the station to play - a play list.

As our radio station, which has started operating since April 2000, delivers music from unknown artists, we presume that our private chan­nel service is for a user generate a personal channel which delivers her his favorite music stream.

This paper reports our idea on a private channel, a novel service on the Internet Radio, and its design as well as our implementation of a prototype system. The next section introduces our radio service system with its operational model and the system configuration of the station. Section 3 gives our idea on the private channel service with its model, the system design, and its prototype system. Section 4 describe future work, and Section 5 gives some conclusions.

2. Flip Over Radio(FOR) Our radio station is called "Flip Over Radio (FOR)." The station is

managed by the university students. At the moment we operate FOR on an experimental basis. We broadcast Indies music which is made originally by unknown artists who work independently from record com­panies. They have a limited opportunity in publishing their music such that the listeners can obtain the information only from specific maga­zines and music stores in Japan.

A Novel Internet Radio Service for personal ... 81

Listeners I+-IC..aI.IILIiO:cII~ ~~wt:.=~~ Artists

Figure 1. The operational model of FOR

2.1. The operational model Fig. 1 shows the model of our radio operation. Our Internet radio

station provide such an opportunity for both artists and listeners to exchange the information on music and artists. Our radio site is a media for this exchange. The artists provide the music that they composed and played as well as the related information. We provide them with tools such as the one to make their home page as well as the message board so that they can communicate with the listeners. Commercial promoters could make use of the information we provide to find a new artist and music, so that an artist could have an opportunity to get a commercial contract.

2.2. The system configuration

Table 1 shows the configuration of the server system of FOR.

CPU Pentium2 333MHz MEMORY 80M

OS Kondara Linux HTTPserver Apache Streaming Icecast [5J

Application Shout [5J Icedj [7] Liveice [8J

Table 1. The server system of FOR

82 Kensuke Arakawa, Yasushi Ichikawa and Yuko Murayama

-1M .It

I\. - I "-\81 -- \/ --- r \I. .. ...., 11M .. .A.. I at '- ,-..... I -• • I • l • II 10,. 12 I t I 41 I • 1 • II III,. U I I I 4 I - - -

-1-

Figure 2. User access per month (2000 - 2002)

lcecastis used to broadcast music. Shoutselects a music to broadcast, and passes the music data to lcecast. Liveice is a real-time re-encoder and passes the encoded data to lcecast. We can mix several MP3 streams and audio inputs from mic(microphone) and Liveice. lcedjis used to run an I cecast radio station such that broadcasts a music at a certain time as scheduled in a program. It can be used together with Icecast to show the information on music being broadcast on the radio station's WWW page.

3. The Operation Report

We have been operating FOR since April 2000. Fig. 2 shows the number of total user access per month. We have not had so many users, presumably it is not because of Indies music, but due to poor amount of contents. Users would not listen to an Internet radio station if it broadcast the same songs repeatedly. During July and August in 2001, we revised icecast in the latest version, so that the facility of the regis­tration function started working well, which registers our radio server to the access ranking server on the Internet. 3 percent of connections were from our university, and 20 percent of connections were from Japan.

4. The Private Channel Service

4.1. The operational model

Just as we give a copy of our favorite music on a tape cassette to some friends, we include the case that one may want to share herhis private channel with some others.

We propose our private channel service in three different ways, the very private channel, the shared channel and the public one. The very

A Novel Internet Radio Service for personal ... 83

private channel is used only by a user who generate the channel - the channel owner. The shared channel can be used by a group of users whom the channel owner give an access to the channel. The public channel is allowed anyone to listen to.

The operational model of our private channel include the followings four participants: an Internet Radio operator, a music sound originator, the private channel owner and a user. A music sound originator provides an Internet Radio operator with the original music sound as well as its copy right for delivering it to the users. The operator delivers it to the users in turn. The users give the music sound originator some feedbacks. Those feedbacks can be used for other users to select music possibly for a private channel.

On the shared channel, the channel owner is a user as well as a service provider together with the Internet Radio operator, because the user gives herhis channel information - the play list of music to some others based on the access control list; the channel owner generates the list and manages it. The others who are given an access to the private channel can listen to the music played according to the play list. Such users could give the channel owner some feedbacks on the play list as well as on a particular piece of music; the latter could be passed on to the music sound originator.

Like the channel owner, those users give the Internet Radio operator the information on the music they have chosen as well as the channel they have listened to.

In case of the public channel, there is no control over the channel access as the channel is publicly available. Other uses than the channel owner may well tell the owner their willingness to recommend the chan­nel to their friends; this may rank the channel. Accordingly this public channel looks like an Internet Radio station operated by an individual, however, the difference is that the former does not require the user to deal with copy rights and all that associated with the original music sound. From this viewpoint this type of private channel is best suited for a user who likes to set up herhis personal radio station easily.

4.2. System design

An Internet Radio system is based on the server and client architecture with a station as a server and a user site as a client. The server has the functions for streaming distribution and delivery, channel generation, and music selection. Moreover, it maintains the information on the use of a channel as well as music sound database. It provides users with a WWW interface for access to a private channel. We design the system

84 Kensuke Arakawa, Yasushi Ichikawa and Yuko Murayama

Figure 3. the system configuration

so that the server provides a user with a private channel. The client system has the functions for receiving music stream and playing it.

A user who would like to own a channel passes the information such as the server the user's name, the private channel name, channel type and the play list. Making a play list, the channel owner may well look up the whole list of music maintained by the server. The more music the server has, the more difficult for the channel owner to find the one shehe likes. In this case, the server may recommend some to the owner based on the hit chart as well as the list of music recommended by the other owners and users who have the similar tastes. In case of the shared channel, the owner needs to specify the access control list for the channel.

The users including the channel owner get the URL of a private chan­nel from the server, comes to the URL site - i.e. the private channel, receive music stream and play it on the client system.

4.3. A prototype system

We implemented a prototype system, and the figure 3 shows the sys­tem configuration.

With our prototype system, a user can generate a private channel. The information on the users and the music sound originator is main­tained as a database using PostgreSQL. We use PHP4 for a WWW interface.

There is a problems with this service. If we provide users with private channels on demand, we will require to run as many private channel

A Novel Internet Radio Service for personal ... 85

processes as the number of user requests. The more private channel processes we have, the more loads the server gets and the slower the system operates. We may need to explore the tradeoff between the performance of the server and the number of private channels. We may well need dynamic channel management.

5. Campus Radio with a Wireless LAN

Some universities have the campus radio stations with the traditional radio wave, run by students. They are broadcasting towards the campus areas and have their studios located within the campus. Their listeners include students, teaching staffs, and those listeners use traditional radio tuners.

Our radio station, FOR is an Internet Radio, but has system structure similar to a the traditional campus radio. One needs a PC and the a network connection to the Internet instead of a radio tuner, to listen to FOR. Within the campus, a listener needs a LAN connection such as the Ethernet.

Indeed one could use various system nowadays to listen to the Internet Radio, such as a PDA, a Handhelds computer and PC. We would need the plural wireless LAN access points in a campus so that listeners have wireless connections throughout the campus area.

6. Related Work

MP3.com,Inc. [9]provides users with a service called my music which is similar to our private channel service. The service lets a user to create a play list on the mp3.com web site. The user downloads the play list in which each entry of music is specified in terms of the URL of the mp3 file. A connection to the server from the user site has to be renewed each time a music file is downloaded,whereas in our system the initial connection is used throughout the operation of a play list.

MP3.com, Inc. is not a radio station, but offers a transfer service for mp3 files. In our private channel service, the play list stays on the radio station site.

They do not provide the shared channel service. This is presumably due to the difference in the viewpoint of the user between the file transfer service site and the Internet Radio station. Just as one gets a music CD from a store, the primary purpose of the user is to get a music file in the MP3.com service. There is less desire for a user to share the CD which the user get, with the other users. On the other hand, on Internet Radio users get into a radio station site and are likely to share their interests in music.

86 Kensuke Arakawa, Yasushi Ichikawa and Yuko Murayama

7. Conclusion We introduced on our Internet radio station and report its operation.

We presented our idea on a novel service, a private channel, on the Internet Radio in terms of model, system design and a prototype system.

Future work includes implementation of the other two types of private channels, the shared channel and the public channel. Moreover, a hit chart based on the users' feedbacks could be useful for the channel owner in selecting music. We may look into some visual interface for this, and incorporate CSCW filtering tools for recommending music based on the use history. Kansei database could be used as well.

References [1] Flip Over Radio : http://radio.comm.soft.iwate-pu.ac.jp last access : Jun. 30,

2002 [2] Ponce Bob , WWW Artists Consortium , The Impact of MP3 and the Future of

Digital Entertainment Products, IEEE Communications MAGAZINE September 1999 Vol.37, No.9, pp.68-70 (1999)

[3) Real system: http://www.realnetworks.com/

[4) Shoutcast: http://www.shoutcast.com

[5) Icecast: http://www.icecast.org

[6) Fraunhofer Research: http://www.iis.fhg.de/amm/

[7) Icedj : http://www.remixradio.com/icedj/

[8) Liveice: http://star.arm.ac.ukrspm/software/liveice.html [9) MP3.com,Inc.: http://www.mp3.com

[10) Alexandra Uitdenbogerd and Justin Zobel, Melodic Matching Techniques for Large Music Databases, ACM MULTIMEDIA '99, pp.57-66 (1999)

[11) JMF: http://www.java.sun.com/products/jave-media/index.html

BUFFER CONTROL /RECEIVER

Tools for On-Door Communications on WWW

Keishi Suzumura Graduate School of Software and Information Science Iwate Prefectural University kc@comm.soft.iwate-pu.ac.jp

Hiromi Gondo Faculty of Software and Information Science Iwate Prefectural University gon@comm.soft.iwate-pu.ac.jp

Yuko Murayama Graduate School of Software and Information Science Iwate Prefectural University murayama@iwate-pu.ac.jp

Abstract In this research we try and implement communication systems using the metaphor of a door on the World-Wide Web (WWW) as a media for informal communications. We call those informal communications through a door "on-door communications." We presume a scenario such that we would visit our firend's room in a student hall of residence; we would knock on the door and have a chat with the friend, and may leave a message on a message board on the door in case of the firend's absence. Accordingly we introduce two tools for such communications. One is a chat system with some awareness features, viz shadow and auditory sig­nals, and the second one is a message board for a handwritten-message exchange. When users visit a net door site, their shadows would appear on the door. A user would knock on the door to notify the others that the user would like to start a chat. The other is a whiteboard system for asynchronous communications, in which each handwritten messages are coded as a set of lines. We incorporate a feature for time visualization into the message board, so that messages in past would be fading out gradually. This paper reports on our design and implementation issues of those two systems.

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

88 Keishi Suzumura, Hiromi Gondo and Yuko Murayama

Keywords: WWW application, multimedia application

1. INTRODUCTION

Figure 1 The Model of On-Door Communications

The motivation of this research is based on the experience of an exten­sive use of a message board on a door in one of the postgraduate student halls of residence. It was meant to be a message board for friends to leave messages only to the owner of the room. On the contrary, due to the fact that the room with the board was located in front of an elevator entrance, the board became soon to be known to many and used exten­sively as a message exchanging media by the other owners for many to many asynchronous communication with anonymous writers. The mes­sages on the board included more of the entertaining content such as a joke and a quiz game enjoyed by many than the personal communication.

The objective of this research is to bring such communication systems using a door as a media into the WWW environment, and to examine the difference from the ones in the real world. Our ultimate goal is to observe the use of those systems and investigate interesting and novel applications.

We introduce two systems. The first one is a system for a chat at a net door, with two features for awareness, viz. shadow and auditory signal such as a knock sound. When users visit a net door site, their shadows would appear on the door to let the others to notice of the arrival of a visitor. A knock on the door would be used to notify the others of one's intention to start a chat.

The second one is a message board for asynchronous communications. Just as the one in the student hall, we designed and implemented it on WWW, which provides users with simple tools for drawing. Letters are coded as a collection of lines. On this board, any message can be written by hand making use of a mouse and a tablet.

Tools for On-Door Communications on WWW 89

This paper reports on our implementation of those two systems. We present the design of our systems as well as the implementation of the prototype systems. Finally we report on our discoveries on the difference from the real one in terms of a system structure, and discuss the issues to be dealt with in future.

1.1. AN OVERVIEW This section introduces a net-door chat system. Two awareness tools

are provided, knock and shadow. A knock on the door is used to let the others to notice of the arrival of a visitor at the door with an auditory signal. A shadow indicates the existence of visitors as well as that of the door owner.

The idea of knock is to let the receiver notice of the sender's urgent wish to communicate by audio signal, just as telephone bells and various types of the auditory signal for email arrival[?]. Our system provides users with a simple tool for knocking with a click using a standard input device such as a mouse. A click produces a sound of a knock on the door, so that anyone who has been opening that door page can hear. A chat system is provided for further message exchange.

Shadow works as follows. When the owner is in the room, the light can be seen through the door window as well as the owner's shadow. A visitor gives a knock on the door and starts chat with the door owner. If the owner is not available but any other visitors are out there, their figures can be seen outside the door, so that the new visitor can join the others in chat. If no one is around, the visitor can leave a message on the chat board. When a visitor comes across at the door, his/her shadow is generated outside the door so that the owner and the other visitors would notice of the new comer.

1.2. THE PROTOTYPE SYSTEM We used JAVA for coding. The server is a JAVA application and

a client is provoked as a JAVA applet. The server can deal with as multiple doors. The client side door knock system interface is written in the HTML, and can be used on the WWW. The visitors need WWW browser. After setting up the applet, the visitor inputs his/her handle name. The handle name is used to indicate the writer of the messages.

Starting of an applet of a caller or owner generates a shadow. Click­ing the image of a door with a standard input device such as a mouse produces a knock. A client produces a sound of a knock on the door, so that anyone who has been opening that door page can hear. The knock sound is prepared in the Audio File Format (AU) and sent to a user

90 Keishi Suzumura, Hiromi Gondo and Yuko Murayama

site as an initial data together with the JAVA applet for a client in the beginning of the communication. When an owner has connected with a door, his/her shadow appears on the door window, whereas the shadow of visitor appears in the foreground of a door. Fig. 2 shows a door on WWW at a client system.

Figure 2 Knock-on-the-Door Communication system on WWW

2. THE ON-DOOR MESSAGE BOARD SYSTEM

2.1. AN OVERVIEW A message board on the door has the following characteristics: l.

Messages on the board are short and written by hand. 2. It provides asynchronous communication. 3. Anyone can read and write; anything can be written on the board, including a message addressed to the others than the owner. 4. The board is write-only and has no eraser for the ordinary users; this is merely due to the fact that the original message board in the student hall did not have any eraser. 5. There is no authentication and users stay anonymous. 6. There has to be a way for the others to know of the existence of the message board; i.e. a feature of awareness is needed. The motivation for anonymity, 5 in the above, is this. The original message board in the student hall was used in such an extensive way that not only for friends left a message to the owner of the room, but also some other owners started leaving messages presumably to the room owner but anyone else enjoyed reading and replied. The writers stayed anonymous with code names such as Doracula and the board worked as an entertaining media for many-to­many communications.

Tools for On-Door Communications on WWW 91

2.2. VISUALIZING TIME

We added a feature so that old messages are presented in a faded color according to the time when they were written.

We locate this function at the client sites, so that it will be left for users to control fading as they like; some may be interested only in the messages written in a few days, whereas other may like to see the message written in a few weeks. A client maintains a cache of drawing information with time stamps, which has been obtained from the server. The client provides a user with a function to control the lightness of the color of the messages according to how long ago they were written. Fig.3 shows an example produced by our prototype system. The messages are washed out gradually according to the time they were written, and those written before the time user specified are faded out.

Figure:1 Fading messages away

Those colors are expressed in HSB(Hue Saturation and Brightness) and we control the saturation value to fade a message as follows. A mes­sage is composed of a set of LINE objects. Each LINE object has been time-stamped by the server. The color information of LINE objects is transformed into HSB from RGB at a client and the saturation parame­ter is recalculated according to the time stamp of a LINE object. Users specify the time duration to change the saturation. If one specifies a few days as the time duration, only those messages which were written in a few days would appear in a faded fashion. Those which had been written before the few days ago would disappear.

2.3. THE PROTOTYPE SYSTEM We used Java for a server, and Java Applet for a client. A owner, the owner of the board, has a link to the message board

server on his/her home page on WWW. Through the link to the server, client systems are generated automatically with the use of Java Applet. A server and a client use our original connection-oriented protocol, the On-Door Communication Protocol(ODCP) for exchange of drawing in­formation. When client system starts operating, it sends the server a

92 Keishi Suzumura, Hiromi Gondo and Yuko Murayama

request for an ODCP connection. In practice, the server can have as many connections as possible as long as its system resource is available.

Figure 4 The owner's Web page and the message board

Figure 4 shows the message board attached by the control panel the pen selection. The size of the current message board is fixed as 1000 x 1000pixels. A user selects a pen with the favorite color and thickness, and start drawing with an input devices such as pentablet and a mouse. When a user starts drawing, the client system starts sampling points of a line. Sampling is performed whenever the client system gets a mouse event which is a standard function of JAVA. Many users requested that they would like to own their own message board; some users requested that they would need multiple message boards for different purposes. With the current prototype system, a server can provide users with more than one board.

2.4. DISCOVERIES AND FUTURE WORK There are several differences between our message board on WWW

and the one in the real world. As we put it in one of the designed features of the system, the aged messages could be faded away. This feature of time visualization would not be possible on the board in the real world. Moreover, in the real world, the owner of the board manages the board, whereas in our system, the owner of the board is not necessarily the manager of the board server.

Writing figures could give us some indication of writers. We investi­gated an algorithm to deal with security issues on how one can guarantee that a user who leaves a message stay anonymous. The idea is to move some drawing points so that the written message would have no more individual writing characteristics in handwriting, however, it would be still readable. This sort of modification would not be possible with a message board in the real world. We shall also start looking at the use of the LCD panel on the real door of a room, which will display the web page on the network as well as to let the users to write the messages.

Tools for On~Door Communications on WWW 93

Moreover, we have started connecting a real whiteboard with a special pen holder and a sensor which detects the movement of the pen holder. The sensor is controlled by a PC on the network, so that messages writ­ten on this board appear on our message board on WWW.This will enable people who had no access to the Internet to leave messages on our system.

3. RELATED WORK

3.1. THE CHAT SYSTEM WITH AWARENESS FEATURES

Ubique[l] is a system that aims to create virtual places on the WWW. This system has such functions as "who is on-line", chat and instant message. The system also dose not use auditory signal in chat.

CyberWindow[2] implemented the metaphor of a window on WWW, and interconnects the real world and the WWW environment with sound awareness. In this system, users in the real world carry hand held com­puters with a wireless connection to the network. The visitor uses one of the three types of message volume: "whispered", "said" and" shouted" . The received messages are transformed into voice messages by means of a text-speech engine. A visitor can also show his/her emotions to the others by using various types of sound; laugh, applause and knock. In our system, not only awareness of sound but also visual awareness is given to a visitor.

3.2. MESSAGE BOARDS

The idea of time visualization is nothing new. The concept of aging messages was explored by Seligmann[4],[5]. Their Metaphorium project produced various messaging systems for novel types of user interaction and text less information navigation. In one system, a bulletine board is a beach, and a message written on the sand in a virtual island would be removed when a wave washed it away. In another system, a message written in the virtual sky with smoke would be faded away gradually in a few minutes. The idea with those systems is the aging process is supposed to be done by the nature although in the system it is done randomly. In any case it is not controlled by a user. The purpose of our time visualization is for a user to select the recent ones out of all the messages written on board. The user has a control on fading.

Flatland by Mynatt[3] is a computer-augmented whiteboard system for asynchronous communication, however, it is designed to support in­formal work in a personal office environment in which the board owner

94 Keishi Suzumura, Hiromi Gondo and Yuko Murayama

leaves some memo and ideas. Time visualization in their use is originally designed to see some memo and ideas in past, whereas the one in our system has a different objective to see all the messages on the board at a glance with their freshness; that is why they are washed out gradually rather than a snapshot.

4. CONCLUSIONS We reported the design, implementation and operation of the system

as well as some issues to be dealt with in future. The differences from the board in the real world, include time visualization and the need to structure of the management. We shall work on sorting out those issues as well as to try and extend the overall communication environment to include the real world, by making use of LCD panels. Since our system can work with client systems with WWW access and the JAVA environment, it could make use of PDAs and mobile phones as client systems.

In the future we would like to distribute our software widely and investigate interesting applications of our system. We started working on other types of on-door communication systems such as under-the­door information passing system as well.

References

[1] K. Scott: Ubique's Virtual Places: Communication and interaction on the World Wide Web, Workshop on WWW and Collaboration, pages 16-20, September (1995)

[2] O. Liechti, M. Sifer, and T. Ichikawa: Supporting Social Awareness on the World Wide Web With the handheld CyberWindow, CSCW'98, (1998)

[3] E. Mynatt, T. Igarashi, W. Edwards and A. LaMarca: Flatland: new dimensions in office whiteboards, pp. 346 - 353, Proceeding of the CHI 99 (1999)

[4] D.D. Seligmann, C. Laporte and S. V. Bugaj: The Message Is The Medium, Proc. of the 6th In-ternational World Wide Web Conference (1997) http://www.scope.gmd.de/info/www6/technical/paper119/paper119.html

[5] D.D. Seligmann and S. V. Bugaj: Live Web stationery: virtual paper aging Proc. of ACM SIGGRAPH 97 Visual Proceedings: The art and interdisciplinary programs of SIGGRAPH97 pp.158-158 (1997)

Buffer Control using Adaptive MQAM for Wireless Channels

Anh Tuan Hoang and Mehul Motani Institute for Communications Research, Electrical and Computer Engineering Department, National University of Singapore, Singapore 119260. engpI629@nus.edu.sg, motani@nus.edu.sg

Abstract: We consider the problem of buffer control in wireless communication using adaptive transmission techniques. In our model. a transmitter at the physical layer has a buffer with fixed length and bits arrive into the buffer according to a Poisson distribution. A server takes bits out of the buffer. maps them into M-ary quadrature amplitude modulation (MQAM) symbols for transmission through a wireless link. We assume that the server can vary the data rate by changing the constellation size of its MQAM modulator. Our objective is to maintain the reliability of transmission. with respect to the required bit error rate (BER) at the receiver and the required buffer overflow probability (BOP). while minimizing average transmit power. We obtain the solution to this problem using dynamic programming and use simulation to show that the obtained solution provides a considerable gain in terms of average transmit power relative to less-adaptive transmission schemes.

Keywords: buffer control, adaptive modulation. dynamic programming.

1. INTRODUCTION

In mobile wireless communication, guaranteeing performance in terms of delay, loss and throughput is an important issue. To do so usually requires a trade off between quality of service and scarce system resources such as bandwidth, buffer space and transmit power. This motivates us to look at the problem of buffer control for wireless communication.

We consider a simple server in which bits are mapped into transmitted symbols using M-ary quadrature amplitude modulation (MQAM). Modulated

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

98 Anh Tuan Hoang and Mehul Motani

symbols are transmitted at a fixed rate through a wireless channel. Suppose that for reliable communication, a certain received bit error rate (BER) is required. If we want to increase the rate at which bits are transmitted from the buffer, the signal constellation size needs to be increased. This will result in more transmit power needed for maintaining the required BER. Of course, how fast bits are transmitted from the buffer affects the buffer overflow probability (BOP) and the objective is to maintain a certain BOP and BER while minimizing average transmit power.

Our research is inspired by the works of Berry and Gallager [1] and Goldsmith and Chua [2]. In [1], the problem of adapting transmission rate and power based on channel and buffer states under queuing delay constraints has been studied. Concentrating on the regime of asymptotically large delay, the authors have characterized simple buffer control policies which exhibit optimal performances. In [2], a variable-rate, variable-power MQAM modulation scheme for high-speed data transmission over fading channels has been proposed. By adapting the signal constellation size to the channel condition, the authors have shown that there is a 5-10 dB power gain relative to less adaptive transmission schemes.

What makes our work different from [1] is that we study the problem in a more practical context in which the buffer length is reasonably small and the data transmission rate is controlled by an adaptive MQAM modulator. What makes our work different from [2] is that we take into account both channel state and buffer occupancy. We use dynamic programming [3], [5] to solve for the optimal adaptive transmission policy and show, by simulation, that the policy obtained gives us a performance gain over less-adaptive schemes. We first solve the problem for an A WGN channel and then show that the policy obtained can be extended for a fading environment.

Our paper is organized as follows. In section II, we give the system model and briefly describe how the server works. In section III, the optimization problem is defined. In section IV, which is the main part of this paper, we present the dynamic programming approach and a simple algorithm that can be used to search for the optimal adaptive policy. In section V, numerical results and discussion are provided. Finally, we give a conclusion for the paper and indicate potential avenues for further research.

2. SYSTEM MODEL

The system model in Figure 1 represents a mobile terminal transmitting data traffic through a dedicated wireless link. We first start with a simple A WGN channel model. Later, we will argue that the solution obtained can be extended for fading environment. We define the following parameters:

A is the average arrival rate (bits/sec), the arrival process is Poisson; L is the buffer length (bits); M is the signal constellation size of the MQAM modulator; Ts is the symbol period (sec); W is the channel bandwidth (Hz) and; d is the power spectral density of the A WGN channel;

Buffer Control using Adaptive MQAM ... 99

Buffer

Pooson -31 \ II H'-------r--'se~r H'------r-----'~~; H L-......,----JReceiw, [---+ & I ~ I : I

, I ' L _ .. _ .. ___ ______ .. _..1 ..... __ _____ . __ __ .. ____ J

C hannellBuffer Stale Infannalion

Figure I. System model of a mobile terminal transmitting data over a wireless channel.

We assume the transmitter functions as follows. The time axis is divided into periods of length Ts. During the !!' period, bits arrive at the buffer and queue up. At the end of this period, a decision is made on how many bits should be mapped into the QAM symbol $k for transmission in the next period. During the (k+llh period, while the symbol $k is being transmitted, new bits arrive and queue up in the buffer and the process is repeated.

3. PROBLEM DEFINITION

Our objective is to minimize the average transmit power while maintaining the required BER and BOP. Since it is often difficult to get a closed-form expression for the BER and BOP, we use upper bounds and numerical techniques instead. As mentioned above, we ftrst start with an A WGN channel model.

3.1 Maintaining Required BER

In [4], for MQAM modulation and ideal coherent phase detection, given the received signal to noise ratio (SNR), the BER for an A WON channel is bounded by

BER ~ 2e -1.5SNR I(M -I) (1)

It is also stated in [2] that for M ~ 4 and 05 SNR 530 dB, a tighter bound forBER is

BER ~ O.2e-1.5SNRI(M-l). (2)

Without loss of generality, we assume that the noise power is unity. From (1), the transmit power needed for transmitting i bits of information while guaranteeing certain value of BER is

Pw(i) ~ log( -Q.5BER)(2' -1) 11.5. (3)

Similarly, from (2), we have

PW(i) ~ log( -5BER)(2 i -1) 11 .5. (4)

100 Anh Tuan Hoang and Mehul Motani

In this paper, we will use (3) and (4) to approximate the power needed to transmit i bits of information using MQAM.

3.2 Maintaining Required BOP

The buffer can be represented by a discrete-time Markov model in which a state i (i = 0, 1, ... L) means there are i bits in queue at the end of a symbol period. Let r = [rd be a policy that transmits rj bits when the buffer is in state i. Clearly, we must have ri less than or equal to i. The transition probability from state i to state j is

Pij = Aj-i+rI for (i - r;) ~ j < L. £-/+ri-1

PiL=l- LAk for j=L. t=o

Pu=o otherwise. (5)

Here Ak is the probability that k bits arrive at the buffer during a symbol period. For an arrival process with a Poisson distribution

At = .!..eAT• (AT.)k . k!

(6)

Let Pi (i = 0,1, ... L) be the steady state probability that there are i bits in the buffer and p = [pJ. Also, let P = [Pijl be the transition probability matrix. The quantity p can be calculated from

p=pP or p(P-I) =0,

with I and 0 are the LxL identity matrix and Lx] zero vector respectively. The steady state buffer overflow probability is

L ...

Po! = L(pi LA). 1=0 j=L-I+l

3.3 Optimization Problem

(7)

(8)

Using all of the above derivations, our optimization problem can be defined as follows:

i=L

minimize over all policies r: Pavr = LPw(n)pl, 1=0

subject to: Pof~BOF (9)

Due to high complexity, we cannot solve for a closed-form expression for r. In the next section, we propose an algorithm that can be used to search for r numerically.

Buffer Control using Adaptive MQAM ... 101

4. DYNAMIC PROGRAMMING APPROACH

4.1 Average Cost Problem

Instead of solving (9) directly, let us consider a related problem of finding a policy r which minimizes a weighted sum of the average transmit power and the buffer overflow probability (given the same constraint on the BER):

minimize over all policies r: C(f3) = P av, + fJP of . (lO)

Here /3 is a positive scalar. Let r* be the transmission policy which minimizes (10) for a certain value of /3 and Pav,* and Pof* be the corresponding values of the average transmit power and the buffer overflow probability. It is easy to see that Pavr* is the minimum average transmit power needed to guarantee that the buffer overflow probability is not more than Pol. So for each value of /3, by solving (10) we obtain a point on the optimum power-BOP curve.

The optimization problem of (10) is called an average cost problem. With finite numbers of buffer states and control policies, there is always a stationary solution to (10). In fact, (10) can be solved using dynamic programming techniques [3], [5].

4.2 Proposed Algorithm

The basic idea behind the dynamic programming approach is to convert our optimization task into a shortest path routing problem. This can be done by considering each buffer state as a node in a graph and assigning a cost to each transition from one node to another. The cost here is the weighted sum of transmit power and probability of buffer overflow. By using various shortest path routing algorithms that are available in the literature, we can find the optimal solution to (10).

Here, we propose an algorithm that searches for the policy r that minimizes (10) over a period of length T. For a mobile terminal, T can be thought of as the length of a particular active period. We usually have T to be quite large; it is of the order of thousands of the symbol period Ts. A shortest path search is carried out in the reverse direction of time, starting from t = T and going back until t = o.

We initially set t = T and set the cost of each node c,JT} = 0, k = 0, 1 .. .L. At time t, suppose that the buffer is in state k and we decide to transmit i bits. The incurred cost is the sum of the transmit power, PJi}, the overflow probability when there are k - i bits in the buffer, (scaled by /3),

and the expected costs that have been accumulated up to time t + 1, L

L P(k - l)jCj(t + 1) . j=O

102 Anh Tuan Hoang and Mehul Motani

At each time t and for each buffer state k, the number of bits being transmitted i should be chosen so that the incurred cost is minimized. The proposed algorithm is:

Algorithm Parameters:

r(t) = (rolt). rit) •... rdt)) is the transmission policy obtained at time t. ,.., is the final policy obtained by the algorithm. e(t) = (colt). cllt) •... edt)] is the cost of each node at time t. P.Ji) is the power to transmit i bits of information, calculated by (3) and (4). PI/is the transition probability from state ito statej, calculated by (5). A, is the probability of I bits arriving to the buffer during a symbol period, calculated by (6).

Initialization: t = T, and cJT) = 0 for k = O. 1 ... L;

Iteration steps: Step 1: t = t - 1; Step 2: For k = 0.1 •... L,

L ~

Ck(t) = Pw(n) + L P(k - n)jCj(t + 1) + P L Al j=O I=L-k+l'Ir+l

Step 3: If t = 0 set"" = r(O) and finish, else go back to Step 1;

We have run numerical tests and the above algorithm always converges to a fixed policy after a small number of the iteration steps. Therefore, we need not run the iteration until t = O. The fixed policy to which the iteration converges is the stationary optimal solution to (10).

5. NUMERICAL RESULTS AND DISCUSSION

5.1 Numerical Results

First, we compare the performance of the adaptive scheme (obtained by the algorithm proposed above) with that of the "less-adaptive" ones. We consider two less-adaptive transmission schemes which are called "Dummy" and "Fixed". The "Dummy" scheme transmits bits at a fixed rate. In case there are not enough bits in the buffer, dummy bits will be transmitted. It is clear that this scheme is very inefficient. In the "Fixed" scheme, a maximum rate is set for the server. If the number of bits in the buffer is not enough for transmission at the maximum rate, the server only transmits whatever is available in the buffer.

Buffer Control using Adaptive MQAM ...

•• It ........... .

..... .. ~..... "......... Dummy

, .................. . " Fhc:ad . .. .... ...... . ...

""'<, ...... .... ,o~ L-....,...."---:':--~--c'=------,--,'-,..---'=-~.,,.--'

, . 14515155 161651717.5 ,e Power (dB)

103

Figure 2. Overflow probability versus average transmit power for adaptive, fixed and dummy schemes (L = 12 bits, A. = 30 kbps, Ts = 0.1 ms, W = 10 kHz and BER = JU3).

'0' ..--_-_--~-~--~---,

..... 10" --- - ... -.. .... --.~ _____ FI>t.d ..... ..... ...... ..... ..... ..... ..... .......

10'",7. ------;-:"'-;-.5 -~'5;----:-:'5.-:-5 ----7;:,s;----:,c1:"s 5:0-------:1'"1

Pow.r(dB)

Figure 3. Overflow probability versus average transmit power for adaptive and fixed schemes (L = 30 bits, A. = 30 kbps, Ts = 0.1 ms, W = 10 kHz and BER = JU.3).

In Figure 2, the performances of the three schemes are plotted for the following system parameters: L = 12 bits, A = 30 kbps, Ts = 0.1 InS, W = 10 kHz and BER = JO.3.

It is clear that the adaptive scheme always performs better than the "Dummy" and "Fixed" ones and the "Dummy" scheme is the most inefficient policy. However, we note that the gain of the adaptive scheme relative to the "Fixed" one is not very large. When the buffer overflow probability is from 10'3 to 10-4, the gain in the average transmit power is only from 0.5 to 1 dB.

Next, we plot the performances of adaptive and "Fixed" schemes under the case when the buffer length is very large. For the buffer length of 30 bits and the other system parameters are kept as before, the performances of the two schemes are shown in Figure 3. As can be seen, the advantage of the adaptive scheme over the "Fixed" scheme is much clearer in this set up.

5.2 Extending the Solution for Fading Channels

The optimal policy we have obtained for A WON channels can be extended for block fading channels. Under the block fading assumption, within one block of channel use, the channel gain is unchanged. In this case, within each fading

104 Anh Tuan Hoang and Mehul Motani

block, the channel can be treated as an A WGN channel with the received SNR being scaled by the constant channel gain. If the requirement is that at every fading state, the upper bound of the BOP must be met then we can use the same algorithm proposed before to search for the optimal policy within each channel state.

For the cases when the required BOP is averaged over all fading states or when a channel is under fast fading, work is currently being carried out. In Figure 4, the performances of three different schemes under fading environment are shown. The fading channel is represented by a ten-state Markov chain. The three schemes being tested are "fixed-rate, fixed-power", "fixed-rate, inverted­power" and "adaptive-rate, adaptive-power". The "adaptive-rate, adaptive­power" has been obtained by a search algorithm similar to the one proposed above. As can be seen, the adaptive technique offers a performance gain in terms of transmit power relative to less adaptive ones.

o .s

O.

~ O.lS 1i e 03 .. ~ 025 'E ~ 02 o U.pIMt.fhlt •.

o IS Ad.pltn-Po"".,

01

/-\ .. Fhc.d .ShI8, Flx.d .Pow.,

O~~~~~~IO~~12--~14--~~~1~8--~~ Pow.r(dS)

Figure 4. Overflow probability versus average transmit power for three different schemes under ten-state Rayleigh fading channel.

6. CONCLUSION

In this paper, we study the problem of buffer control for wireless communication. We have found out that by adapting the transmission rate (via changing the MQAM constellation size) to the buffer state and the channel condition, the power needed to reliably transmit data and to keep buffer overflow probability low can be minimized. Finally, we note that further work is needed in terms of analysis and dealing with fast fading channel conditions.

REFERENCE [1] R. A. Berry and R. G. Gallager, "Communication over Fading Channels with Delay Constraints", To Appear IEEE Transactions on Information Theory. [2] A. 1. Goldsmith and S. G. Chua, "Variable-Rate Variable-Power MQAM for Fading Channels", IEEE Transactions on Communications, vol. 45, no. 10, October 1997. [3] E. M. L. Beale, Introduction to Optimiwtion, John Willey & Sons Ltd., 1988. [4] G. 1. Foschini and 1. Salz, "Digital Communications over Fading Radio Channels", Bell Syst. Tech. 1., Feb. 1983. [5] D. P. Bertsekas, Dynamic Programming and Optimal Control: 2"d edition, Athena Scientific, 2001.

SATELLITES !HIGH ALTITUDE PLATFORMS STATION

A Low Complexity Iterative Receiver based on Successive Cancellation for MIMO

Holger Claussen 1, Hamid Reza Karimi 2, Bernard Mulgrew 1

I Signals & Systems Group, University 0/ Edinburgh, UK 2 Bell Labs Research, Lucent Technologies, Swindon, UK

Abstract: Turbo-encoded mUltiple-input multiple-output (MIMO) systems have recently been proposed for the support of high-speed downlink packet access (HSDPA) in UMTS, where the re-use of spreading codes across the transmitter antennas results in high levels of interference. The state of the art receiver for this system incorporates a channel equalizer, followed by an a posteriori probability (APP) detector and a turbo decoder. However, the complexity of APP detection can become prohibitive since it grows exponentially with the number of transmitter antennas and the modulation order. In this paper, a MIMO receiver is proposed which replaces the optimum but complex APP detector by successive interfer­ence cancellation (SIC) incorporating sub-optimal matched filter detection. Using convolutional encoding at the transmitter, the receiver performance is sustained via iterations between the simplified detector and the convolutional decoder. In combination with a proposed novel soft-output combining scheme, it is shown that the new receiver can outperform the APP-based receiver at a much lower complexity and with no need for channel equalization.

Key words: Iterative detection, MIMO, successive cancellation, serial interference cancella­tion, SIC, soft output combining, order metric, ordering

1. INTRODUCTION

Turbo-encoded multiple-input multiple-output (MIMO) systems have recently been proposed for the support of high-speed downlink. packet access (HSDPA) in UMTS [1]. The concept here is to increase the achievable data rates for a particular user through a combination of code re-use across transmit antennas and higher-order modulation schemes. The code re-use inevitably results in high levels of interference at the mobile receiver, even under non-dispersive channel conditions. In order to tackle such high interference levels, receivers based on the optimal a posteriori prob­ability (APP) detector [2] followed by turbo decoding have been proposed [3][4]. To cope with dispersive channels and in order to avoid sequence estimation, it is neces­sary to use an APP detector preceded by a matrix channel equaliser.

Essentially, the APP detector operates by computing soft-outputs for the trans­mitted bits which most closely match the received signal in an Euclidian sense. The

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

106 Holger Claussen, Hamid Reza Karimi and Bernard Mulgrew

Figure 1. Equalized APP based receiver for a 4x4 MIMO system (Reference)

computational complexity of the APP detector is an exponential function of the total number of bits transmitted during a symbol epoch, which is equal to the product of the number of transmit antennas and the number of bits per symbol. Consequently, the complexity of the APP detector can become prohibitive for increasing numbers of transmit antennas and (perhaps more importantly) modulation orders. This inflex­ibility of the optimal APP detector has resulted in renewed interest in the use of sub­optimal but less complex MIMO detectors.

Successive interference cancellation (SIC) schemes have been considered for many years in the context of multi-user detection for the COMA uplink [5][6][7]. These schemes combat interference by successively detecting and cancelling the in­fluence of data streams from the received signal. The more reliable data streams are detected and cancelled first. In the context of MIMO receivers, the original BLAST detector [8] is essentially a SIC architecture incorporating ordering and detection based on the minimum mean-squared error (MMSE) criterion. Furthermore, signifi­cant performance improvements have been demonstrated through iterations between the BLAST detector and a convolutional decoder [9].

In this paper, a bit-based SIC scheme incorporating simple matched filters (MF) as the basic detection unit is considered as a receiver for a convolutionally-encoded MIMO link. The MF-SIC detector performs iterations with a convolutional decoder in conjunction with a novel soft-output combining technique. Convolutional coding is used, since it provides better convergence than turbo coding in iterative schemes. The combining acts to suppress instabilities caused by erroneously detected and cancelled bits. The resulting receiver architecture is highly scalable in terms of dealing with growing numbers of transmit antennas and high-order modulation schemes.

The proposed MF-SIC receiver is compared with the APP-based receiver consid­ered for an equivalent turbo-encoded MIMO link [4] and is shown to achieve superior performance at a much lower complexity. The performance loss due to the use of a sub-optimal detector is regained via iterations with the decoder, enabled by the novel soft-output combining technique.

2. SIGNAL MODEL

Figures (1) and (2) illustrate the transmission scheme for the MIMO system under investigation. At the transmitter, user data is convolutional or turbo encoded and inter­leaved. The coded data stream is de-multiplexed into NT sub-streams, corresponding to the NT transmit antennas. Each sub-stream is then modulated on to NK 4-QAM symbols and subsequently spread by a factor Q via a set of K orthogonal spreading codes prior to transmission. Each transmitted spread stream then occupies N symbol

A Low Complexity Iterative Receiver based on SIC for MIMO

rlmlCMt>ring

MIMOchlmol "

Figure 2. Iterative MF-SIC based receiver for a 4x4 MIMO system

107

intervals. Also note that the same set of K codes are re-used across all transmit an­tennas. Therefore, the MIMO propagation environment, which is assumed to exhibit significant multipath, plays a major role in achieving signal separation at the receiver. The transmitted signals are received by NR receive antennas after propagation through dispersive radio channels with impulse response lengths of W chips. The received signal vector can then be modelled as follows:

K

[=HLCk:!k+~ k- I

where: mr. E C(QN+W-I)xl Signal at Rx antenna m. my E C(QN+W-I)XI

E CNxl xn ::::.t

C' E CQNxN k

Noise + inter-cell interference at Rx antenna m. Symbol sequence [x:(l) ... x:(NW at Tx antenna n spread via kth spreading code. Spreading matrix for kth spreading code fA: E CQxI .

c; =f! !.l N Tim ..

mHn E C(QN+W-I)xQN Channel matrix from Tx antenna n to Rx antenna m.

(I)

(2)

and y is a vector of iid complex Gaussian variables, R" =E{ vv" }=NoI. The 4-QAM

modulation mapping is such that x: (t) = b:.o (t) + jb:.1 (t) with b:.; (t) E {+ I, -I}.

3. APP RECEIVER As indicated in Figure (1), in this receiver the signal vector r. is applied to an a

posteriori probability (APP) detector following a process of channel equalization. The soft outputs from the APP detector are then applied to a turbo decoder which generates reliable estimates of the transmitted bits.

A full space-time APP detector implies joint detection of KNT transmitted symbols per symbol epoch. For 4-QAM modulation, and for dispersive channels with lSI ex­tending over L symbols, this requires a search over a trellis containing 22(L+l)KNT states.

J08 Holger Claussen. Hamid Reza Karimi and Bernard Mulgrew

The computational complexity is clearly inhibitive for typical parameter values. Note that, in fiat fading conditions (L=O) and for K orthogonal codes re-used over the trans­mit antennas, the number of trellis states reduces to a more realistic value of 22Nr. As a result, a strategy for dealing with dispersive channels is to apply APP detection after a process of space-time equalization which effectively eliminates dispersion. The equalization process inevitably causes noise colouring, which needs to be accounted for in APP detection.

Space-Time Equalization The received signal over N symbol epochs may be written as

K

!:=HI,CI!1 +~=BC!+~ (3) t-l

where o\:=C,! is the vector of spread symbols. A minimum mean-square error (MMSE) equalizer represents a space-time matrix V which minimizes the term E{I~-VrIl2}. It is easy to show [10], that the solution to this problem is given by

V = R.B" (BR.H" + R" r' (4)

where R.=E{uH }= 2C(ti since E{,I,IH}=21. The equalization process Vr essen­tially eliminates the effects of the channel matrix H. As a result, assuming orthogonal spreading codes, the contribution of symbols transmitted using the /(tb spreading code can be retrieved at the output of the equalizer via the de-spreading operation

and k=l...K (5)

where Y.,,=TiJ!. is coloured noise. Due to excessive computational complexity, space-time equalization is usually performed over a block of NE <N symbols and re­peated NINE times to cover the overall transmission period.

APP Detection

Vector z." consists of the equalized and de-spread contributions of N.;v symbols transmitted via the lC" spreading code over a total of N symbol epochs. Considering only the NT rows ofEq. (5) corresponding to the fh symbol epoch, we have

and k=l...K (6)

The APP algorithm can then be applied to derive log-likelihood ratios for each (equally probable) transmitted bit, bA;t(t) n=1...NT ie {0,1}, in the form of soft out­puts

(7)

where 14(t)=TI(t)y. is coloured noise and RuJ,t)=E{14(t)14(t)H}=No Tt(t)TI{t)H. The soft outputs are then applied to a Turbo decoder whose constituent decoders operate based on the max-log MAP algorithm.

A Low Complexity Iterative Receiver based on SIC for MIMO 109

4. ITERATIVE RECEIVER At the receiver of Figure (2), the signal vector!. is fed into a successive interfer­

ence canceller incorporating matched filter detection (MF-SIC). The received signal ofEq. (I) observed over the fh symbol epoch may be written as

(8)

where xt"(t) is a transmitted symbol at the fh symbol epoch and fJ.k"(/) is its code­channel signature at the receiver. The output of the MF-SIC is then de-interleaved and applied to a convolutional decoder. This represents the first iteration of the receiver. Soft outputs from the decoder are then re-interleaved and applied to the MF-SIC for further iterations.

Iteration 1: Here, the MF-SIC operates at a symbol level. The first step is to determine, at each

symbol interval I, the most reliable symbol according to a reliability criterion. Ideally, the symbol with the lowest error probability is selected [6]. Lacking such information, the symbol xt(/) k=1 ... K n=I ... NT with the highest signature energy, l~n(/)12 (or least mean-square estimation error), is selected. The next step is to estimate the selected symbol (soft-output derived via matched filter detection), make a hard decision on the estimate, reconstruct and cancel its contribution from the received signal:

y:(t) = ~=(tt !:(t) r:<t) = r(t)-~:(t){sgn{Re[y:(t)]}+ jsgn{Im[y:(t)]}}

(9)

(10)

The process is then repeated for the next most reliable symbol. If the decision on the selected symbol is correct, then its interference towards other symbols can be completely suppressed. However, a wrong decision doubles the level of interference caused by the erroneously detected symbol. Consequently, the reliability criterion used for the ordering of symbols is of critical importance in any form of successive cancellation.

After the MF-SIC detection of a complete code-block, the corresponding soft-out­puts'Yk.o"(/)=Re[Yk"(/)] andYk.t(/)=Im[Yt(/)], are multiplexed into a single stream for de-interleaving and convolutional decoding (max-log MAP algorithm). The decoder output is fed into the soft-output combiner and an interleaver prior to re-application to the MF-SIC for subsequent iterations.

Iteration 2 and beyond: In the second iteration of the receiver, the MF-SIC has access to reliability infor­

mation at a bit level, in the form oflog-likelihood ratios, A(bk.,"(/», generated by the soft-output decoder in the previous iteration. As a result, at each symbol interval I, ordering can be performed at a bit level (rather than symbol level) based on the log­likelihood ratios (LLRs). In other words, the bit bk,,"(/) with the largest LLR value IA( bk,I"(/»1 (or minimum estimation error probability), can be selected as most reliable. Since bit estimates corresponding to a particular symbol can have different reliabili­ties, the use of LLR values represents an optimum ordering policy. The cancellation

110 Holger Claussen. Hamid Reza Karimi and Bernard Mulgrew

process at the fh symbol interval is based on the more reliable hard bit estimates de­rived from the LLR values:

yZ,i (t) = 2> {!!: (t)" ret) + (-IY r(t)"!!: (t)}

[.(/) = [(I) - / !!: (t) sgn { A (b;,i (I) )}

(11)

(12)

where i=O or 1 depending on whether the bit of interest forms the real or imaginary part of the 4-QAM symbol. The process is again repeated for the next most reliable bit. After the MF-SIC detection of a complete code-block, the soft-outputs y, are again multiplexed into a single stream for de-interleaving and decoding (max-log MAP al­gorithm). The performance of the MF-SIC (and hence the receiver) should improve at each iteration as the quality of the decoder output improves.

Soft-Output Combining

In the proposed iterative receiver, mutual information is exchanged between the MF-SIC detector and the convolutional decoder. Therefore, at each iteration, soft es­timates (in the form of LLR values) at the output of the decoder are fed back to the detector for purposes of interference cancellation. Consequently, new and hopefully more reliable soft-output values are made available at the output of the decoder after each iteration. However, in some cases, the interference cancellation process can lead to poorer soft-outputs for certain bits. This can result in error propagation and there­fore unstable bit-error rate performance in subsequent iterations,

Such instabilities can be avoided by combining the soft-output values computed in the current iteration with those computed in the previous iteration(s). The combining weight factors have a significant influence on the stability and the speed of conver­gence of the iterative receiver. Using this combining process, reliability information already gained for a certain transmitted bit is not lost in the next iteration.

While soft-output combining can be performed either at the output of the detector or that of the decoder, simulations indicate that a combination of both is most effec­tive. If q indicates the iteration index, then soft-output combining may be described as

yZ,;Ct)[q] = a yZ,i(/)[q] + (I-a) yZ,i(/)[q -1]

A;,i(t)[q]=!3 AZ,;Ct)[q] + (I-!3)AZ,i(t)[q-l]

(13)

(14)

Good performance results were found to be achieved via combining factors of a=0.9 and f3 =0.75. Soft-output combining is an essential element of the proposed iterative receiver.

5. PERFORMANCE AND COMPLEXITY COMPARISON

The performance of the APP-based receiver for a Turbo-encoded MIMO link [4] is considered as reference for comparison with that of the proposed MF-SIC based receiver for an equivalent convolutionally-encoded MIMO link.

A system with Nr=NR=4, Q=16 and K=16, similar to the HSDPA specifications is considered. In addition to a flat Rayleigh fading channel, a dispersive channel with 3 equal-power, chip-spaced taps is also considered. The assumed mobile speed is 3

A Low Complexity Iterative Receiver based on SIC for MIMO 111

IER (o"'~mo"'o' I ·tlp dh pi 1"1 lit. (MMtI

..... 1)APP,1UJboI

-.~ r--.... -& .2) Mf -S Ie. col'?'l'64 + Il) Mf·$ Ie, (0l'1li16 -+- "'IIN ·SIC. corwl

- d)MI ·SIC.cOl'NtNC

,. -e-- bl, EOJ2+APP, .. bol

,~ ~ .... bJ) EOJ4.APp. _a + b.)APp._a -€I- b4)Mf·SIC. c""'4 -+ b5,Mf ·SIC, co,,,ms -+- b6) MF ·SIC. COM

10

,,'\.

" ~ '" , l~ \.\

'\ \. ,. -. ~ '\ ,. \'\ \\

'\.'\. . ~" . \ '\ \\ -6 -5 -4 ., . J • , 10 -6

,. ... J 2: 1 E~o (de) .feiCh Nwlvtr anOlI'll'll (bAlo [dB. II uo[h NwlYer 'ftWnna

(a) (b)

Figure 3. Perfonnance comparison

kmIh and the receiver has perfect knowledge of the average channel conditions dur­ing each transmitted data block. For the turbo-encoded MIMO link, a 8-state rate 1/3 turbo encoder is used in accordance with the HSDPA specifications, resulting in a block size with up to 5114 information bits. A total of 6 iterations of the turbo decoder are performed in the receiver. For the convolutionally-encoded MIMO link, rate 1/3, 8- 16- and 64-state convolutional encoders are considered to allow a comprehensive comparison in terms of performance and complexity. A total of 4 iterations between the MF-SIC detector and the convolutional decoder are performed. Soft-output com­biners with the coefficients a=0.9 after detection and /3=0.75 after decoding were used.

The BER performance comparison is presented in Figure (3) for flat (a) and dis­persive (b) channels, while Figure (4) illustrates the corresponding computational complexities in terms of the number of real multiplications.

For flat fading channels, the proposed iterative MF-SIC based receiver outper­forms the APP-based receiver by approximately 1 dB, dependent on the memory size of the convolutional code, consistently at a lower total computational complexity (a2-4). Simulation result (as) clearly demonstrates the degradation in performance when soft-output combining is not used.

For the dispersive channel, the MF-SIC detectors offer again significant im­provements in BER. In fact, the performance improvement over the equalized APP reference is even higher than for flat fading (result b4-6). Even with a simple 8-state convolutional decoder (b6), the proposed receiver offers improved BER results at ap­proximately only 20% of the APP-based receiver complexity. The small performance differences for the equalized APP detector, between using equalizer block sizes of 32 or 24 chips (b2, bl), shows that the edge effects are negligible for this channel.

".lOII ,- ~_""SIC ........ """" Ia ::::-' .... - ISIm->XIII

011000-

11""'""'-Figure 4. Complexity comparison: Multiplications per infonnation bit

112 Holger Claussen, Hamid Reza Karimi and Bernard Mulgrew

6. CONCLUSIONS

In this paper, a low complexity iterative receiver for a convolutionally-encoded MIMO system is proposed and compared with an APP-based receiver for an equiva­lent turbo-encoded MIMO system. The proposed iterative receiver utilizes a succes­sive interference cancellation architecture based on simple matched filters (MF-SIC). Despite the use of a simplified detection scheme, a high level of receiver performance is achieved via iterations with the decoder. In fact, it is shown that, via a combina­tion of bit-based ordering/detection/cancellation and the use of a novel soft-output combining technique at the decoder output, the proposed low-complexity iterative receiver outperforms the APP-based receiver. Depending on the complexity of the convolutional decoder, the BER performance can be improved by up to 2 dB, always at a significantly lower computational complexity. In contrast to the APP detector where the computational complexity grows exponentially with the nurnber of anten­nas and the order of the modulation scheme, the complexity of the proposed iterative solution only grows linearly. This makes the proposed solution highly scalable and even more attractive for 16- and 64-QAM. Furthermore, unlike the APP-based re­ceiver, the proposed solution does not require a matrix-channel equalizer to cope with dispersive propagation environments, making it attractive from an implementation standpoint. Finally, the iterative concept can be exploited to improve the channel esti­mation, which can further improve the performance in comparison to a non-iterative APP detector.

REFERENCES [I] 3GPP TSG RAN WGl, "PARC with APP Decoding for HSDPA", TSG-RI(02)0549,

April 2002, Paris, France. [2] Benedetto, S.; Divsalar, D.; Montorsi, G.; Pol/ara, F., "A Soft Input Soft-Output APP

Module for Iterative Decoding of Concntrated Codes", IEEE Communictions Letters, Volumel, Issue 1, pp. 22 -24, January 1997.

[3] 3GPP TSG RAN WGl, "Further link level results for HSDPA using multiple antennas", TSG-Rl#17(00)1386, November 2000, Stockholm, Sweden.

[4] 3GPP TSG RAN WGl, "Link Level Results for HSDPA using Multiple Antennas in Cor­related and Measured Channels", TSG-l#l 9(0 1)0302. February 2001, Las Vegas, U.S.A.

[5] Verdu s., "Multiuser Detection", Cambridge University Press, 1998. [6] Claussen, H.; Mulgrew, B; Karimi, H.R., "Performance Optimization of Successive

Cancellation Detectors", World Wireless Congress, pp. 797 -802, May 2002, San Francicso, U.S.A.

[7] Guinand, P.S.; Kerr, R.W.; Moher, M, "Serial Interference Cancellation for Highly Correlated Users", IEEE Pacific Rim Conference on Communications, Computers and Signal Processing, pp. 133-136, August 1999, Victoria, Canada.

[8] Foschini G.J., "Layered Space-Time Architecture for Wireless Communication in a Fading Environment when using Multielement Antennas", Bell Labs Technical Journal, pp. 41-59, Autumn 1996.

[9] Li, X; Huang. H.; Foschini. G.J.; Valenzuela, R.A., "Effects of Iterative Detection and Decoding on the Performance of BLAST", GLOBECOM '00. Volume 2, pp. 1061-1066, November 2000, San Francisco, U.S.A.

[10] Klein, A.; Kaleh, G.K.; Baier, P.w.. "Zero Forcing and Minimum Mean-Square-Error Equalization for Multiuser Detection in Code-Division Multiple-Access Channels", IEEE Transactions on Vehicular Technology, Volume 45, No.2, May 1996.

Dedicated Bandwidth Approach for Channel Allocation in a Multi-Service UpJDown Link of a Low Earth Orbit Satellite Constellation

Rima Abi Fadel 1 ,2 , Samir Tohme2

2Ecole Nationale Su¢rieure des Telecommunications 46, Rue Barrault, 75013, Paris, France {abifadel, tohme @ inf.enst.fr} 'Ecole Superieure d'Ingenieurs de Beyrouth Mkalles, Mar Roukoz, BP 11514, Liban frima.abifadel@fi.usj.edu.lb }

Abstract: Different users with diverse requirements need to share effciently the UpJDown Link UDL of a Low Earth Orbit LEO satellite constellation. Three types of services are assumed: real time (voice) services with strict constraints over the delay and the bandwidth, non real time (data) services delay tolerant but with bandwidth guarantees requirements and Best Effort services with no guarantees requirements. The scheme proposed consists in dedicating a certain amount of the total bandwidth for serving real time services, while the remaining bandwidth is shared according to a strict priority scheduling scheme between the two data services considered. All services are served as ''bursts''. Markovian ON-OFF source models are considered, together with a Pareto distribution for data packet lengths with a cut-off at 1502 bytes. The impact of the amount of dedicated bandwidth for real time services on the performance measures of all traffic classes is evaluated.

Key words: Access Networks, Performance Analysis, Satellite Networks.

1. INTRODUCTION

LEO Satellite Constellations are seen as a suitable mean for providing mobile users a global access service to terrestrial networks. Previous work [1,7] studies the connection admission control CAC problem at the air interface level of the satellite constellation using an enhanced trunk reservation technique and considers circuit switching for voice and burst switching for data. In this paper, a complete burst switching scheme is

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

116 Rima Abi Fadel and Samir Tohme

considered. The network considered assumes an integration of three service classes. The highest priority class is the real-time service (voice). Data traffic is separated into two priority classes. The higher priority service has guaranteed bandwidth, and will be referred to as data service, while the lowest priority service called "Best Effort" BE has no guarantees requirements. The scheme proposed considers a division of the total bandwidth into two sets. The fIrst is dedicated for the exclusive use of voice calls. The second is shared between data and BE in a strict priority scheduling manner. A certain boundary separates these two sets. An ON-OFF model is assumed for all sources. All data packet lengths are assumed to be Pareto distributed with a cut-off at 1502 bytes. An active period of the voice service is allocated a unit bandwidth within the dedicated bandwidth set, thus allowing the service of numerous active periods simultaneously. In the second set however, the whole bandwidth is assumed to be equivalent to only one server. The impact of the boundary choice is evaluated.

The "movable boundary" technique has been proposed in [2) for integrating voice calls and data packets. It has been applied in [3) to a broadband network integrating three service types, video, voice and data. [4) proposes to sub-divide network components into multiple sub-components, each having a dedicated bandwidth and carrying only traffic with the same Grade of Service GoS requirement. In [5] and [6], a routing method is developed, based on a base policy, a direct link routing requiring that each call type has a portion of link bandwidth dedicated to it. Priority scheduling, on the other hand, is seen in the literature as one of the simplest forms of scheduling in the context of service differentiation. In[8], static priority scheduling SPS is compared with other scheduling techniques in a General Packet Radio Switch system. In [9), priority is given to real time applications over non real time ones in a WCDMA environment, also calls experiencing handoffs are given priority over newly generated calls. A differentiated service Diffserv model is considered in [10] where flows are partitioned into different classes with different priorities. More general priority assignments than the classical ones are studied in [11]. A non-static priority assignment is also proposed in [12]. Priority scheduling has also been considered in ATM and high speed networks [13]. Delay for CBR traffic is computed in [14). Finally in [15] a probabilistic priority discipline is proposed in order to avoid the starvation problem.

The paper is organized as follows. In section 2, we introduce the source models and the parameters used for the simulation as well as the admission scheme proposed. In section 3, the impact of the amount of dedicated bandwidth over performance measures of the real time service, for different voice source types, is considered. In section 4, performance measures of data and BE, for different boundary choices and different source types, are evaluated. Section 5 discusses the impact of varying data and BE queue lengths. Section 6 concludes the study.

Dedicated Bandwidth Approach ...

2. SYSTEM AND SOURCE PARAMETERS DESCRIPTIONS

117

The French Reseau National de Recherches en Telecommunications Satellites Constellation Project [2], proposes a mobile communication access service for mobile users using a LEO satellite constellation. The total bandwidth serving pedestrian users on the uplink is of 72 kbitsls, with a granularity of 2.4 kbits/s. The cellular system that will be considered in this paper is based on cells fixed to the earth. Furthermore, the user mobility is considered as negligible in comparison with the satellite mobility because the typical diameter of a satellite cell is of order of few hundred kilometers, which is large at the space scale of a terrestrial mobile user (pedestrian, car). Thus, the issues associated with the Handover HO will not be considered. Since the bandwidth on the downlink is usually higher than the one on the up link, we focus on the channel allocation on the up link. Three call classes are considered. A voice source is an ON-OFF source. We assume no packet arrivals during the OFF period. And the ON period is served as a burst. Two voice sources types are considered. For both types, ON and OFF times are exponentially distributed with mean times of 3s (respectively 0.352s) for the ON period and 3s (resp. 0.650s) for the OFF period. Since with the parameters given above, thirty communications (N) can be simultaneously established, the number of active voice sources will be assumed to be equal to thirty in the simulation. An ON burst will be allocated the unit bandwidth of 2.4 kbits/s. BOUND, the boundary value represents the number of ON periods that can be simultaneously served. Since BOUND < N, a finite queue will be considered where ON bursts will be entitled to wait for service according to the simple First Come First Serve policy. Web sources are described in the ETSI technical report ETSI TR 101 112. They are modeled as ON-OFF sources with both ON and OFF times exponentially distributed with means for the UDD 8kbits/s model of respectively 12s and 412s. During the ON time a mean of 25 packets are generated with packet inter-arrival times exponentially distributed of mean 0.5s. Packet lengths are distributed following a Pareto distribution with cut-off at 1502 bytes. Throughout the simulations, the same source models are assumed for both data and BE sources. This represents a rather pessimistic case of study, especially for the BE traffic since the latter is usually described with more sporadic models. Other source parameters will be considered in further simulations and will be discussed in the relevant sections. Two queues are considered, one for the data bursts and the other for the BE service. These queues are served according to a strict priority scheduling scheme. A burst will be given the total bandwidth available for data services once admitted into the service, a bandwidth of (N-BOUND)*unit bandwidth. As mentioned above, the service unit is the burst. Figure 1 represents the system.

118 Rima Abi Fadel and Samir Tohme

3. IMPACT OF THE BOUNDARY CHOICE OVER THE REAL TIME TRAFFIC

In this section, the impact of the choice of the boundary value on the voice traffic performance is evaluated. The queue length is also varied and two source types are considered, the 3s/3s ON/OFF model and the 0.352s/0.65s one. Two performance measures are considered : the burst loss probability and the burst waiting time in queue. A new burst is lost whenever it finds no room in the real time service queue. This study aims to evaluate the benefits of the statistical multiplexing if, instead of considering a pure circuit switched system, the service unit is the burst. Figures 2.a) and 2.b) represent respectively the evolution of the loss probability and of the waiting time for the 3s/3s voice bursts with the number of places in queue m and for different values of the parameter BOUND. Figure 2.a) shows a clear decrease in the burst loss probability for voice as the number of places in queue increase, a decrease that is more significant for smaller values of BOUND. In fact, as BOUND increases, a greater number of bursts can be simultaneously served. A smaller size of the queue is then needed to meet the same loss probability. As for the waiting time in queue represented in figure 2.b), it is observed that the smaller the value of BOUND the bigger the delay. In fact, for higher values of BOUND, the mean arrival rate into the system remains the same while the mean departure rate increases since more channels can be simultaneously used. This leads to a less loaded system and the cases where the queue is full occur less frequently. This means that under the same arrival conditions, the mean number of customers in the system decreases and so does the delay. Another observation is that the delay is more sensitive to m for smaller values of the parameter BOUND. If no queue limitation is done for such systems, the mean customer number that would result is bigger than the chosen queue size. Of course this means that the queue will tend to saturate. If more places were available in queue, the clients would go on fIlling them, thus increasing the delay, and so on until the queue size becomes quite larger than the mean number of clients expected in the equally loaded infinite queue system.

A comparative study of the two systems with 3s/3s sources and 0.352s/0.65s sources respectively regarding the loss probability for bursts shows that for the same boundary value and the same queue size, the second source type largely outperforms the first one. This can be explained by the fact that the proportion of the ON period regarding the overall activity period is of one half in the first case while it is of almost one third in the second one. This means that the relatively longer OFF periods can be better exploited in this case, thus resulting in a higher statistical multiplexing gain. This is a direct consequence to the fact that for the same peak: rate for both source types, the 0.352s10.65s represents a smaller mean rate. On the other hand,

Dedicated Bandwidth Approach ... 119

regarding the waiting delay, much smaller values were noted for this source type as a consequence to the fact that the ON period is shorter.

To conclude this section, an important result is that the choice of the boundary will be subject to the voice source nature as well as the performance measures required for this strict constrained class of service. This choice also has to consider the data services behavior.

4. IMPACT OF THE BOUNDARY CHOICE OVER DATA AND BEST EFFORT

Different loads for the data and the BE classes are considered. This ftrst approach to the problem evaluates the effectiveness of the head of the line HOL scheduling scheme considered. Figures 3.a) and 3.b) represent respectively the loss probability and the waiting time in queue for data bursts, such as WEB bursts. These ftgures show that for a given number of Web active sources, the loss probability of Web bursts is very slightly affected by the number of active BE sources. The slight increase noted is due to the non­preemptive priority scheme. The delay increase with the number of BE active sources is more visible since the residual BE service time is a component of the waiting time of Web bursts. Simulation results not represented here show that for a given number of BE clients, loss probability as well as waiting time in queue for BE clients signiftcantly increase with the number of Web active sources, due to the strict priority scheme chosen. This leads to a starvation problem with high loss probabilities (around 0.5) and large waiting time in queue (around 40s for a service time around 3s) that get more and more critical as the number of Web active clients increases. Figures 4.a) and 4.b) represent the performance measures for the data service when BOUND is varied. An important observation is that for higher values of BOUND, an additional increase affects more signiftcantly the performance measures of the lower classes of service. Only the Web performance curves are represented, but simulation shows that the same conclusions hold for the BE service. In fact, BE is in an even worse situation than what is described for Web when BOUND increases, especially when the load of the Web class increases. Other simulations were run in order to evaluate the impact of varying the source models for data and BE sources. The same mean rate of 266.415 bits/s per source is considered, while the peak rate is varied. Considered values are 64 and 144 kbits/s. An increase of the data service with guarantees loss probability with the increase of the peak rate of the sources is noted. Even though the mean rate is the same, a rapid queue saturation problem may arise when different sources are in their ON state, simultaneously transmitting at their peak rate. The performance degradation is however not drastic since the mean rate is relatively small. Another

120 Rima Abi Fadel and Samir Tohme

consequence of this small value is that the mean waiting delay of data is almost unaffected with the peak rate variation.

5. IMPACT OF DIFFERENT QUEUE LENGTHS FOR DATA AND BE

In this section, data and BE queue lengths are varied. Two values are considered, 5 and 10 places. A "place" in queue is assumed to be large enough to receive a whole burst. Figure 5.a) shows that data loss probability is very sensible to the data queue length. It considerably decreases, for any amount of BE active communications, as the data queue increases. The price to pay, however, is the important increase in the waiting delay for this service, as shows figure 5.b), especially when more active sources compete for the system resources. Increasing the BE queue length however has almost no effect over the performance of data since absolute priority is affected to the latter. Other simulation results show that, for a light load associated with the data service, increasing the data queue length does not affect the performance of BE bursts. On the other hand, the waiting time of BE bursts is significantly penalized when data bursts represent a significant load since BE bursts can only be served when the data queue evolves to an empty state. It can, as the number of active BE sources increases. attain an order of a minute. Its increase however with the BE queue length is far less important than what is observed when the data queue length increases, while leading to better blocking conditions. The data queue should thus be carefully dimensioned since it affects the data service and can contribute in the establishment of a starvation situation for BE bursts. Other simulations show that if a cutoff at 10 kbytes is chosen, a drastic degradation of all performance measures is noted.

6. CONCLUSION

In this paper, a dedicated bandwidth scheme, reserving an amount of bandwidth for voice, is combined with a HOL mechanism that controls access to the remaining bandwidth of the lower priority data and BE bursts. Simulations show that the boundary choice is function of the voice and data source types, as well as of the number of simultaneously active sources of each class. The HOL mechanism applied to the data services shows that the number of active sources of each type is to be controlled in order to guarantee Quality of Service parameters for each service class, in accordance with the system parameters, that is the amount of available bandwidth and buffering space. The dimensioning problem must be carefully performed since various, and conflicting parameters should be considered, and serious problems such as the severe starvation problem should be excluded.

Dedicated Bandwidth Approach ...

2 .• kbiISlIO

Voicobums _____ --, o - ------'

o '----./

N'Iavm- BOUND

Tot. BW· BOUND'14 kbiWI

Figure l.a) Dedicated channels for voice.

i

I !

0, I. 11 12 1S 1. 1. lMemorr .tr:.

Figure 2.a) Voice bursts loss probability as a function ofm and BOUND.

121

0&1. bums ______ ., - -------' (N·BOUND)*24 kbilSl.

o S .. ll!fTon b .... "U'----____ -, -Figure l.b) Head of the Line scheduling for data service with guarantees and Best Effort

service. OII-wow ...... boie-lIonvtm

25a ····-l-· ··· r···· ·f ·····j· ····r·····f·····f····· I

200 -- " '!' - '-- 1- --"-~------ r- - · --~-----t .. -...

'" .... : ..... ~ ... ... ~ .... . : .. ... ~ .. . I : \ 1 \

lIDO .... + .. = .. l-;.......~ .. :-== .... :+.. r:::-:'" ·:t:-r=·· .rr· '~"1 DOI~~--~.= .. *.:= .. ~ .. ~ .. ~ .. = .. ~:-~ .. ~ .. ~.:~ .. ~ .. ~.:~ .. ~ .. ~'1

o • 1 • n '2 '3 Memory .Ize

1. 1.

Figure 2.b) Voice bursts' waiting time as a function ofm and BOUND.

II

\ m ~ ~ DO m ~ m • ~ NunoD.I alGI •• elM ...

Figure 3.b) Web bursts' waiting time

~ ~ m ~ ~ ~ • ~ • • ~

Figure 4.a) Web bursts loss probability for different values of BOUND.

"*-n ... 4IIcWAcIWII.

Figure 4.b) Web bursts' waiting time in queue for different values of BOUND.

122

Figure 5.a) Web bursts loss probability for different queue lengths.

REFERENCES

Rima Abi Fadel and Samir Tohme

~·~.~~~=~~~~~~~=c~m~~~~=~~,~ ,..,... r:lldil. chnII

Figure 5.b) Web bursts' waiting time in queue for different queue lengths.

[1] R. Abi Fadel, S. Tohme. Connection Admission Control CAC and Resources Allocation RA on the Up/Down Link UDL in a Low Earth Orbit LEO Satellite Constellation. Proc. IEEE Int. Symp. Comp. and Commun., Jul.200 1. [2] P. Zafiropoulo. Flexible Multiplexingfor Networks Supporting Line-switched and Packet­Switched Data Traffic. ICCC 1974. [3] O. Starnatelos, J. Hayes. An integrated system for video, voice and data communications. Proc. IEEE SUPERCOMMlICC'92, Vol. 1 , 1992. [4] H. Ji, J. Hui, E. Karasan. GoS-Based Pricing and Resource AI/ocation for Multimedia Broadband Networks. Proc. IEEE INFOCOM 1996, Vol. 3,1996. [5] A. Kolarov, J. Hui. Least Cost Routing in Multiple-Service Networks. Proc. IEEE INFOCOM, Toronto, Canada, June 1994. [6] A. Kolarov, J. Hui. Least Cost Routing in Multi-Service Networks. Proc. IEEE INFOCOM 1995, Vol. 1 , 1995. [7] R. Abi Fadel, S. Tohme. Connection .Admission Control CAC and Differentiated Resources Allocation RA in a Low Earth Orbit LEO Satellite Constellation. Proc. of the IFIP Networking 2002 Conf., May 2002. [8] Q. Pang, A. Bigloo, V. Leung, C. Scholefield. Service Scheduling for General Packet Radio Service Classes. IEEE Wireless Communications and Networking Conf., vol.3, 1999. [9] M. Kazmi, P. Godlewski, C. Cordier. Admission Control Strategy and Scheduling Algorithms for Downlink Paclcet Transmission in WCDMA. IEEE Vehicular Techn. Conf., vol.2, 2000. [10] S. Wang, D. Xuan, R. Bettati, W. Zhao. Providing Absolute Differentiated Services with Statistical Guarantees in Static Priority Scheduling Networks. Proc. Seventh IEEE Symp. Real-Time Technology and Applications 2001. [11] S. Wang, D. Xuan, R. Bettati, W. Zhao. PrOViding Absolute Differentiated Services for Real-Time Applications in Static Priority Scheduling Networks. Proc. IEEE INFOCOM 2001, Vol. 2 , 2001. [12] K. Siriwong, R. Ammar. QoS Using Delay-Synchronized Dynamic Priority Scheduling. Proc. IEEE Int. Symp. Comp. and Commun., Ju1.200 1. [13] C. Li, R. Bettati, W. Zhao. Static Priority Schedulingfor ATM Networks. Proc. 18th IEEE Real-Time Systems Symp., 1997. [14] K. !ida, T. Takine, H. Sunahara, Y. Oie. Delay Analysis for CBR Traffic under Static­Priority Scheduling. IEEEI ACM Transactions on Networking, vol.9, no.2, April 200 I. [15] Y. Jiang, C.K. Tham, C.C.Ko. A probabilistic Priority Scheduling DiSCipline for Multi Service Networks. Proc. IEEE Int. Symp. Comp. and Commun., Jul.2001.

Softer Handover Schemes for High Altitude Platform Station (HAPS) UMTS

Woo Lip Lim, Yu Chiann Foo and Rahim Tafazolli Mobile Communications Research Group. Centre for Communication Systems Research. University of Surrey. Guildford. Surrey GU2 7XH. UK

Abstract: An important property of high altitude platform station (HAPS) Universal Mobile Telecommunications Systems (UMTS) is that unlike terrestrial tower based systems. interference is dependent on the antenna characteristics of HAPS rather than the terrain features (i.e., shadowing) in the service area. In this paper. we exploit this unique HAPS property to propose simple and effective adaptive softer handover schemes for HAPS UMTS. Simulation results obtained show that our proposed handover schemes provide improved system performance as compared to the existing conventional non-adaptive softer handover scheme proposed for terrestrial UMTS.

Key words: HAPS, UMTS. soft/softer handover

1. INTRODUCTION

Soft/softer handover are used in code division mUltiple access (CDMA) systems due to their various advantages over hard handover. When considering handover in a single platform HAPS CDMA system, we note that in concept, the HAPS geometry is similar to a very tall terrestrial tower projecting hundreds of sectorised cells. The handover between cells of a HAPS CDMA system is thus similar to the handover between sectors of a terrestrial tower based CDMA system. Hence. the handover process is faster and softer because a single timer can be used to synchronize all cells [1]. In this paper. we use the term softer handover (SHO) to refer to handover between cells of a single platform HAPS CDMA system.

When designing SHO schemes for HAPS UMTS, we should note that an important unique characteristic of HAPS UMTS is that all base station (BS) transmit

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

124 Woo Lip Lim, Yu Chiann Foo and Rahim TaJazolli

antenna beams essentially originate from the same phased array antenna onboard the platform. As the altitude of the HAPS is much larger than the dimensions of the phased array antenna, the wanted and interfering signals traverse almost the same path and hence undergo similar path loss and shadowing. Therefore, the received signal-ta-interference ratios (SIRs) of the mobiles in HAPS UMTS are dependent on the antenna radiation pattern rather than the channel characteristics (path loss and shadowing) [1],[2].

In wideband CDMA (WCDMA) systems, a mobile continuously tracks the received energy per chip to interference power density ratio (EJ1o) of all the downlink common pilot channels (CPICHs) from the BSs in the service area and report this information to its serving BS. For HAPS UMTS, due to the collocation of BS antennas, the CPICH signals transmitted by the BSs to the mobile experience the same path loss and shadowing. Thus, if we assume that fast fading can be averaged out due to its short correlation length, then, the differences between the received EJ10 values from the mobile's serving BS and the neighbouring BS are basically the differences in antenna gains between the BSs. These antenna gain differences are deterministic and can be utilised to implement simple and effective adaptive SHO schemes.

In this paper, two adaptive SHO schemes for HAPS UMTS are formulated based on the unique HAPS interference property. The performances of the proposed adaptive SHO schemes are evaluated via simulation in terms of quality of service and resource utilisation and compared to the corresponding performances of the conventional non-adaptive SHO scheme (NADS) discussed in [3].

2. DESIGN STRATEGIES FOR HAPS UMTS SOFTER HANDOVER SCHEMES

Softer handover schemes employ signal averaging, SHO margins and the Time­to-Trigger (Lin mechanism to trade off between quality of service and resource utilisation. Since mobiles travel with different speeds and directions, the conventional SHO scheme using fixed SHO margins, signal averaging window and LiT will not yield optimum system performance. This is because fast moving mobiles tend to handover at distances further away from their serving BSs than slower moving mobiles, leading to higher call outage probabilities. Slow moving mobiles on the other hand utilise the limited system resources (downlink BSs' output powers) unnecessarily due to their long stay in the SHO area. To illustrate, we assume that mobiles A and B, both served by BSh are travelling at the same speed in the directions of OA and OB respectively as shown in Fig. 1(a). In this scenario, mobile A will experience a higher rate of change of the difference between the received EJlo values from BS) and BS2 as compared to mobile B. Mobile A will also stay in the SHO area for a shorter duration of time as compared to mobile B since it

Softer Handover Schemes for High Altitude Platform Station (HAPS) 125 UMTS

crosses a smaller SHO area. Hence, if mobile A does not initiate the SHO process early enough, it will be more susceptible to call outage and hence call dropping as compared to mobile B. On the other hand, if mobile B initiates its SHO process too early, it will utilise the limited power resources unnecessarily.

Due to the unique characteristics in HAPS UMTS, the rate of change of the difference between the received EJlo from a mobile's serving BS and the strongest received EJlo from its neighbouring BSs (ROC.dpl/o,) can provide reliable information on a mobile's relative speed and travelling direction for the design of adaptive handover schemes since ROC.dpi/ot is only influenced by the BSs' antenna radiation pattern rather than the propagation environment. If the mobile's SHO add margin (8...t.t) and drop margin (4m,p)1 can be dynamically adjusted based on the information on ROC.dpilolt a better system performance can be achieved as compared to the conventional fixed threshold non-adaptive SHO scheme. Note that this method is not suitable for terrestrial tower based UMTS SHO as CPICH signals transmitted by different BSs to a mobile experience different levels of shadowing and path loss. Hence, tracking the ROC.dpilot will not provide an accurate and reliable indication of the mobiles' travelling speeds and directions in this case.

m or-~--~----~----~_~

i 0' ,"r----~'I I! -10 / --. BS2 i 0(/2

i .., ·15

g; " 'i\!~ . 20 / Travelling

BS / direction OA 1 BSz

Z_25~~~'~========~·~~~ o Q5 1 1~ Distance Ikm) region

(a) (b)

Figure lea). HAPS UMTS handover scenario for mobiles travelling in different directions. (b). The intersection of the antenna radiation patterns of BS I and BS2 in OA direction.

a) EstabUshing the maximum and minimum ROC.dpiloI (ROC.dpIlot.1fIIIJt and ROC.dpIlot.min): A mobile travelling with the fastest speed in the direction OA and a mobile travelling with the slowest speed in direction OB in the service area will experience the maximum ROC.dpUotand the minimum ROC.dpUot respectively. Since the differences between the received EJlo values from the mobile's serving BS and the neighbouring BSs are basically the differences in antenna gains between the BSs, we can establish ROC.dpllot,mmc and ROC.dpl/ot.mtn of the system approximately using the HAPS antenna radiation pattern specified in (1)

assuming that the maximum and minimum mobile speeds in the service area are

I The definitions of &..!d and ~rop in [3] are used in this paper.

126 Woo Lip Lim, Yu Chiann Foo and Rahim TaJazolli

known. As shown in Fig. l(b), ROC4pilol.max = ILlpilol,• - LlpilolaVL1t where Llpilol,• and Llpilola are the differences between the normalised antenna gain levels in dB at the angles under which the fastest moving mobile is seen from the boresights of BS.'s and BS2'S antennas at time II and 12 respectively. L1t is the difference between 12 and 11 which is equal to the simulation time step. ROC4pilol.min can be obtained with the same approach using the slowest moving mobile travelling in direction OB.

b) Softer handover margin variation factor (o_ROC4pi/ot): Depending on the ROC4pilol the mobile experiences, a handover margin variation factor is added to the fixed handover margins to obtain the adaptive handover margins for the mobile. The SHO margin variation factor (O_ROC4pilol) for mobile i is:

1 ROC' -ROC . J o ROC'. = 6pilol 6pilol.1DID +0 ROC. . - 6pilol ROC . _ ROC . - 6p.IoI.1D1D

6p.IoI.max 6pilol.1DID

(1)

where P = 0 _ ROC bpilol.max - 15 _ ROC bpilol.min •

t5_ROC4pilol.max and t5_ROC4pilol.mln are the maximum and minimum SHO margin variation factors corresponding to ROC4pilol.max and ROC4pilol.mln respectively. O_ROC4pilol.max and t5..flOC4pilol.min are design parameters and the relationship between O_ROC4pilol and ROC4pilOI is shown in Fig. 2.

(LROC4pilol

o '-i--i----7"''-------r---... ROC4pilol

ROC4p'/O~mtJX

c5J?OC4p'/O~mln _

Figure 2. Softer handover margin variation factor vs. ROC4pilol'

c) Proposed adaptive softer handover schemes: Two adaptive schemes for HAPS UMTS are proposed in this paper. For the first adaptive scheme (ADSI), only the add margin is adaptable. A mobile with a larger ROC4pilOI will have a higher add margin as compared to mobile having a smaller ROC4pilol' The drop margin remains unchanged regardless of the values of ROC4pIlOl' Hence, for ADSI, the add and drop margins for mobile i can be written as:

t5:w".adopl = I5D1ld +15 _ROC~ilol

t5~.adopl = I5drop

(2)

(3)

where Oadd and I5t1rop are the add and drop margins used for the conventional non­adaptive SHO scheme as explained in [3]. For the second adaptive scheme (ADS2). both add and drop margins are adaptable and each mobile is assigned

Softer Handover Schemes for High Altitude Platform Station (HAPS) 127 UMTS

with individual add and drop margins according to its ROC""ilot. ADS2 ensures that a mobile having a larger ROC""ilot has a higher add margin and a lower drop margin as compared to a mobile having smaller ROC""ilot. The add and drop margins of ADS2 for mobile i are:

3. SIMULATION MODEL

(4)

(5)

We evaluate and compare the performances of ADSI, ADS2 and NADS under the following simulation conditions: a) HAPS system model: A HAPS carrying a WCDMA communications payload

and a multi-beam phased array antenna with beam/gain shaping capability is positioned at an altitude of 22 km in the stratosphere. It projects spot beams on the ground within the service area in a pattern similar to that created by a traditional cellular system to provide mobile communications services. Any residual pointing error due to the movement of the HAPS is assumed to be compensated by appropriate station keeping mechanisms or by steering the beams electronically [1]. The antenna radiation pattern used for cell projection has a sharp roll off of 60 dB/decade and conforms to the specifications proposed in [1]. The gain at cell boundaries is taken to be -13 dB with respect to the maximum main lobe gain (Gm).

b) Cell model: The simulation area consists of 19 cells located near the nadir that are approximated to be equally sized and circular in shape. With Gm = 36.7 dB, the cells projected on the ground have a radius of 1 km. The BSs are assumed to transmit only the CPICH and traffic channels. The transmit power for the CPICH is fixed at 33 dBm. The BS maximum output power is set at 42 dBm and the channel power limit is set at 30 dBm.

c) Traffic model: 32 kbps real time speech service is considered. Calls are generated according to a Poisson process with a mean call duration of 120 s. The speech service is modelled as an on-off model, with an activity factor of 0.5.

d) Mobility model: A newly generated call is assigned a uniformly distributed random location in the simulation area. Each mobile arriving to the system chooses the BS that provides the best link gain as its serving BS. The initial speed of a new user is generated by the uniform distribution U{50 km/h, 120 kmIh] and is assumed to remain unchanged throughout the call. The initial direction of a new user is generated by the uniform distribution U{00, 360°]. A mobile will travel an average distance of 2 km before changing its travelling

128 Woo Lip Lim, Yu Chiann Foo and Rahim TaJazolli

direction. The new direction is generated by a uniform distribution U[ -45°, 45°] with reference to the old direction.

e) Power control model: Centralised transmit power based call admission control is implemented, where calls are only allowed to enter the network provided that in maintaining the Ei/1o requirement, i.e. (Ei/lo),hreslrold of the new and existing calls, there is a non-negative power vector that accommodates the new mobile, and that the output powers of all BSs in the service area do not exceed their respective limits [4]. Furthermore, each forward link channel output power should not exceed an allowable limit. Otherwise, the call is blocked. Similar conditions are applied when adding a new BS to the mobile's active set (SHO mode). The SHO request will be denied if the above conditions are not met and mobiles will continue to try to execute the SHO process in the subsequent time step as long as the mobiles' add margins meet the SHO criteria. When a mobile is in SHO mode, we assume that all the BSs in the mobile's active set will transmit approximately equal amounts of power to the mobile. Fast fading is assumed to be averaged out due to its short correlation length and is not considered in our evaluation. M received samples of Ec!lo are averaged over a rectangular window before being compared with the SHO margins. Due to link variations caused by the mobility of the mobiles and/or varying channel and traffic conditions, even if no new mobiles are admitted, a feasible power vector might not be found at a particular instant. In this case, a simple step-wise removal algorithm is used to identify one by one the mobiles having the worst link gain conditions to be outaged (i.e., have their downlink traffic channels switched oft) until the required Ei/1o value is achieved in the remaining links [4]. A mobile that is in outage continuously for 1 s will be dropped.

t) Simulation parameters: The simulation parameters are summarized in Table 1.

Tabe 1m atlOn parameters l 1 IS' ul .

Parameter Value Patameter Value

Radio access WCDMA Chip rate 3.84 Mcps

Speech service bit rate 32 kbps (E"tlO)llrabdd 7dB

Mobile speed 50-120 kmIh Active set size 2

M (averaging number) 8 Max. BS output power 42dBm

i1T (adding. dropping 2.5 s Max. traffic channel 30dBm

and replacing a link) output power

Simulation time step 0.5 s CPICH transmit power 33dBm

4,., 2dB &"", 5 dB

.s ROCdpU«, .... 1 dB Ii ROC~m'" -1 dB

g) Performance measures: The performance indicators used to evaluate the SHO schemes are:

I. Quality of service

New call blocking probability (Pb): The probability that a new user is denied access to the network by the call admission control mechanism.

Softer Handover Schemes for High Altitude Platform Station (HAPS) 129 UMTS

Call dropping rate (P d): The rate at which ongoing calls are dropped from the network due to the calls being outaged continuously for more than I s.

2. Resource utilisation Mean active set number: The average number of base stations in a mobile's active set throughout its call duration. Active set update rate: The average number of updates (add, drop or link replacement) in a mobile's active set per second.

4. RESULTS AND DISCUSSION

The antenna gains evaluated between 0.6 km and I km (where SHO is normally initiated and executed) is used to determine ROCiJpi/ot,max and ROCiJpi/ot,min' Any ROCiJpiiot values that are larger than ROCiJpi/ot,max or smaller than ROCiJpi/ot,min are fixed at ROCiJpi/ot,max and ROCiJpi/ot,min respectively. The performances of ADSI, ADS2 and NADS are evaluated using the HAPS system level simulator and the results are shown in Fig. 3.

Among the three schemes, NADS gives the worst quality of service. This is because NADS adds BSs to the fast speed mobiles' active sets later than the adaptive schemes. Since fast speed mobiles move towards the cell edge where interference is most severe very quickly, if these mobiles are not in SHO mode, BSs will need to transmit higher powers to these mobiles in order to maintain their received E,/Io requirement. This will result in the system being unable to meet the power requirements, with traffic channels' and BSs' output powers reaching their respective limits. Furthermore, since NADS allows slow speed mobiles to add an additional BS to their active sets earlier than the adaptive schemes, the mean active set number for NADS is higher than the mean active set numbers for the adaptive schemes. This means that NADS will utilise more power resources leading to new calls being blocked and existing calls being removed from the network. In contrast, the proposed adaptive SHO schemes allow mobiles travelling at higher speeds to initiate the SHO earlier and mobiles travelling at slower speeds to initiate the SHO process later so that after the duration of L1T, all the mobiles with different travelling speeds and directions will be able to add the second BS to their respective active sets at about the same distance away from the cell centre. Hence, a more uniform quality of service for all mobiles can be achieved with less resource utilisation.

Comparing the two adaptive schemes, ADS2 has a slightly higher mean active set number than ADS 1. This is likely due to ADS2 dropping the weaker BSs in the slow speed mobiles' active sets later than ADS1. Since slow speed mobiles will not be able to move out of outage conditions as quickly as the high speed mobiles after the weaker BSs are being removed from their active sets, it might be more beneficial to drop the weaker BSs in the slow speed mobiles' active sets slightly later. This will ensure that the slow speed mobiles can have good link quality with their serving BSs once the weaker BSs are removed from their active sets and prevent the system from reaching the traffic channels' and BSs' output power limits. As a result, ADS2 is able to achieve better Pb and Pd as compared to ADSI as shown in Fig. 3, We also

130 Woo Lip Lim, Yu Chiann Foo and Rahim TaJazolli

note that ADS! and ADS2 do not cause any increases in the active set update rates as compared to NADS.

In conclusion, the proposed two adaptive SHO schemes for HAPS UMTS outperform the conventional non-adaptive SHO scheme in both quality of service and resource utilisation. ADS2 is able to achieve a much better quality of service as compared to ADS!. However, ADS2 utilises slightly more resources and is also more complex to implement as compared to ADS! since both add and drop margins are dynamically adapted to the mobiles' ROCI1pi/ot. These proposed adaptive SHO schemes are simple to implement since information on the received Eel10 values from the mobile's serving BS and the neighbouring BSs are readily available.

0.1 r;:::=~:;:;::;--~----/Il

:?!'O.06 :8 <---.:'---'.-=-=~ .s 0.06 6:

i O.04

05 0.02

14 16 18 20 Traffic Load (Er1angs)

~ 1.225 ~ '---.:::........:.==~V

J! 1.22

i 1.21

22

CI

.~ 0.01

15 30.005

14 16 18 20 Traffic Load (Erlangs)

22

0.04451-~--~--;==:::::;::;:;::;-]

.",

-;; 0.044 Oil It: .!! 0.0435

i ::. 004 ;g . q>---.:::"flo;;::==-~

10.0425

1 .20~L4--~16--~18---20~--l22 0.042,L.4---:16-:----'6---:20::-:----l22

Traffic Load (Erlangs) Traffic Load (Erlangs)

Figure 3. Blocking probability, call dropping rate, mean active set number and active set update rate over different traffic loading.

REFERENCES

[1] lTU, "Revised technical and operational parameters for typical IMT-20oo terrestrial systems using high altitude platform stations and CDMA radio transmission technologies," ITU Document 8-1/307-E, March 1999.

[2] Y.C. Foo, W.L. Lim, R. Tafazolli and L. Barclay, "Other-cell interference and reverse link capacity of High Altitude Platform Station (HAPS) COMA system," Electron. Lett., vol. 36, pp. 1881-1882, Oct. 2000.

[3] W.L. Lim, Y.C. Foo, R. Tafazolli and B.G. Evans, "Softer handover performance of high altitude platform station W-CDMA system," in Proc. of WPMC'Ol, Aalborg, Denmark, pp. 99 -104, Sep. 2001.

[4] Y.C. Foo, W.L. Lim and R. Tafazolli, "Centralized downlink call admission control for high altitude platform station UMTS with onboard power resource sharing," accepted for IEEE VTC Fall 2002, Vancouver, Canada, Sep. 2002.

QUALITY OF SERVICE (QoS)

Adaptive QoS and Handover Issues in Wireless Multimedia Networks Using a Dynamic Adaptive Architecture: DYNAA

Rola Naja and Samir Tohme Computer and Network Departement Ecole Nationale Superieure des Telecommunications 46, Barrault Street, 75634 Paris {rola.naja,samir.tohme}@enstfr

Abstract: Adaptive Multimedia services are promising in wireless mobile networks since they can improve the quality of service (QoS). In our paper, we propose a dynamic adaptive architecture DYNAA which is based on a cal1 admission control (CAC) and bandwidth adaptation algorithms. The architecture proposed manages to adapt dynamically the bandwidth adaptation to the user's mobility and the traffic load. DYNAA tries to establish an application-network col1aboration that can cope with high variability in network conditions but can continue to transport multimedia content. Simulation results show the performance of the proposed scheme.

Keywords: Call admission control, adaptive quality of service, handover.

1. INTRODUCTION

In an end-to end QoS framework for multimedia wireless mobile systems, the major issue to be addressed is the high level of fluctuation in resource availability due mainly to mobility. There is a growing consensus that adaptive quality of service presents a viable approach to this issue. Hence, QoS provisioning is the responsibility of the network and the application in order to deliver multimedia content to a wireless mobile terminal in the most acceptable form. For adaptive multimedia services, the existing QoS parameters become very trivial to be guaranteed. The reason is The work reported in this paper has been supported by the ITEA AMBIENCE project.

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

134 Rola Naja and Samir Tohme

that the adaptive framework moves into the problem of the bandwidth degradation caused by adaptation. Note that degradation is obtained when the assigned bandwidth is less than the required bandwidth.

A new QoS parameter, the Degradation Period Ratio DPR, is proposed in [2]. It represents the portion of a call's lifetime that a call is degraded. However, DPR does not characterize the bandwidth degradation. In order to fully characterize the bandwidth degradation and to provide better QoS to service users, the authors in [3, 6] propose two novel QoS parameters: the degradation ratio DR and the degradation degree DD, which designate respectively the frequency and the degree of degradation.

In the classic adaptive framework, the call dropping probability P drop

becomes very trivial to be guaranteed at the expense of the application degradation. It is true that the forced termination of a call is a very frustrating phenomenon that may happen to a user. However, service degradation can be very annoying especially when that arrives frequently. One of the critical tasks of a mobile computing environment is to prevent frequent adaptation, by reducing DD and DR, due to the dynamics of resource and mobility of flows while still optimizing the network performance (i.e. still having Pdrop less than the required P drop ( Pdrop,qos»'

Another important task is to take into account the current load conditions when adapting the calls' bandwidth. Even, if the HO load decreases, CAC in [6] always accepts the HO request and that leads to the degradation of other calls in order to satisfy the incoming request. The above mentioned tasks are achieved in our proposed scheme DYNAA. In order to reflect the current load network conditions, DYNAA dynamically adapts the amount of bandwidth's adaptation based on the current network conditions (based on the average P drop).

2. THE DYNAMIC ADAPTIVE ARCHITECTURE: DYNAA

Figure I shows the dynamic adaptive architecture DYNAA proposed. In this paper, the focus is on application and network adaptive layer.

2.1 Application Layer

The resource specification for a flow of a class i specifies the minimum bi,min, the required bi,req and the maximum bi,mtIX granted bandwidth. Thus, the network has the ability to adjust the granted rate of the flow within the range of the resource specification. Multimedia applications must adopt the layered coding approach in such a way that they can accept varying degrees of

Adaptive QoS and Handover Issues in Wireless Multimedia Networks Using a Dynamic Adaptive Architecture: D YNAA

135

network guarantee levels. When the adaptation handler informs the multimedia scaler about the granted bandwidth, the scaler selectively chooses a subset of the hiemrchical coding depending on the resource availability. A customer in class i uses one bandwidth among the range [bI,J,b/,2, ... ,biJ, ... ,bi.KJ, where b/J< b/J+Jfor j=1,2 ... ,Kr l. Note that Ki is the number of class i multimedia layers and b/J the llayer bandwidth of class i. According to the flow's resource specification, bi,l is the minimum bandwidth bi,min' blJ the required bandwidth bl.req and bl.K1 the maximum bandwidth bl,max of class i.

Appllcalion Layer

! Adaptation Handler

• QoSJBandwidth Renegotiation, Standard Network Laye"

TCP Signalling for IP adaptive 1 networking

Network Adaptive Layer

Network Monitor '[ Call Admission Control

Scheduler Adaptation Controller

Figure 1. DYNAA Architecture

Applications derme a softness profile that allows an efficient match of application requirements to network resource availability. According to the softness profile, we consider two classes of applications: 1. The Hard Adaptive class (HA): This class regroups applications that are

adaptive with stringent constraints. 2. The Soft Adaptive class (SA): This class includes applications such as

email and HTTP that are adaptive with soft constmints.

2.2 Network Adaptive Layer

The network adaptive layer implements specialized modules that support the multimedia requirements. These modules are the network monitor, the call admission control, the scheduler and the adaptation controller. In order to provide adaptive service, various algorithms interact in the following sequence of events (figure 2):

The network monitor computes periodically the amount, the mtio of application degmdation and the current network load. These measures are used by the CAC when accepting new and HO calls. Application notifies the network that it wishes to set-up a flow between end-points and provides the flow specification. At this point, the network performs admission control using measures computed by the network monitor. After accepting a request,

136 Rola Naja and Samir Tohme

CAC communicates with the scheduler which schedules the waiting requests.

An incoming flow may cause a resource conflict between competing flows. Hence, the scheduler intemcts with the adaptation controller in order to perform a resource adaptation among the existing flows. Next, the adaptation controller calls the bandwidth allocation algorithm (BAA) or the bandwidth adaptation with no degradation algorithm (BNDA) for the distribution of bandwidth among competing flows.

DYNAA is a centralized/distributed architecture, using a centralized adaptation controller and distributed adaptation handlers. The adaptation controller notifies the adaptation handler about the bandwidth allocated through signalling. As for the adaptation handlers, they determine whether or not the application will adapt to any portion of the available bandwidth.

2.3 CAC and Bandwidth Adaptation Algorithms

The class i degradation pammeters DDt> DRi and P drop measured periodically by the network monitor are passed to the CAC. We refer to DDi.qos and DRi.qos as the upper-bound values of the degradation parameters of a class i. Let x;{t) stands for the number of calls of class i users in a cell at time t. bi.ass;{s.t) denotes the assigned bandwidth for a call s of class i users at time t where bi.ass;{s.t) E [bl•J.bi•2 ..... bij ..... bi.KJ and 1 ::; s ::; x;{t). Let I(f) be the indicator function which returns 1 iff is true and 0 otherwise. If n is the number of classes, L1 T the measurement time interval, and l' a time variable, the degradation parameters DDI and DRi of a class i are such that [3]: For i=l ..... n:

X;(I)

1 r L (bt.req - bt.ass;(k,t))I(bt.a3s;(k,t) < bt.req) D[);(r)=- J k=1 dt

ll.T x;(I) r-llT (bt.req - bt.l) L I(bt.ass;(k.t) < bt.req)

k=1

xi(l) 1 r L I(bt.ass;(k,t) < bt,req)

DRi(r) = - J k=1 dt ll.T r-llT Xi(t)

In our CAC, a new call is accepted only if the degmdation parameters of all classes are less than the corresponding upper-bounds values. For HO calls, we introduce two thresholds for the call dropping probability: P drop_min and Pdrop_max. If P drop is greater than Pdrop_max then the HO load is relatively high. Hence, the HO call is accepted without testing the degradation parameters and BAA is applied. On the other hand, if P drop is less than Pdrop_min (respectively between Pdrop_min and Pdrop_max), we accept the call if the performance degmdation pammeters are less than the upper-bound values and BNDA (respectively BAA) is applied. The scheme proposed aims to have a compromise between P drop and the degradation performance parameters:

Adaptive QoS and Handover Issues in Wireless Multimedia Networks Using a Dynamic Adaptive Architecture: DYNAA

Flow's QoS Specification

Figure 2. Network adaptive layer

Conflict resolution

Application

137

In fact, the adaptation envisioned in this paper is adapted to the mobility through measuring P drop .

BAA and BNDA will decide the changes for the calls' bandwidth in a cell adaptively when there is a call arrival. In order to prioritize HA class, the adaptation is perfonned over the SA class fIrst and then over HA. In our paper, class I refers to HA and class 2 to SA.

With BAA, if the available bandwidth (A) is less than the required bandwidth (b/,req) of a new call belonging to the class i, our algorithm tries to lower the bandwidth of some calls belonging to SA class to bz.req• Then, the same procedure is applied to some calls of HA class. If A is still less than b;.req, the call is rejected whenever it is a new call. In case of a HO call, BAA tries to squeeze to bZ•mln the bandwidth of some calls belonging to SA class and then that of some calls of HA.

As for BNDA, some adaptation is made without call's degradation. In fact, BNDA behaves as BAA but differs in the HO handling. If after reducing to the required bandwidth, A is still less than bl•req, no degradation is perfonned. More details ofthe implemented algorithms are reported in [8].

3. NUMERICAL RESULTS

3.1 Scenarios and Performance Evaluation

The fIrst scenario implements DYNAA as proposed in the previous section. In this scenario (referred as DYNAA), we don't assume that the HO requests can wait in queues. We have shown by simulation in [8] that DYNAA improves the overall perfonnance. The second proposed scenario (referred as DYNAA_wait) implements DYNAA architecture with a waiting alternative. In this scenario, the HO requests belonging to SA and HA class (SA_HO and HA_HO requests) wait in two distinct queues.

138

0."

0"

3: ~ 010

OOS

3a 0010

rl~g=-::j 0 ....

0 ...

j (>.10004

0002 , 0000

00 0,1 02 0,3 0.4

Col ortgIna1Ing .... denoIIy(caI per Me per km2)

Rola Naia and Samir Tohme 3b

_ --'D---o------4

-O-P_ -o-P_DYNAA_wllit ---..-p ... OYNAA

0.1 0..2 0.3 Q4

Col originating _ denIIIy (cal per sac per 1cm2)

Figure 3. 3a. Mean average delay, 3b. Call dropping probability

A queued HA_HO request is deleted from the queue when it passes through the HO area before getting a channel or if its communication is completed before passing through the HO area. SA connections typically are more tolerant to delay as compared to HA connections. Thus, whenever a SA_ HO request isn't satisfied within the current cell, it is transferred to the SA queue of the target cell.

Whenever there are some free resources due to a call departure, the new available bandwidth must be used in order to satisfy the HO requests and the degraded existing calls. Therefore, we apply the call admission control algorithm which tests Pdrop and the degradation parameters. Then, CAC communicates with the scheduler. In order to schedule the SA_HO and HA_HO requests, the Queue Length Threshold (QLT) scheduling policy is applied: QLT gives priority to SA traffic whenever the number of queued SA_HO requests is above some threshold (Lth) [9].

Afterwards, the scheduler interacts with the adaptation controller and applies BAA or BNDA according to the decision of the CAC. If after serving the HO requests, there are still some available resources, existing calls are upgraded as follows. First, we pick up the most degraded calls in the HA class and we increase their bandwidth to the required bandwidth. We keep doing this until there is no available bandwidth or until every call in the HA class has a bandwidth larger than or equal to b I .req. Then, the same procedure is applied to SA class. Next, we try to increase the bandwidth of the calls with the smallest bandwidth in HA class. Same thing is done to SA class if the available bandwidth is still greater than zero.

Computer simulations have been derived by assuming a seven cell network. The edges of the simulated space wrap around to the opposite edges with each cell having a complete set of interfering cells so as to avoid the border effect. The considered cells are assumed to have a radius of lKm, and a capacity of 60 channels. Users are vehicular with an average speed of 40Kmlh. The unencumbered session duration of a voice call is 120s. The bandwidth requirements are (bl.l ,bl.2,bl,3)=(l,2,3) for class 1 (HA) and (b2.l,b2,2,b2,3)=(2,4,6) for class 2 (SA) such that (bl,req ,b2,req) =(2,4).

Adaptive QoS and Handover Issues in Wireless Multimedia Networks Using a Dynamic Adaptive Architecture: DYNAA

4a

139

4b

0.006 I :::=g~~_waitl 0.000II I -<>-g=_walt I 0.004 0.0004

0 - i§-0

0.002 0.0002

0.000 0.0000 D.. 0.2 0.3 U 0.' 0.2 0.3 0.'

calt originating rate density (can per sec per km2) can originattng rate denolty (..a per sec per km2)

O,lZ 4c 4d

-<>- OYNAA_ wolt o.OS 0.10

__ OYNAA I -<>-OYNAA_walt I OYNM 0."

0. ..

0 .... 0.08 om

ti" 0 0

0." 0 .02

O.~ 0.0'

0.00 0.00 0, 1 0.2 0.3

can origInattng rate density (calt per lee per km2) 0 .. O.t G.2 0.3

can originating rate density (catt per I8C per km2) 0 ..

Figure 4. 4a,4b,4c,4d: Degradation Parameters ofHA and SA classes

Note that the upper-bound values for DD, DR are respectively (DRJ,qos,DDJ.qoJ = (0.01,0.01) for HA and (DR2.qos,DD2.qos) = (0.1,0.1) for SA. (Pdrop_min'pdrop_max)= (0.0075,0.009), Pdrop,qOS = 0.01, L,h=2 and LfT=5s.

The HA class in the simulation is represented by voice traffic. Its arrival process is assumed to be Poisson with mean rate A.nv• The SA class considered in our paper is represented by a typical WWW session of type UDD 64 Kb/s that consists of a sequence of packet calls [10]. The HTTP session arrival process is Poisson with mean rate A.nd.

In this subsection, we try to compare the performance of DYNAA with DYNAA_wait scenario. The mean average waiting time Tdelayincreases with DYNAA_ wait (figure 3). This is quite logical because HO requests of the SA class wait before being served. On the other hand, with DYNAA_wait, more chance is given to the HO request to be served: P drop decreases due to the diminution of the forced termination of the calls. At the same time and as expected, DYNAA and DYNAA_wait manage to maintain Pdrop less than Pdrop_qos' At the call originating rate density of 0.1 calls per sec per Km2, P drop

with DYNAA_wait becomes stable at the value of 0.9% . The priority of the HA class is reflected by its degradation values lower

that those of the SA class (figure 4). The degradation parameters with DYNAA_wait are relatively less than those of DYNAA. In fact, since with DYNAA_wait the call dropping probability is reduced, the bandwidth with

140 Rola Naja and Samir Tohme

no degradation algorithm (BNDA) is applied more frequently. On the other hand, the HO requests are served whenever there is a departure. Therefore, the HO requests will use the free resources to satisfY their need while HO requests in DYNAA do not have necessarily free resources to be served with. Thus, even if some priority is assigned to HO requests, degradation is relatively attenuated with DYNAA_wait scheme. Consequently, DYNAA_wait may improve the performance ofDYNAA.

4. CONCLUSIONS

In this research, we investigated the adaptive resource allocation for multimedia applications. We proposed a dynamic adaptive architecture based on a CAC, a scheduler and bandwidth adaptation algorithms. It is shown that, by adjusting the amount of bandwidth degradation based on the current network conditions, our proposed scheme can be dynamic and consequently achieve better QoS. The proposed scenario has been extended by permitting to HO requests to wait in the HO area before being served. As a result, the overall performance has been improved.

References

[1] A.T.Campbell, R. R.-FLiao ,G. Bianchi, On Utility-Fair Adaptive Services in Wireless Networks, Sixth International Workshop on Qual1ty of Service(lWQoS 98), 1998. [2] T.Kwon, Y.Choi, C.Bisdikian, M.Naghshineh, Call Admission Control for Adaptive Multimedia in Wireless/ Mobile Networks, Proceedings of ACM workshop on Wireless Mobile Multimedia,WOWMOM'99, Page(s): 51 -58. [3] Y.Xiao, C.L.P. Chen, Y.Wang, Fair Bandwidth Allocation for Multi-class of Adaptive Multimedia Services in Wireless/mobile networks, IEEE 53rd Vehicular Technology Conference VTC 2001, Page(s): 2081 -2085 vol.3. [4] S. Kim, T. Kwon,Y. Choi, Call Admission Control for Prioritized Adaptive Multimedia Services in Wireless/mobile Networks, IEEE 51 st Vehicular Technology Conference Proceedings VTC 2000, Spring Tokyo, Volume: 2,Page(s): 1536 -1540 vol.2. [5] J.Wang, P.Cao, X.Yang, Adaptive Mobile Multimedia QoS Control and Resource Management, Ninth IEEE International Conference on Networks Proceedings, 2001. [6] Y.Xiao, C.L.P Chen, QoS for Adaptive Multimedia In WirelesslMobile Networks, Proceedings. Ninth International Symposium on Modeling, Analysis and Simulation of Computer and Telecommunication Systems 2001 Page(s): 81 -88. [7] S.K.Das, S.K.Sen,P.Agrawal, K.Basu, Modelling QoS Degradation in Multimedia Wireless Networks, IEEE International Conference on Personal Wireless Communications, 1997 Page(s): 484 -488. [8] R.Naja, S.Tohm6, DYNAA: DYNamic Adaptive Architecture for Quality of Service Provisioning in Wireless Mobile Multimedia Networks, submitted to IFIP WG6.7 Workshop and Eunice Summer School on Adaptable Networks and Teleservices. [9] R.Naja, S.Tohm6, QoS Provisioning and Handover Issues in Mobile Wireless Multimedia Networks, IEEE Workshop on Applications and Services in Wireless Networks 2002, Paris. [10] European Telecommunications Standards Institute ETSI-Universal Mobile Telecommunications Systems UMTS; Selection procedures for the choice of radio transmission technologies of the UMTS-TRIOI 112 V3.2.0-1998-04.

Dynamic QoS Guarantee with Repeater in Power Controlled WCDMA Urban Environment

Mohammad N. Patwary 1, Predrag Rapajic 1, Ian Oppermann 2

J School of Electrical Engineering and Telecommunications, University of New South Wales, Sydney, Australia.

2 Southern Poro communications 6A, Nelson Street, Annandale, Sydney, Australia

Abstract: In dense urban environment, the possibility of signal reception failed (receiving NAK) repeatedly at the receiving end of the conventional ARQ scheme is non­zero even if with strong FEC as 113 turbo code and sometime become noticeable. More robust and modified ARQ protocols are required which may adapt the FEC code rate according to the channel condition. We propose to use repeater as a reliable alternate in this critical channel condition. On the other hand ARQ scheme is efficient only for non-real time communication. Comparing with the ARQ scheme, reliable multipath from repeater will serve with lower time delay, lower SINR requirement and hence more network throughput with the cost of repeaters and multi-user detection scheme (when higher BER performance required).

Keywords: Quality of Service (QoS), Line of Sight (LOS).

1. Introduction

As the bandwidth of the third generation mobile communications systems increase, radio channel time dispersion can produce noticeably frequency selective fading within the band. The capacity demand on such systems is also increasing, leading to smaller cell size within the network. Multipath scenario has been studied from the very beginning while the CDMA system has been launched. A deterministic approach of multi-path tracking has the advantages of using models of real-network that yields an environment a realistic spread of the signal both in time and angle. Digital transmission using antenna diversity in frequency selective fading channels has been investigated in [1]. The investigations were based on fully decorrelated branch with uncorrelated reflectors with the same average power delay profile at each branch. In [2] the radio channel has been sampled

This work has been supported by Southern Poro Comnwnications, Sydney Australia

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

142 Mohammad N. Patwary, Predrag Rapajic and Ian Oppermann

simultaneously in the spatial and temporal domains on two antenna branches and the antenna signal correlation, dispersion and fading for the practical system has been analyzed. A set of antenna diversity for rural macro cell and urban microcell have been shown with array antenna in defuse channel model in [3]. In this paper we analyzed a way to assist with cell layout and to combat interference and fading degradation with optimum label of transmit power and time delay, by using base station diversity using repeater. Our simulation shows that

1. QoS in terms of time delay can be improved significantly by using repeater comparing with the performance of ARQ in Dense urban environment.

2. Using a repeater that provides a strong multipath, there will be 3dB gain in BER performance with close loop power control as WCDMA.

We organize the paper in following way. Section 2 describes a System model that has been used for simulation. Section 3 shows the comparison of the time delay performance of the network with repeater and ARQ scheme, section 4 describes the system capacity along with BER performance with and without repeater, section 5 deals with the power control scenario and the network stability and finally the conclusion.

2. System Model

For our simulation model we considered a 7-cell cluster, and the central cell serves the test mobile. Mobiles within the network are uniformly distributed and there are regular grids of Base station [4]. To increase the coverage 6 repeaters are placed in every cell in such a way that the coverage radius in the direction of the side-lobe in every sector becomes approximately equal to the coverage radius in the direction of the main lobe radiation. In doing so repeaters two repeaters are placed in the side-lobe radiation edge of every sector of the parent cell (Fig 1). All six repeaters are 60° apart from each other and 30° apart from the center of the base station sector main lobe such a way that the signal around the extended radius within the cell is easy to reach. Repeaters are selected such a way that the downlink sensitivity level of the repeater donor antenna is equal to the sensitivity level of user equipment and for uplink repeater antenna sensitivity level is the same as the base station antenna sensitivity level. Repeater uplink amplifier gain is 5dB greater than that of downlink. (These assumptions are made from the repeater specification from the manufacturer) .We also consider the terrain is uniform.

Dynamic QoS Guarantee with repeater ...

Main lobe

Base StationT"~ti;jt:;~~~

Fig: 1- Single cell with three sectors and six Repeaters

143

Repeater 1

,!t~; "Y,~ LJ ,--(..-p--Ms

BS x2 y2 Repeater 2

Fig. 2: geometrical representation of LOS and Multipath

If the base station antenna height ~s • the MS antenna height HIDI• and the horizontal distance between the base station and the Mobile station is d • then the effective distance between the BS transmitter and MS receiver is

D=~d2 + (HbS -HmsY ... (1)

for Urban microcell the difference between the BS and MS antenna height is 15m and in rural macro cell 30m. [4]. If the cell radius d» (~s - HIDI) then the effective distance can be considered as the cell radius d.

3. Time delay prediction

We considered the scenario in a particular sector where there are 2 repeaters been inserted and the repeaters are equidistant from the base station. If the distances between the base station and repeaters are XI & X2. the distances between the repeaters and the test mobile are Yl & Y2. the angles between the LOS and the multipath to the repeaters are <Xl & . <X2 and the angles between the LOS and the received signal path from repeaters to the MS are 131 & 132 respectively then:

D·sin PI ... (2) x2 =

D·sin P2 ... (4) XI =

sin[n-(al +PI)] sin[n-(a2 + Pz )] D·sinal ... (3) Y2 =

D·sina2 ... (5) YI=

sin[n - (a l + PI )] sin[n - (az + pz )]

Hence the LOS signal will travel a distance d and the multipath signal will travel a distance of (XI + Yl) and (X2 + Y2) for their respective repeater. Let the respective time delay between LOS signal and the reflected paths are 'tl and 't2. While the network is designed with the numbers of repeater considered in the system model then the maximum time delay for the receiving process completion will be'tl and't2

144 Mohammad N. Patwary, Predrag Rapajic and Ian Oppermann

T = t + max('tl , 't2) ... (6)

The relation among the t, 'tl and 't2 can be given as follows:

.!....= t+TI = t+T2 (7) d (XI + YI) (X2 + Y2) .. ,

From rectangular geometry it can be found that max (XI + YI) = max (X2 + Y2) = l.4d, i.e. transmitted signal can be available to the receiver with sufficient strength only 40% additional time after the arrival of LOS signal through the repeater. On the other hand, if the signal could not received correctly due to the fading and other attenuation; in particular we are concern about dense urban environment and if there is no available strong reflected path, the conventional solution for this problem is requesting for retransmission of the signal, which is well known as ARQ scheme.

S ..

A

S ..

t t+11 t+~ T [unit timer t 2t 3 t T (unit tim/!}

Fig. 3(a): Delay spread scenario with repeater

Fig. 3(b): Delay spread scenario with ARQ

fA-'" .. - --" I • ,, ' ... - -, -I--'

--r--llA It. I'

-t" .... --,-' ~i~ .. ~j ~' ;,;' ~l--::: ......... ""1~ .......

..... of 1'-

• ~ I t I: - -l '" - --I ,

I ! ; i ~ u ~ ~ ~ ~ u ~ ~

...,.,..,--.tGII ... _ ......... ~ ... ~

Fig. 4: QoS performance of the network with Repeater and ARQ

Fig 4 shows the test mobile distance vs QoS (in terms of time delay). Comparing with result in [2,5& 7] we may conclude that for the same amount of BER requirement the system repeater will outperform the system with ARQ scheme and with this delay performance it is possible to serve both real-time and non real-time services.

4. System Capacity With Repeater From the very beginning of the introduction of the repeater for cellular repeater for CDMA [8], to date repeaters are using as a tool to extension of coverage. In urban areas due to the shadowing there is a high probability of deep fading. We for a solution of the shadowing problem is to provide more than one LOS signal to both Uplink and Downlink. In every instant the total

Dynamic QoS Guarantee with repeater ... 145

received power is the vector sum these paths. In the proposed system model it there are two repeaters have been placed in every sector. The carrier to interference plus noise ratio (CINR) of the system for reverse link: can be given as

S CINR=- ... (8)

10 Where, S is the Power of test mobile at the BS receive and In is the total interference and noise. If I represents the total Interference N represents the total noise (both of them in dB) then 10 can be written as follows

10 = N +1 ... (9) where N = k . W . FBS • T (Without repeater) ... (10) N = k . W . (FBS + Faep)· T (with repeater) ... (11)

where k is boltzman constant, W is bandwidth, FBS and Frep stands for base station noise and repeater noise respectively. The interference of the cellular system consists of two components, which are intra-cell interference and inter-cell interference. The measure of these interferences as follows as in [12]

I ={M(l+~)-l}·VAF·S ... (12)

p·G ·G S = m BS m (without repeater) ... (13)

L p·G ·G ·G·G ·G S = m BS m d rep C (with repeater). (14)

L· Lteeder

Where, M, V AF, Pm, G, L and ~ represents the number of users, Voice activity factor, Power transmit from the mobile, different antenna gain (as suffix), path loss (as suffix, in this analysis we used Hata model) and reuse fraction respectively. Received EJNo can be measured by the following equation

(Eb) M W - = 'i.CINR. ·-·VAF No ;=1 I R

ror

... (15)

We consider the signal received at the base station receiver is directly from the mobile is of equal strength to the signal from the reflected path through the repeater. The multipath from the repeater has the same delay profile property as from the direct path [8,9] except certain delay whose sustainable limit is maximum 20J.ls in UMTS (wideband channel)[ll] without equalizer. In Fig-6 it has been shown that with scenario of both signal with equal strength gives the doubled capacity with the repeater than that of the capacity without repeater. Comparing the scenario with the soft hand off it is possible to avoid the performance degradation by using optimal maximal ratio combining diversity instead of selection diversity in forward link: and

146 Mohammad N. Patwary, Predrag Rapajic and Ian Oppermann

the best solution of this problem is adaptive multi-user detection for uplink. The BER of the QPSK modulation can be given by:

BER.". = ~ J~: ) .. (16)

With different percentage of fading through the channel the system capacity varies in between the dotted and solid line in Fig 5, for every specific bit rate.

Fig-5 : Number of user vs Received EblNO with and without repeater for different bit rate.

5. Power Controlled Scenarios

Fig-6: Number of user vs BER curve for different bit rate with and without repeater.

In our analysis we placed three repeaters per cell to provide an extra multipath with the adaptive directivity by the repeater antenna, where the direct path signal may undergo deep fading due to the shadowing. Each signals suffers from a distance dependent attenuation that can be calculated using the appropriate propagation law.

Base station

Fig: 7 -showing the schematic signal path from the base station and the repeater to the mobile station. Point A, B and C represent the positions of the mobile.

In Fig 8 Points A and B are outside the shadowing region and C in the shadowing region [6]. For every scenario, total power received by the mobile is the vector sum of all considerable multipath signals. However to demodulated the received signal we used ML multi-user detection. If S) is the signal received through the direct path and S2 is the signal received from the reflected path through the repeater then the total instantaneous received power Sins can be written as

... (17)

Dynamic QoS Guarantee with repeater .. . 147

If we consider Sins is the instantaneous required power and Slot is the total received power, A factor a introduced as a regulating factor to decide about the transmit power command (TPC) in the closed loop power control technique and the value of a is directly proportional to channel fading.

S'O' =a ·S/ns ... (18)

So the controlled received power is then it is always expected to adjust the transmit power to reach the required level according to suitable TPC. It is possible when there is at least one LOS path does exist. Repeater does the same for the shadowed region with the deep fading without degradation of QoS.

Fig-8: Cell radius vs received power with and without repeaters

In Fig-8 the dotted and the solid line represents the total received power with minimum to maximum value fading in one the multipath channel. It is found that with no fading there is also at least 10% increment in coverage when repeaters are placed to increase the system capacity. The performances of the repeaters depend on its antennas isolation and the amplifier gain [2,9] . Although the amplifier with the initial stability, the feedback conditions can render the system as a whole unstable which is due the receiving the firing signal from the coverage antenna back to the donor antenna. So as long as any signal leaving the amplifier is attenuated by an amount greater than the amplifier's total gain before it reappears at the amplifier input as feedback or interference, its effect will eventually die away and the system remain stable. However, if the signal attenuation is less than the amplifier gain, results a positive feedback and hence leads to instability [3]. This paper concern about urban areas and we proposed to use repeater to increase capacity only in urban environment. In UMTS closed loop power control tracking which keeps a continuous power monitoring in every 667J.ls, ensures the lower probability of being the network unstable.

6. CONCLUSION

In urban area where capacity problem is one of the most common issues that the operators are facing even from the beginning of the cellular system to

148 Mohammad N. Patwary, Predrag Rapajic and Ian Oppermann

serve the network with least percentage of blocking probability. We proposed to solve this problem by using repeaters within the cellular network to provide an extra multipath (LOS), which not only ensures the minimization of the blocking probability but also increase the instantaneous capacity as well as extension of the coverage. In our future work we are investigating the performance of Adaptive Multi-user detection with repeaters in WCDMA.

REFERENCE

[1]. B.Glance and L. J. Greenstein, -"Frequency Selective fading effects in digital mobile radio with diversity combining" IEEE Trans. Com vol. 31, no 2, pp 1085 -11094 Sept. 1983.

[2]. P. C. Eggers, J. Toptgard & A. M. Opera - " Antenna system for Base station Diversity in Urban Small and MocroceU" in IEEE Journal on Selected Area in Communications Vol. 11. No. 7 September 1993 pp 1046 - 1057.

[3]. O. Ncjlrklit, & J. B. Anderson - "Defuse Channel Model and Experimental Results for Array Antennas in Mobile Environments" in IEEE Transactions on Antennas and Propagation Vol. 46, No.6, June 1998 pp. 834 - 840.

[4]. Erceg, D. G. Michelson, S. S. Ghassemzadeh, L. J. Greenstein, A. J. Rustako Jr. P. B. Guerlain, M. K. Dennison, R. S. Roman, D. J. Barnickel, S. C. Wang and R. R. Miller -"A Model for the Multipath Delay Profile of Fixed Wireless Channels" - in IEEE Journal on Selected area in Communications vol. 17. No. 3 March 1999, pp 399 -410.

[5]. Michele Zorzi -"Some Results on Error Control for Burst-Error Channels Under Delay Constraints" IEEE Transactions On Vehicular Technology, Vol. 50, No. I, January 2001

[6]. J. P. Decruyenaere and D. Falconer- " A Shadowing Model for prediction of coverage in Fixed Terrestrial Wireless Systems" in VTC 1999 pp 1427-1433

[7]. K. H. Tsioumparakis. T. 1. Doumi and J. G. Gardiner -" Delay spread Statistics in a two­path Radio Environment" in VTC 1996 pp- 642 - 646.

[8] William C. Y. Lee and David J. Y. Lee -" The Impact of repeaters on CDMA System performance" in VTC 2000, pp 1763-1766.

[9]. W. T. Slingsby and J. P. McGeehan - " Antenna isolation Measurements for On-Frequency Repeaters"-IEE Antenna and Propagation conference 4-7 April 1995, pp 239-243

[10]. R. S. Larkin - "Multiple signal Intermodulation and Stability Considerations in the use oflinear repeater" in VTC 1991 pp 747-750.

[11]. K. H. Tsioumpara1ds, T. 1. Doumi and J. G. Gardiner-"Delay-Spread Considerations of Same Frequency Repeaters in Wideband Channels"- in IEEE Trans. on Vehicular Technology Vol. 46, No 3, August 1997, pp

664-675. [12]. J. S. Lee and L. E. Miller·" CDMA System Engineering Handbook:'- Artech House Publishers.

UMTS !WIRELESS LANs

Very Tight Coupling of Wireless LANs and UMTS Networks: A Technical Challenge and an Opportunity for Mobile Operators

Manfred Litzenburger, Hajo Bakker, Stephen Kaminski, Klaus Keil Alcatel Research and Innovations, D-70430 Stuttgart, Germany

Abstract: Coupling of Wireless LANs like IEEE 802.l1a1b or HIPERLANI2 systems with Public Land Mobile Networks (PLMNs) like GSM or UMTS offers benefits for both operators and users. Coverage and capacity of an operator's network can be extended with equipment expected to be considerably cheaper than for PLMN coverage. Thus, the operator can participate in the expected WLAN boom. Its customer experiences homogenous access to services with high data rate. PLMNs and WLANs should be considered as complementary systems: PLMNs provide universal coverage and high mobility support, while pico-cellular WLANs will be applied in hot spot areas, offering high data rates (up to about 45 Mbitls/cell in HIPERLANI2). Several levels of interworking are currently being defined by the relevant standardisation bodies. Tight and Very Tight Coupling integrate the WLAN into a unified Radio Access Network. But PLMNs and WLANs are based on different design philosophies. Chances, problems, and possible solutions are discussed in this paper.

Key words: Wireless LAN, Tight Coupling, Very Tight Coupling, IEEES02.11, HIPERLANI2, Multistandard Radio Access, Radio Resource Management

1. INTRODUCTION

Coupling of WLANs and PLMN systems can be implemented in different ways. 3GPP defines six stages of coupling with an increasing level of interworking [1]. The scenarios range from Open Coupling (levell, basically constituting two separate access systems with common billing only) over Loose Coupling (level 2 and 3, additionally common Authentication, Authorisation, and Accounting (AAA) services) to Tight

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

152 Manfred Litzenburger, Hajo Bakker, Stephen Kaminski, Klaus Keil

(level 4 and 5) and Very Tight Coupling (level 5 and 6). Several field trials currently are implementing and testing the Open and Loose approach (e.g., [2]). In this paper we consider coupling with UMTS. With Tight Coupling the WLAN Access Point (AP) is connected like a Radio Network Controller (RNC) to the Serving GPRS Support Node (SGSN) in the Core Network (CN), while with Very Tight Coupling it is connected to the RNC and, thus, gets an integral part of the UMTS Radio Access Network (RAN). An Interworking Unit might be added, which will be explained later. The complexity of implementation, of course, increases with the level of interworking. Tight or Very Tight Coupling means a single account for the subscriber and at least service continuity when roaming between the two access technologies. This requires, of course, the use of dual mode user terminals. The basic architectures of WLAN-UMTS Tight and Very Tight Coupling are shown in Fig. 1. We focus here on Very Tight Coupling. It provides the possibility to offer a homogenous service to the subscriber and to perform inter-system Radio Resource Management (RRM).Very Tight Coupling is an important step towards a Multistandard RAN (also called heterogeneous or co-operative networks) that integrates separate air interfaces into a single common network [3].

WLAN Very Tight Coupling

t. tile -Intcmot

WLAN Tight Coupling

Figurel. Architecture of Tight and Very Tight WLAN coupling to UMTS

An important issue is the ownership of the WLAN. While with Open and Loose Coupling the two access networks may be owned by separate operators, the most probable solution for Very Tight Coupling is that the UMTS operator owns also the WLAN part and, thus, keeps its customers within its own network.

Very Tight Coupling o/Wireless LANs ... 153

2. DESIGN CRITERIA

Important criteria when designing a Very Tight Coupling architecture are: - Use of off-the-shelf WLAN APs and Wireless Terminals (WTs), or, at

least, no hardware modifications should be necessary. This enables the user to access WLAN systems, independently of the coupling scenario, just by loading individual software drivers. As few as possible changes in the UMTS infrastructure to offer the UMTS operator a cost-efficient coupling solution. Independence from the actual WLAN standard, i.e., the requirements for the Very Tight Coupling architecture should be defined by UMTS standardisation bodies (i.e .. 3GPP) while extensions of the WLAN standard should be specified individually by the different WLAN standardisation bodies.

- A common Radio Resource Management to enable a better utilisation of the UMTS operator's radio access network by load distribution between UMTS and WLAN.

- Transparency towards the end user with respect to services which are billed independently of the used radio access system, otherwise a manual selection has to be offered.

- Seamless service continuity with respect to data loss and break time during the switch between the different radio access technologies.

WLANs can be applied for coverage extension of UMTS, especially where ubiquitous UMTS coverage is not available, e.g., within bUildings. All necessary functions for connection set-up, paging, etc. will be provided via the WLAN air interface as well. Thus, all UMTS transport channels and their mapping to MAC-d or MAC-c/sh, resp., have to be provided and transported by the WLAN systems. When coupling a WLAN to a UMTS infrastructure, two different philosophies collide. WLANs pursue a plug­and-play approach for fast and flexible reconfigurability. This enables the operator to cope with changes in configuration and topology, which may be required quite often (e.g., at fairs, exhibitions, offices). The reconfigurability is supported by the Dynamic Frequency Selection (DFS) [5] which at least is mandatory for WLANs operated in Europe. The DFS algorithm assures that each WLAN AP selects out of the range of available frequencies the best suited one with respect to interference. The frequency selection made by each WLAN AP is based on filtered interference measurements performed by the WLAN AP and its associated WTs. The DFS is carried out during start-up of the WLAN network, removal or addition of WLAN APs or the occurrence of external interferers. UMTS networks, on the other hand, are

154 Manfred Litzenburger, Bajo Bakker, Stephen Kaminski, Klaus Keil

assumed to have a rather static topology. The cell- and frequency-planning is performed offline and every change requires Operation and Maintenance (O&M) activities for reconfiguration. Thus, a UMTS network operator will not accept that each modification within a WLAN network is visible inside the RNC, i.e., some filtering function hiding the internal structure of the WLANisrequired.

In general a WLAN AP is rather intelligent and autonomous compared to a UMTS Node B, because in the UMTS RAN the intelligence is concen­trated in the RNC. E.g., if we consider mobility: In a WLAN the handover is initiated by the terminal, which selects a new AP. The AP has its own RRM and connection admission control functionality. In UMTS the RNC initiates and controls handovers based on link qUality measurements provided by the terminal and the Node B. In order to realise a common RRM it is necessary either to modify the WLAN handover procedures, which might be difficult with respect to the different WLAN standardisation bodies, or to 'hide' the mobility inside the WLAN network towards the RNC in such a way that the RNC still can control the traffic in the radio cells. Another important issue is the transport of user data traffic. Compared to UMTS access networks, WLANs provide a much larger bandwidth to the subscriber. This traffic will have to be transferred through the network the WLAN is connected to. A higher degree of interworking involves a larger number of UMTS network elements that have to provide the necessary capacity with respect to pro­cessing power and interface capabilities, which can be currently considered as to be up to around 30 Mbitls per connected WLAN AP for the advanced WLAN standards. Emphasis should be taken to concentrate this impact only to the user plane, e.g. by separation of the user and control plane in the RNC.

3. IMPLEMENTATION APPROACHES

There are several possibilities for the realisation of the Very Tight Coupling. Generally, an RNC controls a large area in a UMTS network which may contain several hot spots to be covered with WLANs, possibly located far away from the RNC location. Due to the pico-cellular nature of WLANs this means a huge number of WLAN APs has to be connected to and controlled by one RNC. Therefore it makes sense to introduce an Interworking Unit (IWU) between the RNC and the WLAN APs. This IWU, located at the WLAN hot spot, can either act as a pure traffic concentrator, i.e., without own intelligence, or additionally an IWU control plane is introduced to enable a control and supervision functionality of the WLAN APs. However this approach does not correspond with the UTRAN network topology, as 3GPP does not defme any further active network element (with

Very Tight Coupling of Wireless LANs ... 155

respect to processing and conversion of protocols) between the RNC and an UMTS Node B. Therefore, we are proposing that the IWU represents towards the RNC some kind of Node B serving one or more 'WLAN-UMTS radio cells' which correspond to the WLAN hot spot area. In this case the WLAN APs might be seen as remote installed layer 1 modules of a UMTS Node-B. For large hot spot areas the IWU should act as a Node B with multiple sectors consisting of multiple 'WLAN-UMTS radio cells'. This solution is applicable in the case of grouping several APs within a floor of a building or any other similar configuration. Each 'WLAN-UMTS radio cell' is controlled individually by the RNC. Thus a better supervision of the WLAN traffic in specific areas is achieved. The interface between the RNC and the IWU is very similar to the UMTS lub interface between RNC and Node B and should be based on available technology, e.g., ATM. The IWU functionality might be included in one specific WLAN AP or the IWU represents a standalone device. As the distance between the IWU and the WLAN APs is limited, Ethernet connections could be applied.

Compared to WLAN APs which are directly connected to an RNC, the approach based on an IWU offers the following benefits: - The kind ofWLAN network (lflPERLAN/2 or IEEE 802. 11 alb [6,7]) is

hidden towards the RNC, i.e., there is a defined split between the UMTS responsibility and the WLAN responsibility with respect to standardisation. The interface between the RNC and the IWU has to be specified by 3GPP, while the interface to the APs has to be specified by the WLAN standardisation bodies, based on interworking requirements which have to be defined by 3GPP. This means that the IWU controls common interworking procedures within the WLAN network, the implementation of theses procedures might differ between the individual WLAN systems. Other UMTS functions, e.g., authentication, security etc., which are terminal related and which are required for the very tight interworking have of course to be defined by 3GPP. As these latter functions are located within higher protocol layers, they are independent of the underlying WLAN technology. These functions should be transparent to the IWU to reduce the complexity of the IWU as much as possible.

- The WLAN topology and any configuration changes (removal or addition of APs) inside the WLAN network is hidden towards the RNC.

- The WLAN mobility is hidden towards the RNC, as all mobile WLAN terminals within the area of an IWU are still in one or more 'WLAN­UMTS radio cells'. Thus, there is no need to modify the handover algorithm of the WLAN systems.

- The Very Tight Coupling can be realised without hardware modifications of the WLAN terminals and the WLAN APs.

156 Manfred Litzenburger, Hajo Bakker, Stephen Kaminski, Klaus Keil

WLAN Very Tight Coupling can be considered as a step towards a common RRM. The RNC has to control resources in the WLAN in order to support and control mobility and connection admission. Forced handovers from UMTS to WLAN and from WLAN to UMTS should also be possible to perform load balancing. Thus, for a common RRM, the IWU has to provide continuously a load report to the RNC which comprises the current resource utilisation of the complete 'WLAN-UMTS cell'. This load report varies as it is based on internal WLAN functionality like, e.g., change of the modulation scheme, handover of WLAN terminals between 'WLAN-UMTS radio cells' (which is not controlled by the RNC), or 'on the fly' installation or removal of WLAN APs based on the plug and play principle.

To enable a seamless handover in the case of a UMTS-to-WLAN or WLAN-to-UMTS transition all WLAN data and control information should be mapped to UMTS logical channels. · The protocol stacks defmed in this paper are based on the approach to use the IP layer provided by the WLAN network to transport UMTS MAC-d and MAC-c/sh PDDs. Based on this assumption, each WLAN system offering the possibility to transport IP traffic can be used for this Very Tight Coupling approach.

II' (SoMoo)

Ft)OP Ft)OP

OlPU ur.m; ur.m; R!: R!:

we<! we<! MJ>C<f", WCod'"

1.0' 1.0' 1.0'

CN RNC IWU WLAN AP WLAN WT

Figure2 User plane protocol stack for Very Tight Coupling

A protocol stack for the user plane depicted in Fig.2 for the case of a WLAN-only WT. For details of the individual UMTS protocol entities we refer to [4]. Above the UDP layer the normal UMTS stack is applied, i.e., the architecture is completely transparent, the IP-tunnelling mechanism (PDCP, GTP-U) between the terminal and the CN is not affected. The WLAN is just used as another air interface. For a dual mode terminal equipped with a WLAN and a UMTS interface a second protocol stack below the MAC-d layer should be drawn, but as there is no difference to the standard UMTS stack, it is omitted in Fig.2. The frame protocol (FP) between the RNC and

Very Tight Coupling of Wireless LANs ... 157

the Node B, which in UMTS is necessary for the multiplexing of the MAC PDU is removed. Instead the individual IP connections between the RNC and the IWU are used for this functionality. The corresponding control plane for the "intelligent IWU" approach is depicted in Fig.3.

The generation and the analysis of the load report between the IWU and RNC is processed within the 'IWU Control' layer. The NW AP protocol (Node-W Application Part, an extension of the UMTS NBAP protocol) might be used as transport function (see Fig.3). The load report represents the total load of the overall 'WLAN-UMTS radio cell', but there is still the possibility that the traffic inside a 'WLAN-UMTS radio cell' is not distributed homogeneously, e.g., a huge number of WLAN WTs commu­nicate via one AP while other APs serve only a few WTs. In this case the IWU has to distribute the load by means of a forced handover within the 'WLAN-UMTS radio cell'. This function is included in the WLAN RRC layer. The IWU also supports the handling of the complementary handover strategies of UMTS and WLAN systems: Within UMTS the handover is controlled by the RNC for all CS and PS services transported in the DCHs. In contrast, the handover decision algorithm in WLAN terminals is completely independent from the WLAN AP, i.e. the AP is not involved, a WLAN terminal contacts without any control of the WLAN network the selected new AP. We propose the following concept: - The RNC shall not have any control with respect to handover of inside a

"WLAN-UMTS radio cell", i.e. horizontal handover. - If a WLAN terminal moves from one "UMTS-WLAN radio cell" to

another controlled by the same or another IWU the RNC is informed. - Each vertical handover is controlled by the RNC, i.e., the RNC informs

the IWU that a dual mode terminal will leave or enter the 'UMTS­WLAN radio cell' .

4. BENEFITS FOR OPERATORS

The Very Tight Coupling approach integrates WLANs together with UMTS into one single access network, owned and operated by one operator. It enables common optimisation of bandwidth resources within the network as well as on the air. It is to be expected that WLANs will proliferate in the coming years and take a considerable share of the wireless data traffic. With Very Tight Coupling the operator can build upon his established customer base and keep its customers within its own network to offer broadband services complementary to UMTS.

158 Manfred Litzenburger, Hajo Bakker, Stephen Kaminski, Klaus Keil

- --... ~c:c lEll\7l~IE ~ ~/ lJ,IIS

~ ~ lM1S ...... ~ "" -- ....,..

""'" lJ,IIS - ~ \\UN u./IS

~ II'C RC \\UN -

INC<! \\UN RC INC<!

MIC<I'" FtC ~ r-- - - t---lD' lD' lD' lD' lD'

P~ ~ I----

p 1P P 1P

Nl5 ~ \\UN -&E«23 E«23 wow: waw: AlN ~ - ~ - \\UN U U ..., ...,

!IU-PS I ~'lrlnlllffaJ (N RIle IWU Wl.AN p,p WlANWT

Figure 3 Control plane protocol stack for Very Tight Coupling

s. SUMMARY

Tight and Very Tight Coupling are the most interesting options for PLMN operators for integrating WLANs into their networks and extending the services offered to their customers. However, a lot of issues remain to be solved, so it is not to be expected before UMTS Release 6, whereas Open and Loose Coupling can be realised with today's equipment. With Very Tight Coupling UMTS and WLAN air interfaces are integrated into one single RAN, allowing a real common Multistandard Radio Resource Management for overall optimisation of the usage of radio as well as components resources. The end user experiences an universal and homogenous access to services without having to care about different technologies, accounts, or procedures.

6. REFERENCES

(1] 3GPP (TSG SA) TS 22.934 V.1.0.0: "Feasibility study on 3GPP system to Wireless Local Area (WLAN) interworking", March 2002.

[2] J. Ala-Laurila, J. Mikkonen, J. Rinnemaa: "Wireless LAN Access Network Architectures for Mobile Operators," IEEE Commun. Mag., Nov. 2001, pp. 82 - 89.

[3] Wireless World Research Forum (WWRF): "The Book of Visions 2001", http://www.wireless-world-research.org, Dec. 2001.

[4] 3GPP (TSG RAN), TS 25.401 V3.8.0: "UTRAN Overall Description (Release 99)", Sept. 2001.

[5] ETSI; HIPERLAN Type 2; Data Link Control (DLC) Layer; Part 2: Radio Link Control (RLC) sublayer; TS 101761-2 V1.3.l (2002-01)

[6] http://www.etsi.orglframesetlhome.htm?/technicalactivlHiperlanlhiperlan2.htm [7] http://standards.ieee.orglwireless/overview.html#802.11

Dynamic UMTS Simulator for Congestion Studies and Evaluation of Resource Management Techniques

Sami Nousiainen, Krzysztof Kordybach, Paul Kemppi, Veti-Pekka Kroger VIT Information Technology, Tekniikantie 4B, FIN-02044 Espoo, Finland Tel: +35894564415, Fax: +35894566027, Email: sami.nousiainen@vtt.ji

Abstract: A dynamic UMTS system level simulator has been designed and implemented in Windows NT/2000 environment using Visual C++. The objective has been to design and implement such a simulator that can be used to create and study different traffic load scenarios and resource management techniques for tackling congestion situations.

An UMTS system level simulator that is comprehensive and flexible and takes into consideration 3GPP recommendations is presented in this paper. It contains very detailed models of many important UMTS features. Simulator features designed for creating different load situations in the network are described. Furthermore, a congestion/admission control framework for UMTS is presented. Finally, simulation results illustrating the congestion causation features and congestion detection are shown.

Key words: UMTS, simulator, congestion, resource management

1. INTRODUCTION AND BACKGROUND

The operation and resource management in the air interface of CDMA based networks, such as UMTS, is considerably more difficult than in networks employing FDMAfI'DMA technology, such as GSM. Accurate system level simulations are required for the assessment of the network performance under different conditions. Some previous studies related to load control and system level simulators can be found in [1], [2], [3], [4], [5], [7].

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

160 Sami Nousiainen, KrzysztoJ Kordybach et al

2. STRUCTURE AND MODULES OF THE IMPLEMENTED SIMULATOR

Cells and UEs. Each cell has e.g. sector number, neighbors, TX power limitation and antenna coordinates. Each DE has e.g. momentary TX power, coordinates, speed, direction, mobility class, service type, QoS level, terminal type and access class.

Channel model. Common pilot channel (CPICH) transmitted in the downlink direction is modeled in the simulator. It serves as a reference for e.g. cell selection and soft handover algorithms. In addition to that, dedicated channels with data and control parts are modeled in the uplink and downlink directions. Power offset between the data and control parts can be given separately for each service type. The data part of the dedicated channel is turned off in the simulator when a DTX period occurs and in non-primary cells in SSDT mode of the soft handover [9].

Interference and measurement model. The signal-to-interference ratio (SIR) is calculated in the simulator based on the instantaneous transmission powers. The partial orthogonality of the OVSF codes in the downlink is taken into account with orthogonality factor. The SIR is calculated for dedicated channels and for common pilot channel; the common pilot channel quality can be used in handover algorithms. In soft and softer handover, signal combining is modeled with selection or maximum ratio combining.

Power control. The initial transmission power for DE in the beginning of a call and for Node-B is chosen based on open-loop power control. After that, the inner-loop power control executed at every simulation step adjusts the uplink and downlink transmission powers with a fixed step based on the SIR value [10]. Transmission power limits (maximum and minimum) can be defmed for Node-B. The DE maximum transmission power is dependent on the terminal type [11].

Cell selection and soft handover. The cell selection is based on common pilot channel (CPICH). The soft handover operations can be based either on CPICH EclIo or received power and both of these possibilities have been implemented. The link operations supported are link addition, dropping and replacement [12]. Two different algorithms have been implemented: one using relative thresholds (WCDMA as proposed in [13]) and another using absolute thresholds (no replacement operations). Also, softer handover between sectors of the same base station is supported. The signal combining in soft handover is in the uplink selection combining and in the downlink maximum ratio combining; in softer handover, the uplink combining is maximum ratio as well. Optionally, site selection diversity transmission (SSDT) can be switched on [10] influencing data channel transmission in the DL.

Dynamic UMTS Simulator for Congestion Studies ... 161

Admission and congestion control framework. The framework for admission and congestion control consists of three different thresholds for load indicators: admission, start congestion and end congestion. The first threshold is used when the decision about admitting a call is made; optionally, an estimate about the effect of the new calIon the load indicator can be used. The two latter thresholds are used in the congestion or load control and a timer is used with them: a cell is marked as congested in one direction if the load indicator remains above the start congestion threshold for long enough (timer) and the cell is said to be in normal state again after the load indicator has remained below the end congestion threshold long enough (timer). If the cell is congested in either direction (UL or DL), the cell is marked as congested. If any cell is marked as congested, a network level congestion status is changed to "congested".

In principle, any feasible and meaningful load indicators could be used in connection with the described framework but we support the usage of two different load indicators for both directions (uplink and downlink):

• sum of lISF (uplink and downlink) • total transmission power of the cell (downlink) • received interference in the cell (uplink)

In admission control, irrespective of how high the threshold values has been set, the physical resources are always checked because lack of physical resources would ultimately lead to the rejection of the call. Essential physical resources include the number of OVSF codes in the downlink and total transmission power of the cell and are thus related to two of the load indicators.

User generator and streams. The user generation and arrival in the simulator is based on defining one or multiple user streams. For each stream, the arrival intensity is given and the user arrival process is Poisson. Also, the starting time and maximum number of users can be separately specified for each user stream enabling one to create congestion peaks in time. Service type and terminal type are assigned for each user stream and thus it is possible to create e.g. such a situation, in which most of the offered traffic is speech but then at a certain instant of time some WWW users arrive at the network.

In real networks, if a user is blocked or dropped prematurely, he is likely to attempt to call again increasing offered traffic. This behavior is also modeled in the simulator by enabling one to specify the maximum number of call retries per user. The actual number of call retries is then drawn from discrete uniform distribution between zero and the given number.

Service types and QoS levels. Different service types can be defined in the simulator and a traffic model can be assigned to them. For each service type, multiple QoS levels with different parameters can be specified.

162 Sami Nousiainen, KrzysztoJ Kordybach et al

Requested user data rates are defined in the uplink and downlink, EblIo requirements are given and RAB priority is specified for each QoS. The simulator keeps track of "bad quality" time of users by observing the time when the SIR is below the target value and a maximum tolerable consecutive bad quality time can be defined for each QoS level. Exceeding this results in call dropping. Also, accumulated bad quality time is recorded during simulations for each call even if it call dropping does not occur.

Traffic model. The call duration is implicitly determined by the traffic model used in the simulator. The generated traffic characteristics such as activity and asymmetry between uplink and downlink are dependent on the service type used and thus the traffic generator should be very flexible and configurable to support many kinds of service types.

The traffic generator relies on the traffic model of ETSI [14] and consists of session, packet call and packet levels. However, the same generator can be applied for generating traffic arising from e.g. speech calls by associating the packet call level to activity or inactivity periods in the speech. Thus, by changing the traffic generator parameters it can be adjusted to produce load corresponding to a wide range of different service types.

The uplink and downlink directions can be set to be coupled, which is the normal case e.g. in WWW service type where traffic in the uplink is generated in the form of a request and then the requested data arrives in the downlink direction.

The traffic generator is a relevant module for generating congestion since it has an influence on the transmission activity in the air interface, which in turn affects the interference situation in the network.

User distribution and mobility model. The user distribution and mobility model has been described in detail in our previous paper [6]. However, in this paper we are integrating that work to a more extensive simulator framework containing plenty of other features. The most relevant features are summarized here:

• arbitrary 2D user distributions can be given in a matrix • users' mobility can be constrained to streets and the street network

can be given as input in vectored format (MapInfo format) • different mobility classes (pedestrian, vehicular) can be defined • turning probabilities at crossroads can depends on street widths or

hotspot locations

The geographical user distribution has an effect on the offered traffic load distribution between base stations and the mobility of the users influences handover rate and handover directions. Thus, different spatial congestion situations can be created using this feature.

Input and output data. Data describing the simulation scenario and parameter values are given as input to the simulator. Node-B locations, their

Dynamic UMTS Simulator for Congestion Studies ... 163

neighbors and propagation losses for each cell are given in input files. The propagation losses can thus be calculated with any proper tool (e.g. ray­tracing). Related to the simulated congestion situation, user distribution matrix, street network, service type percentages, user stream starting times and traffic model parameters are given as input. All data defining a momentary state of the simulator is stored into number of output files enabling extensive analysis.

3. SIMULATIONS AND CONCLUSIONS

Three different aspects of congestion are illustrated with example simulations. The simulations shed light on different simulator features that can be used for creating various traffic load scenarios and congestion situations [8] and show how the resulting congestion situations are detected with the congestion control framework. The general simulation parameters are listed in Table 1.

Table 1. General simulation parameters Parameter Value Background noise -108.1 dBm Congestion timer 5 s Congestion start threshold 0.3 Congestion end threshold 0.2 Power control rate every 5 slots DPDCH activity, speech 0.5 Req. EblIo in UUDL, speech 4.5/6.0 dB DPCCH-DPDCH UL, speech 3.0 dB

Parameter Active set size limit Handover timer Add threshold (relative) Drop threshold (relative) Replace threshold (relative) DPDCH activity for data Req. EblIo in ULlDL, data DPCCH-DPDCH UL, data

Value 3 1 s 4 6 2 1.0 1.515.0 dB 6.0 dB

Non-unifonn user distribution. In the first simulation, non-uniform user distribution created in such a way that the peak. of the distribution was placed between the centers of three cells and so that the user density was 7 times as high in the peak area as in other places. The radius of the peak was 200 meters. The generated users are shown in Figure 1. The active set size for each user is also marked in the same figure with color and symbol. The average DL load indicator values from all cells and for one congested cell are shown in Figure 2.

164 Sami Nousiainen, KrzysztoJ Kordybach et al

Figure 1. User locations and used active set size.

User streams. In the second simulation, two different user streams were used to create temporary congestion peak: a base stream representing constantly offered traffic and a congestion stream modeling sudden increase in the number of calls. The arrival intensity for the former stream was 1.5 and it was initialized with 100 users. In the latter stream, the arrival intensity was 3.0, it was started at 300 seconds and the maximum number of users to generate from it was 300. The total number of users in the network and separately for each stream is shown in Figure 3. As a result of these streams, in one cell the status was changed to congested at 299 seconds and back to non-congested at 399 seconds. The load indicator only barely exceeded the start congestion threshold but since it was marked as congested, the congestion status remained as long as the load indicator dropped below the end congestion threshold for at least 5 seconds.

o.s 0.8

•••• ~.totTI ... e'll' - cong.sttd eel - cong.S:lion IdIllUl

100 200 :oJ «J)

-('J

Figure 2. Average load indicator from all cells and for one congested cell.

Dynamic UMTS Simulator for Congestion Studies ... 165

Service types. In the third set of simulations, two different service types were used separately: normal speech and high data rate streaming service type. The data service type used spreading factor of 16. The initial number of users was 50. The average number of users in the simulations was 56 and 60 for speech and data, respectively. The resulting average load calculated from all cells is shown in Figure 4.

°o~~'m~~~=-~~~--~~--~~~~ Urn.!.)

Figure 3. Total number of users and number of users in each stream.

'I ! 0.2

] O.IS

t • 0 .1

0.Q5

~~~'m~~m~~~~~~~~~~~~ lime (I)

Figure 4. Average load from all cells for speech and data

Conclusions. Based on the extensive implemented simulator core, it is possible to design and implement various resource management algorithms. The algorithms could concern changing the requested QoS levels of new calls or already ongoing calls, constraining the access to the network using access classes or prioritizing e.g. emergency calls or real-time service types using RAB priorities. Soft handover algorithms could be evaluated in

166 Sami Nousiainen, KrzysztoJ Kordybach et al

different scenarios (different users distribution and mobility) and using either power based or SIR based handovers with absolute or relative thresholds.

Subsequent work concerning the simulator will focus on implementation of resource management techniques on top of the UMTS simulator described in this paper and evaluation of those resource management techniques.

Acknowledgements. This work has been partially performed in the framework of the project 1ST CAUTION, which is partly funded by the European Community. The Authors would like to acknowledge the contributions of their colleagues from National Technical University of Athens, VTT Information Technology, Cosmote Mobile Telecommunications S. A., Telia Mobile and Motorola S.p.A.

REFERENCES

(1) R. De Bernardi, D.lmbeni, L. Vignali, M. Karlsson, "Load control strategies for mixed services in WCDMA", IEEE Vehicular Technology Conference, vol. 2, pp. 825-829, Spring 2000.

(2) S. A. Ohorashi, E. Homayounvala, F. Said, A. H. Aghvami, "Dynamic simulator for studying WCDMA based hierarchical cell structures", 12th IEEE International Symposium on Personal, Indoor and Mobile Radio Communications, vol. 1, pp. 32 -37, 2001.

(3) S. Hl1ml1l1iinen, H. Holma, K. Sipill1, "Advanced WCDMA radio network simulator", PIMRC, 1999.

(4) R. Hoppe, H. Buddendick, O. WOlfe, F. M. Landstorfer, "Dynamic simulator for studying WCDMA radio network performance", IEEE Vehicular Technology Conference, vol. 4, pp. 2771-2775, Spring 2001.

(5) J. Muckenheim, U. Bernhard, "A framework for load control in 3 rd generation CDMA networks", IEEE Global Telecommunications Conference, vol. 6, pp. 3738-3742, 2001.

(6) S. Nousiainen, K. Kordybacb, P. Kemppi, "User Distribution and Mobility Model Framework for Cellular Network Simulations", 1ST Summit, Thessaloniki, Greece, 2002.

[7) W. Rave, T. Kohler, J. Voigt, O. Fettweis, "Evaluation of load control strategies in an UTRAlFDD network", IEEE Vehicular Technology Conference, vol. 4, pp. 2710-1714, Spring 200 1.

(8) CAUTION IST-2000-25352, "0-3.3: Traffic Load Scenarios and Decision-Making", October 2001.

[9] ETSI, 30 TS 25.211 v 4.0.0, "UMTS; Physical channels and mapping of transport channels onto physical channels (FOD)", 2001.

(10) ETSI, 30 TS 25.214 v 4.0.0, "UMTS; Physical layer procedures (FOD)", 2001. [11] ETSI, 30 TS 25.101 v 4.0.0, "UB Radio transmission and Reception (FOD)", 2001. (12) BTSI, 30 TS 25.303 v 4.0.0, "UMTS; Interlayer Procedures in Connected Mode", 2001. [13] ETSI, 30 TR 25.922 v 4.0.0, "UMTS; Radio resource management strategies", 2001. [14] ETSI,. TR 101.112 v 3.2.0, "Selection procedures for the choice of radio transmission

technologies of the UMTS", 1998.

Capacity and Coverage Increase with Repeaters in UMTS

Mohammad N. Patwary I, Predrag Rapajic I, Ian Oppermann 2

1 School of Electrical Engineering and Telecommunications, University of New South Wales, Sydney, Australia. 2 Southern Poro communications 6A, Nelson Street, Annandale, Sydney, Australia

Abstract: Due to the limitations of the radio wave propagation. there are possibilities of having zones in the mobile communication system where the direct signal from the base station does not reach the Mobile Station. These zones are conventionally called "dead zones" such as tunnels. shopping malls and other indoor venues. Repeaters are the most effective and an efficient way to provide the mobile service guarantee from both operator and user point of view. Besides that. operators use repeater to extend network coverage. which leads to a capacity cut off due to the repeater noise. We proposed to consider repeater to increase capacity for urban areas where path loss exponent is more than 3.4. In this paper we analyze the system capacity and coverage for both uplink and downlink with the 3GPP recommended repeater.

Keywords: Universal Mobile Telecommunication System (UMTS). Quality of Service (QoS).

1. Introduction

In UMTS cells are designed with a layered structure e.g. picocell, microcell, macrocell. The available radio resources also vary from layer to layer but remain the same QoSs demanded[lJ. One of the benefits of UMTS is improved and continuous QoS guarantee with extended capacity and coverage compared to the existing systems such as GSM. Even in regions with sufficient link budget, due to the terrain variety and the dense urban structure there are places that cannot have as good coverage as the network is designed. From a QoS perspective, the call dropping is more problematic than call blocking.

This work has been supported by Southern Poro Communications, Sydney Australia

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

168 Mohammad N. Patwary, Predrag Rapajic and Ian Oppermann

One of the most cost effective engineering solutions for this situation is to insert a repeater [2][3], which may incur only 15-20% of cost of a new base station. It is expected that the insertion of repeater reduces the capacity of the system [4]. Repeaters are usually used to extend the coverage in rural and suburban environments. We proposed to insert repeater in dense urban areas to increase the coverage as well as system capacity. We analyze the network coverage and capacity with and without repeater in different propagation condition. Our simulation shows that

1. The CDMA system capacity with repeater is the trade off with coverage, up to the path loss exponent of 3.4.

2. Beyond this point, presences of repeaters provide the system capacity and the coverage increment.

3. Doubled System capacity can be achieved by inserting repeaters to extend the coverage in the propagation environment with the exponent of in between 3.7- 3.8 for ITU pedestrian A channel and in between 3.8 - 3.9 for ITU Vehicular A channel.

Even if the operators use the repeater to extend or improve the coverage, the total interference reduces due to the reduction of intercell interference. This reduction is due to the increase of the cell coverage radius, which leads to a capacity improvement. In the following section of the paper we introduce a system model in section 2, clarify the effect of repeater on system capacity in section 3, we examine the extension of coverage scenarios with repeater in section 4, finally the conclusion in section 5.

2. System Model

For our simulation model we considered a 7 -cell cluster and the central cell serves the test mobile. Mobiles within the network are uniformly distributed and there are regular grids of Base station [1]. To increase the coverage 3 repeaters are placed in every cell in such a way that the coverage radius in the direction of the side-lobe in every sector becomes approximately equal to the coverage radius in the direction of the main lobe radiation. In doing so repeaters are placed in between the side-lobe radiation edge of two-neighbor sector of the parent cell (Fig 1 (b». All three repeaters are 1200 apart from each other and 600 apart from the center of the base station sector main such a way that the signal around the extended radius within the cell is easy to reach. Repeaters are selected such a way that the downlink sensitivity level of the repeater donor antenna is equal to the sensitivity level of user equipment and for uplink repeater antenna sensitivity level is the same as the base station antenna sensitivity level. Repeater uplink amplifier gain is 5dB

Capacity and Coverage Increase with Repeaters in UMTS 169

greater than that of downlink. (These assumptions are made from the repeater specification from the manufacturer)[5]. We also consider the terrain is uniform.

Fig: l(a) - hexagonal cell geometry with 7 cell cluster

Fig: 1 (b) - Single cell with three sectors and three Repeaters

3. Effect of repeater on system capacity

For our simulation we have used following data in both uplink and downlink.

Downlink Bit rate 12.2,144 , BS transmit power 40dBm 384··· kbps BS antennagain 15dB Repeater donor 20dB

TargetSNR 5dB ,1.5dB , 1.0dB···

antenna~ain UPLINK Feeder loss 4dB MS transmit power 24dBm Sensitivity level of -7OdBm MS antenna gain OdB Repeater donor Repeater Uplink 92dB antenna ~lifier gain DL frequency 2140MHz BS sensitivity level -96dB DL amplifier gain 87dB Repeater antenna -96dB MS sensitivity level -7OdBm UL sensitivity level Fading margin 10dB UL frequency 1950 MHz BS noise figure 5dB Test Mobile distance 200m Repeater Noise 3dB figure Chip rate 3.84 Mcps

VAF 0.4

the target EblNo of downlink radio link for any specific service can be determined by the following equation from [5]:

W·p,

(~) =~ ... (l) No '..... R·N

170 Mohammad N. Patwary, Predrag Rapajic and Ian Oppermann

Where W is chip rate, P, is the transmitted power at the Base station for specific service radio link, R is bit rate, L.n is path loss between the serving base station and test mobile, N is total interference and noise.

N =N,h+NFBs+(NFrept +1 ........... (2)

where Nth, NFBS and NFrep are thermal noise, base station noise figure and repeater noise figure respectively. Note that the repeater noise (*) will be added only when the system is considered with repeater.

1= P( X+ Y) .... (3)

Where x = (I-a) and y = r _1_ Lm LII

P is the budgeted transmit power from the base station. X is the same cell interference and Y is the other cell interference, in our simulation we considered only cells within the active set, a is orthogonality factor (orthogonality factor is the uniqueness of the transmitted signal from each base station) which is 90% for ITD Pedestrian A and 60% for ITU Vehicular A, (a is 100% for a single propagation path, we assume another 10% decrement in orthogonality factor when the system with repeater has considered), and Ln the path loss from the neighbor base stations within the active set to the test mobile. For simplicity, we can rewrite the equation (2) as follows:

N = N'+l .... (4) Where N'= N,It + NF BS + (NF rep)" From equation (1) it can be determined the power required to maintain a radio link for a specific service as the following equation:

P, = (~ ) .: .[Lm·N ] ..... (5) o ' .... e'

Substituting the value of N from equation (2), (3) and (4) in equation (5) it can be rewritten as follows:

P, = (; ) . : . [p«l-a)+Y'L".)+L"..N') ] .•• (6) o target

Since it is very difficult to define the location mobile within a practical network, hence we introduce a factor x that will be use to calculate the average instantaneous distance of any mobile in the network. Average distance of any mobile from other neighbor base station in the active set (we consider no of cells in active set are 6 as recommended for UE in [9]) is approximately equal to the distance between two base stations if they are placed in uniform grid in the network and if we consider serving cell is in the center of the 7 cell cluster and rest 6 are the members of active set. The

Capacity and Coverage Increase with Repeaters in UMTS 171

value of x have been according the above assumption as the following table where n is the propagation constant:

n x 2.0 0.850 2 .. 5 0.825 3.0 0.800 3.5 0.775 4.0 0.750 4.5 0.725 5.0 0.700

From the above assumption the value of Lm.Y of equation (5) can be determined with the relation

{d)n 1 y '4" =Z Ii .-; ... (7) where Z is the number of cell within the active

set. Also in equation (7) d is the distance of the test mobile from the serving base station and D is the distance between two base station in uniform grid network and Z is the no of cells within the active set . If we consider the transmit power budget is constant from every base station then with that amount of power total number of radio link can be supported can be find from the following relation where V AF is the Voice Activity Factor.

No _ of _ Radio _link = _P- ., .. (8) VAF·P,

In our system model repeaters are inserted in such a way that there is a probability of hearing maximum of six nearest base through the repeaters. This signal strength gives an additional increment in I (In eq. 3). On the other hand, the insertion of repeaters [with the parameters used in our system model] allowed for an increase in the distance of 60-80% in different environments, which allows a significant decrement in I. Besides that due to the additional path through the repeater (for the user under the repeater coverage i.e. if the test mobile with this region), the orthogonality factor (l reduces considerably. The capacity of the system model has been measured for different bit rate in terms of number of radio link service capability. Simulation results are given below.

172 Mohammad N. Patwary, Predrag Rapajic and Ian Oppermann

Fig 2: (a)

I-I __ _

1- -•

, , I . --t--_.

. ...-

~~~,-~--~~ •. --~~~~.=-~ ---Fig 2: (c)

.. ::.':~~.:--.:r ••

. I I _~_ I

- ' -t . , - : I

I , f 1 :

1- .. ;. '1 J ' "" I ,

.. ....... -Fig 2: (b)

Fig 2 - (d)

t ""'-__

Our simulation shows that for ITD Pedestrian A channel the repeaters within the environment of path loss exponent 2 - 3.4 (2-3.3 for ITV Vehicular A channel) increase coverage with trade off of no of radio link capacity. But within the region with this path loss exponent operators are thinking to minimize coverage cost as there are lower numbers of users within this region and which cannot make them to compensate the network installation cost. On the other hand beyond the path loss exponent of 3.4 the system capacity of the network with repeater starts to dominate the capacity of the network without repeater. A doubled system capacity can be obtained for the path loss exponent in between 3.7 - 3.8 for lTV pedestrian A channel and for lTV Vehicular A channel it is in between 3.8-3.9, which is very common in dense urban area. This gain is comes due to the capability of making possible the base stations considerably apart from each other and thus reducing intercell interference significantly. This lowered interference allows the equipments to reach their target SNR with a low power requirement from the base station to maintain the radio link. Hence with the same budgeted transmit power from the base station able to more radio link than that of the system with out repeater. From the result shown above it could be conclude that for rural and suburban areas where path loss exponent is low and coverage is most considerable issue, repeater can give significant solution. On the other hand for urban areas where path loss exponent is high, repeater can dominate the capacity of the network without repeater as well as the coverage significantly. It should be noted that even though we have considered a 7 cell hexagonal cluster, the

Capacity and Coverage Increase with Repeaters in UMTS 173

behavior of the number link supported with respect to path loss exponent will remain same regardless of the cell geometry if the network with repeater designed such a way the extended radius uniform in each direction.

4. Extension of coverage with repeater

The coverage extension is dependent on the repeater coverage antenna gain, the donor antenna gain and the repeater amplifier gain. In our system model we consider three sectors per cell each 120-degree. With ideal directive property in sector antenna, signal attenuation is lower towards the direction of the main lobe than that of the side lobe. So we proposed to install repeater in intracell hand off region i. e 60° apart from the main lobe center. Due to the additional noise in the repeater amplifier the repeater should not inserted just along the cell boundary [2]. This threshold is the function of the repeater noise figure. Usually the coverage is the most important issue of consideration while designing the network for the rural and suburban areas where the path loss exponent is lower than that of the urban areas. To overcome this coverage problem we investigated the performance of the repeater. The most effective positioning of the repeater when it is intended to extend the coverage is in between the sector boundary (in intra cell hand over region) and within the coverage threshold of the base station. We performed our simulation for different environment. In our simulation we have used the assumed data in the table I.

... •• _ ... -

:r .. . .. ~12 " t", • I· . .

~

... --~,..-" ...

Fig 3: (a) Coverage comparison in different Fig 3: (b) coverage comparison in different environment environment

Our simulation shows that the extension of coverage with the repeater is more efficient in rural and suburban areas than that of the urban area. The coverage radius can be extended 60-80% in different area but doing that there is a cut off in the capacity of about 20-50%. This cut off in capacity is found due to the repeater noise. However, statically within these environment operators are concern about coverage and available capacity could be enough to serve the existing user. A base station and three sets of

174 Mohammad N. Patwary, Predrag Rapajic and Ian Oppermann

duplex repeater can serve a coverage for which at least three base stations is to be installed.

5. Conclusion In this paper we analyzed both the system capacity and the coverage in UMTS environment where fast power control has been adopted in both the uplink and the downlink. From the simulation results it could to be conclude that the repeater, which is usually used in extension of coverage could be used to increase the system capacity as well as the coverage in dense urban area. Due to the dynamic control in each radio link power of different service and environment to meet the target SNR, it is possible to design UMTS network with repeater and control the repeater with the algorithm that have been proposed in [3] which will help to optimize the capacity cut off in rural and suburban areas. In our future work we are targeting adaptive SDMA possibility with repeater for layered structured UMTS network.

Reference: [1]. Harri Holma and Antti Taskala - "WCDMAfor UMTS" -Jhon Wiley & Sons, LTD,2oo1 edition. [2]. Moussa R. Bavafa and Howard H. Xia - .. Repeaters For CDMA System" in VTC 1998 pp- 1161-

1165. [3]. William C. Y. Lee and David J. Y Lee - "The impact of Repeaters on CDMA system" pp 1763-

1767 in VTC 2000. [4]. Wan Choi, Bong Youl Cho and Tae Won Ban "Automatic on-off switch repeater for DSlCDMA

reverse link capacity improvement' in IEEE Communication Letters Vol-4, No.5 April 2001 PP 138-141.

[5]. K. S. Gilhousen, I M Jacobs, R. Padovani, A. J. Viterbi, L. A. Weaver(Jr), C. E. Weatiey (III) - "On the capacity of cellular system" in IEEE Transaction on Vehicular technology May 1991, pp-303-312.

[6]. K. Sipila, Z. C. Honkasalo, J. L. Steffens, A. Wacker - "Estimation of capacity and required Transmission Power of WCDMA Downlink Based on a Downlink Pole Equation" in VTC 2000, pp 1002-1005.

[7]. E. H. Drucker -" Development and Application of Cellular Repeater' in VTC 1988, pp 321-325. [8]. W. T. Slingsby and J. P. Mcgeehan - "A High Gain Cell Enhancer' in VTC 1992, pp 756-758.

3GPP TS 25.133: "Requirements for Support of Radio Resource Management (FDD)". [9]. 3GPPTS25.106, "UTRA Repeater: Radio Transmission and Reception". [10] 3GPP TS25.143, "UTRA Repeater Co'!formance Testing". [11]. "Repeater Site Survey - Highway coverage with repeater" from AUgon Repeater System

Equipment.

Pre-Authenticated Fast Handoff in a Public Wireless LAN based on IEEE S02.1x Modell

Sangheon Pack and Yanghee Choi School o/Computer Science & Engineering, Seoul National University, Seoul, Korea Telephone: +82-2-880-1832, Fax: +82-2-872-2045, E-mail: {shpack, yhchoi}@mmlab.snu.ac.kr

Abstract: With the popularity of portable devices, public Internet access service using wireless LAN has started in many countries. In the public wireless LAN network, since re-authentication latency during handoff affects the service quality of multimedia applications, minimizing authentication latency is very important in order to support real-time multimedia applications on the wireless IP network. In this paper, we proposed a fast handoff scheme using the predictive authentication method based on IEEE 802.1x model. In our scheme, a mobile host entering an area of an access point (AP) performs authentication procedures for a set of multiple APs instead of the current AP. Multiple APs are selected using a Frequent Handoff Region (FHR) selection algorithm considering users' mobility patterns and their service classes. Since a mobile host is authenticated for FHR in advance, the handoff latency due to the re­authentication can be minimized. Simulation results show that the proposed scheme is more efficient than other schemes in terms of delay.

Key words: Wireless LAN, Fast Handoff' Authentication, FHR, IEEE 802.1x

1. INTRODUCTION

Public wireless Internet services based on IEEE 802.11 wireless LAN technology are becoming popular in hot spot regions such as hotels, airports, shopping malls, and so on. Unlike the existing wireless Internet service, the public wireless LAN system can provide fast Internet access at speeds up to

I This work was supported in part by the Brain Korea 21 project of the Ministry of Education, and in part by the National Research Laboratory project of Ministry of Science and Technology, 2002, Korea.

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

176 Sangheon Pack and Yanghee Choi

IIMbps using portable devices such as laptop computers and Personal Digital Assistances (PDA). In this public wireless LAN system. the user authentication and mobility support between Access Points (AP) are one of the critical issues.

To overcome some drawbacks of the existing authentication scheme, IEEE has suggested an alternative authentication scheme based on the IEEE S02.Ix model [1]. In IEEE S02.1x, a network-to-client authentication mechanism utilizing EAP (Extensible Authentication Protocol) is used as the encapsulation protocol for upper-layer authentication information [1]. Since IEEE S02.1x provides a network port access control scheme, it is more scalable and robust than other schemes. The authentication mechanism may impact network and device performances. Because mobile hosts should to be authenticated during and after handoff, the used authentication mechanism need to be responsive to the handoff time-scale required in micro-mobility environments [2]. However, since AAA servers are located at locations far away from the AP, current handoff schemes cannot meet all requirements of the real-time multimedia applications.

In this paper, we propose a fast handoff scheme that minimizes the authentication latency in a public wireless LAN. This algorithm is a centralized method based on traffic patterns and user mobility characteristics. In terms of architecture, we assume that the public wireless LAN system is based on IEEE S02.1x and uses IETF standard authentication servers. The remainder of this paper is organized as follows. Section 2 outlines the IEEE S02.1x model. In Section 3, we propose the fast handoff scheme using FHR selection. Section 4 describes the simulation results. Section 5 concludes this paper.

2. BACKGROUND

In this paper, we assumed a public wireless LAN architecture based on the S02.1x model [1]. Fig. I shows the basic components and the port-based access control mechanism. The Supplicant system is an entity at one end of a point-to-point LAN segment that is being authenticated by an Authenticator attached to the other end of that link. The Authenticator system is an entity at one end of a LAN segment that facilitates authentication of the entity attached to the other end of that link and the Authentication server system is an entity that provides an authentication service. Port Access Entity (P AE) is the protocol entity associated with a port.

In Fig. 1, the Authenticator's controlled port is in the unauthorized state and is therefore disabled from the point of view of access to the services offered by the Authenticator's system. The Authenticator P AE makes use of

Pre-Authenticated Fast Handoff. .. 177

the uncontrolled port to communicate with the Supplicant P AB, using EAPOL protocol exchanges, and communicates with the Authentication Server using Extensible Authentication Protocol (EAP). The communication between the Authenticator and the Authentication Server may make use of the services of a LAN. The public wireless LAN architecture based on 802.lx is in Fig. 2. In this architecture, the Supplicant is a user host requesting the authentication and moving from one AP to another AP. The corresponding AP and AAA server play the roles of Authenticator and Authentication server, respectively.

Recently, new authentication scheme for fast handoff is proposed [3]. This is called preauthentication scheme. In this scheme, stations can authenticate with several APs during the scanning process so that when association is required, the station is already authenticated. As a result of preauthentication, stations can reassociate with APs immediately upon moving into their coverage area, rather than having to wait for the authentication exchange. Preauthentication makes roaming a smoother operation because authentication can take place before it is needed to support an association. However, since this scheme doesn't predict where the MH moves in the future, the preauthentication may be useless in some cases and cause unnecessary authentication procedures in the wireless link.

, , , , , , , l ____ ........ ________ .. __

Wi .... LAN

Fig.I. Port-based authentication scheme

/7\ .... .... - --Autbmtiatar ArAIaIbt .. (.\_I'OW) (A~_

'I UP'*ftf'mlJJ ~

"'"" --UN

Fig. 2. The basic IEEE 8OO.Ix architecture

3. PRE-AUTHENTICATED FAST HANDOFF

In this section, we propose a fast inter-AP handoff scheme. In the scheme, a mobile host performs authentication procedures not only for the current AP but also for neighboring APs (Frequent Handoff Region), when it handoffs.

178 Sangheon Pack and Yanghee Choi

3.1 Frequent Handoff Region (FHR) Selection

The FHR is a set of adjacent APs. It is determined by the APs' locations and users' movement patterns. Namely, the FHR consists of APs with which mobile hosts are likely to communicate to in the near future. Although there are a lot of APs in a public wireless LAN, the movement ratios between each AP are not same. The handoff probability for specific APs can be calculated by the movement ratio. The movement ratio is usually affected by the AP's location and user mobility. For example, if two APs are installed within a large conference room, users may move from one AP to another AP frequently and the movement ratio will be high. However, if there are some obstacles between the APs, users will seldom move between the APs. Therefore, to find the correct movement ratio between APs, these factors should be considered.

To measure movement ratio between APs, event logging database system can be used. Table 1 shows an example of a database that records the users' login and handoff events.

Table 10 Example of Event Log Database Number UserID Login Time Handoff Time Handoff Time 1 2314 07:54:57/3 2 3452 08:00:5512 08:05:18/5 3 1093 08:04:23/3 08:14:03/6 08:15:17/4

After recording the events, we should find out the handoff ratio between APs. The handoffratio is calculated in Eq. (1) using information in the event database.

H (i 0) = N (i, j) ,J R(i,j)

(1)

H(i,j) and N(i,j) denote the handoff ratio and the number of handoff events from AP(i) to APm, respectively. RO, j) denotes the residential time in AP(i) of handoff events from AP(i) to APO). The weight values between APs are determined by the handoff ratio. Eq. (2) shows the weight value function between AP(i) and APO). wei, j) denotes the weight value.

(i = j)

(i "# j, AP(i) and AP(j) are adjacent)

(AP(i) and AP{j) are not adjacent)

(2)

Pre-Authenticated Fast Handoff. .. 179

As in Eq. (2), the weight value is inversely proportional to the handoff ratio. The weight value in the path from AP(i) to AP(i) is set to zero and the weight value between non-adjacent APs is infmite. To select the FHR, the user's service level as well as its mobility pattern should be considered. Some users may be satisfied in spite of the session disconnection during handoff. But, other users may want more seamless connectivity without any data losses during handoff. To support these users, more neighboring APs should be pre-authenticated. To consider the user's service level, we defined the weight bound value according to the users' service class. According to the value, the number of selected APs for each user is limited.

Using Eq. (1) and (2), we can obtain an N by N weight matrix, W. N denotes the number of APs. W represents the weighted bi-directional graph of AP placements. Using the W, the identifier of current AP, and a weight bound value, the FHR for a user can be selected. Detailed procedure is presented in [6]. The procedure is similar to that of Dijkstra's algorithm.

3.2 Modified Key Distribution in IEEE 802.1x Model

Since IEEE S02.Ix supports only one to one message delivery, the modified key distribution is required. Fig. 3 and 4 show the proposed key distribution. The one-time password scheme is used for the user authentication. Although a mobile host sends an authentication request to the AAA server, the server sends multiple authentication responses to all APs within an FHR. Mter receiving responses, APs except the current AP keep the authenticated information during a specific time period (soft state). If there is no handoff event during that period, the information expires and the mobile host should perform re-authentication when a handoff event occurs. In the Diameter protocol, the valid time period value of a session is delivered using an Attribute Value Pair (A VP) [4]. In addition, mUltiple keys can be distributed to the mobile host using multiple A VPs.

Fig. 4 shows the re-authentication message flow after handoff events. We assumed the AP(B) is an AP belonging to the FHR. If a mobile host hands off to APCB), since the AP(B) receives session information in advance, further message exchanges are not needed. In S02.Ix model, the controlled port changes into the authorized state after authentication procedures. In our scheme, since ports are in the ready state for fast handoff after receiving a grant response message from the AAA server, the port in the ready state can be changed into the authorized state just by checking the identifier of the mobile host, without further interaction with the AAA server. Therefore, the total handoff latency can be decreased.

180 Sangheon Pack and Yanghee Choi

r&PPU~ --lrA~Ul;i~l~;-: :-A~u.;;tk;~;i :-A~~~ 1 : PAE :: PAE(A) :: Serlin' :: PAE(8) :

: ~--+--+:: :; : : ~IAP.""""""":: :: :

JW.........,~ __ -4 ___ ..:: : ~m, '. I. '1 I

" II " I

I I, ---t:-~~:: : 'I 'I " I " I, II I

~', II II f ot1'IM' ~---~---.:: :

" II " I It - '"""TT'-- I:AJ''''--<t''\--- - .. I

:: I I'W1.u~I:: ::IPw1~I: ," ,I, I I<':';.==~.

~ .: ---:t-- wvx.y -t: -.. : =' :: n :: : _ .. _____ .... --too __________ ..... ___ ..... ___ ..... _1 ____ .. _______ 1

ri.~';';':: rA~~;'1 :"A~Um;"::-~PPiI~t--: PAE(A):: Server :: PA~8) :: fAE'

:I~~I:: ::I~ .... a.t.4I:: :. (Eq»"'O ): :: :: --.w..

: : : :: +---rr v.r . .....-: :: :: +---tt-....,Mom, I " " " : :: :: IAP~ I,

I " " " I " " I, ............ : :: :: ::........, I " " 'I~ I " " " t " ., f' : :: 1: ::DlllaT~1 I " " " .-". I I " " I. I I " " " I I '. " II I

'-- -------- -! '-- -_ .. -- -- --! ,---------- -- ~ ' .. ---- ---- .. -~

Fig. 3. Message Flow before Handoff Fig. 4. Message Flow after Handoff

4. PERFORMANCE EVALUATION

4.1 Simulation Environment

For the performance evaluation, we assumed a simulation environment in Fig. 5. In this environment, AP(4) is the current AP of a mobile host. In this simulation, we assumed that there are three types of services: Class 1, 2, and 3. Each class has three weight bound values, 1, 2, and, 3, respectively. User i denotes a user in the class i .

Fig. 5. Simulation Environment

We used the independent and identically distributed (i.i.d.) mobility model [5]. In this model, time is slotted and a mobile host can make at most one move during a slot. If a host is in AP(i) at the beginning of a slot, then during the slot it moves to AP(i+l) with probability p, moves to AP(i-l) with probability q, or remains in AP(l) with probability I-p-q, independently of its movements in other slots. Each transition probability can be found based on Eq. (2). To consider the weight value of stable hosts, we used the stability factor, a . If a = 0 , the mobile host hands off to another AP with probability

Pre-Authenticated Fast Handoff. .. 181

1. On the other hand, if a = 00, the mobile host stays in the current AP. Eq. (3) shows the transition probability between APs. P(i, j) is the transition probability from AP(i) to APG) and G is the normalization constant.

{

II

P(i, j) = ~ w(1,)

-·a G

4.2 Result & Analysis

(i '¢ j)

(i = j)

1 ( G=L-.. +a ) j"i W(lJ)

(3)

In this section, we compare the handoff latency in the proposed fast handoff scheme, the preauthentication scheme, and the general handoff scheme. The total latency is the summation of the latencies in both the wireless network and the wired network. Each latency is proportional to the hop delay in each link and the number of message exchanges. We didn't consider any processing time in the AAA server.

Fig. 7 shows the average latency when the AAA server is located in the local domain. According to the FHRSelect algorithm [6], user 1 with the lowest priority authenticates only three APs. On the other hand, user 2 and 3 authenticate five and eight APs, respectively. The handoff latency of the proposed scheme is about a half of that of the general scheme. The latency of the preauthentication scheme is similar to that of class 2. However, it requires more network resources. Fig. 8 shows the result in case of the remote AAA server. We found that the average latency in this case is much higher than that of the local sever in the general scheme. However, the latencies in the proposed scheme remain same. This is because the handoff is completed by message flows only in the wireless link. There are no re­authentication message deliveries and further server processing.

,.... 14 <> "-~ 12

~ -Normal -;:: 10 -.... - Class 1 u CI 8 .. :----.... 3 6

-----Class 2 - - Class 3 :t:: 4 ~

0 "0

2 -Pre ~ :I: 0

0 1 2 5 10 20

Stability Factor

Fig.7. Average HandoffDeJay (Local AAA Server)

182

'0 50 u

.§.40 ;... g 30 ~ j 20 It: i 10

= 0

+-

;, ~ -o

Sangheon Pack and Yanghee Choi

~

~

2 5 !O

Stability factor

--. ....

20

~Normal

-Class!

-'-Class2 - Class 3

-Pre

Fig.B. Average HandoffDeJay (Remote AAA Server)

5. CONCLUSION

In this paper. we proposed a fast handoff scheme for a public wireless LAN system. Since the handoff and re-authentication procedures are essential in public wireless LAN, we focused on the minimization of the authentication latency during the handoff. In our scheme, multiple APs selected by the predictive algorithm. The algorithm utilizes traffic patterns and users' characteristics, which are collected and managed in the centralized system. Simulation results show that the total handoff latency of the proposed scheme is much less than that of the general handoff scheme and the preauthentication scheme. In the case where the AAA server is located in a remote domain, there is an even greater decrease in handoff latency.

REFERENCE

1. IEEE Standards for Local and Metropolitan Area Networks: Port based Network Access Control, IEEE Std 802. 1 x-200 1 , June 2001.

2. A. T. Campbell and J. Gomez, "IP Micro-Mobility Protocols," ACM Mobile Computing and Communication Review, Oct. 2000.

3. Matthew S. Gast, "802.11 Wireless Networks -The Deftnitive Guide," O'Reilly, 1st Edition, April 2002.

4. Pat R. Calhoun et al., ''Diameter Base Protocol," Internet draft, draft-ietf­aaa-diameter-lO.txt, April 2002.

5. A. Bar-Noy, I. Kessler, and M. Sidi. "Mobile Users: To Update or Not to Update?" ACMJBaltzer Journal of Wireless Networks, July 1995.

6. Sangheon Pack and Yanghee Choi, "Fast Inter-AP Handoff using Predictive-Authentication Scheme in a Public Wireless LAN," Networks 2002 (Joint ICN 2002 and ICWLHN 2002), Aug. 2002.

Service Integration Multiple Access (SIMA) A Protocol for Supporting Voice and Data in Wireless LANs

Apichan Kanjanavapastitl ,2 and Hassan Mehrpour2

J Department o/Telecommunication Engineering, Mahanakorn University o/Technology, Bangkok 10530, THAILAND. E-mail: apichan@ee.unsw.edu.au

2School 0/ Electrical Engineering and Telecommunications, University o/New South Wales, Sydney 2052, AUSTRAliA. E-mail: h.mehrpour@unsw.edu.au.

Abstract: We propose a wireless medium access control (MAC) protocol based on IEEE 802.11 called Service Integration Multiple Access (SIMA) for supporting voice and data transmissions over wireless local area networks (WLANs). This proposed protocol is a hybrid protocol between time division multiple access (/'DMA) and carrier sense multiple access with collision avoidance (CSMAlCA). Since voice traffic generated by some standardized encoding schemes is classified as constant-bit-rate (CBR) trqffic, we use TDMA protocol to support the transmission 0/ the CBR voice trqffic. On the other hand, CSMAICA is still used to support the transmission 0/ data traffic. The maximum number 0/ voice calls in the same local area network when using various standardized voice encoding schemes is computed. In addition, the total transmission delay between terminals is discovered.

Keywords: IEEE 802.11, IDMA, Voice.

1. INTRODUCTION

The research area in real-time traffic transmission over wireless local area networks (WLANs) has been much interest in the past few years. One of the key drivers of this research field is the deployment of the IEEE 802.11 wireless LAN standard [1]. Real-time voice traffic can be transmitted in the IEEE 802.11 network by using point coordination function (PCF) in which polling technique is used to support the voice transmission. However, the transmission of data traffic uses distributed coordination function (DCF), which actually is a protocol named carrier sense multiple access with

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

184 Apichan Kanjanavapastit and Hassan Mehrpour

collision avoidance (CSMAlCA). Many studies of voice transmission over IEEE 802.11 have been performed. The effect of overhead for voice frame on the number of possible voice conversations has been studied in [2]. Discovery of various voice payload sizes and effect of polling scheme have been performed in [3]. In addition, an analysis of voice transmission in IEEE 802.11 has been carried on in [4].

Normally, some standardized voice encoding schemes always generate fixed-length voice packets in certain interval. The voice traffic generated by these schemes can be classified as constant-bit-rate (CBR) voice traffic. The CBR voice traffic transmitted using polling technique as used in IEEE 802.11 may lead to channel inefficiency resulted from large overheads of the polling packets. Although packing a number of voice packets generated within an interval into one large-sized packet as proposed in [4] can reduce the inefficiency, voice quality might be degraded due to a chunk of combined voice packets lost in a wireless medium. In addition, combining CBR voice traffic generated within a period might increase delay and delay jitter. Therefore, we introduce a wireless medium access control (MAC) protocol named Service Integration Multiple Access (SIMA), which is based on IEEE 802.11, for supporting CBR voice traffic generated from a standardized voice encoding scheme. We use time division multiple access (TDMA) to establish a circuit-switched connection-oriented service for the CBR voice traffic. CSMAlCA, however, is still used since it supports the transmission of data traffic. Therefore, in this paper, we only concentrate on the performance of voice transmission in SIMA. To discover the effect of various standardized voice encoding schemes, we analyze the maximum number of voice calls in the same local network when using G.729 and two encoding rates ofG.723.1. We also compute the total delay to transmit voice packets in two directions within the same local network.

The paper is organized as follows. Section 2 presents SIMA frame structure and access method. Section 3 describes delay components in voice encoding scheme. Analytical results of the protocol with different voice encoding schemes are provided in Section 4. Conclusion is presented in Section 5.

2. SERVICE INTEGRATION MULTIPLE ACCESS

2.1 Frame structure

To support voice and data traffics, channel transmission time of SIMA is separated into two periods: contention-free period (CFP) and contention period (CP), as IEEE 802.11. These two periods can be grouped into

Service Integration Multiple Access (SIMA) 185

superframe structure as shown in Figure 1. During a CFP, voice traffic is transmitted using TDMA. Thus channel transmission time in a CFP is divided into contiguous time slots. The number of TDMA slots per CFP is set to a certain number; the length of every CFP is fixed. Like IEEE 802.11, CSMNCA protocol is still used to support data transmission in a CPo Some studies use voice activity detection (V AD) to improve channel efficiency by allowing other terminals to use temporarily idle TDMA slots during silent periods. However, we did not consider the exploitation of the V AD in that way because it is difficult for the other terminals to detect the silent intervals especially in wireless environment. The other terminals may misinterpret the status of a temporarily unused slot; this results to packet collision in that slot. On the other hand, we designed that a terminal who has established a voice call has to use its own TDMA slot if it has data traffic to transmit. Hence, the data traffic of that terminal must be fragmented into small data packets in which the lengths of the packets are equal to the payload size of the voice traffic. This can improve channel efficiency and also reduce the congestion of data traffic in a CP.

~(fuallqIIl) ~(fuaI~)

ConIeni .... tm: period Cono:IJion period __ 1I:IUIIl~1qIIl -. fuallqIh variJbIe Iqth (per....,.m.n:)

- 1DMA I CSMAICA .... - ~ 1DMA I 1 I I

Figure 1. Superframe structure

To keep the delay time between two consecutive voice packets (delay jitter) from the same terminal within an acceptable range, the interval time of a superframe should be set according to the interarrival time of the voice packets depending on voice encoding scheme to be used in the protocol. Note that the interarrival time can be known from the total voice encoding delay as discussed in Section 3. Like IEEE 802.11, however, the medium may be busy at the end of a CP; this results to a foreshortened superframe length for the next period as shown in Figure 1. In this case, the quality of voice at a called terminal might be affected by the varied delay jitter.

2.2 Access protocol

Each terminal can establish a call by notifying a called terminal via an access point (AP) during a CPo After the called tenninal has confirmed the call setup, the access point assigns an uplink TDMA slot and a downlink TDMA slot in a CFP to the caller terminal (and called terminal if they are in the same local network) if some slots are still available. On the other hand,

186 Apichan Kanjanavapastit and Hassan Mehrpour

voice calls are blocked if all TDMA slots are occupied. The TDMA slot length is equal to voice payload size depending on voice encoding scheme to be used plus some headers and a short interframe space time (Tsifs), as used in IEEE 802.11 . Each tenninal who has established a call can transmit voice packets in the assigned slot for future transmission until the call is terminated.

After assigned some slots to voice calls, the unused TDMA slot may be used to transmit data traffic by any tenninal. Data transmissions in these available TDMA slots use slotted ALOHA random access protocol. A tenninal sends a data packet in any of the available TDMA slots, which is randomly selected from a discrete uniform distribution, to a destination tenninal. If the data packet is successfully received, the destination tenninal transmits an acknowledgement packet back to the source terminal in the next CP. If the source terminal cannot receive the acknowledgment packet, it attempts to send the data packet again in the next CFP. Note that the successful source terminal cannot reserve the random selected slot in the subsequent CFP for subsequent data transmission.

B ~ Beacon packel CF .... ~ CFP end packet U ~ Uplink TDMA slots D ~ Downlink TDMA slots T .... : Soon inlcrframc .pa.e time

as used in IEEE 802.11

Figure 2. Operation of SIMA protocol

As illustrated in Figure 2, the access point broadcasts a beacon packet to begin a CFP. The use of beacon packet can eliminate the problem of shifting physical position ofTDMA slots due to the medium being busy at the end of a CFP. This implies that the position ofTDMA slots in SIMA is defined as a logical position instead of physical position. In addition, the beacon packet is transmitted to allow any terminal to locate the position and identify the status of the TDMA slots. In this way, any tenninal can transmit data packets in the available TDMA slots as mentioned. The number of TDMA slots in a CFP and the status of TDMA slots are indicated in a status field as shown in the figure. We modified the ordinary beacon packet used in IEEE 802.11 by adding some extra bits to be the status field. Some bits out of the total extra bits are used to indicate the number of TDMA slots in a CFP while the remaining bits are used to specify the status ofTDMA slots.

Service Integration Multiple Access (SIMA)

3. DELAY COMPONENTS IN VOICE COMPRESSION

187

To determine the superframe length of SIMA, delay components in voice compression must be discovered. In general, analog voice signal is digitized into pulse code modulation (pCM) signals by a PCM encoder. Then the PCM samples are passed to a compression algorithm, which analyzes the samples in a block format. The size of a block depends on the compression technique to be used. Normally, a processor is used to compress a block of PCM samples. Hence, there is a delay time called processing delay occurred by the operation of the processor. This delay varies with the processor speed. However, we considered processing to be fast enough so that processing delay could be ignored. In addition to the processing delay, since the compression algorithm must have some knowledge of the PCM samples in block N+1 to accurately reproduce sample in block N, an additional delay called lookahead delay could be occurred. Therefore, the encoding delay consists of the sum of the lookahead delay and the processing delay. After compressed, the encoded/compressed bit stream is packetized incurring a packetization delay. This delay is a function of the sample block size and the number of blocks placed in a single voice packet. Therefore, the total voice coding delay is the summation between the encoding and the packetization delays. Table 1 shows some voice encoding schemes and their parameters.

Table 1. Voice encoding standards Standard Coding Bit Block Look Packeti- Total Payload

type rate size ahead zation delay (bytes) (!hl!s} {ms} {ms} delax{ms} {ms}

0.711 PCM 64 0.125 0 20 20 160 0.726 ADPCM 32 0.125 0 20 20 80

0.729 CS- 8 10 5 20 25 '20 ACELP

0.723.1 MP-6.3 30 7.5 30 37.5 24 MLQ

0.723.1 MP-

5.3 30 7.5 30 37.5 20 ACELP

4. ANALYTICAL RESULTS

In this section, we analyzed the maximum number of voice calls in which the caller terminal and called terminal are in the same local area network. In addition, the total transmission delay between the terminals was computed

188 Apichan Kanjanavapastit and Hassan Mehrpour

Table 2. Parameters for anal:xsis Parameter Symbol Unit Value

Duration of superframe TSF ms 37.5(0.723.1) and 25(0.729) Transmission rate R Mbps 2 (FHSS) and II (DSSS)

Transmission rate ofPHY header Rp Mbps I Higher layer and MAC headers H bytes 57

Physical layer header P bytes 16(2Mbps)

24(11Mbps) Max.. number of calls Ne Computed in section 4.1

Min. value ofCP TCP-lflin ms Computed in section 4.1 Max.. size Service Data Unit SmoxSDU bytes 2304

Fragment threshold size f bytes 2304 Beacon size B bytes 73 CF-Endsize CF."d bytes 24

Voice payload size V bytes 24 and 20 SIPS interval T./ft ms 0.028 A slot time Talot ms 0.05

Time to handle a voice call Ty ms Computed in section 4.1 Transmission time ofRTS

Tru Computed in section 4.1 (20 bytes) ms

Transmission time of CTS Tea Computed in section 4.1 (14 bytes) ms

Transmission time of ACK Tack Computed in section 4.1 {14 bytes) ms

4.1 Maximum number of voice calls in the same network

We derived an equation to fmd the maximum number of calls in a CFP for two encoding schemes: 0.729 and 0.723.1. The parameters for this analysis are shown in Table 2. Since IEEE 802.11 states that the header of IEEE 802.11 physical layer must be transmitted at the rate of 1 Mbps, the transmission time for sending the physical layer header should be considered separately from the transmission time of the MAC information. We followed the parameters as used in [4] for the analysis except the beacon packet size, which was adapted to use with our protocol by adding extra 264 bits in a beacon packet to be the status field as mentioned in Section 2.2. Eight bits out of 264 bits are used to indicate the number ofTDMA slots in a CFP. The remaining 256 bits are used to specify the status of each TDMA slot whether it is available or occupied. Moreover, we still used the same MAC header size, as stated in [4], for information transmission in the two periods.

If the caller and the called terminals are in the same local network, the total number of TDMA slots for voice communication is four since the communication from a terminal is sent first to the access point and then from the access point to the other terminal. Time to handle a call in a CFP is

_ 4.(V +h) 4·P 4.T Tv - + + .Ift

R Rp (1)

Service Integration Multiple Access (SIMA) 189

As [4], to find the maximum number of calls, the minimum period ofCP, 1'.:P' is required. The time needed for a maximum packet size stretch,

1'.:P-SITeICh , is also included in the calculation of 1'.:p • Therefore,

1'.:p = 1'.:p-min + 1'.:p-strelch , where

1'.:p-min = 21'./ft + 21'.101 + 8T..ck + T max

1'.:P-SITeICh = 1',.13 + 1'./ft + 1'.:13 + 1'./ft + T max

( )([ f+h] P ) S_sDu-f(m-l)+h P T .... = m-l -- +-+Tock +2T./IS + +-+T,d +T./IS

R Rp R Rp

20·8 P 14·8 P where m =r SmaxSDU /11, T"ck = 1'.:13 =--+-, and Tcts =--+-

R Rp R Rp

(2)

(3)

(4)

(5)

A beacon packet and a CF end packet are sent in a CFP, therefore the overhead of these packets, T"h' is also required to determine the maximum number of calls. Note that we let the number of beacon packets generated by the access point to have one packet per superframe. As mentioned in Section 2.1, the superframe length should be equal to the total encoding delay of the encoding scheme to be used in the protocol. To fmd the maximum number of calls, we used the following equation.

N = _Ts_'F_-_r:..:...p_-_T_O_h c T. v

(6)

As seen in Figure 3, G.723.1 provides the maximum call higher than G.729. Also, the maximum call when using G.723.l at the coding rate of 6.3 kbps and 5.3 kbps is not much different. However, quality of voice provided by an encoding scheme to be used should also be considered .

.. 35

0 .729 cs.AECLP

0 .723.1 MI'·MLQ

31.11

0 .723.1 MP·ACELP

Figure 3. Maximum number of calls in various encoding schemes

190 Apichan Kanjanavapastit and Hassan Mehrpour

4.2 Total delay of voice transmission

We computed the total delay of the voice transmission between two terminals (A and B) in the same network. When considering a typical voice packet generated, the delay suffered by this packet should include: (1) the time between its generation and the position of uplink slot, (2) the queuing time, (3) the transmission time, and (4) the propagation delay. The queuing time could be ignored because there is one voice packet per frame only. The propagation time was also neglected since the radio link is short. Therefore, the delay from terminal A to B (and vice versa) is:

T SF Tv < D < T'sF 1'"p-streclo Tv 2+"2- A++B- 2 + 2 +"2 (7)

As seen, there will be no stretching period in the lower bound. Since the superframe is fIxed, the average time of the generation time to the position of uplink slot is T SF 12. Note that the position of the uplink slot was assigned

in uniform distribution. Also, the position of downlink slot was assumed to locate next to the position of the uplink slot. When assuming there will be stretching period in the upper bound, the delay from the fIrst component is TSF 12 + Tcp-strech 12. As [4], given the delay jitter, the maximum total delay

for the two directions of the call is shown in equation (8). As seen, the smaller superframe length, the smaller total delay of the voice call.

(8)

5. CONCLUSION

Service integration multiple access (SIMA) was proposed. Voice traffic generated by voice encoding scheme is transmitted using TDMA, while data transmission uses CSMAlCA. We analyzed the maximum call for some encoding schemes. We also computed total transmission delay of the call.

REFERENCES

[1] ISOIIEC and IEEE Draft International Standards, "Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications," ISO/IEC 8802-11, IEEE P802.111D1O, Jan. 1999.

[2] M. Visser and M. Zarki, "Voice and data transmission over an 802.11 wireless network," Proc. IEEE PIMRC'95, 1995, pp. 648-652.

[3] B. Crow, I. Widjaja, J. Kim, and P. Sakai, "IEEE 802.11 Wireless Local Area Networks", IEEE Comun. Mag., Sept. 1997, pp. 116-126.

[4] M. Veeraraghavan, N. Cocker, T. Moors, "Support ofvoice services in IEEE 802.11 wireless LANs", Proc. IEEE INFOCOM, 2001, pp. 488-497.

Spatial Variation of Digital Television Signal in an Indoor Environment

AlP Ong Jin Teong, Ms. Yan Hong, Mr. Shanmugam Ganeshkumar GPS Centre, College of Engineering (BLK S2-B4c-17) Nanyang Technological University, Singapore 639798 Tel: +65 6790-5372 Fax: +65 6790-6059 Email :jtong@ntu.edu.sg

Abstract: This paper describes the equipment used and the procedures adopted for the measurement of received signals and other quality of service (QOS) parameters from a single transmitter in two identical unoccupied apartments situated at the third and fifth levels above the ground. The spatial variations of the received signal in the apartments have been analyzed and the main results presented. The behaviour of other parameters will be briefly discussed and would be presented in more details in later papers. The objective of this study is to refine the measurement procedures, which will be used later to characterize the received signal (and other QOS parameters) in different indoor environments in the dense urban Housing Development Board (HDB) environments in Singapore. The results will be used for the planning of the future fixed digital TV service in Singapore.

Key words: spatial variations, quality of service (QOS), digital TV

1. INTRODUCTION

For the planning of the future fixed wireless digital TV service in Singapore, the reception of signals and other QOS parameters in both the indoor and the outdoor environments must be characterized to be able to the determine the number and characteristics for transmitters required to provide a minimum acceptable service to be provided. Since approximately 90% of the Singapore population in lives in the high-density urban HDB environment, the characterization of QOS parameters in such environment is extremely important. The reception in such environments has not been investigated before.

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

192 Ong lin Teong, Yan Hong and Shanmugam Ganeshkumar

This paper will concentrate on initial measurements in the indoor environment [1]. For this preliminary study two unoccupied three bedroom housing units - one at level three and another at level five were used for the first campaign of measurements. These two units are part of 18-unit on a 12 story block situated in the campus ofthe Nanyang Technological University. There are several other blocks of housing blocks in the vicinity. This particular block was chosen because there is line-of-sight from level five to the TV transmitter. At level three the transmitter is blocked by housing block about 220 meters away.

In section 2 the equipment used for the measurements and the procedures will be briefly described. Measurements were made along a line in the apartment and, at selected locations, detailed measurement were made within a one square meter. Some spatial variations of the received signals are shown in section 3. Over small-scale spatial distances there are significant variations of the received signals, however if the signals over a short but larger-scale distance are averaged, the average signal varies across the room. The small-scale spatial variations of signals are similar to the temporal variations (fading) encountered in a mobile environment. In section 4, the cumulative distribution function (CDF) of the received signals is analyzed. Other interesting observations are also briefly reported in section 5.

2. Measurement Set-up and Procedures

During this experiment, transmission from the transmitter located at Bukit Batok was used. The elevation of this site is 80meters m.s.l. The antenna is located 214 meters above ground level. The transmission channel selected is Channel 37, with 602MHz as central frequency and 8 MHz bandwidth. The transmitter EIRP is around 47.78 dBW.

To obtain a reference receiving signal outside the apartment, the equipment used consisted of lIP 8595E Spectrum Analyzer, Log-Periodic Antenna, laptop computer. The Labview V5.1 software preinstalled in the laptop controlled the lIP 8595E Spectrum Analyzer via a GPffi-PCMCIA interface and recorded the measurement results (Figure 1 ). With this equipment set-up, the antenna was held outside the room window, pointing towards the transmitter for 3 minutes. A total of 256 measurements were made and the median value was taken as the outside reference receiving signal value.

Spatial Variation of Digital Television Signal ...

HP8S9S SpecIJlIm Analyzer

1.0,-_ .......

Figure 1. Equipment set-up for obtaining the signal strength outside the window

193

The indoor measurement system shown in Figure 2 consists of the following:

• TV & Satellite Analyzer (PROMAX, Prolink-7) • Indoor TV Omni-directional antenna (SOUND TECH, HL-669A) • Tripod stand • Laser pointer • Attenuators (IOdB. 6dB, 3dB) • Personal Computer

Omni-directional antenna

RS · 232

Prolink - 7 TV & Satellite Analyzer

Figure 2. Indoor Measurement equipment setup

Firstly, the measurement along a line (I-D) was conducted in the living area (on both the 5th floor and 3rd floor). A long piece of masking tape with 5cm markings was pasted on the floor from one end of the living room to the family room (1135cm in length). The layout of the apartment is shown in Figure 3.3. The antenna was mounted on a tripod stands at a height of 127cm and moved point by point along the line. For all measurement the

194 Ong lin Teong, fan Hong and Shanmugam Ganeshkumar

antenna was orientated in the direction of the transmitter. A total of 227 values were collected and used to represent the statistics of the signal strength variation along the line.

Secondly, an area measurement (2-D) was done in the master room (on Sth floor only). A 1m x 1m grid paper, with 400 points evenly distributed at Scm spacing each, was placed near the master room. The same indoor TV antenna facing transmitter site was placed over the grid paper. To determine the point of measurement accurately, a laser pointer was attached to the tripod stands and the laser beam was shone on the grid paper, coinciding with the desired point. The total amount of 400 data was collected along the path pre-assigned for further analysis.

3. SPATIAL VARIATION OF RECEIVED SIGNALS

3.1 Measurement along a line

A total of 227 values were taken along the line on both of Sth floor and 3rd floor and displayed in Figure 3 & Figure 4 respectively.

JQ

lSi) liD 11Jl lliD ;m PoIrt. on I Suoi8hC lIne(ll5 .".~~ ___ ---'

Figure 2. Signal Variation for Straight Line on Fifth floor

Spatial Variation of Digital Television Signal ... 195

Signa' V,nltiDrI for 'l1ut:I F!ocI1' ~r-----r---~~----~----~----,

~65 on

1m eli " ,tss

50

·E n /ill 100

Figure 4. Signal Variation for Straight Line on Third floor

From the figures presented above, it can be seen that there is a large variation of the received signal with small-scale changes in location. The signal varies as much as 35dB on Fifth Floor (27 dB on Third Floor) due to the effects of wall transmission loss and multipath propagation. It is observed that the signal strength drops when the received signal has to travel through more walls (Refer to Figure 5).

Drranamltte,

Figure 5. Apartment Layout

196 Ong lin Teong, Yan Hong and Shanmugam Ganeshkumar

3.2 Measurement within a square

15 ID

- ... _ .. ::.; • .: ... " .. __ .... 6

5 0 0

"""" Figure 6. 3D Plot of Signal Strength Distribution

A 3-D distribution of signal strength for the 1m by 1m square is shown in Figure 6.

The x-y axis of the graph plotted indicates the measuring points within the square while the z-axis shows the signal strength. The origin is at the bottom right hand comer of the graph. From the graph, it can be seen that the maximum and minimum signal values of the area occurring at the point (6.5,20) and point (20, 7.5) respectively.

The result shows that at points with line-of-sight (LOS) to the transmitter there is lower multipath "fading" and with no wall transmission loss, the signal strength is stronger. At positions of the points blocked by the wall or other obstruction there are high multipath "fading" and wall transmission losses; hence the signal strengths are weaker.

4. STATISTICAL DISTRIBUTIONS OF RECEIVED SIGNAL AND QOS

Figure 7 depicts the cumulative distribution function (cdf) for the signal strengths measured along the line on the fifth floor.

Spatial Variation of Digital Television Signal ... 197

CDF VI Signal Slrfnglh levellllfll1ou! Runnng Mnn

"Uit •. ;nJIG dIU' ~. .. i C(f-''IIIS: _ , 1 ;

-.. : ~ ~ i.: : .. ~', . _, ~ . .. ,-.~.- . ... _ ........ . ...... ··f •

. , w _ _ u.n I ..~: . ~ ~ . .. o..t ~ l~

: i .

0..0001 -+-.--, .• -j-...-..............-.....-.---.;.-. r ·;...,....,....-, ....... ~ :!O AO 50 «I 7Q

S9III s..n", L_ (dl..V)

Figure 7. The Cumulative Distribution Function

By using MatLab programming, the mean of the distribution is found to be 63.83 dBJ..I.V. This represents the average signal strength across the line. The median, which is the 50% point of the cdf, is also found to be 64.7dBJ..I.V. Since the mean and median are not ofthe same value,[thus it can be concluded that the signal is not log normally distributed (Note: The received signals are in dB).

The cdf of the raw data collected for the line on the fifth floor was plotted on the probability paper. It can be seen that between 26% to 95% there is a straight-line fit. This suggests that as a rough approximation a lognormal distribution could be used to model the variation of signal strengths up to approximately 95% probability. The standard deviation for this cdf is 7dB. In the definition for area coverage for DVB-T [2J, 95% coverage over a small area is defined as "good" and 70% is defined as "acceptable". To obtain a more accurate lognormal distribution, the received signals should be averaged over a short distance (to remove the large variations of the signal strength with small-scale distance changes). The analysis is not yet completed.

s. OTHER SIGNIFICANT OBSERVATIONS

In this experiment, other parameters such as the statistical distribution of BER, correlation of Rx signal with BER, time variation (small scale effects due to the presence of personnel) were investigated. When one person walked across the area between the transmitter and receiving antenna the signal strength varied; there were also amplitude variations across the video

198 Ong lin Teong, fan Hong and Shanmugam Ganeshkumar

spectrum. Difference results were obtained when directional antenna instead of omni-directional antenna were employed. The data are being analysed and will be presented in details in later papers.

6. CONCLUSIONS

In this paper, we presented preliminary results of the spatial variation of the received signal strength of the DVB-T TV received signal operating at UHF band in an indoor environment.

The signal strength on the third floor has greater variation as compared with that on the fifth floor. This is could be because that the receiving antenna on the third floor does not have any line-of-sight (LOS) to the transmitter, the multipath signals gives rise to Rayleigh "fading". On the fifth floor, there is a dominant signal, giving rise to Ricean "fading" which is not as deep as Rayleigh.

There is a rise in the wall transmission loss when there is an increase in the number of buildings and walls between the transmitter and the receiver.

The received signal could be roughly approximated by a lognormal distribution over a range of probability.

7. REFERENCES

1. NTU IHPT (In-House-Practical- Training) Report, ''EOO2 Modeling For Area Coverage For Digital TV In Singapore, December 200 1

2. Chester 1997 Multilateral Coordination Agreement Relating to Technical Criteria, Coordination Principles and Procedures for the Introduction of Terrestrial Digital Video Broadcasting (DVB-T), Chester, 25 July 1997.

8. ACKNOWLEDGEMENTS

The authors wish to thank the IHPT students who made the measurement and many colleagues for their contributions to this work.

The authors would like to thank the Singapore Broadcasting Authority for supporting this project.

MULTIPLE ACCESS TECHNIQUES

Development of a Strong Stream Ciphering Technique Using Non-linear Fuzzy Logic Selector

Ahmed M. Al-Naamany and Afaq Ahmad Information Engineering Department College of Engineering, Sultan Qaboos University

p. O. Box 33, Postal Code 123; Muscat, Sultanate o/Oman Tel.: (968) 515 327, (968) 515 328; Fax. (968) 513 416E-mails: naamany@squ.edu.om; afaq@squ.edu.om

Abstract: This paper introduces a new methodology of producing Pseudo- Random Binary Sequence (PRBS) of high complexity while with lesser burden of the processing requirements. The key idea behind the implementation of the proposed methodology is the use of a nonlinear fuzzy logic selector. A set of Linear Feedback Shift Registers (LFSRs) of nominal but varying lengths is used to reduce the processing as well as hardware burdens. In the proposed scheme, the LFSRs can be loaded with different configurations according to changed characteristic polynomials and polynomial seeds. This facilitation is being arranged such that they are to be selected by fuzzy logic selector switches. Further, the inputs to the fuzzy logic selector are based on user­defined parameters.

Key words: stream ciphering, linear feedback shift register, pseudo-random binary sequence, fuzzy logic, crypto analysis

1. INTRODUCTION

Security systems today are built on increasingly strong cryptographic algorithms that foil pattern analysis attempts [1]. The use of pseudo-random processes to generate secret quantities can result in pseudo-security [2]. The leading cryptographic concept is Data Encryption Standard (DES). It is a computerized extension of the old traditional methods, where, substitution

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

200 Ahmed M.AI-Naamany and Afaq Ahmad

and transpositions are being applied in series. Another major class of cryptosystems, popular in military applications is stream ciphers [1] - [7]. Most practical stream-cipher designs are centred on Linear Feedback Shift Registers (LFSRs). Although the Pseudo- Random Binary Sequence (PRBSs) generated by LFSRs satisfy all the randomness criteria, but the Beriekamp-Massey algorithm may recover the unknown state of a simple LFSR, and its feedback connection, with just 2n known bits of the PRBS. So that, the need to generate a more "complex" sequence led to the idea of using multiple LFSR's and somehow mixing their generated sequences so that the ultimate complexity would be the product of the individual complexities of the sequences. At present, most practical stream cipher designs centred on the non-linear combination ofLFSRs [8] - [15].

In view of above discussed issues in the field of cryptography it is the need of the day that a strong stream cipher keys must be designed and analyzed. This is actually the prime justification to carry out this work. The aim of our work is to propose such LFSR based stream ciphers designs, which have stronger keys. To make keys as strong as possible needs the analysis of PRBS used in the stream ciphers. This paper introduces a new methodology of achieving highly complexed PRBS while with lesser burden of the processing requirements. The key idea behind the implementation of the proposed methodology is the use of a non-linear fuzzy logic selector. A set of Linear Feedback Shift Registers (LFSRs) of nominal but varying lengths is used to reduce the processing as well as hardware burdens. In the proposed scheme, the LFSRs can be loaded with different configurations according to changed characteristic polynomials and polynomial seeds. This facilitation is being arranged such that they are to be selected by fuzzy logic selector switches. Further, the inputs to the fuzzy logic selector are based on user-defined parameters.

2. STREAM CIPHERING

The one-time pad inspired the technique of stream ciphering. A stream cipher has perfect security, with the obvious disadvantage that it need unlimited amount of keys. And, thus suggests to encrypt the plain text by adding "Pseudo-Random" Binary Sequence (PRBS) generated by a deterministic algorithm. To decrypt the cipher-text, the process of subtraction of the added sequence is to be carried out [2], [7], [10], [12]. Linear feedback shift register, most often used in hardware designs, is the basis of generating PRBS for the use of stream ciphers. A string of bits is stored in a string of memory cells, and a clock pulse can advance the bits one space in that string. The XOR of certain positions in the string is used to

Development of a Strong Stream Ciphering ... 201

produce the new bit in the string for each clock pulse. It is possible to choose the positions in the string to XOR so that, as long as the memory cells are not initially loaded with all zero bits, the period of the sequence of bits produced by that XOR may go up to 20 _1, where n is the number of cells in the string [8], [9]. Figure 1 illustrates an LFSR (n = 4) associated with the characteristic polynomial 1+ x + X4. If initial state, the contents of SI, S2, S3,

and s4 is 1000 then, the output (PRBS) will be 000111101011001 ... which them repeats. The characteristic polynomial of this LFSR is 1 + x + x\ which is primitive. This is so in general, an n-stage LFSR has maximal period 20 - 1 if and only if its characteristic polynomial is primitive.

-. Sl --. S2 ---. S3 ---. D-FF-l D-FF-2 D-FF-3

CI C2 C3

.. XOR ... ..... EB ....

Figure 1: A 4-cell LFSR

S4 D-FF-4

C4

.. ... Outp ut

S) (PRB

Any such generated PRBS is ultimately periodic, and so closely related to cyclic codes. In application to cryptography some essential properties of the sequence should be required. These properties are: (i) period should be very large (say 1050), (ii) sequence should be easily generated through any deterministic algorithm, and (iii) Knowing a portion of plain text should not enable the cryptanalyst to reproduce the whole sequence. Unfortunately one can show (due to linearity) that 2n consecutive bits determine the whole sequence. Just 100 consecutive bits for the sequence of length 250 -1 gives the coefficients of the primitive polynomial hence the key LFSR. Using another or string of LFSRs to make the sequence nonlinear thereby strengthens the output from an LFSR. This technique can be applied in many ways like for the telecipher devices the use of LFSRs in pinwheels, Geffe generator [5] etc. Philosophy behind all such techniques is stem from the following fundamental facts: (i) choosing which combination or combinations of LFSRs is used to contribute to the output, (ii) being XORed with the output of the chosen LFSR or LFSR blocks, (iii) choosing the different characteristic polynomials of the LFSRs, and (iv) selecting a set of initial state of the LFSRs. Keeping in view of the above, for the first time we are exploring the feasibility of strengthening the output sequence by exploiting the technique of fuzzy logic.

202 Ahmed M.AI-Naamany and Afaq Ahmad

3. FUZZY LOGIC SELECTOR - PROPOSED MODEL

The fuzzy selector and the selected set of LFSR's each with different configurations are depicted in Figure 2. Figure 3, shows the details of fuzzy logic selector where, fuzzification transforms the crisp values of Mutual and Internal Keys into fuzzy sets. This transformation is governed by evaluating the membership functions represented in Figures 4 and 5 respectively. The fuzzy rule base contains the set of fuzzy rules in linguistic form, which is used to select a specific output from a given sets of input scenarios. The fuzzy rules used in this case are enumerated in Figure 6. The inference engine evaluates the rules and then based on the preconditions thus finally, recommends a set of actions at specific degree. The overall fuzzy output is the union of the outputs resulting from each rule. The defuzzification produces a non-fuzzy control action that represents the best of recommended fuzzy actions. The fuzzifier performs a mapping from a crisp point X = (xhxnl E U ( universe of discourse) into a fuzzy set A' in U. We used the triangular shaped fuzzifier (the most commonly used). The membership function for each term in the partitioned spaced is expressed mathematically as of [1,2] and given in Equation (1) below:

I1A(X) = {1-IX~XJ(l (1)

I x-xJ> a-I otherwise

Where Xo and a are the center and the slope of the triangular membership function respectively. In the case of this model the two variables considered; are the Mutual Key and Internal Built-In key whose each values are fed into the fuzzy selector to achieve the desired non-linear selection of the LFSR to be used. For simplicity each variable's universe of discourse is divided into three input membership functions, Zero, Middle and High. This is seen to be adequate as the inputs ranges from zero to 15 only.

Development of a Strong Stream Ciphering .. . 203

M::~U.I r

Inter:;==::::! Key

Dala

64 Blls LFSR 1

-: 64 Blls LFSR I 4---1----l~ Select .. d L-____________ ---' Encrypted 18

8481

~ I I t----t---~t_ ___ ~64~~B~I~18~L~F~S~R~ __ ~ ~-4---~

,. 0

1

,;.

64 Blls LFSR I ---L--____________

Figure 2: The Fuzzy Selector system with the LFSR's

Figure 3: Details of Fuzzy Selector

--.

Figure 4: Fuzzification of Mutual Key

Input Yortoble .,r.teir)Olt('Y"

Figure 5: Fuzzification of Internal Key

De,.

Output

204 Ahmed M.Al-Naamany and Afaq Ahmad

4. IMPLEMENTATION CRITERIA

4.1 The Fuzzy Rule Base

The rule base is set up using IF-THEN rules based on the fuzzy model. The IF-THEN rules describe what action is to be taken if a certain set of conditions is met. It incorporates information about every possible condition that the system can encounter. Note that when a fuzzy rule base recognizes information that is partially true it can partially invoke more than one rule at a time. If two or more rules are to be invoked that have the same value, than the rule that is invoked to the greatest degree is chosen (the maximum), and the rest are discarded [16], [17]. This is the union fuzzy operation. In our model of the system fuzzy logic rule base is represented as a two-dimensional table look up as shown in Figure 6.

1 . II (MutueLKey is zero] end (lnterneLKey is zero] then (outputl is zero] (1] 2 . If (Mutuel Key is zero) end (InterneLKey is m id) then (output1 is low) (1) 3 . II IMutueCKey is zero) ... nd (InterneLKey is high) then (outputl is high mid) (1 J 4 . If (MutueLKey is mid) end (lntern .. LKey is zero) then (outputl is low) (1) 5 . If (MutueLKey is mid) end (InterneLKey is mid) then (outputl is mid) (1) S. If (MutueLKey is mid) end (InternaLKey is high) then (outputl is high) (1) 7 . II (MutueLKey is high) end (InterneLKey is zero) then (outputl is lowmid) (1) 8 . II Mutuel Ke is hi h end Internel Ke is mid then out utl is hi h 1

Figure 6: Fuzzy rule base of Fuzzy Inference System.

4.2 The Fuzzy Inference Engine and Defuzzification

In evaluating the rule we used the minimum operator (the fuzzy

AND operator), i.e. Jii ol = Jill (XI) A Jili (x2 ) where d denotes the output 1 2

region of rule i, and I; denotes the input region of Rule i for the j

components, Xl and X2 are the two inputs, e and e in our case. For the defuzzification an average (Centroid) defuzzification formula (Equation 2) is used to determine the outputs.

M . .

L f.t' o~ y~ ;=1 J

M . L f.t' o~ ;=1

(2)

Where j denotes the jth component of the output vector (Oij is the region of

Rule i for the jth output component, Y~ denotes the center value of region d j' M is the number of fuzzy rules in the combined fuzzy rule base. Dividing it into nine memberships allowing specific selections to be chosen for every

Development of a Strong Stream Ciphering ... 205

input configuration performs the output defuzzification. The memberships are depicted in Figure 7.

:z .

... olo.A.tput ...,.rt.~ .. o .... put1 -

Figure 7: Defuzzification of output selection

5. RESULTS & CONCLUSIONS

- - .. ~...- -.-

Figure 8 shows the result where output, internal and mutual keys can be clearly related with each other. Through this paper we put forward an approach of generating PRBS of high complexity with nominal overhead of processing cost and time. We have compared our approach with other existing methodologies for the achievement of same objective. It has been noted that this method is capable of giving very much justifiable positive indicator towards the direction of feasibility of using the Fuzzy logic theory to achieve the more complexed sequences for the use of stream ciphering.

Figure 8: Output and keys selection

6. REFERENCES

[1]. Jovan Dj • Golic, 'Recent advances in cryptanalysis', publications de 1 'lnstitutMathematique, vol.64n8, pp.183-204, 1998

[2]. Johansson, T., 'Collation Attacks On Stream Ciphers And Related Decoding Problems', Information Theory Workshop, pp.156-157, 1998

[3]. Blackburn, S.R., Brincat, K., Mirza, F. and Murphy, S., 'Cryptanalysis of Stream Cipher', Electronic Letters, Vol. 34 no. 12, pp. 1220-1221, 11 June 1998.

206 Ahmed M.Al-Naamany and Afaq Ahmad

[4]. Chi-Chun Lo, and Yu-Jen Chen, 'Secure Communication Mechanisms For GSM Networks', IEEE Transaction On Consumer Electronics, vol. 45 no. 4, pp. 1074-1080,1999.

[5]. Geffe P.R, 'How to protect data with Ciphers that are really hard to break', Electronics, pp. 99-101, Jan. 4, 1973

[6]. Alan G. Konheim, 'Cryptography: A Primer,' A Wiley-Inter-science Publication, John Wiley & Sons, 1981

[7]. Chi-Kwong Chan and L.M. Cheng, 'Design of key stream generator Electronics Letters lEE, June 1998, vo1.34, no. 12, pp 1206-07, 1998

[8]. Solomon W. Golomb, 'Shift Register Sequences', Aegean Park Press Revised Edition 1982

[9]. Wayne G. Barker, 'Cryptanalysis of Shift-Register Generated Stream Cipher Systems', Aegean Park Press, 1984

[10]. Zeng, K., Yang, C. -H., Wei, D. -Yo and Rao, T.R.N., 'Generators In Stream - Cipher Cryptography' , Computer, vol 24 no. 2 pp. 8-17, Feb. 1991

[11]. Chi-Kwong Chan, Cheng, L.M., 'The CHNN Non-linear Combination Generator' , IEEE International Conference On Electronics, Circuits and Systems, vo1.2, pp.257-260, 1998

[12]. Dachselt, F., Kelber, K., Schwarz, W. and Vandewalle, J., 'Chaotic Versus Classical Stream Ciphers- A Comparative Study', IEEE International Symposium on Circuits And Systems vol.4, pp.518-521, 1998

[13]. Chaoping Xing, and Kwok Yan Lam, 'Sequences With Almost Perfect Linear Complexity Profiles And Curves Over Finite Fields', IEEE Transactions On Information Theory, vol. 45 no. 4, pp. 1267-1270, 1999

[14]. J. OJ. Golic, 'Cryptanalysis of Three Mutually Clock-Controlled Stop/Go Shift Registers' , IEEE Transactions on Information Technology, Vol. 46, No.3, pp 1081-1090, May 2000.

[15]. Ahmad A., Al-Mushrafi M. J, Al-Busaidi S., Al-Naamany A., Jervase J., 'An NLFSR Based Sequences Generator for Stream Ciphers', SETAOI conf. pp. 11-12 May 13-17,2001 Bergen

[16]. Wang, L. ; "Adaptive fuzzy systems and control, design and stability analysis"; 1994, Prentice Hall.

[17]. J. Jang, "Neuro-Fuzzy and Soft Computing" Prentice Hall, New Jersey, USA, 1997

Wireless MAC Scheme for Service Differentiation A Distributed Protocol

Abdulla Firag and Harsha Sirisena University of Canterbury, New Zealand

Abstract: This paper presents a fully distributed Multiple Access Control (MAC) scheme that supports service differentiation in a wireless LAN environment. In the scheme, stations use CSMA for channel access, with collisions between stations having different priorities resolved by sending beacons in a predefined manner. The scheme includes an authentication protocol for stations to join a Basic Service Set and notify their priority level. Both analytical and simulation results are presented for evaluating the performance of the MAC scheme in a two priority level scenario involving voice and data traffic. The proposed scheme has a saturation throughput of about 0.84 with 1000 byte data packets, almost independent of the number of stations. The results show that good service differentiation is achieved among different priority traffic, including support for voice traffic with Quality of Service bounds on the packet latency.

Key words: priority, quality of service, collision resolution beacon, authentication protocol

1 INTRODUCTION

Mobile multimedia communication is evolving, aided by recent developments in wireless networks and portable devices like notebook computers. Multimedia applications impose requirements on communication parameters, such as data rate, drop rate, delay and jitter. Providing the Quality of Service (QoS) required for these applications is challenging.

IEEE 802.11 [1]-[2] is the most widely used WLAN standard today, but it has drawbacks with regard to QoS. In addition to the distributed (DCF) access method, the IEEE 802.11 standard also dermes a centralized (PCP) access mode for giving real-time services higher priority than non-real time services [3]-[5]. However, the centralised scheme requires the existence of

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

210 Abdulla Firag and Harsha Sirisena

access points with specialised functions, and moreover in [6] it is concluded that the centralised mode performs poorly. Recently, methods to implement QoS in IEEE 802.11 via service differentiation have been proposed [7]-[9].

This paper presents an alternative fully distributed Multiple Access Control (MAC) scheme that supports service differentiation in WLANs. The scheme is described in Section 2 and a throughput analysis is provided in Section 3. Simulation results are given and discussed in Section 4.

2 MAC PROCEDURE

In the MAC scheme, each station determines the number of stations in the same Basic Service Set (BSS)[2]. The station also learns the Identification numbers (ID) given when stations join the BSS. This information is used in achieving collision resolution as explained in Section 2.1.

Traffic can be prioritised to many levels. If a station wants to send Data of Priority level X (DX) data then it has to wait until the channel is idle for a time AIFSXN (Arbitration Inter Frame Space for priority X New data). If the station senses that the channel is idle for AIFSXN time then it can send data after a RTS/CTS (Request To Send I Clear To Send) exchange as in [1].

If CTS is not received, the station has to perform collision resolution. However if CTS is received but not an ACK (Acknowledgement) then the station has to try to resend the data again after an AIFSXN channel idle time.

When DX collision occurs and then the station senses an idle channel for AIFSXC (Arbitration Inter Frame Space for priority X Collision resolution) time, it sends a collision resolution beacon (CBX), of length AIFSXC. The timings obey: AIFS (X -1)C < AIFS (X -1)N < CAIFSXC < AIFSXN

The choice AIFSXC < AIFSXN ensures that collided data has a higher priority than new data and AlFS (X -1)N < AlFSXN so that new data D(X-l) has a higher priority than new data DX. Also, beacon length AIFSXC has to be greater than the RTS duration so that CBX has priority over new data that starts being sent at the same time. Collided stations send CBXs at the same time for the same duration, if they have collided data of the same priority.

2. 1 Collision Resolution

In collision resolution, a collided station sends a set of beacons to inform other collided stations that it has data to send. If the BSS has M stations and a DX collision has occurred, then after CBX, the collided station sends M-l beacons of length NPB (No Packet Beacon) and 1 beacon of length PPB (Packet Present Beacon) in the following order: if the collided station's ID number is N (1 ~ N ~ M), then the N'h beacon has to be a PPB. Between

Wireless MAC Scheme for Service Differentiation 211

each beacon, there must be at least CRIFS (Collision Resolution Inter Frame Space) channel idle time. CRIFS is the maximum time for a station to switch from transmitting to receiving and back again. This is shown in Fig. 1.

r- - r

lfi lOI,ruon -

"~ .~ z -- -- - +- "'-r-- I--- ~-u U) U) U)

[ZI~ ~ U) U) U) !)j x !:!: m !:!: CD~ CD ~ U) [Z U) u. U) !!: u. LL

l~ B II: Il. II: ~ ~ &: l~ u. u. 0 u.

l~ « I iii iii U Il. Q iii iii iii U) I iii

ST At 021..-- ~ r-- - f-- f- -_. I 0 -.

TP I A

StA2~~ 01 'R r- - - . -- r- -0

:;':~t- t6t C A -i- _ -

IR - I • I • T 0 I I I I I I I A

OX : Prio!!!y X Data an1val STAX = Stalion ID number X. R .. RTS. C . CTS I ~LAS8S 01 energy_inspec;t'led (belion TP • Tollen Pass. TR .. Token Received. O. DATA, A .. ACK

. -Figure 1. Colhslon Resolution for a BSS that has 3 statlOns

After the [(h NPB the station senses the channel for beacons. If a beacon is present, this means the station with ID K has sent a PPB announcing it has DX data to send. In this way collided stations learn which stations have DX data to send, and this data is then sent in the order of the stations' priority.

Each station (STA) involved in collision resolution sets a counter to the number of stations that have transmitted PPB beacons prior to it. For example, in Fig. 1, after collision resolution STAI sets its counter to 1 and ST A3 sets its counter to 2. At the end of collision resolution, a token is generated at the station that has the counter value 1, giving it the right to transmit its data. Then the token is sent to the other ST As involved in the collision. A different token is used for different collisions, so each time a collision occurs a new token is created. Stations decrement the counter after sensing an idle channel for a time SDlFS (Scheduled Data Inter Frame Space). If a station does not receive the token when its counter reaches 0, it assumes that the token is lost and will try to send the data again after the channel is idle for a time AlFSXN.

When collision resolution is over, the station with the lowest ID number that has data to send will start to send the data after the channel is idle for a time SDIPS. SDIPS has to be greater than SIPS but less than AIFSIC so that newly arriving data will not collide with scheduled data. When the station completes the data transfer it sends a TP (Token Pass) message to the station that has the next lowest ID number, from which it must receive a TR (Token Received) message. If a TR is not received and a timeout (specified channel idle time) occurs then the station sends a TP to the next STA in line. The timeout duration has to be greater than SIPS (Short Inter frame Space) but less than SDIPS so that the stations' counters will not decrement.

212 Abdulla Firag and Harsha Sirisena

Furthermore, if timeout has not occurred but a TR is not received correctly then the TP has to be sent to the next station in line as before. After reception of the TP, the receiving station has to send a TR to the sender of the TP. After sending the TR, if the next channel idle time is less than SDIFS then the station has to assume that the token has been lost.

2.2 New Stations and ID Numbers

New ST As use a similar method to the IEEE 802.11 DCF protocol to access the channel. The new ST A senses the channel before initiating a packet transmission. If the channel is idle for a time greater than an Authentication Initiation Inter Frame Space (AIIFS), then the new ST A broadcasts the authentication initiation frame (AIF). (AIIFS must be greater than the largest AIFSXN to avoid overloading the BSS with new stations). Otherwise, transmission is deferred and a backoff process is entered.

The new STA accesses the channel when its backoff timer expires. An Authentication Conftrmation Frame (ACF) is used to notify it that the frame it sent has been received. If an ACF is not received before the channel is again idle for an AIIFS time, the new ST A reenters the backoff process.

The ST As in the BSS must respond when they receive an AIF from a new ST A. If the authentication is successful, an ACF must be sent to the new STA after SIFS channel idle time but if unsuccessful, an authentication rejection frame must be sent. Each of these frames must be acknowledged by the new STA.

If an acknowledgement frame is not received from the new ST A and timeout occurs then the ST As in the BSS enter collision resolution after the channel has been idle for AIFSAC (Arbitration Inter Frame Space for Authentication Collision Resolution) as shown in Fig.2. AIFSAC must be less than AIIFS and greater than the largest AIFSXN.

I eolobn_kAIon

~ I II) '" ~ '" p - - - -, I'" '" ,OJ

-s

srT 02 - c-

~ -I ~I- - - c-- .!:!:~ -R~ A IAiF

I-I-U -- ~ ~

A1F ALChonII>allon initiation Fr1Ime A Ad<

OX 0.,. 0' prioriIyX arriYlII NSTA New STA eddk1g irIormation

ACF ALChonII>atlon conIrmation Frame STAID STA 10 ....-Wormatlon ,,.me

. . Figure 2. Addltion of new STA mto the BSS

Wireless MAC Schemefor Service Differentiation 213

The ST A with the token after collision resolution sends an authentication response frame to the new ST A. If authentication fails, the process ends when the acknowledgement frame is received from the new ST A.

After successful authentication of the new ST A, the ST A holding the token has to inform all other ST As in the BSS that a new ST A is being added to the BSS. This information includes new STAID number and new ST A address. This process is shown in the Fig. 2.

If the new station adding information could not be sent to any of the stations in the BSS, then the station sending this information has to try again after AISFAC channel idle time. Any station that does not respond again is removed from the BSS by broadcasting a removal information frame.

Each station keeps a list,including addresses and ID numbers, of stations in the BSS. When a packet is received the receiving station checks whether the sender' s address is in the list. If not, the receiver informs the sender that it is no longer a member of the BSS and so has to try to join the BSS again.

3 SATURATION THROUGHPUT

The saturation throughput is defined as the throughput achieved by the system when all stations in the BSS have non-empty transmission queues. To calculate it, assume the saturation condition is reached in the system so that every station in the BSS has the same priority data ready for sending. This situation is shown in the Fig. 3 where the BSS has 3 stations and every station always has data of priority 1 to send.

- -- -- - I- - -rl

ISTAI 01 _

Figure 3. Saturation condition of the system

In the Figure, the same situation is repeated during the next period. The period duration for a BSS that has M stations and has priority 1 data is given by:

Period = AlFStN + RTS + AlFSIC + CBt + M x CRIFS + M x PPB + (M -l)(SDIFS + ACK + D + TP + TR + 3x SIFS) + (SDIFS + ACK + D + SIFS)

214 Abdulla Firag and Harsha Sirisena

Then the Saturation Throughput = M(D-h)/period, where h = No. of Header bits / channel rate in seconds.

4 SIMULATION RESULTS AND DISCUSSION

To evaluate the performance of the MAC protocol, simulation results are obtained using MA TLAB. The parameters used are summarised in Table. 1. These values are based on the IEEE 802.11 wireless LAN standard. All the packets include MAC and PRY headers.

Table 1. Parameters used in the Simulation

Data packet payload 8000 bits MAC header 272 bits PHYheader 128 bits

ACK 112 bits + PHY header Slot Time 20ms CBI 70ms

RTS 180 bits + PHY header SIFS 10ms AIFS2C 1I0ms

CTS 112 bits + PHY header CRiFS 30ms AIFS2N BOrns

TP 112 bits + PHY header SOIFS 50ms CB2 1I0ms

TR 112 bits + PHY header AIFStC 70ms PPB 30ms

Channel Bit rate 2 MbJ>S AIFSIN 90ms NPB IOms

4.1 Average voice packet delay

Two types of voice traffic, Continuous Bit Rate (CBR) and ON-OFF, are simulated. In the ON-OFF case, the on and off periods average 300 ms and are exponentially distributed. With CBR, and during the on periods of ON­OFF traffic, stations generate 160 byte packets at 32 kb/s. Thus the inter­packet time is 40 ms and the voice frame duration (with headers) is 840 J.1s.

The results obtained are shown in Fig. 4. The average delay shown is the average duration between voice packet generation at the station and the receipt of the packet's ACK. This Figure shows that, as expected, CBR traffic experiences more delay than ON-OFF traffic. This is because with CBR more voice packets are generated in a given time interval.

!:~ ---ON/OFF voice .--l .... L .. .L. .. ~ 30 -CBR voice .., 325 <> :120

.~ 15 o ! 10 .,. j : ..... ~ ..... ~ .... ~ ..... : ...... .

o 4 8 12 16 20 24 28 32 No . of voice stations

Figure 4. Average delay of voice packets

I::r.=_~=o~.=la~~=.~d~o.~o ~--~1'~ ... i 30 _Dala ~adO .224

~ _Dala ~BdO.448 [,::::' : . :::',:::::: ~ 25 . ; .

:1 20 .i 15

CIt 10 .,. ~ Sl -·,~- - -~'''~·· ~-''~· ~~~~~ ~ 0 +-.:

o .. 8 12 16 20 24 28 32 No . 0 fvoice stalions

Figure 5 Both voice and data traffic

Wireless MA C Scheme for Service Differentiation 215

4.2 Performance of voice with data traffic present

In this simulation every station generates ON-OFF priority voice traffic. In addition, stations also generate lower priority 2 data traffic at a Poisson rate A. / M packets/so M is the number of stations in the BSS. During each simulation the data load PD = )J]data / rc of the channel is kept constant. Here b dolO is the number of bits in the data packet (including headers) and rc is the channel bit rate. The data frame length is 4.1 ms. Simulations are run for 3 different data loads and the average voice delay is shown in Fig. 5.

The Figure shows that the average voice packet delay increases with an increase in the data load. Thus, for example, if the maximum tolerable delay of a voice packet is 10 ms, then 19 voice stations can be supported with a data load of 0.448, and more than 32 voice stations with no data load.

4.3 Saturation Throughput

In the simulation, saturation is achieved by generating packets faster than the channel can serve so that all stations have non-empty transmission queues. Results for different data packet lengths are shown in Fig. 6. It is seen that the saturation throughput is not very dependent on the number of stations in the BSS but depends on the data packet length. The simulation results agree closely with the analytical result obtained in Section 3.

0.9 -r--~---~------...,

0.8 :; ~0.7 '" g" 0.6 ........ ;. . . .. . ~ ........ : ........ ~ .... ... . o " . ..c 0.5 ... g 0.4 . . . .

.~ 0.3 · .. -----;---------:----.-' . ~ 02 ...... : ......... : ...... __ 2statlons (J) _ 10 stations

0.1 ....... . ................ __ 30 stations

O ~--~--~-=~==~~ o 200 400 600 800 1000

Packet payload length (bytes)

Figure 6. Saturation throughput

4.4 Service Differentiation

4 8 12 16 20 24 28 32 No .ofvolce stations

Figure 7. Two voice priorities

In this simulation every station in the BSS generates two streams of voice traffic, one with priority 1 and another with priority 2. In each case, voice traffic is of the ON-OFF type. The results obtained are shown in Fig. 7. It is

216 Abdulla Firag and Harsha Sirisena

seen that priority 2-voice stream experiences more delay than does the priority 1 stream. This confirms that service differentiation is achieved among traffic of different priorities.

5 CONCLUSIONS

A distributed multiple access control scheme was proposed that was shown to provide service differentiation in wireless LAN environment. The MAC scheme presented can support any desired number of priority levels. A new approach is used to resolve the collisions between stations. With this new approach a saturation throughput of about 0.84 is obtained with packet data load length of 1000 bytes and this does not depend very much on the number of stations in the network.

A simulation study was also presented to assess the impact of data traffic on the voice traffic. It was shown that for a wireless LAN operating at 2 Mbps with ON/OFF voice using coding rate of 32 kbps can support 19 voice stations with simultaneous data utilization of 0.448 with a maximum accepted voice packet delay of IOms. Furthermore, this increases to 30 voice stations when data utilization is decreased to 0.224.

References

[1] Wireless LAN Medium Access Control (MAC) and Physical Layer (PRY) Specifications, IEEE Standard 802.11, June 1999.

[2] B. P. Crow, I. Widjaja, J. G. Kim and P. T. Sakai, "IEEE 802.11 Wireless Local Area Networks," in IEEE Communication Magazine, Volume: 35 Issue: 9, Sept. 1997, Page(s):116-126

[3] M. Veeraraghavan, N. Cocker, T. Moors, "Support of voice services in IEEE 802.11 wireless LANs," in INFOCOM 2001. Proceedings. IEEE, vol: 1,2001, Page(s): 488 -497

[4] Jing-Yuan Yeh, C. Chen, "Support of multimedia services with the IEEE 802.11 MAC protocol," in Communications, 2002. ICC 2002. IEEE International Conference, vol: I, 2002, Page(s): 600-604

[5] A. Petrick, "Voice Services over 802.11 WLAN," in IIC, Taipei, Conference Proceedings, Page(s): 61-63

[6] M. A. Visser and M. EI Zarki, "Voice and data transmission over an 802.11 wireless networks," in Proceedings ofPIMRC'95, Toronto, Canada, September 1995.

[7] J.L. Sobrinho and A.S. Krishnakumar, "Distributed Multiple Access Procedures to Provide Voice Communications over IEEE 802.11 Wireless Networks," GLOBECOM '96, Volume: 3, 1996, Page(s): 1689 -1694

[8] I. Aad, and C. Castelluccia, "Differentiation mechanism for IEEE 802.11," INFOCOM 2001. Proceedings. IEEE, Volume: 1,2001, Page(s): 209 -218

[9] A. Veres, A. T. Campbell, M. Barry, " Supporting Service Differentiation in Wireless Packet Networks Using Distributed Control," Selected Areas in Communications, IEEE Journal, Volume: 19, Issue: 10, October 2001, Page(s): 2081-2093.

Packet Acquisition Evaluation of Slotted Spread ALOHA Data Networks

Waseem Jibrail and Ranjith Liyana-Pathirana University o/Western Sydney School 0/ Engineering and Industrial Design Locked Bag 1797. Penrith South DC 1797. NSW, Australia

Email: w.jibrail@uws.edu.au r.liyanapathirana@uws.edu.au

Abstract: In this paper packet acquisition performances of both slotted spread ALOHA and spread ALOHA multiple-access schemes are examined and compared. The probability of packet acquisition of both schemes are measured as users compete to gain access to the network hub receiver in the presence of additive white Gaussian noise (A WGN) and narrowband interference. Moreover. channel load sensing protocol (CLSP) is employed at the receiver hub in order to improve packet acquisition performance.

Results clearly show that the slotted spread ALOHA scheme outperforms spread ALOHA in terms of the probability of packet acquisition. It is also shown that this probability may be improved further by employing the CLSP at the receiver hub.

Key words: Spread Spectrum Synchronisation. Spread Aloha

1. INTRODUCTION

A spread ALOHA hub station in a data network need only be capable of synchronising to the received signals, all of which use the same Pesudonoise (PN) code, a much simpler problem than that faced by CDMA hub, where PN-codes received all different. On the other hand, the problem of packet acquisition (also known as PN-code acquisition) imposes a limitation on the network capacity, as the capacity of wireless data-networks is essentially limited by the number of simultaneous users that can achieve and maintain

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

218 Waseem librail and Ranjith Liyana-Pathirana

packet acquisition, rather than by the number of users that can maintain a certain bit error-rate during the data-demodulation process [1]. Using only one pseudonoise (PN) code, spread ALOHA multiple-access scheme users compete to gain access of the receiver hub at any time, while in slotted­spread ALOHA users are only allowed to transmit their packets in the beginning of each time slot as shown in Fig.!. Recently, packet acquisition performance (in terms of the probability of packet acquisition P a) of a spread ALOHA scheme for wireless data-networks has been evaluated in the presence of AWGN and narrow band interference [2]. Each user in the network is initially required to gain access to the receiver hub by transmitting a preamble (i.e. a packet without data).

The network hub employs one simple serial-search receiver based on a non-coherent digital sliding correlator [3,4]. The serial-search receiver operates by receiving the direct sequence (DS) signal, then searching serially for the correct signal phase using a detector, namely, a digital digital sliding correlator. It was shown [2] that the probability of successful packet acquisition depends on the operating characteristics of the receiver (Le. probability of detection P d and probability of false alarm Pf ), the number of users r and the code length and it is expressed as:

Pa = ( )( } ( n )"" ....... (1) 2 - Pd 1 + r.P! 2r 2 -1

For a given receiver parameters such as, bandwidth B, signal-to-noise ratio (SNR), threshold level Vnth and dwell-time (Le. integration time) Td, Pd and Pf may be expressed as [5].

P f = Q[(V nth -1)~ BT d ]. ................................. (2a)

Pd = QkVnth -1-SNR)~BTd/(1+2.SNR)J ...... (2b)

Where

1 00

Q(x) =- Jexp(-u 2 /2)du 2n 0

Furthermore, the type of receiver employed by the hub and its key parameters have either a direct or indirect effect on packet acquisition

Packet Acquisition Evaluation of Slotted Spread ... 219

performance and hence the probability of packet acquisition. Moreover, it was shown that [2-4] adaptive threshold detection at the receiver hub plays a significant role in improving the probability of packet acquisition.

To further improve the packet acquisition performance of spread ALOHA, a channel load sensing protocol (CLSP) [6,7] may be employed at the hub station receiver. In CLSP the hub senses the channel load status (i.e. number of packets being transmitted). If channel load is over a certain

• • Time Slot

Fig.1 Slotted Spread ALOHA

Channel

UI

U2

U3

Cbannel

Load

threshold, then packet access to the hub is rejected until the channel load falls below threshold. In this way a maximum number of users are allowed to access the network hub for synchronisation purposes.

2. PERFORMANCE COMPARASION

The up-link part of the network, shown in Fig.2, is implemented using MATLAB. As network users transmit their packets simultaneously, they are also kept at different distances from the hub receiver by randomly varying the initial phase of the transmitted PN-codes in each run. Variable initial phase is realised by randomly delaying the outputs of all transmitters as shown in Fig.2. Packets transmission are only allowed at the beginning of each time slot using gating. Extensive simulations are conducted of the slotted Spread ALOHA scheme to measure the probability of packet acquisition in the presence of A WGN, narrowband interference and multiple-access-interference (MAl). Fig.3. shows the probability of packet acquisition as a function of correlation length of the digital sliding correlator

220 Waseem librail and Ranjith Liyana-Pathirana

Narrowband Interference

Fig.2 Block Diagram of Simulated Network

Network

Hub

Receiver

for both slotted spread ALOHA and spread ALOHA. Results were obtained for a signal-to-noise ratio (SNR) of 0 dB, interference-to-signal ratio (ISR ) of 12.5 dB and PN-code length ofL=255 chips.

c: o

0.1 r;=::::::;;::t:::::;~~=::::::;----:----:-----r----:--;:-l ..... Sprlld ALOHA -+- Stallod Sproad ALOHA , ,

OJ19 ------,------ .------,------t------;-----'.c:I I I I I I I I

:§ 0.00 ------:------~------~------f- ----r------:----- ..... ::J I I I I • 0" 'I I I I V • I I I I I I

« "" " j 0.07 ------~-----~------!-- ---!- ---~-----~------i-----~ ::'.::: 0. • I I I I ,

- 0.00 ------:..-----.:--- -,.- ----~------~-----..:------~-- .. --o I I I • I I ,

~ ::.:::: =- I I • I I I I

..g 0,05 ------:...---- -- ---!------~------~---- - ..:--- ---~-----,0 I " 1 I I I

e: :::: a.. I ::::

0.04 - -- ----~------:--- .. -:------~-----~------~-----I I I I I I , I I I f I I I I I I f I I I I I ,

15 20 25 :ll 35 40 45 50 Correlation Length "Chips"

Fig.3 P a as a function of correlation length

Fig.4 shows P a as a function of ISR for adaptive and constant threshold detection at SNR= 0 dB, for both spread and slotted spread ALOHA schemes. while, Fig.5 illustrates P a as a function of number of users for both spread and slotted spread ALOHA schemes and at SNR= 0 dB and ISR = 12.5 dB. Clearly, simulation results demonstrate that the performance of the slotted spread ALOHA scheme is superior to the spread ALOHA scheme in

Packet Acquisition Evaluation of Slotted Spread ...

0.'8 r----iF=;;;=::::;:::=:=?======:===i, • Spread Aloha (Constanl Threshotd)

I: 0.' o E ..!! 0.14 ::s IT u 0( 0.12 ~ ;x :il 0.1 II..

'0 ~0.08

j 0.06

e 0.. 0.04

~ Spread Aloha (Adaptive Threshold .. - - - - - - • • -.. • •• + S-Spread Aloha (Constant Th",shold) -

-A- S-Sproad Aloha (Adaptiv. Threshold) ... --- --... ... ... ..... ----~ ... -_ ... --_ ... -_ ...... ... ... _ ...... ~ ... --_ ... -............ --- -_ ...

.... - -- -- ------r-···-···~········t·······:!: · · · ---· , , , ,

.. .. ... _-_ ... -.. "' ''' ... --- ~ ...... -_ ... -- -- --_ ... --_ ... ~ - ..... -_ ... ... -_ ... --_ ... ......

• • 0.025:----- -'':--0 ------"5'-------120

ISR(dB)

Fig.4 P a as a function of ISR

221

terms of the probability of acquisition for the same receiver at the network hub. This gain in the probability of packet acquisition is significant when adaptive threshold detection is employed at the receiver. Table I illustrates this improvement in P a for different ISR 's and different number of users r.

0.04 r-;--r==:===~=;:===::,c::::::;;=~;=::::::;-, : + Sp, .. d Aloa (emont Thr •• hold

0.035 I: o ~ 0.03 "5 IT

:J. 0.025 ... .. ;x :il 0.02 Q. ... o ~0.O'5

j 0.01

e 0.. 0.005

: ~ Spr •• d Aloh. (Adoptlv. Thr,""old) •• ~ • • • - .. . • S .Spread Aloh. (COMlOnl Th,.IIl.leI)

: -A- S-Spr .. d Aloha (Adaptlv. Thr,""oldl

...... - [-· ··· ---f--- - SN~.:OdB:l~~1i6-d~·······

--- ... +-- '-- - ...... --- -- ~ - ... .... - ...... ... t ... -_ ... --- ... t ...... --- ..... -t ...... ---- ... " , " , , ,

Fig.5 P a as a function of number of

users

Generally, results indicate that as MAl (number of users) and narrowband interference (ISR) both increase, the gain in P a also increases. This stems from the fact that adaptive threshold detection, employed by the receiver at the network hub, enjoys better performance, in contrast to constant

222 Waseem librail and Ranjith Liyana-Pathirana

threshold, particularly in high interference environments [4]. In this way network capacity may be enhanced.

r Gain IS Gain users in P~ RdB inP~

3 2 5 1.7 5 1.7 7 1.9 10 1.7 9 2.31 15 2 11 2.14 20 2 13 2.1 25 2.5 15 2.13 30 2.6 17 2.25

Table I Gain in P a of Slotted Spread

ALOHA over Spread ALOHA using Adaptive

Threshold

3. SLOTTED SPREAD ALOHA WITH CLSP

To further improve the performance of slotted spread ALOHA scheme, in terms of the probability of packet acquisition, channel load sensing protocol (CLSP) [6, 7] is employed to control the maximum number of users competing to gain access of the network receiver hub station in a given cell at one time. This is performed by monitoring the channel load and applying a threshold to limit the number of users that are allowed to access the channel at anyone time. Two threshold levels are examined ,these are; (2/3) r and (1/3) r , where r is the total number of users competing to gain access to the same receiver, for acquisition purposes, at the network receiver hub.

Fig.6 illustrates the probability of packet acquisition as a function of ISR with and without CLSP using adaptive threshold at the hub receiver. This indicates clearly that slotted ALOHA scheme may perform much better with CLSP (i.e. as the number of users are controlled during initial synchronisation).

Packet Acquisition Evaluation of Slotted Spread ...

0. 161-----,---;:=:=~~=s:::;::;==~==~ + S·Aloha willi Adaptlw Thrnshold

0.15 r: ~ .l!l 0.14 :::I

10.13

i ~ 0.12

CI. '0 ~ 0.11

li ! 0.1 e CI. 0.09

- ---- - - - - - - 1- - --

, .. S·Aloha wUh (1/3) r CLSP o S·Aloha willi (2/3) r CLSP

r .. 30 Users ___________ L ______ _ __ _ _ 1 __________ _ , , , , , - - - - -0 - - - - - ,- - - - - - - - - - - -,- - - - - - - - - - -.. ' , , - - - - - - - - - - -1- - - - - - - - - - - - 1- - - - - - - - - - -

+ : - - - - -oF - - - - -'?- - - - - - - - - - - -:- - - - - - - - - - -, , -----------f ----~ -----6 ----------

, + ... 0 - - - - - - - - - - -,- - - - - - - - - - - -+- - - - - - - - - - - t

, '+ , , O·08SL---------', '-O ------'15'------..320

4. CONCLUSION

ISR (dB)

Fig.6 Probability of Packet Acquisition

against ISR with CLSP

223

The probability of packet acquisition for both slotted spread ALOHA and spread ALOHA schemes are compared in the presence of AWGN, narrrowband interference and MAL Results signify the superior performance of slotted ALOHA scheme for the same serial-search receiver employed at the network hub. Thus, network capacity may be improved.

It is shown that incorporating channel load control protocol (CLSP) with slotted spread ALOHA further enhances its performance in terms of the probability of packet acquisition.

REFERENCES

[1] Madhow U and Pursley M.B," Acquisition in Direct-Sequence Spread-Spectrum

Communication Networks: An Asymptotic Analysis", mEE Trans. On Information

Theory,Vo139, No.3, May 1993, pp. 903-912.

[2] librail W and Liyana-Pathirana R "Packet Acquisition Performance of Spread ALOHA

Wireless Data Networks", Proc. Of the third International Conference on Information,

Communications and Signal Processing (ICICS 2001), ISBN 981-04-5149-0, P012S, Oct.

15-18,2001, Singapore.

224 Waseem librail and Ranjith Uyana-Pathirana

[3] Jibrail W, Liyana-Pathirana R, Robert FRY," A Modified Serial-Search DS Code­

Acquisition Scheme", mEE Sixth International Symposium on Signal Processing and its Applications (ISSPA 2001),13-16 August, Catalog No. OlEX467C, ISBN 0-7803-6704-9, 2001 IEEE, Kuala Lampur, Malaysia.

[4] Jibrail W,"DS Spread Spectrum Signal Acquisition Using Adaptive Threshold Techniques·, Int. Jour. Commun. Syst., Vol. 11, 1998, pp. 297-303.

[5] Simon,M.K., Omura, J.K, Scholtz R.A and Levitt B.K,"Spread Spectrum

Communications, Vol. III, Rockville,MD: computer press 1985.

[6] K. Tohimitsu, T.Yamazato, M. Katayama and A. Ogawa,"A Novel Spread Slotted

AWHA System with Channel Load Sensing Protocol", mEE Jour. On Selected Areas in

Communications (JSAC), Vo1.12, No.4, May 1994, pp. 655-672.

[7] H. Okada, M. Saito, T.Yamazato, M. Katayama and A. Ogawa, "Performance Evaluation

of COMA ALOHA Systems with Channel Load Sensing Protocol", Global

Communications Conference, Vol. 2, Nov. 12-18,1996, pp. 1291-1295.

CODE DIVISION MULTIPLE ACCESS CDMA

On Erlang Capacity of CDMA Systems

Samad S. Kolahi Member, IEEE Unitec Institute of Technology, Auckland, New Zealand

Abstract: In this paper, the Erlang capacity of mobile protocols such as FDMA (Frequency Division Multiple Access), TDMA (Time Division Multiple Access), and CDMA (Code Division Multiple Access) systems are compared assuming perfect power control. Methods of calculating the traffic capacity of such systems are discussed. A simulation and modeling method is used and the results are compared with previous analytical methods. The simulation result is close to analytical result although a different approach is adopted. CDMA can provide up-to 20 times more traffic capacity than FDMA and 5.3 times more traffic capacity than TDMA.

Key words: CDMA, Capacity, Erlang

1. INTRODUCTION

Mobile communications has enjoyed continuous growth in terms of number of mobile phone users in the last several years. Traffic management of the network is becoming more important as the number of mobile phones increase. Tele-traffic engineering issues are vital in planning, design, and dimensioning of mobile CDMA networks. Traffic engineering of FDMA and TDMA systems is a straightforward matter as each user occupies a slice of the bandwidth or time slot respectively. If each user occupies 30KHz of bandwidth, with frequency reuse of 7 and 3 sectors per cell, the number of channels per sector in 12.5 MHz is 19. For a TDMA system, the system uses the 30KHz band for three calls, each taking turns in using the 30KHz range. The number of channels will be N = 57 Channels/sector. Using the Erlang­B formula (1), assuming blocking probability of 0.02, the traffic intensity is calculated as A = 12.3 for FDMA systems and A = 46.8 for TDMA systems. Note that the result implies that although the number of channels is 3 times greater in TDMA compared to FDMA, it can carry 3.8 times more traffic to that of FDMA.

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

228 On Erlang Capacity of CDMA Systems

(A)N IN! Pblocking = -N---- (1)

L(A)k I k! k=O

The traffic engineering issue is not straightforward in CDMA as all calls use the same bandwidth (the whole bandwidth spectrum). The traffic capacity of the CDMA system can be investigated by varying the traffic loads (arrivals versus departures to the system) and determining the probability of call losses.

The organization of this paper is as follows. In the next section the system model is discussed. The blocking probability calculations are introduced in section 3. Simulation and analytical results are reported in section 4 and some concluding remarks are given in section 5.

2. SYSTEM MODEL

A cellular CDMA network with 37 cells is considered (a home cell and three tiers of neighboring cells) with a base station located at the center of each cell. All cells are assumed to be homogeneous in every respect. The reverse link (from mobile to cell site) is modeled as it is the limiting link due to its inferior performance compared to forward link [1]. The calls to the CDMA system are modeled as Poisson [2] with mean arrival rate of A calls/sector/second and mean call holding time of 1I~ seconds per call. In queuing terms, this is a MIMI 00 system which is being used for CDMA systems modeling. Once the mobile call has been admitted, it stays in the system during its call holding time which is modeled as negative exponential with probability density function:

f(t) = pe-Jil (2)

Traffic density (offered traffic load), it I J1 , represents the excess of the arrival rate versus departure rate. it I J1 is measured in Erlang.

All calls are allowed into the system (soft capacity) if they meet the required Quality of Service (QoS). Any calls not meeting this required quality are not permitted to enter the system but are blocked.

3. CDMA BLOCKING PROBABILITY

In CDMA all calls use the same frequency range. These calls therefore interfere with one another. CDMA capacity decreases with the amount of interference. Consider a CDMA home cell and its neighboring cells, each cell site not only receives interference signal from mobiles in the home cell

Samad S. Kolahi 229

(intra-cell interference) but also from mobiles located in neighboring cells (inter cell interference).

The power of signals received is the product of the transmitted power, mth power of the distance and a lognormal shadowing parameter (~) with mean zero and standard deviation of u ~. This shadowing parameter varies with different terrains. Assuming St and S are the transmitted and received power respectively, we have:

S = St.T-m1O~110 (3)

The interference from the jth mobile in neighboring cell i is expressed as [4]:

Tmm 10~o110 (/) ij = S ~ 110 • m

10 m To (4)

(5)

where S is the received signal strength at home base station, r m is the distance to corresponding home cell base station (figure 1), ro is the distance to the neighboring cell, ~o and ~m are lognormal (Gaussian in dB) random

variable distribution with zero mean and standard deviation a ~ representing

shadowing parameter in neighboring and home cell, and m is path loss exponent.

Total other cell interference /0 is interference produced by all users who

are power controlled by other base stations. Assuming a CDMA system with M outer cells and N users per cell, then the total other user interferences-to-signal ratio (// S) 0 is:

M N

(II S)o = II/ij / S (6) i=l j=l

On each arrival of a new call, the total interference is determined from which the blockage condition can be checked. This involves repeatedly generating Tm between 0 and 1 and uniform random variable () between 0

and 21t. Using figure 1, '0 can be calculated for each user as:

To = ~ fro 2 + d 2 + 2 d fm Cos B (7)

230 On Erlang Capacity of CDMA Systems

Figure 1. Interfering call distance to home cell

Using equations (5) and (6), the total received power from interfering cells at the home base station is calculated by considering the path loss exponent m and shadowing parameter ~o-~m. If independent lognormal

variables~m and ~o have average zero and variance of U/' ~o-~m has

mean zero and variance 2u ~ 2 • For each interfering call, a lognormal

shadowing parameter, ~o -~m' is generated with mean zero and standard

deviation (J" ~..fi. The transmission quality of a CDMA call may then be

calculated in terms of the energy per bit over total interference [4] spectral density Eb IN o.

Eb SIR SIR WIR - = --= = ---------No IIW «N-I)S+Io+T])IW (N-I)+(I1S)o+T]IS

(8)

(11 S) 0 is the ratio of other cells interference to the received signal strength

(S) at home base station, N is the number of active users in the cell, T] is background noise, WIR is Processing Gain, W is available spread bandwidth, and R is data rate.

Taking voice activity into consideration, we have:

Eb WIR - = -----------No a(N -I) +a(I1 S)o +1// S

(9)

Samad S. Kolahi 231

On each call arrival Eb IN 0 is determined (6,9) to decide if the call is

accepted to the system or is blocked. The blocking probability can then be calculated for a given traffic load.

4. RESULTS

4.1 Simulation Results

The simulation is performed for one million arrivals. On each arrival, the total interference at the home base station is determined and from this the blocking condition can be checked. If the required call quality ofEb I No ~ 7dB, or BER<JO-3, is not achieved, the call is blocked. The

blocking probability is then calculated for a given Erlang traffic (AI f.J) , by taking the ratio of the total number of blocking events to the total number of call arrivals.

For values of (j, = 8dB, WI R = 125 ,Eb I No = 5.012(7dB),

a = 0.4, S I TJ = -ldB , and m = 4 and the blocking probability of 2%,

and assuming perfect power control, traffic intensity (A I f.J) is determined by simulation as 27 Erlangs which corresponds to approximately 36 voice channels per sector per 1. 25MHz. The 12.5MHz spectrum utilizes nine 1.25MHz CDMA frequency blocks [5]. The number for channels must therefore be multiplied by 9. Total number of channels supported will be 324 which corresponds to 309.7 Erlangslsector.

4.2 Analytical Results

Using the analytical method [3] with perfect power control, the traffic capacity can be calculated as:

A (I-T)(W I R)F(B,(j p) = Erlangs/sector f.J a(1 + f)(Eb / No)

where

(Eb I No )[Q-l (Pblocking )]2 B = -------""--

(W I R)(I-T)

and

00 1 -/ Q(z) = J.J e-2 dy

l 2n

(10)

(11)

(12)

232 On Erlang Capacity of CDMA Systems

for perfect power control:

BN F(B,O) = 1+-(1- 1+-) 2 B

(13)

where T is the background (thermal) noise to total acceptable interference, (f p is the standard deviation of power fluctuations, f is the ratio of other

cell interference to own cell interference. Numerically for Pbloc1cing=O.02, WIR=1250, T =0.1, (f p =0, /=0.55, Eb I No = 5.01(7dB) and a =0.4, we

obtain: Q-J(Pbloc1cing)=2.057, B=0.019, F(B,O)=0.871, AI JI. = 315.3 Erlangs/sector.

This analytical Erlang capacity of 315.3 is slightly higher but close to the simulation results of 309.7 Erlangslsector obtained above.

Imperfections in power control schemes can reduce this capacity by approximately 20% [3]. The simulated CDMA traffic capacity is therefore approximately 247.8 Erlangs. This is 247.8/12.3 =20 times more than an analogue system and 5.3 times (247.8/46.8) more than the TDMA Erlang capacity calculated above. These results are very close to the results determined analytically in [3] although a different approach is adopted.

s. CONCLUSIONS

The traffic capacity of a COMA system is analyzed using simulation and modeling tools and compared to analytical methods. The analytical results are slightly higher but close to simulation results obtained in this work.

COMA can provide up to 20 times more traffic capacity than its predecessor FDMA systems and up to 5.3 times more traffic capacity than TOMA systems for the same frequency range.

6. REFERENCES

[1] J.S. Evans and D. Everitt, "Analysis of Reverse Link Traffic Capacity for Cellular Mobile Communication Networks Employing Code Division Multiple Access," Australian Telecommunication Networks and Applications Conference, Melbourne, pp. 775-780, 1994.

[2] J.S. Evans and D. Everitt, "On the Teletraffic Capacity of CDMA Cellular Networks," IEEE Transaction on Vehicular Technology, vol 48(1), pp. 153-165, 1999.

(3) A.M. Viterbi and A. J. Viterbi, "Erlang Capacity of a Power Controlled CDMA System, " IEEE Journal on Selected Areas in Communications, vol 11(6), pp. 892-899, 1993.

[4) K.S. Gilhousen, I.M. Jacobs, R. Padovani, A. J. Viterbi , et al, "On the Capacity of a Cellular CDMA System," IEEE Transaction on Vehicular Technology, vol 40(2), pp. 303-311, 1991.

(51 R. Padovni, "Reverse Link Performance of IS-95 Based Cellular Systems," IEEE Personal Communications, third quarter, pp. 28-34, 1994.

Power and Spreading Gain Allocation in CDMA Data Networks for Services with a Relative Priority

Kwang-Seop Jung School of Electronic and Electrical Engineering Kyungpook National University, Taegu, Korea

Sun-Mog Hong School of Electronic and Electrical Engineering Kyungpook National University, Taegu, Korea

Eun-Young Park School of Electronic and Electrical Engineering Kyungpook National University, Taegu, Korea

Abstract In this paper, we consider algorithmic approaches for effectively pro­viding data services for mobile users with different levels of priorities in a DS-CDMA system. The priority level of a user is specified by a factor that is a weighting on the instantaneous data throughput of the user. We define the weighted instantaneous aggregate data through­put (or the weighted channel efficiency) and use it to characterize the performance of the prioritized data service. Our prioritized data ser­vice is implemented so that the weighted instantaneous aggregate data throughput is maximized via efficient power and spreading gain allo­cation. We obtain properties of an optimal allocation that provides the maximum weighted instantaneous aggregate data throughput. Us­ing these properties, we develop an efficient algorithm for computing the optimal power and spreading gain allocation. We also propose two suboptimal algorithms that are very fast and simple to implement.

Keywords: CDMA data networks, prioritized data services

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

234 Kwang-Seop lung, Sun-Mog Hong and Eun-Young Park

1. Introduction As demands increase for data services over mobile wireless networks,

the effective use of limited radio resource (channel bandwidth) becomes an imperative in wireless mobile data communications [1, 2]. Often the non-real-time data services are less sensitive to delays compared to real­time services, and can be provided on a best-effort basis. There has been significant work on exploiting the delay tolerance of the data services to achieve high aggregate data throughput via an efficient allocation of the radio resource. Specifically, attempts have been made on power control and spreading gain allocation to maximize the aggregate non-real-time data throughput in a direct-sequence code-division multiple access (OS­COMA) system [2, 3, 4].

In this paper, we consider algorithmic approaches for effectively pro­viding data services for mobile users with different levels of priorities in a OS-CDMA system. The priority level of a user is specified by a factor that is a weighting on the instantaneous data throughput of the user. Firstly we define the weighted instantaneous aggregate data through­put (or· the weighted channel efficiency) and use it to characterize the performance of the prioritized data service. Our prioritized data ser­vice is implemented so that the weighted instantaneous aggregate data throughput is maximized via efficient power and spreading gain alloca­tion. This allows us to avoid dictating a strict priority service in which a lower-priority user is serviced only if services for all higher-priority users have been completed. Obviously this strict priority service can lead to a waste of the limited radio resource, since it cannot adapt ef­fectively to time-varying channel quality due to fading, shadowing, and distance loss [5]. Secondly, we obtain properties of an optimal allocation that maximizes the weighted instantaneous aggregate data throughput. Using these properties, we develop an efficient algorithm for computing the optimal power and spreading gain allocation. We also propose two suboptimal algorithms that are very fast and simple to implement.

2. System Model We consider the uplink in the OS-COMA mobile packet data network

that consists of M mobile data users and a single base station. The mobile users transmit streams of data packets over a common radio channel using their respective terminal-specific spreading codes. Let us denote by gi (gi > 0) the instantaneous channel gain between user i and the base station at an instant. Also, let us denote by Wi (Wi> 0) the weighting factor on the instantaneous throughput of the user i. The

Power and Spreading Gain Allocation in CDMA Data Networks 235

value of Wi specifies the priority level of user i. The larger relative to other users' the value Wi is, the higher the priority level of user i is. The priority levels are discrete. Let us denote the number of priority levels by L and the weighting factor of the levell by wI, respectively. Assume that wI > ... > w L . Obviously, Wi E {wI : 1 = 1, ... , L}, i = 1, ... , M.

We assume that without loss of generality wI9I > ... > WM9M. Each user can use the transmit power in an interval of zero to the maxi­mum power Pmax. The bit-energy-to-interference-power-spectral-density Eb/10 of user i is given by

(1)

where Pi and Ni are, respectively, the power and spreading gain of user i, q is the background interference and thermal noise power, and a is a constant depending on statistical characteristics between the spread­ing codes of users. The weighted instantaneous aggregate throughput (weighted channel efficiency) is defined by the weighted sum of each user's instantaneous throughput. That is,

M C(p, N) = L (3 ~i f( Pif:tNi ),

i=I N~ q + a L-j=IJi:iPj9j (2)

where (3 is the ratio of the channel code rate to the number of bits per packet. The function f represents the probability of a successful packet transmission. We assume that f is differentiable and non-decreasing in Eb/10. In this work, we assume that spreading gains can be real numbers for the simplicity of our analysis. Since the priority levels are specified by the weighting factors, the ratio of a pair of weighting fac­tors becomes a quotient of relative prioritization between the pair of corresponding priority levels. Note that the weighted channel efficiency (2) with Wi = 1, i = 1, ... ,M, becomes the instantaneous aggregate throughput in [4].

3. Jointly Optimal Power and Spreading Gain In order to maximize the weighted channel efficiency in the DS-CDMA

networks, it is essential to mitigate the impact due to the multiple ac­cess interference (MAl) and time-varying channel quality. Specifically we maximize the weighted channel efficiency C(p, N) by efficiently al­locating the power and spreading gain. This results in an optimization problem to obtain the optimal combination of power and spreading gain

236 Kwang-Seop lung, Sun-Mog Hong and Eun-Young Park

allocation such that maxC(p,N) p,N

(3)

subject to 0 ::; Pi ::; Pmax and Ni > 0 for i = 1, ... M. We can fortu­nately apply results in [4] to represent the optimum spreading gains of the optimization problem (3) as a function of the respective transmission power of M users.

Proposition 1 [4]: The optimal spreading gain Nt of user i is

N~ = ",*(q + a E~1,#iPjgj) t ,

Pigi

i = 1, ... M, where 1

",* = arg max - 1(",). 112':1 '"

(4)

(5)

Substituting (4) into (2), we can see that the weighted channel effi­ciency depends upon p only. As a consequence, Proposition 1 allows us to reduce the optimization problem into a much simpler power control problem as follows. C (p) denotes C (p, N* (p) ).

maxC(p) p

subject to 0 ::; Pi ::; Pmax, i = 1, ... M.

(6)

We obtained properties of an optimal solution to this power allocation problem. Two most important properties (necessary conditions) of the optimal solution are summarized as follows. Due to the limited space, proofs are omitted.

Lemma 1: Suppose that p* = (Pi, ... ,PM) solves the optimization prob­lem (6). Then, pi = 0 or pi = Pmax for all i E {I, ... ,M}.

Lemma 2: Suppose that p* = (Pi, ... ,PM) solves the optimization prob­lem (6), and suppose that pj = Pmax for some j E {I, ... ,M}. Then, pi = Pmax for all i E {I, ... , M} such that both Wigi > Wjgj and Wi ~ Wj are satisfied.

Lemma 1 reduces further the power control problem into a simple power assignment problem, which can be solved more efficiently. Lemma 2 helps to reduce significantly the number of candidate solutions, thereby the computational cost required to verify their optimality. The follow­ing two lemmas are also useful in implementing an efficient algorithm for solving the optimization problem. Let gmax = max{gi : i = 1, ... , M}

Power and Spreading Gain Allocation in CDMA Data Networks 237

and gmin = min{gi : i = 1, ... ,M}.

Lemma 3: If Pm~",gl ~ u:,.l!;;, pi = Pmax and pi = 0, i = 2,3, ... , M.

L /. If pma.,9ma", < 1 wL * - fior all- E {I M} emma.lf. q - a MwL(M-2)wL' Pi - Pmax • , ... , .

We use the results of Lemmas 3 and 4 in implementing power alloca­tion algorithms for obtaining an optimal or suboptimal solution to the problem (6). The allocation algorithms are described in the following section.

Recall that the weighted channel efficiency (2) without prioritization becomes the instantaneous aggregate throughput in [4]. Accordingly, the instantaneous aggregate throughput maximization in [4] is a special case of the weighted channel maximization problem (6). It is easy to show that Lemmas 2-4 for no prioritization reduce to their corresponding results in [4], respectively.

4. Power and Spreading Gain Allocation Algorithms

Our algorithm OPT solves the power assignment problem (6) to ob­tain an optimal solution. It generates efficiently all the candidate solu­tions satisfying the necessary condition of Lemma 2. OPT then com­pares their respective weighted channel efficiencies to determine an op­timal solution. We also propose two suboptimal algorithms SAl and SA 2. They are very fast and simple to implement. We denote by Ip and I~, respectively, the index sets of users withpmax assigned and with no power assigned. It is convenient to represent by irin the index of the user in I~ with the highest channel gain in priority level I, i.e., irin = min{i E I~ : Wi = wi}, I = 1, ... ,L, if it exists. It is also conve­nient to represent by imin the index of the user in I~ with the smallest index without concern about priority levels, i.e., imin = min{i E I;}. The algorithms begin with Ip = cf> and return Ip as a suboptimal solution.

Suboptimal Algorithm 1 fSA1} Step 1: Set Ip = cf> and Cmax = O. Step 2: For I = 1, ... , L, find, if it exists, irin , assign Pmax to users in Ip U {iFin} and compute its weighted channel efficiency 0, for this assignment. Step 3: Find f = argmax{a, : I = 1, ... , L}. Step 4: If Or < Cmax , stop and return Ip. Otherwise, update the index set Ip by Ip U {irin }. Update I; by the complement of Ip and amax by

238 Kwang-Seop lung. Sun-Mog Hong and Eun-Young Park

Gr, respectively. Step 5: If the cardinality of Ip is equal to M, stop and return Ip. Oth­erwise, go to Step 2.

Suboptimal Algorithm 2 JSA2} Step 1: Set Ip = 4> and Cmaz = O. Step 2: Find i min • Assign Pmaz to users in Ip U { imin} and compute its weighted channel efficiency Go for this assignment. Step 9: If Go < Gmax , stop and return Ip. Otherwise, update the index set Ip by Ip U {imin }. Update I; by the complement of Ip and Gmax by Go, respectively. Step 4: If the cardinality of Ip is equal to M, stop and return Ip. Oth­erwise, go to Step 2.

SAl and SA2 are not necessarily intended to generate a solution sat­isfying the necessary conditions, but they give a solution satisfying the conditions. In Step 2 of SAl and SA2, we compute the weighted chan­nel efficiency for a given power assignment p using the optimal spread­ing gain N*(p) of (4) corresponding to this assignment. In fact, the spreading gain allocation of (4) is always assumed for a given optimal or suboptimal power allocation in this paper.

When wi are identical for all 1 = 1, ... , L, OPT, SAl, and SA2 are basically identical and give an optimal solution. This is also true when wl+1 jwl , 1 = 1, ... , L - 1, are sufficiently small.1 We implemented an algorithm (MAX) for maximizing the unweighted instantaneous aggre­gate data throughput, and a strict priority service algorithm (STR). STR implements a strict priority service in which a lower-priority user is serviced only if services for all higher-priority users have been com­pleted. Note that MAX and STR solve the problem (6), respectively, for the above two special cases: wi, I = 1, ... ,L, are identical and wl+1 jwl ,

1 = 1, ... ,L - 1, are sufficiently small.

5. Numerical Experiments Numerical experiments were performed in C-language on a 933-MHz

Pentium III PC. First we compared the performances of our OPT, SAl, and SA2 algorithms for M = 6 and M = 12 with L = 3. The weighting factors of the three priority levels were set to 1.0, 0.5, and 0.25, respec­tively. The number of users in each level is assumed two and four for M = 6 and M = 12, respectively. We obtained the computational time and weighted aggregate instantaneous throughput for each allocation

Power and Spreading Gain Allocation in CDMA Data Networks 239

Table 1. Performance Comparison of OPT, SAl, and SA!! Algorithms.

Weighted channel efficiency Computational time (J.ls)

OPT

1.52 6.31

M=6

SAl

1.52 1.31

SA!! OPT

1.52 0.16

2.63 35.1

M=12

SAl

2.63 1.48

SA!!

2.63 0.16

Table!!. Aggregate and Weighted Aggregate Data Throughputa of MAX, SA!!, and STR Algorithms.

M=6 M=12

MAX SA!! STR MAX SA!! STR

Weighted Aggregate Throughput 1.13 1.24 0.88 0.68 0.95 0.43 Aggregate Throughputb 3.01 2.75 0.88 2.25 2.10 0.43

a(xlO-1). bpackets/user/s.

update. For this comparison each user is assumed to have an infinite supply of packets. The distance model in [2, 6] with log-normal fading is assumed with the parameter set similar to the one given in [2]. We assumed q = -97 dB, Pmax = 100 mW, and O! = 1.0.

Table 1 indicates that SA2 is 39.4 and 219.4 times faster than OPT, respectively, for M = 6 and M = 12. They also indicate that SA2 is 8.19 and 9.25 times faster than SAl, respectively, for M = 6 and M = 12. Clearly, the computational advantage of SA2 is more significant as the number of users increases. Compared to OPT, the weighted aggregate instantaneous throughput was 99.5% and 99.2%, respectively, for SAl and SA2 for M = 6. They were 99.3% and 99.2%, respectively, for SAl and SA2 for M = 12. These results indicate that the computational advantage of SA2 is significant and it generates a suboptimal allocation that gives the weighted aggregate instantaneous throughput very close to the theoretical upper limit that we can achieve.

We also implemented the strict priority service algorithm (STR) and an algorithm (MAX) for maximizing the unweighted instantaneous ag­gregate data throughput. The weighting factors were set to 1.0 and 0.25, respectively. Only one user was assumed to be serviced with level 1. The numerical results of our experiment are presented in Table 2. The ratios of the weighted channel efficiency were 0.71 and 0.91 for (STR)/(SA2) and (MAX)/(SA2), respectively, for M =6. The ratios of the unweighted channel efficiency were 0.29 and 0.91 for (STR)/(MAX)

240 Kwang-Seop Jung, Sun-Mog Hong and Eun-Young Park

and (SA2)/(MAX), respectively, for M =6. The ratios were 0.45, 0.72, and 0.19, 0.93 for M =12. These ratios show that SA2 is superior to STR in terms of both the weighted and unweighted channel efficiency. They also show that SA2 performs a trade-off between channel efficiency and strict prioritization. In fact, SA2 implements a relative prioritiza­tion at the cost of a graceful loss of channel efficiency. This relative prioritization might be useful for providing a certain level of premium data services against transmission delays due to congestion while keeping channel efficiency reasonably maintained.

6. Conclusion

In this paper, we considered algorithmic approaches for effectively providing data services for mobile users with different levels of priorities in a DS-CDMA system. Our algorithms implement a trade-off between channel efficiency and strict prioritization.

Acknowledgments

This work was supported by Korea Ministry of Information & Com­munication under University Research Program.

Notes 1 w'+l < aMpm .. ,,9mjn l 1 L 1 W h w'+l th t th" al't . -;;;r _ q+aMpmu9rn .. ,,' = , ... , -. e can c oose -;;;r so a IS mequ I y

is satisfied, since gmaz, g;;;!n and q are finite in practice. If q < OI.M Pmaz gma:z:, the inequality . 1 d w l+1 JlmU... approxImate y re uces to -;;;r < 9rn .....

References

[1] Garg, V. K. and Wilkes J. E. {1996}. Wireless and Personal Communications Systems. Upper Saddle River, NJ: Prentice-Hall.

[2] Oh, S.-J., T. L. Olsen, and Wasserman, K. M. {2000}. "Distributed power control and spreading gain allocation in CDMA data networks," in Proc. IEEE Info­com'OO, pp. 379-385, Tel Aviv, Israel.

[3] Oh, S.-J. and Wasserman, K. M. {1999}. "Dynamic spreading gain control in multi-service CDMA networks," IEEE J. Select. Areas Commun., vol. 17, no.5, pp. 918-927.

[4] Oh, S.-J. and Wasserman, K. M. (1999). "Optimality of greedy power control in DS-CDMA mobile networks," in Proc. ACM/IEEE 5th Annual Int. Conj. Mobile Comput. and Network. (MobiCom'99), Seattle, WA.

[5) Guo, Y. and Chaskar, H. (2002). "Class-based quality of service over air interfaces in 4G mobile networks," IEEE Commun. Magazine, vol. 40, no. 3, pp. 132-137.

[6] Stuber, G. L. (1997). Principles of Mobile Communication. Boston, MA: Kluwer Academic Publishers.

Adaptive Closed-Loop Power Control Using an MMSE Receiver in DS-CDMA Systems*

Lian Zhao and Jon W Mark Centre for Wireless Communications Electrical and Computer Engineering Department University of Waterloo

Abstract A closed-loop power control algorithm, which includes both power control

and power allocation functions for DS-CDMA systems, is proposed. The target power level is updated by an iterative algorithm for a minimum mean square error (MMSE) linear filter receiver. The received power is compared with the target to generate the current power control command (PCC). The transmitted power is adjusted by applying a variable stepsize generator that utilizes the PeC history and the estimated mobile speed. Simulation results are presented to show the improvement of the proposed adaptive power control over the conventional fixed stepsize power control.

1. Introduction Power control is one of the most important factors in DS-CDMA systems.

Transmit power control (TPC) aims to compensate for the power fluctuations between the transmitter and the receiver so as to maintain the received power at a desired level. TPC is performed by using an open loop (OLPC) and a c1osed­loop (CLPC) strategy. OLPC attempts to eliminate the slowly varying factors such as path loss and shadowing effects. CLPC is used to compensate for a wide range of fast fading effects. Our research is focused on using the CLPC to mitigate Rayleigh fading [1, 2]. The first research objective, therefore, is to minimize the standard deviation (SID) of the received power by using TPC.

Conventional approach for TPC is to use a sequence of 1 bit power control command (PCC) to regulate the mobile's transmit power using a fixed stepsize. It has been investigated in [3, 4] that there exists an optimal stepsize in terms of tracking ability for a given mobile speed. An improvement is made in [5],

-This work has been supported by the Natural Science and Engineering Council of Canada under grant no. RGP1N7779.

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

242 Lian Zhao and Jon W Mark

where a different stepsize is used for different mobile speed. Another case of the variable stepsize approach is the inverse update algorithm [6], where the stepsize at each iteration is made equal to the inverse of the estimated channel fade. This algorithm offers better performance at the expense of complexity and increased bandwidth requirements on the downlink to carry the extra PCC bits. All the above schemes belong to the one-step TPC. The shortcoming with the one-step TPC is that the PCC is discarded when the update is executed. Thus, it fails to explore the correlations of the PCC history. The inclusion of the PCC history (multi-step) will generally enhance the tracking ability. Adaptive stepsize based on a fixed look-up table using the most recent several PCC has been proposed in [1]. The shortcoming of [1] is that the selected stepsize is not related to the mobile speed.

In this paper, we propose an adaptive multi-step CLPC approach, which achieves adaptive stepsize by using PCC history and the estimated mobile speed. The advantages are (1) PCC history is exploited for multi-step power control; (2) the stepsizes are associated with the estimated mobile speed; (3) PCC is still one bit each time, no extra PCC bit is required; (4) better tracking ability is achieved. The SID of the received power is greatly reduced with a negligible increase in computing.

Besides TPC, power control has also been used as an efficient approach for resource management by way of "power allocation" [7]-[9]. By employing the advanced receiver structures, the target power can be reduced considerably, leading to a lowered transmit power and prolonged battery life. We consider using a Minimum Mean Squared Error (MMSE) linear receiver [10] to demod­ulate the desired signal in a multiple access environment. An algorithm [11] which iteratively updates the target powers and receiver filter coefficients is applied for the purpose to reduce the target power.

2. Outer Loop Power Control Algorithm Consider a synchronous CDMA system with a processing gain N, and

BPSK modulation. N-dimensional column vectors Si and Ci are used to de­note the preassigned unique signature sequence and the linear receiver filter coefficients of user i, respectively. The received signal at the base station is

(1)

where K is the number of users in the system, bj, pj and hj are the information bit, the transmitted power, and the channel gain of user j, respectively. The term V = (Vb ..• , VN) t is an N-dimensional white Gaussian noise vector with per-component zero-mean and variance E[ vl] = (72. The receiver filter output

Adaptive Closed-Loop Power Control ... 243

of user i is K

Yi = L: Vpjhjbj(c!sj) + ni (2) j=1

where ni = c~V is a Gaussian random variable with zero mean and variance 0'2C~Ci. The signal to interference ratio (SIR) of user i can be written as

SIRi = Pi hi (Cit Si)2 .

(Ci t Ci)0'2 + E#i pjhj (Cit Sj)2 (3)

Our aim is to find optimal powers, p = [P1!P2,·· ·,PK], and the filter coeffi­cients, Ci, for i = 1, ... , !(, such that the total power is minimized while each user i satisfies the quality of service requirement, i.e., S I Ri ;::: 'Y:, where 'Y: is the target SIR for user i.

By using the analysis of standard interference functions [11]-[13], an itera­tive algorithm for user i can be written as [11]

Ci(n)

Pi(n + 1)

where A is a function of the powers and signature sequences of the interferers, and is given as Ai = E#i pjhjsjs~ + 0'21, where 1 is an identity matrix.

3. Inner Loop Power Control Algorithm Fig. 1 shows the log-linear model of the proposed CLPC. The transmit

power Pt (n - 1) dB is updated by 6. (n - 1) dB each Tp seconds, which is called the power control cycle, to obtain the transmit power level Pt (n) at time instant n,

Pt(n) = Pt(n - 1) + m(n - 1- k) * 6.(n - 1), (6)

where m(n - 1 - k) is PCC bit (±1), and the index k is the number of loop delay which accounts for generating, transmitting, and executing the PCC. The corresponding received signal power at the base station is Pr (n) = Pt (n) + x(n), where x(n) is the fading gain. The received signal power Pr(n) and the receiver filter coefficients c* (n) are used to generate the desired power level P* (n) by the functional block C2P, which is an implementation of (5). Then, the desired power level P* (n) is fed back to the functional block P2C by using (4) to update c*(n), which will be used in the next power control cycle. The received power Pr (n) is then compared with the desired level P* (n), and a PCC is generated and fed back to the mobile. The model in Fig. 1 also includes the possibility of return channel errors and loop delay kTp•

244 Lian Zhao and Jon W Mark

Figure 1. Proposed power control model.

Feedback Noise

The working principle of the proposed adaptive stepsize is to apply variable stepsize based on pee bit pattern, which can be written as

m = [m(l), m(2), ... , m(L), ... , m(L + k)], (7)

where m(L + k) is the most recent pee generated at the base station. Due to the loop delay of kTp ' the most recent pee bit is m(L) at the mobile. For one-step power control, only m (L) is used for transmit power updating; while for the proposed multi-step adaptive power control, the contents from m(L) down to m{l) and the estimated mobile speed are used to generate the applied stepsize. The length L is called the pee memory length. After execution of the transmit power updating, m is shifted one bit to the left for the next cycle.

Table 1 is an example of the stepsize look-up table when the mobile speed is 40 kmIh. The pee memory length is set at 7. Thus, the column m(7) is the most recent pee bit at the mobile, and m(l) is the oldest pee. An "X" represents "don't care". When a "-I" is received, it is compared with the previous pee. If the previous bit is a "-1", then the comparison continues, until a "+ 1" is reached or until all the bits in the memory have been compared. The index of the applied stepsize is determined by the number of consecutive "-I"s counted. The look-up table for a "+1" appearing in the column of m(7) is not presented due to assumed symmetry.

The variable stepsize 8(1) to 8(7) in Table 1 is obtained by taking statistical average over fading processes, i.e., 8(1) is the average of the fading gain dif­ference between the lowest fade and the second lowest fade, and so on. When executing TPC, further improvement can be achieved by adjusting these aver­age stepsizes appropriately. More results and explanations will be presented in the simulation section.

In order to estimate mobile speed, we define a parameter, namely, the Aver­age Fading Slope Duration (AFSD), as the statistical average number of pee

Adaptive Closed-Loop Power Control ... 245

during the period when the fade changes from its peak to the valley, or from the valley to the peak. Given the power control frequency, the value of the AFSD is directly related to the mobile speed. Simulated AFSD is shown in Table 2 for various mobile speed. For all the simulated results throughout the paper, the carrier frequency is set at 900 Mhz, and power control cycle is 1.25 ms (800 Hz). When the mobile needs to estimate the speed, the mobile may transmit a constant power level around the target regardless of the pee. Then the mobile counts the consecutive number of + 1 or -1 pee. The resultant aver­age number can be regarded as the average AFSD. The mobile speed can thus be approximately estimated from Table 2.

Table 1. Look-up table to generate variable stepsize when mobile speed is 40 kmIh.

m(1) m(2) m(3) m(4) m(5) m(6) m(7) stepsize 6( i) adjusted 6(i) index

X +1 -1 -1.38 -1.35 1 X +1 -1 -1 -1.40 -1.0 2

X +1 -1 -1 -1 -1.01 -0.74 3 X +1 -1 -1 -1 -1 -0.74 -0.52 4

X +1 -1 -1 -1 -1 -1 -0.52 -0.37 5 +1 -1 -1 -1 -1 -1 -1 -0.37 -0.28 6 -1 -1 -1 -1 -1 -1 -1 -0.28 -0.2 7

Table 2. Simulated AFSD. carrier frequency 900Mhz. PC frequency 800Hz.

Speed (kmIh) 10 15 20 25 30 35 40 45 50

AFSD 19.40 17.65 14.88 12.17 10.48 9.15 7.97 7.22 6.46

4. Simulation Results Fig. 2 shows the target power generated using the MMSE iterative algo­

rithm. Assume that the system supports voice, video and data services, with target SIR being {6, 8, 10} dB, and number of users {lO, 5, 5}, respectively. The AWGN noise power level is (72 = 10-4 • Each user is allocated a Gold sequence with length 31. The applied polynomials for the Gold sequence are 24 and 35 in octal notations. At the beginning of the iterations, the transmit power is set at 13 dB below the AWGN noise power, and the filter coefficients are initialized to be the signature sequences of the users. It can be seen that the power converges after 3 iterations.

Fig. 3 is an illustration of the fixed stepsize power control, where the target power level is determined by the outer loop power control. Here, one of the

246

I I I !

...

....

, ...

"'!-, ----!-----!".--~-----! ',.radon IndIx

Figure 2. Target power levels for the MMSE iteration algorithm.

Lian Zhao and Jon W Mark

'~~-=-~~~~~~~~Q~~-=---:­_ , .... !rOO_IT,)

Figure 3. lllustration of the fixed step-size power control.

voice users is assumed to be the desired user. The parameter values used are: mobile speed equals 10 kmIh, stepsize equals 0.5 dB, no delay and no random errors. Under these conditions, we can see that the CLPC with fixed stepsize 0.5 dB works quite well. The received power SID is reduced to 1.042 by TPC. Simulated optimal stepsize to minimize the SID of the received power is 0.5 dB for mobile speed 5-10 kmIh, 0.75 dB for 15-35 krnIh, 1 dB for 40-45 kmIh, and 0.75 dB for 50 kmIh.

. ,. -..... .. -f2.35 3.01 S.l1111 cI!t

-!boo 2710 2120 27XI :v..a 2710 2110 2770 27tO 2710 2100

Poww oonIroI cycIoo (T pi

(a) Profile for adaptive stepsize

f • 1 • . 'l£io:-::::'Z1':-:-,.-:mo-=:::---::2=7":--::'Z1~""'-::2"':::--:Z7«>-=:::---::mo=--::27=""'-::Z-=---==_

Pow .. oonIlOI cycIoo (T pi

(b) Profile for fixed stepsize

Figure 4. Tracking capability of the adaptive stepsize power control. stepsize = [1.35, 1.0, 0.74,0.5218,0.3737,0.2785,0.2] dB, and fixed stepsize power control at 1 dB, with mobile speed 40 km/hour, 2Tp loop delay.

Figs. 4(a) and (b) illustrate the tracking ability of the adaptive stepsize and fixed stepsize TPC, respectively. In the figures, the transmit power is upshifted

Adaptive Closed-Loop Power Control ... 247

(aimed for 0 dB target received power) and reversed (-Pt) in order to have a good visual effect. It is shown that adaptive stepsize TPC attempts to use curves to fit the variation of the fade, while the fixed stepsize TPC uses straight lines. The selection of the 8 (n) is a tradeoff between the slope-overload distor­tion and granular noise. The main principle for the adjustment in Table 1 is to reduce the first few step sizes for higher mobile speed, and to enlarge the first few stepsizes for lower mobile speed compared with the average. The reason is that for higher mobile speeds, the more frequently appeared fading valleys make the first few average stepsizes quite large. But in practical situations, it is difficult to fully counteract these valleys. As a result, the overall performance can be improved when we reduce these step sizes appropriately to lower the granular noise. However, when the mobile speed is slower, properly increas­ing the first stepsize can compensate for the loop delay to a certain degree. It is noted that the results given in Table 1 have been obtained by test-and-trial based on the above principle. We believe that there should be a large space for the step size adjustment and the PCC memory length selection.

Table 3. Received power STD (dB) with optimal fixed stepsize and adaptive stepsize for dif­ferent mobile speed, and the percentage gain in SID reduction.

speed(km/h) 20 25 30 35 40 45 50

Optimal fixed (dB) 1.576 1.800 2.003 2.194 2.350 2.453 2.613 Adap no adjust.(dB) 1.526 1.685 1.875 1.989 2.067 2.111 2.234 Adap w adjust.(dB) 1.514 1.653 1.841 1.924 2.01 2.076 2.171 % Improvement 3.95 8.16 8.15 12.31 14.59 15.42 16.92

Table 4 lists the received power SID for the optimal fixed step size and adap­tive step size TPC, and the percentage gain relative to the optimal fixed step size case. Significant improvement can be obtained, especially when the mobile speed is high.

S. Conclusions An adaptive power control algorithm which minimizes the target power level

and the standard deviation (SID) of the received power for DS-CDMA system is presented and evaluated. At the base station, an MMSE linear receiver is employed to demodulate the desired signal. At the mobile station, transmit power is updated with variable stepsize, which is generated based on PCC history and the estimated mobile speed.

Simulation results show that the MMSE iterative algorithm can converge to the target power level quickly. The proposed adaptive stepsize TPC exhibits a better tracking ability compared with the optimal fixed stepsize power control. It is conjectured that further improvement can be obtained through fine-tuning.

248 Lian Zhao and Jon W Mark

References

[1] c. C. Lee and R. Steele, "Closed-loop power control in COMA systems," Proc. o/IEEE, ,no. 4, pp. 231-239, Aug. 1996.

[2] H. J. Su and E. Geraniotis, "Adaptive closed-loop power control with quantized feedback and loop filtering:' in Proc. IEEE Vehicular Technology Conf., , no. 1, pp. 76-86, Jan. 2002.

[3] S. Ariyavisitakul and L. F. Chang, "Signal and interference statistics of a COMA system with feedback power control," IEEE Trans. Communications, pp. 1636-1634, Nov. 1993.

[4] F. Adachi, M. Sawahashi, and H. Suda, ''Wide band OS-COMA for next-generation mo­bile communications systems:' IEEE Communications Magazine, pp. 56-69, Sept. 1998.

[5] S. Nourizadeh, P. Taaghol, and R. Tafazolli, "A novel closed loop power control for UMTS:' in 3G Mobile Communication Technologies, 2000, pp. 56-59.

[6] A. Chockalingam, P. Dietrich, L. B. Milstein, and R. R. Rao, "Performance of closed­loop power control in OS-COMA celluar systems;' IEEE Trans. Vehicular Tech., pp. 774-789, Aug. 1998.

[7] 1. W. Mark and S. Zhu, "Power control and rate allocation in multirate wideband COMA systems:' in Proc.IEEE Wireless Communications and Networking Conf., pp. 168-172, 2000, (invited).

[8) 1. T. Wu and E. Geraniotis, "Power control in multi-media COMA networks:' in Proc. IEEE Vehicular Technology Con/. , pp. 789-793, 1995.

[9] L. C. Yun and O. G. Messerschmitt, "Power control for variable QoS on a COMA chan­nel," Proc.IEEE Military Communications Con/., pp. 178-182, Oct. 1994.

[10) U. Madhow and M. L. Honig, "MMSE interference suppresion for direct-sequence spread-spectrum COMA," IEEE Trans. Communications, vol. 42, no. 12, pp. 3178-3188, Dec. 1994.

[11] S. Ulukus and R. O. Yates, "Adaptive Power Control with MMSE Multiuser Detectors," in Proc.IEEE Inti. Con/. Communications, pp. 361-365, 1997.

[12] R. O. Yates, "A framework for uplink power control in cellular radio systems," IEEE J. Select. Areas Communications, vol. 13, no. 7, pp. 1341-1347, Sept. 1995.

[13] P. S. Kumar and 1. Holtzman, "Power control for a spread spectrum system with mul­tiuser receivers;' in Proc. IEEE Intl. Symposium on Personal, Indoor and Mobile Radio Communications, pp. 955-959,1995.

CORDIC based QRD-RLS Adaptive Equalizer for CDMA Systems

Tim Zhong Mingqian, AS Madhukumar and Francois Chin Institute for Communications Research 20 Science Park Road, Singapore 117674 Email:{rtpunq.madhu.chinfrancois}@icr.a-star.edu.sg

Abstract The conventional RAKE receiver employed in CDMA system can not satisfy the desired demand when the propagation channel possesses a considerable number of paths and a deep fading that causes serious inter-user and inter­symbol interferences. This paper proposes a novel means of adaptive equalizer based on QRD-RLS algorithm to substitute the traditional RAKE receiver. Regarding the computational complexity of the RLS, the well-known CORDIC algorithm has been exploited and plays a key role in the hardware implementation of the new approach. The proposed RLS structure is simulated extensively under different channel parameters and performance is compared against conventional RAKE structure.

Key Words QRD-RLS, CORDIC, CDMA, Adaptive Equalizer

1. Introduction

The time-varying multipath propagation and the multiuser interference are the two important issues that limit the capacity and performance of a wireless communication system based on the code division multiple access (COMA) technology. The commonly proposed schemes to deal with these two factors are to use transmitter power control, diversity technique and/or error control coding [1][2]. An alternative approach to combat the distortion brought by the channel characteristic and the multiuser interference is to replace the traditional RAKE receiver with an efficient equalizer using the adaptive algorithms such as Least-Mean-Square (LMS) and Recursive Least­Squares (RLS).

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

250 Tim Zhong Mingqian, AS Madhukumar and Francois Chin

Regarding the time-varying Rayleigh fading channel which is typical for a practical mobile radio system, due to the limitation of number of RAKE fingers, it's hard to reach a satisfied system performance when the number of channel path increases and the channel coefficient changes quickly from time to time. In addition, the system complexity increases dramatically as more RAKE fingers have been appended and a channel estimation module must be included at the receiver side. Time Domain Equalizers based on LMS or RLS can be a solution to solve the issues related to multipath interference and errors due to channel estimation.

The RLS algorithm takes into account all the information that extend back to the initialization and updates the estimation of the tap-weight vector upon the arrival of new data, thus it is preferred over the LMS algorithm due to its superior convergence properties [3]. The orthogonal triangularization of the input matrix via QR-decomposition (QRD) has played a crucial role in RLS filtering. Such triangularization process can be realized through a series of Givens rotations, which is commonly employed to do the QR updating on a sample-by-sample basis [7]. Figure 1 shows the block diagram of the proposed RLS based adaptive equalizer structure for a CDMA receiver.

r'· .. · .. ······· .. ········································ ............................. _ ....................................................................................................... ! i ! ! Pilot ! i Oespreading RLS Weight i ! Generator ! i i i i

~~ I R ·! d i ecel\/e !

! i

Oesnreading FTR i \ ............................................................................................................................................................................................. .1

Figure]: Structure ofRLS Adaptive Equalizer in aCOMA system

Armed with the fast developing VLSI technology, systolic array has become practically feasible when realizing the QRD-RLS in hardware implementation. The CORDIC (Coordinate Rotation Digital Computer) algorithm [4] has been introduced to perform the two-dimension vector rotation instead of the conventional Givens rotations. The main idea underlying this algorithm is to do phase shifting through a series of "micro­rotations" using a fIxed set of elementary rotation angles. Through a proper choice of the elementary angles all computations can be implemented efficiently in VLSI using a sequence of shift and add/subtract operations. Generally, a look-up-table holding the elementary rotation angles is set up in advance to perform the phase shifting replacing the trigonometric functions

CORDIC based QRD-RLS Adaptive Equalizer for CDMA systems 251

exploited in Givens rotations, which are not applicable in hardware realization.

This paper discusses the design and development of a CORDIC based QRD-RLS channel equalizer in a CDMA system and compares its performance with conventional RAKE structure.

This paper is organized as follows, section 2 briefly introduces the mathematical principles of the QRD-RLS algorithm and discusses its limitation due to the high computational complexity. The CORDIC algorithm will be depicted in section 3 and its implementation in hardware realization will also be included. Section 4 discusses the simulation studies.

2. QRD-RLS Algorithm

The basic idea underlying the RLS algorithm can be described in two processes:

1. We calculate the output of a transversal filter produced by a set of tap inputs and then try to obtain the error estimation by comparing the output with the desired response.

2. We find a method to minimize this error by adjust the tap weight of the transversal filter so that we can approach the desired response at the output.

Let d(n) be the desired response vector at time n, u(i) the input signal vector at time i and w(i) the tap weight we are looking for, and then the goal is to minimize the cost function:

n"1 12 E = LAn-, d(i) - wH (i)u(i) (1) i=l

where A is the forgetting factor used to ensure that the effect brought by past signals is reduced or "forgotten".

Eq. (1) can be converted to a matrix format as follows:

e = 1~(n)112 = IIA(n)d(n) - A(n)A(n)w(nf (2)

where the symbol 11-112 stands for the squared Euclidean norm and:

A d· (1~1~ 1) (3) = lag A ,A , ... ,

d H (n) = [d(1),d(2), ... ,d(n)] (4)

A H (n) = [u(I),u(2), ... ,u(n)] (5)

Regarding Eq. (2), since the multiplication by a unitary matrix does not change the norm of one matrix, we apply the QRD to transform the weighted input signal matrix A(n)A(n) into an upper triangular matrix R(n) , the first part to the right of Eq. (2) has been transformed simultaneously into two auxiliary matrices denoted by P(n) and V{n):

252 Tim Zhong Mingqian, AS Madhukumar and Francois Chin

(n)e(n) = -Q [ p(n)] [R(n)W(n)] V(n) 0

(6)

whereQ(n) is a unitary matrix. It's easy to observe from Eq. (6) that the cost functione approaches its

minimum value IIV(n)112 when the following equation satisfies

R(n)w(n) = pen) (7) The process can be proceeded recursively by a series of complex

Givens rotations which give rise to the unitary update transformation matrix Q(n) :

Q(n) =J M (n)"'J 2 (n)J1(n) (8) here J k (n) is the single Givens rotations transformation matrix used to

eliminate the corresponding element in the signal vector that enters the system at time n. With all these transformation matrices, the non-zero vector i on the bottom will be transformed into a zero vector:

JM(n) ... J2(n~l(nm"[o; 1 (9)

J k (n) is given as:

1

o

cosO 0 0 sin· 0 k Jk(n) = 0 0 0 (10)

0

0 0 0

-sinO 0 0 cosO n

k n

It can be seen from Eq. (7)-(9), as the data stream enters the equalizer row by row, it is annihilated to zero while the matrix R(n) and pen) are updated accordingly.

Therefore, it's easy to obtain the desired weight value through Eq. (7).

3. CORDIC ALGORITHM AND ITS IMPLEMENTATION

3.1 CORDIC Algorithm

Considering Eq.(lO), trigonometric functions have been employed here to perform orthogonal rotations. But such functions can't be applied in practical hardware implementation as far as fixed-point value is concerned.

CORDIC based QRD-RLS Adaptive Equalizer for CDMA systems 253

The CORDIC algorithm will be introduced to solve such problem since it utilizes a look-up-table to do the rotation through series of sub-rotations.

There are two modes of CORDIC algorithm: the "vectoring" mode, which is used to determine the phase and magnitude of the input vector and the "rotation" mode, which actually performs the rotation of the input vector. In order to execute the Givens rotations, "vectoring" mode is applied ftrstly to determine the angle to be rotated followed by the "rotation" mode rotating the vector through a set of sub-rotation for certain loops.

Consider a two-dimensional vector v represented by v = x + jy in the complex plane. If the vector is rotated by an angle 0, the new vector is presented as v = vei8 . The angle 0 can be expanded into a set of elementary angles OJ with pseudo-digits qj E {-1,+1} and angle expansion error Zn-l' such that

n-l

0= Lqj ·OJ +Zn-l j=-1

here the sub-rotation angles OJ take on the following values

{ 1! 12 (i = -1)

OJ = arctan(2 -j) (i = 0,1,···, n -1) (12)

(11)

The pseudo-digits qj are used to determine the direction of rotation according to the present value of Z/. Under the vectoring mode, Zj is set to zero during the initialization and q j is decided by the position of the current vector V j in the updating process, once the original vector has been rotated to the abscissa, the phase can be obtained by the value of Zn • Similarly, under the rotation mode, Zj is initialized by the angle desired for rotation and q j is decided by the sign of current Zj in the updating process, once the loop ends, the vector v has been rotated by 0 while Zj approaches zero.

3.2 Systolic array

Another fascinating feature of CORDIC is that the determination of the rotation angle takes the same number of clock cycles as does the actual rotation. This permits a systolic array of QRD-RLS mtering. CORDIC algorithm has been exploited here to realize the function of every small cell in the systolic array as shown in Figure 2. [7] There are two kinds of cells in the ftgure differentiated by their shapes. The cell is always deftned as CORDIC processor element (CPE), with the round one working under ''vectoring mode" and the square one working under "rotation mode".

254

u(3)

u(2)

u(l)

u(2)

u(l)

o

Tim Zhong Mingqian, AS Madhukumar and Francois Chin

u(l)

o o

d(3)

d(2)

d(l)

a

Figure 2: Structure of systolic array

w,la

4. Simulation results

The proposed system is extensively simulated for different channel parameters and the performance is compared against conventional RAKE receiver. The CDMA based transceiver structure used in the simulation is compliant to 3GPP standards [5]: using orthogonal variable spreading (OVSF) as channelization codes and a segment of Gold sequence as scrambling code. Pulse shaping filter has been included in the system to smooth the chip signal and discard higher frequency components. Two receiver side antennas have been utilized to improve the performance and a two-branch structure is employed to reduce the taps, which is set as 4, needed for the RLS equalizer. The channel model used in the downlink simulation system is a Rayleigh fading channel with 10 paths as denoted by L. The maximum delayspread is 20 chips. The Doppler frequency denoted by Fd is relatively low in the simulation thus the time-varying channel changes slowly. The processing gain has been set to 16 and the number of user has been set as the quarter respectively the half of the system load. The system supports a total bandwidth of 5MHz, a chip rate of 3.84Mcps and a data rate of 144kbps. The simulation assumes perfect synchronization and no power control matters are employed.

Figure 3 shows the simulation results of the proposed system. The solid curve in every figure denotes the BER against SNR per bit through the

CORDIC based QRD-RLS Adaptive Equalizer for CDMA systems 255

Rayleigh fading channel when the number of user is set to single (single-user bound) [6] , serving as a preference for comparison. It's obvious that the RLS equalizer outperforms the conventional RAKE by nearly 4 dB at the BER level of 10.2 in the case of quarter system load. It can be also observed easily that the system performance deteriorates as the number of user increases. However, the MUI caused by the increasing of the user number doesn't affect the equalizer as heavily as for the RAKE receiver. For both two cases, a noticeable error bottom can be observed for the RAKE receiver case but the performance for the RLS equalizer improves with EblNo.

10'1 ::.::::;:;:;: ~. * " :: :c:::.:::::::::.:.:::;:::: ::::::::::t:::::::::::.:::::.:::.:: :::::E: . -. ---------. - -- ---... -----------. ---~ -.. --.... -_ ... -_ .... -

· ,

'0· H ~ ~ ~ ~ ~ ~Hn~~~T ~ ~ ~~HHH~ ~ ~ ~ ~~Hm~~~Hn~ ~~;;: ~~ H ••••••• _ •• ___ •••••• _.L ••• __ ••••• __ ••• __ ____ L ••• __ • •••••••••••• · . .... -.-.-.------.---- .. --.------.----- ... ---.- .. --- ........... -. · , .... . __ ....... -... ---:-._------ .. - ----- .. ---:- ... --- ........... -..

· . 1~~--------~·~--------~,~------~~,&

(a) quarter system load

::: : ~ : : : : ~: :: :: =: _ :: ::: . : : : : : : :::: ::: :: : : : : :t :: : : : : : ... :: :: : : :::: . . . ...... --------. -..... ~ .. _ .............. " ........ . ..... . ....... . , . . . 'a-4a~-------"""'·-------~,~ ------..:'---~,·&

(b) half system load

Figure 3: Performance of the proposed method in a CDMA system

256 Tim Zhong Mingqian, AS Madhukumar and Francois Chin

s. Conclusion

This paper has focused on a novel approach of signal receiving in CDMA system: replace the RAKE with an adaptive equalizer based on QRD-RLS algorithm due to its perfect convergence property and robustness against multipath interference. Considering the computational complexities. which hinder it from practical realization. the CORDIC algorithm has been introduced to make it feasible for hardware implementation. The complicated matrix computation has been carried out by a systolic array consisted of CPE cells and driven by a certain clock. It can be observed from the simulation results that the adaptive equalizer outperforms the RAKE to an ideal extent in a multipath fading channel and the system complexity has been reduced since the channel estimation module can be omitted.

Reference

[1] Ozan K. Tonguz, and Melanie M. Wang, "Cellular CDMA Networks Impaired by Rayleigh Fading: System Performance with Power Control", IEEE Trans. on Veh. Tech., Vol. 43, No.3, August 1994.

[2] 1. Boutros and E. Viterbo, "Signal Space Diversity: A Power- and Bandwidth-Efficient Diversity Technique for the Rayleigh Fading Channel, " IEEE Transactions on Information Theory, vol. IT-44, pp. 1453-- 1467, July 1998.

[3] E. Eweda. "Comparison of RLS, LMS, and sign algorithms for tracking randomly time­varying channels." IEEE Trans. on Signal Processing, vol. 42, no. 11, pp. 2937-2944, Nov. 1994.

[4] Y. H. Hu, "CORDie-based VLSI architecture for digital signal processing." IEEE Signal Processing Magazine, 9 (3): 16-35, 1992.

[5] 3GPP TS23.101 V3.5.0: "UE Transmission and reception (FDD )", Release 1999 [6] Proakis, J.G: "Digital Communications", Third Edition, McGraw-Hill Inc.,1995 [7] Simon Haykin: "Adaptive Filter Theory", Third Edition, Prentice-Hall Inc.,1996

Resource Allocation Using Dynamic Spreading Gain Control for Wideband CDMA Networks Supporting Multimedia Traffic

Hailong Huang and Francois Chin Department 0/ Electrical & Computer Engineering, National University 0/ Singapore, 10 Kent Ridge Crescent, Singapore, 119260 engp0921@nus.edu.sg

Institution o/Communication Research, ,20 Science Park Road, #02-34/37,Singapore Science Park II, Singapore, 117674 chinfrancois@cwc.nus.edu.sg

Abstract: In this paper, a resource allocation scheme, Dynamic Power and Spreading Gain Control (DPGC), for voice and data integrated wideband CDMA networks is proposed to achieve multiple QoS requirements. Specifically, using fixed SIR and spreading gain for voice users, both power utility function and the packet throughput are maximized for each data user by adjusting spreading gain and the target signal-to-interference ratio (SIR) in the basestation receiver. Computer simulation has demonstrated that the proposed resource allocation scheme DPGC is effective in increasing throughput of data users with given SIR requirement and power constraint.

Key words: Spreading Gain Control, Wideband CDMA, Multimedia

1. INTRODUCTION

In a voice and data integrated wideband CDMA wireless network, both the power of all the users and transmission rates of data users may be considered as controllable resources. To cater for the requirement of high speed communication especially data applications in 30 wireless networks, we have to maximize data users' packet data rate Ithroughput while still stick to tight power limits due to battery size. Typically, the quality of service

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

258 Hailong Huang and Francois Chin

(QoS) in wideband CDMA can be controlled by an appropriate selection of transmitted power [1] and spreading gain [2].

In the previous papers, the different power control (PC) algorithms fall into mainly two categories: those based on the measured values of received power (signal-level-based PC [3][4]); and those based on the measured signal-to-interference ratio (SIR-based PC, CIR balancing PC [5]). In this paper, we only consider the SIR-based PC because it can provide larger capacity and better grade of service. To make it applicable for data applications in 30 wireless systems that require dynamic high throughput (bit rate), we try to maximize data user's power utility function [6] through network assisted distributed SIR-based power control [7][8] in uplink and simultaneously keep voice user's power consumption at a level such that it can support the voice user's QoS requirement.

The basic idea of spreading gain control is to (in)decrease the spreading gain of data terminals as the MAl level (in)decreases, which can result in great improvements in data traffic throughput compared to fixed spreading gain [9]. The general issue we wish to address in this paper is how to assign powers and spreading gains to different classes of traffic (voice and data) so as to maximize information rate given a fixed number of users and QoS constraints. The QoS performance measure for voice traffic is the target bit error rate (BER), while the data traffic the minimum throughput requirement.

2. SYSTEM MODEL AND PERFORMANCE MEASURES

We now specify a system model for the wideband CDMA network and the corresponding performance measures for it. Assuming that there are two classes of traffic, voice and data, within a single cell, the aim is to determine the throughput performance for different number of voice and data users.

In this integrated wireless system considered, each user (voice and data) generates a sequence of data packets. And a new packet is generated as soon as the preceding packet is successfully delivered. Since the users continuously transmit packets, the number of active users in the system therefore equals to the users in the system (no inactive users). The model for voice users differs from the model for data users in that voice users do not retransmit packs with errors. The same packet length and coding scheme are assumed for both classes.

Resource Allocation Using Dynamic Spreading Gain Control.... 259

The system channel rate is RbIs, where R = W /G. W is the CDMA system chip rate and G is the spreading gain. The DLC layer ftrst converts (segment or combine) the original information packets and ftnally encodes them into the target-length packets using certain error detection coding. After that, the sequence of packets is of length L bits with code rate r ( Lr information bits). We assume that the probability of undetected transmission errors is negligible.

Given N v active voice users and Nd active data users, the signal-to­interference ratio (SIR) for users in each class in the basestation receiver is deftned as:

(1)

(2)

where subscripts v and d indicate voice and data classes respectively. ~ watts is the transmission power , and 0'2 is the noise power in the base station receiver. The power assignment (Pv and Pd ) can mutually change SIRvand SIRd in a drastic way.

For voice users, since there is no need to retransmit the erroneous packets, we only concern about the BER and then assign a ftxed spreading gainGv·

Pbv = kexp(-pyv) (3)

where k and P are parameters which can be adjusted to match a particular coding scheme.

Data transmission is a sort of error-sensitive application, when the receiver detects an error in a packet, a selective ARQ protocol will request the source to retransmit the error packet again. Suppose the SIR is y d ' the packet success rate at the receiver is:

!(Yd) = (1-Pbd)Lr = (l-kexp(-Prd»Lr (4)

Utility function for data traffic is defmed as the number of information bits delivered accurately to a receiver for each joule of energy expended by the transmitter [10].

Ud = Lif(Yd) = Rdr!(Yd) (b/j) (5) PdLlRd Pd

260 Hailong Huang and Francois Chin

where Lrf (y d) is the expected number of successfully transmitted information bits, and PdLI Rd means the energy consumed in transmitting one packet.

Now we define data throughput as the number of correctly received bits per second.

T(y )= Lrf(Yd) =w,.4"(y )/G (6) d L/ Rd IJ d d

Since we adopt SIR balancing power control, all the active terminals have the same optimum targets, and then the same throughput.

3. DYNAMIC POWER AND SPREADING GAIN CONTROL (DPGC) FOR MULTI-QOS WIDEBAND CDMA

Now we describe the resource allocation scheme, DPGC, for this voice and data integrated wideband CDMA system. The resource allocation issue here is to select a proper assignment of powers and processing gains Pv , Pd ,Gv and G d so that QoS requirements are satisfied. The QoS measure for voice user is the BER, while that for data user is throughput. Now we aim to maximize the throughput of data users for fixed N v and N d in the capacity region, subject to an acceptable BER for voice users.

For voice users, since its data rate is very low, we assign a fixed spreading gain Gv = 256 for them. If Ev is the maximum acceptable BER for voice user, we can get an equivalent constraint from Equation (3), i.e., the minimum target signal-to-interference ratioy;.

.. .. 1 k 7) SIRv ~ Yv where Yv = f3 In(;:-) (

The constraint (7) can be combined with (1) to obtain the following: Pd S APv - B (8)

G - (K -1)y" (J'2 where A = v v v and B = _

Kdy: Kd From (2), we can get the other equation:

P _ Yd(KvPv +1) d - CPv +D

Gd -(Kd -1)Yd (9)

where C = KvYd and Gd - (Kd -l)Yd

Resource Allocation Using Dynamic Spreading Gain Control.... 261

2 D = CT Yd

Gd -(Kd -l)Yd

Then we can obtain the relationship between Pd and Y d by substituting (8) into (9),

AD+BC Pd = A _ C = g(y d) (10)

Now we can get the utility function for data users by combining (10) and (5),

U d = Rr f(Yd) (b/J) g(Yd)

(11)

Given voice and data user numbers and the spreading gain for data users, we can get the maximal utility and Y opt by differentiating (11) with respect to Y and set the derivative to zero. Next, we set the other feasible spreading gains according to WCDMA system parameters and get the corresponding Yopt in the same way. To choose the operation point from these feasible spreading gain and Yopt pairs, we need to see which one can get the maximal throughput using (6). At last, Pd and Pv can be obtained from (10) and (9) respectively using the operation pair we choose.

4. PERFORMANCE ANALYSIS AND NUMERICAL RESULTS

Now we will illustrate the analysis above and performance of the proposed DPGC with simulation results. The system parameters are listed in the next table.

In table 1, the QoS requirements for data and voice traffic are totally different. Data aims for maximal throughput with a lower bound, while voice for acceptable BER.

System Parameter Voice Data Chip Rate W 4.096x lOb chips/s

Spreading gain G 256 32 or 64 or 128

Code Rate r 0.75 Code Gain tJ 2 Code index k 0.5 Packet Length L 700 bits ~S:BER < 1O-~ NA

QoS: T NA ~ 2Ukbps Throughput

Table 1. System Parameter

262 Hailong Huang and Francois Chin

In this simulation, we must fIrst determine the system capacity, i.e. the users of both kinds of traffic that the system can support in the term of QoS requirement.

Figure 1 presents the system admissible region using DPGC and NAPC [6] with fIxed spreading gain 32. The line represents the limitation condition and the space below the line is the feasible state space, which means all the combination of two kinds of users in the feasible region can be supported by the system and vice versa. It's obvious that the proposed scheme DPGC can support more data users than NAPC, due to the flexibility to ameliorate the MAl. Figure 2 illustrates the data users spreading gain assignment. '+' and '.' stand for spreading gain 32 and 64 respectively. We can see clearly that when there are fewer users (especially the data users) in the system, it will assign 32 spreading gain to data users for the sake of maximizing the packet throughput. On the other hand, with the increasing of data users, it will assign a larger spreading gain as 64 to data users such that they can have a better packet success rate and therefore more data users can be supported at the same time at the price of lower throughput per user.

Fig 1. System Admissible Region

Fig 3.Data User Throughput (dynamic G vs. fixed G 64)

" j I. ~ •• a

1.$

0.

Fig 2. Data User Spreading Gain

Fig 4. Data User Target SIR

Resource Allocation Using Dynamic Spreading Gain Control.... 263

Figure 3 compares the data users' throughput ofDPGC and NAPe with a fixed spreading gain 64 when there are voice users ranging from 1 to 80. The curves in the figure stand for the throughput of 1 to 18 data users from top to bottom at the step of 2 users. The dashed line is the throughput lower bound for data users. It's obvious that when there are less than 10 users, the throughput of DPGC is much larger than that of NAPe. In case of NAPe with fixed spreading gain 32, it cannot support 18 data users as DPGC. Also note worthy is that the drop of throughput with the increase of data users is much more severe than the increase of voice users.

Unlike voice user, who has a fixed target SIR, data users varies their target SIR to balance the power and throughput. Figure 4 illustrates the target SIR for 3, 8, 13 and 18 data users. The turning points represent the change of spreading gain. Target SIR will influence another performance measure, packet success rate, which will eventually determine the throughput. Figure 5 shows the packet success rate for data users. There exists a direct proportion relationship between target SIR and packet success rate, that's why they have familiar curves. We can point out from this figure that the scheme works well since it can at least obtain a packet success rate of 0.4.

The power utility ratio of DPGC to NAPe with fixed spreading gain 32 is shown in figure 6. The curves in this figure stand for the ratio of 3, 8, 13 and 18 users respectively from right to left. The bigger the ratio, the better the power efficiency. Therefore, DPGC can obtain a better power efficiency than NAPe by 3000 to 6000 times especially when DPGC supports as many data users as possible.

JU I'" lo.

" '.1

, .. ..,. ,.".-

Fig 5. Data User Packet Success Rate

--i olOOC , .... _ " •• ...,. '.11110~ , .. ...,. " 1-....

,. DO " -_ .... Fig 6. Ratio of Utility (dynamic G to fixed G 32)

264 Hailong Huang and Francois Chin

s. DISCUSSION AND CONCLUSION

In this paper, we have addressed issues in the support of multimedia services in a wideband CDMA system. In particular, a resource allocation scheme, Dynamic Power and Spreading Gain Control (DPGC), for multimedia wideband CDMA system is presented. The scheme allocates bandwidth and power for data and voice users, and it differentiates voice from data calls by using higher spreading gain. By dynamically changing spreading gain, DPGC can achieve better throughput, system capacity and power utility than pure power control scheme. The simulation results show that the proposed resource allocation scheme DPGC is quite effective in increasing capacity of data users with higher SIR/throughput requirement and lower power constraint. For practical implementation, it only requires the network to broadcast the current user number in the system so that the mobile terminal can look-up for the corresponding spreading gain; while the basestation receiver will update the corresponding target SIR in its uplink power control without having to inform the mobile terminals.

References

[1] L.C. Yun and O.G. Messerschmitt, "Power control for variable QoS on a COMA channel," Proc. IEEE MILCOM, vol. 1, Fort Monmouth, NJ, pp.178-182, Oct. 1994

[2] C.L.! and K.K. Sabnani, "Variable spreading gain COMA with adaptive control for integrated traffic in wireless networks," Proc. IEEE VTC, vol.2, Chicago, IL, pp. 794-798, July 1995

[3] Oa Rocha Lima, A., Brandao, J.C., "General analysis of downlink power control in COMA systems," Telecommunications Symposium, 1998. ITS '98 Proceedings. SBTIIEEE International, Vol.1, pp. 172 -176,1998

[4] J.F.Whitehead, "Signal-level-based dynamic power control for co-channel interference management," in Proc. IEEE Vehicular Technology Conf.,Secaucus,NJ, May 1993, pp. 499-502

[5] S.Ariyavisitakul, "Signal and interference statistics of a COMA system with feedback power control," IEEE Trans. Commun., Vol. 41, pp. 1626-1634, Nov. 1993

[6] O. Goodman and N. Mandayam, "Network assisted power control for wireless data," VTC 2001 Spring. IEEE VTS 53rd Vol.2, pp.1022-1026,

[7] Hailong Huang, Francois Chin, "Maximizing uplink packet throughput and power efficiency for OS-COMA based wireless data system", 3Gwireless'2002, May, 2002

Resource Allocation Using Dynamic Spreading Gain Control.... 265

[8] Hailong Huang, Francois Chin, "Performance evaluation of various Network Assisted based Power, Packet and Spreading Gain Control scheme for DS­CDMA based wireless data system", VTC2002 Fall

[9] Joon Bae Kim, Michael L.Honig, " Resource allocation for multiple classes of DS-CDMA traffic", IEEE Trans. On Vehicular Technology, vol. 49, pp. 506-519, March 2000.

MOBILITY

On the Fixed-Point Implementation of Turbo Code in 3GSystem

Sun Minying Institute for Communications Research, Singapore

Tan Wee Tiong STMicroelectronics Asia Pacific Pte Ltd, Singapore

Abstract: Fixed-point arithmetic is mandatory in hardware implementation. This paper looks at critical issues of fixed-point arithmetic for turbo code. The techniques presented include approximation method to implement the non-linear item in Log-MAP algorithm, fixed-point representation of the soft input, soft output and internal metrics, sliding window strategy. and iterations stopping criterion. A complete set of parameters for immediate hardware implementation of turbo code is provided.

Key words: turbo code, fixed-point, 30

1. INTRODUCTION

Turbo code has attracted much attention in communications world, especially for cellular communication systems where the coding gains could be matched to bandwidth efficiency. In the standards for next generation mobile communications, turbo code is adopted as the coding scheme for both UMTS and cdma2000 system. In this paper, we discuss some critical issues of the fixed-point implementation of the decoding algorithm, in particular, the input quantization effect, bit-width of the internal metrics

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

268 Sun Minying and Tan Wee Tiong

computations, the sliding window strategy, and the stopping rule. Based on simulation results, efficient solutions are proposed for real time implementation of turbo code.

Most of the applications in 30 systems require real time implementation. For high date rate transmission, the buffer size of the channel coding chain becomes bottleneck for the real time implementation because of the transmit time interval (TTl) based operations (80 ms maximum). For example, the maximum coded data stream size equals to 19200 * N bits for a 384 kbps class user equipment (VE) capability, where N is the number of soft bits to represent one soft decision signal, which is determined by the decoding algorithm. It is obvious that N is a primary factor influencing memory requirement of the system.

Moreover, the memory size of the turbo decoding itself is also large due to the bit-wised interleaver of the coding block and bi-directional computation of path metrics requested by the algorithm. The large external memory brings out processing latency and thus lower the throughput. The number of bits used to represent soft input and internal variables is a very important design parameter that influences the total buffer size and hardware complexity. The optimisation strategies for each variable in the decoding algorithm will be discussed in details.

By using logarithm domain MAP algorithm, real time implementation of turbo code becomes feasible. However, the good performance of this coding scheme is obtained through iterative processing. Better performance can be achieved by more times of iterations, at the cost of processing latency. A stopping criterion is suggested based on the log-likelihood ratio, which is effective and simpler than the convention stopping criteria in terms of implementation.

Our discussions are limited to turbo code specified by 30PP standard, i.e.rate 113 , constraint length 4, with transfer function

(1)

The code block size for the turbo code ranges from 40 to 5112 bits. The next two sections describe the fixed-point implementation of the

decoding algorithm, with Section 2 analysing different approximating methods of the non-linear item in the log-domain computation, and Section 3 discussing fixed-point representation of internal variables and proposing a simplified stopping criterion. A summary is given in Section 4.

On the Fixed-Point implementation o/Turbo Code in 3G System 269

2. CORRECTION ITEM IMPLEMENTATION

In Log-MAP decoding algorithm, all the computations are linear, except the logarithm calculation, called correction item, in MAX* operator which is described in the form:

There are two basic ways to approximately realize this non-linear operation in fixed-point arithmetic - look up table and linear approximation. A fixed-point number is represented by (d, f) with d a total bit-width and f the fraction part. The non-linear function f(x) is then transformed to 2f x f ( x / 2f). Generally 3 bit precision, i.e. f = 3 is sufficient to achieve as good performance as infmitely soft quantization [5]. The least square method was used to approximate the natural logarithm function, and the fixed-point linear function for 3 bit precision was obtained in the form :

{5 -xl 4,

f fix (x) = 0,

o ~ x ~ 16

x> 16 (3)

The look up table method is described in details in [5]. The two methods are identical in terms of performance as well as computation complexity. What attracts us most is a constant approximation method, i.e. the correct item is approximated by a constant value. For a 6-bit input quantization, we select value 3 from look up table to approximate the correction item. The BER performance is shown in figure 1. It indicates that the constant approximation method has less than 0.05 dB loss comparing with look up table method. On the other hand, the decoding algorithm is sensitive to the SNR estimation because of the nonlinearity computation. The performance of each implementation method as a function of SNR estimation offset is displayed in figure 2. It is found that MAX-Log-MAP is not sensitive to the SNR estimation accuracy as expected. In contrast, the other three methods are sensitive to SNR estimation error. And the constant approximation behaves similarly as the other two methods in the presence of estimation error.

270 Sun Minying and Tan Wee Tiong

--eonsU6.3) --eonsU5.2) _ A_ Con.U4.1) -...- rable_(4. 1) -00- rSbIe_(5.2) -+- rable_(6.3)

o " ~=-,,,",~,

'" '"

0.0'

ID '£.4

~ ~ ~ ~ ~ u ~ u ~ ~ u u EbiNO [dB)

Figure I Performance comparison of different correction item implementation for various quantization level

0.,

'E"

.,

~Log --Ma. •. - Const,

. , SNR 0fIs81 (dB)

Figure 2. SNR mismatch effect at 0.5 dB

From these simulation results, we can conclude that the constant approximation Log-MAP algorithm is a good solution with 6-bit quantized soft input. However, if the number of soft input is reduced to 4-bit width. The performance of constant approximation method approaches MAX-Log­MAP algorithm with 0.5 dB loss of Et/No. The look up table method of 4-bit quantization losses less than O.ldB comparing with 6-bit quantization. In 3G system, Viterbi decoder generally co-exist with turbo decoder. 4-bit soft input is sufficient for Viterbi decoder to get desirable performance. Therefore, to save memory budget, 4-bit quantization is a better solution for 3G applications.

To reduce the storage of path metrics, sliding window (SW) technique was introduced to the forward and backward state metrics computation for MAP decoder. It is extremely useful to reduce the working memory. The forward and backward metrics are computed with a shortened window size Ns= NIW, where N is the entire block size and W is the number of

On the Fixed-Point Implementation a/Turbo Code in 3G System 271

segmentations. The SW technique can reduce buffer size of the backward metrics by W times. The disadvantage of SW technique is that a training sequence is necessary for the backward metrics computation in order to achieve reliable distribution of states for processing since the initial states of the backward metrics are unknown for all shortened blocks except the last one. Total redundant backward computations to decode one block of data are Nb *(Ns-l). Therefore, Ns shall be deliberately selected to compromise between memory size and computation overhead. According to the simulation results, Nb=6* (m+ 1) is sufficient to get reliable state distribution of 13. In our implementation, the window size is 100 bits and Nb is 24 bits. For the worst case in 3GPP standard, the code block size for turbo code is 5114, the SW scheme saves one third of the total memory while the computation increases 8% per iteration.

3. INTERNAL VARIABLES AND STOPPING RULE

The maximum difference between different forward state metrics can be loosely estimated as maximum difference between branch metrics. With the decoding block size increase, forward path metrics a. and backward path metrics ft will increase without bound. The magnitudes of a. and 13 increase along with the recursive processing and the magnitude of LLR increase with the number of iterations. Subtraction is commonly used for normalisation because the soft output is only influenced by the difference between state metrics instead of their absolute values. At a given time k, the maximum metric is determined and subtracted from a. or 13 at each state to scale down the values. The number of bits for each variable is then determined by observing its value range which is shown in Figures 3 and 4.

>DO 200 20D 100 100

00

• !! ... :!: • 100

. 100 .... . ,.. .... .... .... Number of Uef'ations

Figure 3 Value range of internal variables VS. number ofiterations

272 Sun Minying and Tan Wee Tiong

-- Alpha_max -.-Alpha_min • Ej ., . -' .-' _.

400 ____ I'!!. -- - •

300 to -------200 ••..

!l'OO~

.~~~ ...•. ·700 c -·000

.1000 ·"00 ·1200 .,300+-~-_r_~ _ __r_-~__r--__,

G,O O.S 10 1.5 2..0

EblNO (dB)

Figure 4 Value range ofintemal variables vs. Eb"No

According to the value range, 12-bit width is sufficient to contain all the received infonnation during the computation. It is found that when the clipping level is reduced from 12 bits to 8 bits, the perfonnance degradation due to quantization distortion is less than 0.1 dB. It is interpreted that, when the value of LLR reaches a certain level, reliable hard decision can be made upon LLR and truncation errors will not influence the perfonnance.

0.1

b •. ", •.

"'" -'" 1&3

~ 'E-< , , , 0

,E,6

'E" -<>~ 0.0 O~ ... 0.1 o.a ' .0 u u u

EbINO[dB)

Figure 5 Perfonnance of variable bit-width representation of path metrics.

Moreover, the LLR trend tells us that after the LLR value reaching its clipping level, e.g. 127 for 8-bit presentation, further computation of LLR is not necessary because the value is overflow and will be truncated to the maximum value. A simplified stopping criterion is suggested based on this numerical property of LLR. If minlLLRI > threshold, the LLR is considered reliable enough for a hard decision. Following this stopping rule, the iteration can stop at any decoding stage for a particular iteration as long as the criterion is satisfied. The advantage of our LLR stopping criterion over

On the Fixed-Point Implementation of Turbo Code in 3G System 273

conventional methods relies on the immediate halting at the current iteration, instead of waiting for comparison of two LLR results. The efficiency of this method is shown in figures 6 and 7, which indicates that the average number of iterations to achieve a certain BER performance is significantly reduced.

The dependence of the value range on the Et/No is presented previously in figure 4. The threshold is selected according to the value range of LLR in terms of following conditions: If the threshold is chosen based on value range at low Et/No, the iteration may halt before a reliable LLR is achieved at high Et/No and consequently, the performance is largely discounted. If the threshold is chosen based on high Et/No value range, extra effort is spent to satisfy the criterion at low Et/No. For example, At 3 dB Et/No, 4 iterations are needed to accumulate a LLR value of 311, as compared to 18 iterations needed to reach the same LLR level at 0.3 dB. Actually, the processing latency of 18 iterations is not acceptable for real time implementation Therefore, the threshold shall be adaptively selected for different Et/No. In our implementation, the threshold is chosen based on high Et/No value range, and at the same time, maximum number of iterations is fixed for low Et/No. The strategy is a compromising of performance and implementation complexity.

00 0.' 10 ,.. 2.0

EbINO[dB)

Figure 6 Average number of iterations vs. Et/No

274 Sun Minying and Tan Wee Tiong

,E-O +-o-"'T"""'"""T"""'"""T""""""T""""""T"""'"""T'"" .................. ......,..........,. ......... .::..-, 0.0 0.1 0.2 0.3 0.4 G.6 0.8 0 .1 0 ,8 0 .1 1.0 t.t 1.2

EbINO [dB]

Figure 7 BER perfonnance of the proposed stopping rule.

4. CONCLUSIONS

The optimisations for fixed-point implementation presented in this paper lead to significant reduction in hardware resource. By reducing the soft input quantization to 4-bit, the buffer size for TTl based operation is saved by 33%. Moreover, by using the (100, 24) sliding window scheme, the state metrics memory only equals to window size which saves around one third of the storage for turbo decoding of large coding blocks. The clipping and stopping method can reduce the computation by 50 % at high Et/No. If the clipping and stopping method is applied to a. computation, the computation of each iteration can be further reduced by 20 %.

5. REFERENCES

[1) Berrou, A. Glavieux, and P. Thitimajshima, Near Shannon limit error-correcting coding and decoding, Proceeding 1993 International Conference on Communication, p.l 064-1070.

[2] L.R.Bahl, lCocke, F.Jelinek, J.Raviv, "Optimal Decoding of Linear Codes for Minimizing Symbol Error Rate", IEEE Trans. Inf. Theory, Vol. IT-20, March 1974, pp.284-287.

[3) S. Wilson, Digital Modulation & Coding, Prentice-Hall, 1996, p. 604. [4) AJ. Viterbi, "An intuitive justification of the MAP decoder for convolutional codes",

IEEE JSAC, vol. 16, pp. 260-264, Feb. 1998 [5] G.Montorsi, and S. Benedetto, "Design of Fixed-Point Iterative Decoders for

Concatenated Codes with Interleavers", IEEE JSAC, vol. 19, No.5, May 2001. [6] 3GPP TS 25.212 version 3.4.0, "Multiplexing and channel coding (FDD)", [7) R.Y.Shao, S. Lin, "Two Simple Stopping Criteria for Turbo Decoding," Electronic

Letters, vo1.35, pp. 701-702, 1999.

Cellular Positioning by Database Comparison and Hidden Markov Models

Trond Nypan, Oddvar Hallingstad UniK - University Graduate Center

Abstract: Filtering of infonnation with respect to position is believed to be an important factor in order to increase revenue of 2.5 and 30 cellular communication networks. We present a technique, which in conjunction with enhanced cell-id positioning methods has the potential to provide relatively inexpensive positioning with improved accuracy, wide coverage, and low system impact. The idea is to compare measurements of location sensitive parameters done by the receiver with a database. This process gives rise to primary position estimates, which is filtered by a secondary estimation procedure based on hidden Markov modeling. The position error is less than 24.1 meters 67% of the time and within 71.3 meters 95% of the time.

Key words: cellular positioning, hidden Markov models, database comparison, pattern recognition, channel impulse response, channel sounding

1. INTRODUCTION

Many network operators see filtering of information with respect to position as an important condition in order to increase revenue of their 2.5 and 30 cellular networks. User terminal positioning in cellular communication systems of today is typically done using techniques based on a) OPS receivers in handsets, b) time-difference measurements from base­transceiver-stations (BTSs), and c) enhanced cell-id methods. None of these techniques have yet proven superior in terms of accuracy, cost, system impact and coverage. For most network operators it is natural to begin using enhanced cell-id in order to provide positioning in their cellular networks.

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

278 Trond Nypan and Oddvar Hallingstad

Enhanced cell-id methods are less expensive and have less system impact compared to the a) and b) methods. The accuracy obtained today in enhanced cell-id systems is from about 75 meters up to kilometers depending on the network in a specific area. Researchers are thus working on various methods of improving the accuracy of enhanced cell-id positioning.

One of the methods with the potential to further improve the accuracy of enhanced cell-id is based on database comparison of location sensitive parameters. In such systems location sensitive parameters along roads are mapped and stored in a database. Later the same parameters are measured by moving user terminals and compared with a database to yield position estimates. In [1], [2] and [3] the received power level (RXLEV) from surrounding BTSs is used for comparison. The parameter used in this paper is the channel impulse response (CIR) envelope as estimated from measurements in wideband mobile communication systems like the UMTS and Digital Audio Broadcasting (DAB).

We have chosen the CIR because scattering environments near the virtual line between the transmitter and the receiver influence the measured wideband CIR as a function of location. Although any location dependent parameters, scalar or vector, measured by the user terminal or the BTS may be used for location using the described set up.

Location database • Mapped ems • Digital maps

Figure 1. Possible system architecture

A possible architecture for a system using database comparison to yield position estimates may look like Figure 1. Here the processing load is put on a location server at the BTS. The user terminal may also perform the positioning processing requiring the location database to be available, e.g. on a CD in a vehicle set-up.

Cellular Positioning by Database Comparison ... 279

We divide the processing performed by the location server into two parts. First the database comparison is carried out. This produces what we refer to as the primary position estimates. These positions are input to the secondary estimation procedure. In [4] we used a Kalman filter (KF) to perform secondary estimation. In this paper we use a hidden Markov model (HMM) based estimator instead of the KF. In the sections below we present the primary and secondary estimation procedures, test system set-up, results and conclusions.

2. DATABASE COMPARISON

2.1 Location sensitive parameters

The most sensitive location dependent parameters measured by the user terminal or the BTS is the RXLEV and the eIR. The RXLEV from nearby BTSs is measured by the user terminal both in idle and in dedicated mode. In cities where the density of BTSs is high a relatively unique pattern is measured with respect to position. This is because the user terminal is monitoring RXLEV from many neighboring BTSs. In environments where the density of BTSs is low, e.g. rural and mountainous areas, the measured pattern is likely to be less unique.

In urban and mountainous areas there is most of the time a non line-of­sight path between the transmitter and the receiver. The main propagation mechanism is therefore by scattering from the surface of obstacles and diffraction around them. In practise energy arrives via several paths and a multipath situation is said to exist at the receiver. When measured by a wideband receiver, the eIR yields an estimate of the number of multiple propagation paths as well as their relative delay and strength. The uniqueness of the eIR depends on the topography of the area and the system bandwidth, e.g. built-up and hilly areas experience more distinct eIRs than flat areas when measured by a wideband receiver. The eIR is measured both by the user terminal and the BTS in dedicated mode.

2.2 Channel measurements

Measurements have been carried out using a channel sounder constructed to perform outdoor eIR measurements in the 900 and 1800 MHz ranges. It was designed and implemented by Siemens AG. A detailed mathematical description is given in [5]. The pulse repetition frequency and bandwidth

280 Trond Nypan and Oddvar Hallingstad

were set to 195.3 Hz and 5 MHz respectively. Four different routes in urban and suburban areas of Munich were measured. The route length varied from 400 to 800 meters. Several measurement runs were performed along the same routes.

2.3 Constructing the database

The database may be constructed in two ways. One is to process real measurements taken from a vehicle moving along the streets in the coverage area of the mobile communication system. The other way is to process predictions using state-of-the-art radio planning tools. The latter enables comparison to be performed also outside the streets. In this paper we use the fIrst method to build a database. The streets are divided into elements of 4 meters. All the measured CIRs within each street interval are averaged to yield one database item for every 4 meters street interval.

2.4 The processing steps of database comparison

The processing steps of the database comparison are described in detail in [4]. Basically the measured CIRs of the user terminal are processed identically to the ones in the database, except that averaging is performed over 0.5 seconds instead of over 4 meters. At time steps, I, each of the averaged CIR vectors, denoted ~ (I) , is compared with the database, denoted U, and a vector cost-function, denoted d(~(l),U), is calculated. This cost­function displays the similarity between the measured CIR and the ones in the database along the road. The primary position estimates, denoted y(l), may now be calculated according to the least Euclidian distance criterion, see [6] section 4.6, at every time step I.

Due to problems with ambiguity and reproducibility of the measured CIRs, the errors of the primary position estimates are relatively large, see discussion in [4]. The use of HMM enables street and vehicle motion modeling. The model is used to design an HMM based estimator, which performs the secondary position estimation. In the following sections we discuss how this technique may be used to obtain a relatively accurate and robust positioning system.

Cellular Positioning by Database Comparison ... 281

3. IDDDEN MARKOV MODELING

3.1 Introduction

HMMs [7] have been used in a wide range of applications, e.g. bioscience, control, communication, and image, speech, and signal processing. A main application of HMMs is speech recognition systems. One characteristic of speech recognition is that the speed of the various speakers is variable. It is thus necessary to compress or expand time in order to match measurements with a recorded database. This feature is normally referred to as dynamic time warping. In our database comparison system a similar situation occurs. The database records are coupled with discrete positions along streets in the coverage area, but the user terminals to be located have different velocities.

3.2 The hidden Markov model

A discrete Markov process (chain) may be described as a system being in one of N distinct states, denoted SI'S2'''' 'SN' as illustrated in Figure 2 (where N = 5 for simplicity). At regularly spaced discrete times, the system

Figure 2. A left-right Markov chain with 5 states and state transitions

undergoes a change of state (possibly back to the same state) according to a set of transition probabilities associated with the state. We denote the time instants associated with state changes as I = 1,2, ... , and we denote the actual state at time I as q(l). The state transition probabilities, denoted aI}, are defined by

al} = p( q(l + 1) = 5) I q(l) = Sj), i,j E {1,2, ... ,N} (3.1)

i.e. the probability that the model will be in state Sj at time 1+1 if it was in state S, at time I, where N is the number of states. This stochastic process is called an observable Markov model because the output of the process, y(l) , is mapped one-to-one to the states. In our research we have used a hidden

282 Trond Nypan and Oddvar Hallingstad

Markov model, which we denote l· (A, B, "'), where the observed output, y(/) , is viewed as a probabilistic function of the state [8]. In our case each state is a position interval along the road. The observation symbol probability distributions, bij, are defined by

bij = P(y(/) = Sj Iq(/) = s,), i,j E {1,2, ... ,N} (3.2)

i.e. the probability of measuring (observing) state Sj when the model is in state Si at time I. The initial state distribution, 1lj, is defmed by

Hi =P(q(l)=s,), iE{I,2, ... ,N} (3.3)

i.e. the probability that the model will be in state Si at time 1=1.

3.3 Parameter estimation

The transition probability distributions, denoted A = [ alj 1, are estimated from the speed distribution of vehicles in the coverage area. the observation symbol probability distributions, denoted B = [blj J, are estimated directly from the cost-functions, resulting from the primary comparison process, described in [4]. The initial state distributions, '" = [H;] , may be estimated from enhanced cell-id positioning methods. In our system we have assumed that the initial position is known.

3.4 State (position) estimation

Secondary position estimation using hidden Markov modeling is performed at time step I, by fmding the "optimal" state sequence associated with the given observation sequence, denoted Y = [y(1),y(2), ... ,y(/)]. Our optimality criterion is simply to choose the states q(l) which are individually most likely, at each time step I. This optimality criterion maximises the expected number of correct individual states. Note that we have not used the more complex Viterbi algorithm, which calculates the most probable state sequence, in our processing. To implement our solution the variable

(3.4)

is introduced. This variable expresses the probability of being in state Sj at time I, given the observation sequence, Y, and the model, ,t. The calculation

Cellular Positioning by Database Comparison .. . 283

of this probability was performed by using the relatively effective forward­backward algorithm, [9] and [10].

4. TEST SYSTEM

Four different routes were measured several times by the channel sounder. For each route the first CIR measurement run was used as database. The second was used to train the observation symbol probability distributions, B = r bi} 1 , of the HMM. The rest were used to estimate position using the previously described estimation procedure.

5. RESULTS

The results are based on about 12 km of distinct positioning trails using the described system set-up. The cumulative error distributions of the primary and secondary positioning process are depicted in Figure 3. The error using the HMM based estimator is less than 24.2 meters 67% of the time and within 71.3 meters 95% of the time.

:: ' •• ,' ", /'t-<~' C"-" :-'-'-----.-.. gOl ..... . ,' B ' ,. :S06 " ... .. I .. ~ . !

.~0.5 / ... .. I ]0.4" / ....

0.3 i ;

0.2 1 I

0.1; .. ,

50

Primary esUmales: •• ... .. ......... _ 61 pen:: 64.4 m

Error 1m)

95 pen:: 446 m

HMM •• Umale.: . - .- 61 pen:: 24.2 m

95 n:: 71 .3 m

100 150

Figure 3. Cumulative error distribution of the positioning process of urban and suburban areas of Munich

6. CONCLUSIONS

We have tested a cellular positioning system based on database comparison of the channel impulse response and hidden Markov modeling

284 Trond Nypan and Oddvar Hallingstad

of streets and vehicle motion. This is believed to be a way of increasing the accuracy of enhanced cell-id positioning systems used by network operators today.

Technology based on database comparison and HMMs is relatively easy and inexpensive to integrate in networks and standard mobile handsets. The accuracy obtained is satisfactory for most of the location-based services due to be introduced in cellular networks during the next years.

REFERENCES

[I] H. Laitinen, J. Uihteenmiki, and T. Nordstrom, "Database Correlation Method for GSM Location," presented at VTC 2001 Spring, Rhodes, Greece, May 2001.

[2] O. Kennemann, "Continuous location of moving GSM mobile stations by pattern recognition techniques," presented at PIMRC, 1994.

[3] S. Mangold and S. Kyriazakos, "Applying pattern recognition techniques based on hidden Markov Models for vehicular positioning location in cellular networks,"

presented at VTC Fall, 1999. [4] T. Nypan, K. Gade, and o. Hallingstad, "Vehicle positioning by database

comparison using the Box-Cox metric and Kalman filtering," presented at VTC Spring, AL, USA, May 2002.

[5] T. Fehlhauer, P. W. Baier, W. Konig, and W. Mohr, "Optimized wideband system for unbiased mobile radio channel sounding with periodic spread spectrum signals," presented at IEICE Trans. on Communications, Japan, 1993.

[6] R. O. Duda, P. E. Hart, and D. G. Stork, Pattern Classification, 2 ed: John Wiley & Sons Inc, 2001.

[7] L. R. Rabiner, "A Tutorial on Hidden Markov Models and Selected Applications in Speech Recognition," Proceedings of the IEEE, vol. 77,1989.

[8] L. Rabiner and B. Juang, Fundamentals of speech recognition: Prentice Hall, 1993. [9] L. E. Baum and J. A. Egon, "An inequality with applications to statistical estimation

for probabilistic functions of a Markov process and to a model for ecology," Bull. Amer. Meteorol. Soc., vol. 73, pp. 360-363, 1967.

[10] L. E. Baum and G. R. Sell, "Growth functions for transformations on manifolds,"

Pac. J. Math., vol. 27, pp. 211-227, 1968.

Architectural Considerations for Personal Mobility In the Wireless Internet

Mazen Malek Shiaa and Finn Arve Aagesen Department of Telematics, NTNU University, Trondheim - NORWAY {malekJinnarve}@item.ntnu.no

Abstract: Personal Mobility, and the need for it, is considered to be the main driving force behind the spreading of the Wireless Internet. Moreover wireless devices and applications are being developed based on a basic assumption that users need to get access to services regardless of their location and used equipment. Wireless Services, on the other hand, are very dynamic and have a vibrant configuration and settings. In order for this to be utilizable a new network architecture is required. In our department we are developing Telecommunication Architecture for Plug-And-play Systems (TAPAS) to be a generic platform and application development environment based on plug-and­play technology, which uses the Web as means for service definition, update and discovery. TAPAS is intended to satisfy three basic classes of properties: Flexible and adaptable, Robust and survivable, and QoS aware. Personal mobility falls into the first class. Aspects of user, application and session mobility are dealt with using different mobility management schemes. The paper discusses trends in the Wireless Internet concerning devices, applications and services, and presents the approach used within TAPAS. Personal mobility in this context is based on the user, terminal and session mobility.

Key words: Personal Mobility, User, Device, Application, Session, Plug and Play

1. TRENDS IN THE WIRELESS (lNTER)NET: DEVICES AND APPLICATIONS

Wireless Internet is the network of radio-connected user devices and servers using various services. Different players, such as governments, standards bodies, manufacturers, and commercial enterprises, are preparing

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

286 Mazen Malek Shiaa and Finn Arve Aagesen

the telecommunication industry for an upgrade of cellular phone services for a wireless Internet. The PC industry, is stepping up pace for providing the mobile users with wireless access through usual desktop environment.

To provide some aspects of the "Big Picture" one should answer the following question: which Platform is suitable for Wireless Internet services, and which devices and technologies are available for it?

There are different kinds of wireless devices capable of participating in wireless Internet services:

• Java phones: Normal cellular phones with the ability to run Java. • Wireless PDAs: Wireless handheld devices. • Communicating appliances: e.g. cameras, wireless printers, barcode

scanners, pens, etc. • Wireless PC: Laptops and notebooks with full desktop capabilities.

All of these devices fall into one of two main categories: Tiny or PC-like computational power. This concerns the onboard processor, memory, display, control, etc. These devices are available on limited scale, however their characteristics need to be improved to run content-rich applications.

In terms of services a variety is available for specialized persons and markets. These services use different mark-up languages to achieve poor content services, mainly browsing, such as WML (for WAP users), cHrML (for i-mode phones), HDML (for handheld), etc.

The key applications of any wireless system will be of complex interactive nature with rich and dynamic content, which cannot be fulfilled by simple browsing means. Applications in Commerce (mobile transactions with high security), Transportation (search for complex airline information, reservations), Airports (queue free check-in), customer relationship (automatic upgrades, remote service checks), etc. could all benefit from a Plug and Play platform that has full support for various mobility scenarios and built on common programming interface.

2. UNDERSTANDING PERSONAL MOBILITY

To see what will be future's networks and services, understanding the nature of people is needed. Users are mobile and what they keep on their person is important. A mobile person drives personal content, wireless applications, and even mobile hardware. Therefore, Personal Mobility is dermed by: "a utilization of services that are personalized with end user's preferences and identities independently of both physical location and specific equipment. "

A model, thus, is needed to represent the mobile user with all its profile and preferences, applications and sessions, devices, location and access

Architectural Considerations for Personal Mobility ... 287

point, underlying operating system and domain policies, etc. There are certain factors that play crucial role in defining the overall structure and complexity of any given model or platform. User-to-device relation could be utilized in different ways; e.g. exclusively one user per device, different users controlled by SIM cards, different users determined at login phase, etc. Another factor could be the mobility of Personal content, which is determined by the flexibility of moving user's specific information and preferences to the user's location and device.

3. TOWARDS A NEW NETWORK ARCmTECTURE

Addressing Personal mobility in a dynamic environment in which both resources and services vary in terms of availability and configuration is a real challenge. As introduced in the previous sections, figuring out what is mobile is half of the problem, keep it functioning is the other half. A vital and initial requirement is the availability of the services, or the loading of the application. When services are subject to dynamic change regarding their functionality, e.g. different versions at different domains, the platform should provide the means for automatic reconfiguration and update at every node and terminal. Once a terminal is accessing a service it should be able to keep on accessing it transparently of any further network and application management tasks. Issues such as resource (re )allocation, service (re)adaptation and dynamic (re)configuration should also be dealt with.

Requirements of platform to exploit the full power of mobility in the Wireless Internet could be:

• To be Person based and not device based. • "Anytime and anywhere" installation and operation. • Freedom for content providers and support for dynamic content. • Automatic discovery of new components and their services. • Support for different capabilities of devices. • Continuous adaptation to the environment.

There are additional requirements that are implementation dependent and depend on the type of development environment and the applications:

• Expressiveness: what interactive methods possible at user device • Availability: How to apply different addressing schemes

(physical and mobile) for different devices and applications. • Reliability: What central components to be replicated. • Personality: How to define User session and User profile.

288 Mazen Malek Shiaa and Finn Arve Aagesen

4. TAPAS AND ITS APPROACH FOR PERSONAL MOBILITY

4.1 What is TAPAS

TAPAS is a generic platform and application development environment that uses plug-and-play technology to provide flexible means for network operational tasks such as installation, configuration and management [1]. The basic model is founded on the concept of the Theatre metaphor, where actors perform according to a predefined manuscript and a director manages their performance. Actors could be software components running in a wireless device providing the user with certain functionality. Once functionality plugged in, it is possible to configure it, adapt it to the environment, manage it and move it. Figure 1 shows how different objects are related to each other in the TAPAS terminology. It is based on the director-actor relation, as well as the node-domain relation. Services and applications (or ServiceSystems) are stored as plays (different manuscripts for different Roles), while executed and performed by ApplicationActors. Dynamic control is provided through the CapabilitySupport that handles all resources and capabilities offered by nodes and domains, e.g. transmission channels and IP addresses. The Plug-in phase of every ApplicationActor, on the other hand, defines the quality at which it is going to execute.

Node mobility is supported through dedicating special purpose actors at each device, Mobility Agents, which are controlled by the Mobility Manager. There are two basic schemes for this kind of mobility management; device location update and node discovery [2]. This paper focuses on the Personal Mobility issues and support provided to applications.

4.2 Representation of Personal Mobility in TAPAS

What is central to our approach is the terminal identification and mobility of functionality. Within the TAPAS framework terms and notions on Personal mobility have been defined and related to active entities in the system. For example:

• person/persona is referred to by an ID and profile, • user-to-device relation is defined at login phase, • a user-application interaction is controlled by a User Agent, • every time a user logs in a user session is maintained, • personal content is defined by applications and is downloadable, • applications are dependable on complex capability system

regarding device, domain and environment characteristics.

Architectural Considerations for Personal Mobility .. . 289

manages

~an ges

User

VISitor

manages

Figure 1. The object model of TAPAS.

---.-----------------------------.----.-...... -.............. --.. --.. -..... ---- -.--- ------------------------------············---···-··-···1

--- !

i:l PlaysIManuscripts Capability Requirements

Web Server

~ Session 8ase

~

(tod~orsof othe~domalns)

_ .... _-_._._-------;-.. -~ User Profile 80.

I --L! C1J

I~a I, AppilcaUon Actors

, Node 3 Node 2 ! 1.._._ .... ___ .•••• ~4. ____ . __ . __ .. _ .. . _ .. . _ .••. _ .•••..•• _ ..•••..••• _ ..... ....... _ ••.... _ .... _ •. _ .•• _0_. __ . ___ .. _ ... _ . .. __ _ .. ____ .)

Figure 2. Terms used for Personal Mobility in TAPAS.

290 Mazen Malek Shiaa and Finn Arve Aagesen

Figure 2 provides a general overview of how Personal Mobility is handled by TAPAS. The devices depicted are characterized by different set of capabilities, that's why certain application components run at networks nodes instead of user's device. For purpose of multi-domain environments the domain's director, as there is only one director per domain, could contact other directors inquiring about visiting user's ID and profile.

4.3 User and Session Mobility

In TAPAS user's interactions are controlled by User Agent and user sessions are maintained by the director in terms of Session descriptions, which is a detailed sketch of running applications, actors and their related data. Figure 3 demonstrates how user's session is managed by the User Agent, and consequently maintained by the director's data base. An example is provided for a session description and a user profile. In this example application actors are distributed on the user's device and a network node, a typical example is a chat client and a server. When a session is suspended information on every actor's data, e.g. user name, connections, type of application and information for child sessions should be stored.

User

I I

{User Login phase, Interactions}

I ---

Director

Node 1 EPi'oIUe '--_______ _ ---' GUI...tIIngs,lIllPl1<:atton ltinga.Peraona.I

Node 3

Figure 3. Session mobility: A user is interacting with the system through UserAgent and its session is stored in the director's Session Base.

User's Login phase is central to the definition of users identity, characterization of device capabilities, resumption of user sessions, and transfer of personal content. It is used by the director to provide users with proper access rights and profiles. Therefore, there are two types of logins in a multi domain environment, as shown in Figure 4. Eventually, users could

Architectural Considerations for Personal Mobility ... 291

access their home domain after some inter-director negotiation and authentication.

Local Domain

-6 --, ,..-__ V_is_ito_r_D_o_m_ai_n ______

dlreCtorl I L I 4-'IUuL~

Oomalnl

Figure 4. User mobility: User A is accessing his home domain from Domain2.

s. EXPERIMENTATION

The basic configuration that we are experimenting with is based on two subnets, within one LAN network, which have two wireless parts. These are referred to by LAN! and LAN2 in Figure 5. Two sets of support systems and applications run at these subnets with two different directors. ill this configuration domains are viewed as an IP address range controlled by a subnet mask, and therefore directors consider plugged in nodes by their IP addresses. A mobile node, laptop computer, is plugged in at these two domains and considered as a mobile device running and participating in the two applications. This mobile node acquires two different IP addresses before the user can login. Different Mobility scenarios are being performed, e.g. user login and logoff, session suspend and resume, visitor domain login, device identification, etc.

6. CONCLUSION

This paper demonstrates Personal Mobility approach in TAPAS. Basic user, device and session mobility issues are introduced. It provides a general view of requirements for supporting this kind of mobility, and some implementation and experimentation features are pointed at. Furthermore, a closer look is provided at Wireless illtemet applications that could be modelled and developed using this platform.

292 Mazen Malek Shiaa and Finn Arve Aagesen

Node 4

Figure 5. Experimentation with two domains and a mobile device.

Future work is meant to concentrate on two issues. First, extending the platform to handle the wireless aspects, such as: signal quality variation, connection recovery, and intelligence on missed nodes and partitioned networks. Second, providing support and basic functions for small devices, such as PDAs and Java phones.

REFERENCES

1. Aagesen, F. A., Helvik, B. E., uwongse, V., Meling, H., Bra:k, R., Johansen, U. "Towards a plug and play architecture for telecommunications". Paper presented at the IFIP TC6 Fifth International Conference on Intelligence in Networks, Bankok, November 1999.

2. Mazen Malek Shiaa and Finn Arve Aagesen. "Mobility management in a Plug and Play Architecture", IFIP TC6 Seventh International Conference on Intelligence in Networks, Saariselka, Finland, April 2002. Available from: http://www.item.ntnu.no/-plugandplay/publicationsl

A Development of Flexible Access Control System for Advanced ITS Networking

Mitsuo Nohara, Sheng-Wei Cai, Hitoshi Inoue, Yoshiro Okamoto and Tadao Saito Toyota Info Technology Center, Co., Ltd.

2-1-23, Nakameguro, Meguro-ku, Tokyo 153-0061 Japan Telephone: +81357232600 Facsimile: +81 357232690 email address:nohara@jp.toyota-itc.com

Abstract: A combined use of the ceIlular and high-speed "hot-spot" communications can provide higher-rate and more convenient network access services to mobile users on-board vehicles. The authors developed an experimental flexible access control system so as to indicate the technical feasibility of such services. The system consists of a set of QoS monitor and mobile router that monitors and selects the optimum wireless access media among the PacketOne, Dedicated Short Range Communication (DSRC) and wireless LAN. It also consists of a pair of flexible access processing module and gateway that provides a seamless data stream to the applications. A series of field experiments has been conducted at KDDI Training Centre, Saitama, Japan, where the access points of DSRC and wireless LAN and other network equipment were placed for the experiment. The experimental results showed that the flexible access control works wen for the optimum link selection and the flexible access buffer control provides a seamless streaming at the average throughput much higher than the cellular one.

Key words: Mobile Network, Wireless Access Network, Wireless Link QoS Control, Mobile Routing, Seamless Applications and Field Experiment

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

294 Mitsuo Nohara, Sheng-Wei Cai et al

1. INTRODUCTION

An enhancement of the advanced Intelligent Transport System (ITS) is expected to make it a major part of the information infrastructure for the network users on-board vehicles. A kind of high-speed wireless access media is expected to become available for the users so that they can access to the network in the same manner they do at their homes and offices using the broadband Ethernet. In addition to the conventional cellular networks having the wide coverage, there have been several high-transmission rate wireless access schemes developed. One example is the Dedicated Short Range Communication (DSRC)[l] which can act as a so-called "hot-Spot." Within its coverage, users can access to the network at a higher rate, compared with some conventional cellular ones. Along with such "hot-spot" deployments, an efficient use of the mUltiple media of high bit-rate wireless access is one of the key issues to achieve an optimum use of them. Meanwhile, some technique shall be applied so as to provide users seamless applications over the wireless media change with a short-period interruption. Following such discussions, we set two key issues to be developed as follows: - automatic optimum wireless access media selection, and - seamless application provisions over the media change.

The authors have developed an experimental flexible access control system so as to achieve those key issues and indicate the technical feasibility of the service to be provided over multiple number of wireless access media. In section 2, the flexible access control and experimental system are described. In section 3, the field experiment results are shown. And fmally section 4 concludes the paper.

2. FLEXIBLE ACCESS CONTROL SYSTEM

For the system design, we assumed a use of the following wireless access media: PacketOne, DSRC and wireless LAN (IEEE802.11b.) The PacketOne represents a packet-switched cellular network with nation-wide coverage in Japan at a transmission rate of 64 kbitls. As for the DSRC, it is expected that the number of road-side access points will be expanded to cover the places such as high-way service areas, petrol-stations, parking lots and shops. The wireless LAN will also be used at home-garages. Those can be regarded as promising media candidates for the advanced ITS use.

A Development of Flexible Access ... 295

Figure 1 presents the system configuration developed. The QoS controller (QoS Cont.) monitors each of received signal level and packet throughput and choose the best medium at the time and place. The mobile router then establishes a network · access path over the selected link, by updating the home agent (HA) registration with the link information. The mobile IP technology[2] is used for this selection. The pair of flexible access processing module (F APM) on-board and flexible access gateway (FAGW) at the network side configures the middleware buffer and provides a seamless, continuous application service to the user. The F APM is installed on one of the Car-PC though which an internet access from the other PC is also provided.

_ .................. _- ---------------................. __ ............ .. /'.-' ----, :

I ! ~--I

Network-side System

Figure 1. A flexible access control system configuration.

2.1 Wireless access media QoS Control

As explained in the previous section, the QoS Cont. monitors the received signal level and throughput of each wireless access medium. Figure 2 indicates the received signal level and throughput of each medium. The media control is achieved using those measured data. In the figure, the left­side column depicts the link-selection parameters. From the top, the parameters are as follows: The priority defines the media selection order. The switch interval defines the minimum link period during when the selected link is kept so as to avoid any media chattering at an unstable link condition. Each threshold value defmes the medium "on" when the received signallevellthroughput is above it. The media switch can be done either automatically or manually.

The figure shows a measured example of the automated media selection in which the received signal level is used as a criterion and each throughput

296 Mitsuo Nohara, Sheng-Wei Cai et al

shows the actual data throughput sent over the link. At first, the W -LAN is used then along with the received level degadation, the link is switched to the PacketOne. Then along with the level recovery, the W-LAN is selected again. Finally, the DSRC is selected when its received level becomes "on."

Media Switch

,. W-LAN r OSRe

Cf.llular

Auco-selftC

Figure 2 . . A measured example of QoS monitor and media selection.

As shown in Fig. 2, the media selection has been correctly performed using the received signal level as a QoS measure among the three wireless media.

2.2 Flexible Access Buffer Control

The pair of F APM and F AGW has been applied so as to provide a seamless application to the user even when the wireless link interruption is occurred during the media switching. Each of the F APM and F AGW bas some amount of middleware buffer memory allocated to each application.

A Development of Flexible Access ... 297

Figure 3 shows an example of the F APM buffer control for the video streaming to a mobile user. The link condition for the measurement is the same as the one shown in Fig. 2. At fIrst, the amount of buffered data increases using the W-LAN, then decreases using the PacketOne, increases again using the recovered W -LAN and fInally decreases using the DSRC. At last as shown in the middle, the data is reset, when the application is terminated. It is observed that an example of 500 kbitls-video stream can be provided with no interruption over the three media switched time-to-time. It is also observed that the amount of buffered data at the F APM varies according to the media throughputs and the application rates output from the buffer. Meantime, the data amount at the F AGW varies according to the wireless media throughputs, as the network throughput is fast enough compared to the wireless ones.

Figure 3. A measured example of flexible access buffer control at F APM.

As shown in Fig. 3, the flexible access buffer control, placed as a middleware pair, has been used for the seamless application service provision.

3. FIELD EXPERIMENT

The experimental system shown in Figure 1 has been developed. In addition to the QoS Cont.lMobile Router and F APMIF AGW pair, the system also consits of a pair of on-board and road-side I-Mbitls ASK DSRC modem,

298 Mitsuo Nohara, Sheng-Wei Cai et al

an MPEG-4 encoder/decoder pair, Web-cameras, and so on. The on-board system has been placed on the experimental vehicle of Toyota Grand-High Ace, on which an UPS power generator and battery set has also been equipped. The car has space wide enough for the on-board system mount including the UPS set and the equipment rack.

The network-side system has been placed at the KDDI Training Center, Saitama, Japan where the network has been prepared. The venue has been built and run by the consortium members for the experiment ofKDDI Labs., Toshiba, Panasonic and Toyota InfoTechnology Center. A series of field trials has been conducted there so as to examine the technical feasibility of the developed system and to conduct the quantitative link-performance evaluations.

Tables 1 and 2 indicates the switching period from one wireless media to the other, in a manner of manual and automatic switching, respectively. As shown in Table 1, the switching period from the W-LAN to DSRC is 3.0 second while the one to PacketOne is 15.5 second. The first one of3 second represents the media switching period consumed for the mobile-IP control with a HA-update. Meantime, the one for the PacketOne further includes the switching over the public cellular network. The values shown in Table 2 varies, as each contains the link-status change detection of about 8 second. Referring to those results, a switching duration of several seconds shall be taken into account for the flexible access buffer sizing.

a e e a SWltc mg peno m sec., manua SWltC 19. T bl 1 M di . h' . d' . hin

Wireless Access ~W-LAN ~DSRC ~PacketOne

Media FromW-LAN * 3.0 + 0.6 15.5 ± 4.5

DSRC 4.1 + 0.5 * 16.1 + 2.9 PacketOne 3.9 + 0.8 3.4 ± 0.7 *

T bl 2 M di 't h' . d' . h' a e e a SWI c mg peno m sec., automatIC SWltC mg. Media, from W-LAN PacketOne W-LAN ~to ~PacketOne ~W-LAN ~DSRC

Measured 8.021:9 5.3 + 2.6 10.0 + 0 data

Also, sets of vehicle-mounted I-DIN sized PC with a touch-paned display and browser have been developed. The one has been mounted on the vehicle front panel for the driver use, so has been the other in the middle of the front-row sheets for the passengers use.

A Development of Flexible Access ... 299

Photo 1. An external overview of the experimental vehicle.

Photo 1 shows an external overview of the experimental vehicle. Also Photo 2 depicts an on-board view taken during a field experiment.

Photo 2. An on-board view taken during the field trial.

300 Mitsuo Nohara. Sheng-Wei Cai et al

4. CONCLUSION

A flexible access control system has been developed, which consists of the set of QoS controller and mobile router, the pair of F APM and F AGW and several experimental equipment and measurement tool. A set of three promising wireless access media candidates of PacketOne, DSRC and W­LAN has been applied. A series of field experiments has been conducted so as to indicate the technical feasibility of such system and to evaluate the link performance quantitatively. The results showed that the pair of flexible wireless media selection and buffer control can provide a seamless application even over the link changes to the application for on-board customers. It is expected that those results are to be reflected for further studies with a variety of wireless media to be newly introduced such as cdma2000 lX-EVDO, IEEE802.11a, etc. It is also expected that the mobile­IPv6 technology, which is to be standardised at the IETF, will be taken into account for further mobile router configuration.

ACKNOWLEDGEMENT

This research was conducted under the contract from Telecommunications Advancement Organization (TAO), Japan.

Thanks are due to Mr. Tadao Mitsuda, President and CEO of Toyota InfoTechnology Center, Co., Ltd. and Mr. Hisanobu Nakagawa, fonner project research member, for their helpful guidance and useful discussions given through the project.

REFERENCES

(1) ARIB Standard STD-T75, "Dedicated Short-Range Communication System," Ver. 1.0, Association of Radio Industries and Businesses (ARID), Sep. 2001.

(2) Mobile IP: The Internet Unplugged. J. D. Solomon, Prentice Hall, 1998.

Ubiquitous Access to Personalised Services

1 Tore E. J{lSnvik, 2 Anne Marie Hartvigsen & 3 Do van Thanh 1. Unik - University of Oslo - Norway - tif: +4790199176 - torejoen@iji.uio.no 2. AgderUniversity College - Norway - tif: +4795755499 - amhartvi@stud.hia.no 3. Telenor R&D - Norway - tif: +4790977102- thanh-van.do@telenor.com

Abstract: Nowadays, the user expects not only to be able to access services and applications anywhere at anytime and on any terminal fixed or mobile; but also in the same way as at home. In order to provide ubiquitous access to personalised services this paper propose to implement the User Profile containing the user's preferences, settings and personal data, as an XML web service which is accessible to the user at anytime anywhere and on any terminal. As an XML Web service, the User Profile will also be accessible from any application over operator's borders, firewall and without the need of a specific middleware platform, e.g. CORBA, RMI, DCOM, etc. With such a User Profile, ubiquitous and customised access to applications and services will be a reality. The paper starts with a brief overview of the XML Web service concept. The User profile is described in details before the main focus, which is about realising the User Profile as an XML service. In the conclusion, further works and activities are proposed.

Key words: XML Web services, Web technology, User profiles, Distributed Computing, Ubiquitous service access, Service customisation, Preferences and Settings

1. INTRODUCTION

Nowadays, the user expects not only to be able to access services and applications anywhere at anytime and on any terminal fixed or mobile; but also in the same way as at home. The functionality, the behaviour, the presentation, the look and feel, the preferences and settings, etc. should preferably be the same or as close as possible to what he/she is used to. Of course, it is always possible for a skilled user to set up and customise an

C. G. Omidyar (ed.), Mobile and Wireless Communications© Springer Science+Business Media New York 2003

302 Tore E. JfjJnvik, Anne Marie Hartvigsen and Do van Thanh

application to his taste, and to restore the original settings after use. However, it is a boring and time consuming task that he/she would prefer to be exempt from For a non-technical user such a task can neither be required nor accepted.

In telecommunications, the User Profile is used to capture the user's preferences, settings and personal data such as address/telephone list, bookmarks, family photos, encryption keys, etc. In order to make the User profile available to any desired application or service at the same time as to allow the user to access and modify it at anytime anywhere and on any terminal, we propose in this paper to make the User Profile available on the World Wide Web as an XML Web service.

As an XML Web service, the User Profile will be accessible from any application over operator's borders, firewall and without the need of a specific middleware platform, e.g. CORBA, RMI, DCOM, etc. With such a User Profile, ubiquitous and customised access to applications and services will be a reality. The paper starts with a brief overview of the XML Web service concept. The User profile is described in details before the main focus, which is about realising the User Profile as an XML service. In the conclusion, further works and activities are proposed.

2. USING XML IN DISTRIBUTED COMPUTING

Distributed Computing is not a new research area. It was addressed already by CORBA, RMI, DCOM, etc. What XML web services additionally offer is an XML based access via transparent Internet protocols like HTTP, relieving distributed computations from the need of middleware platforms and enabling the use of services across company borders, frrewalls, infrastructures and technologies.

Several large software companies and standardisations organisations have gathered behind the Web service concept, putting a lot of work, money and marketing efforts into it. However, there exists a lot of confusion and divergent opinions of what Web Services are and what they should be used for.One might classify Web Services as any kinds of services that are available on the World Wide Web. Often this would include customer focused services that are available through Web browsers, such as tracking a shipment or maintaining a customer profile through a Web interface that connects to underlying systems.

In this paper, by XML Web services it is meant self-contained, modular applications that can be described, published, located, and invoked over a network [1]. Specifically these applications use XML for data description, SOAP (Simple Object Access Protocol) for messaging (invocation), WSDL

Ubiquitous Access to Personalised Services 303

(Web Service Description Language) for service description (in terms of data types, accepted messages, etc) and UDDI (Universal Description, Discovery and Integration) for publishing and discovery. Figure 1 shows how the protocols play together in the Web Service architecture. The idea with XML Web services is that one application should be able to dynamically exploit the functionality of other applications, which is exposed as Web services. The applications offering Web services might themselves use other Web services, either for accessing content and business logic, or to utilise supporting services such as security, billing, orchestration, etc. [2].

1. Publish service

description. including

WSDLflle

Web Service Provider

3. Service Invocation

MeeS'aD'~W"'.

SOAP Extanalon.

SOAP

HTTP SMTP FTP

TCPIIP

Transport Protocolll

2. Discovery of suitable services

and how to

HTTPS

Web Service Client

Figure 1. Web Service Protocol Architecture

3. WHAT IS A USER PROFILE?

In telecommunication systems, in order to allow the customisation of services, a User Profile is introduced to capture the preferences and settings of the users as in the case of UPT (Universal Personal Telecommunication) [3] and in TINA (Telecommunications Information Networking Architecture) [4]. In GSM (Global System for Mobile Communications), such a User Profile is called Subscriber Data or Subscriber Profile [5]. The

304 Tore E. Jf/Jnvik, Anne Marie Hartvigsen and Do van Thanh

User Profile contains information, which is required for service provisions, identification, authentication, routing, call handling, charging, subscriber tracing, operation, and maintenance purposes. Unfortunately, the User Profile as defined has many limitations. The User Profile is intended for the customisation of the main service, namely voice communication or telephony, and its supplementary services, e.g. call forwarding, call answering, etc. It is also stored deep inside the operator's system and is hardly available to the 3rd applications or services.

1 1 lIJeIQIIa

~ UiIeIRoIIIe

""-1<1 ~ L...

-txx:krIr1<S I"'"

1

1 I 1 .~ 1 1 1 I FbJIIgrfo iOBt#rgric ~

lIgef~1caIcJ'f'IU\1e u: 1

~ I~

'1~ 1

0 .1 U.l I U.l 0 .1

I .............

.,.,-......~ .. "' .. tr ' _.......UII) ~_...-.n .....

Figure 2. The User Profile structure

In order to allow the users access to multiple applications and services anytime, anywhere and on any terminal, the content of the User Profile needs to be extended to fulfil the following requirements: • For each user the User Profile shall be expandable to incorporate the

preferences and settings for any additional application or service that the user requires.

• For each application the User Profile shall contain the information necessary for the presentation of the application on the terminal types requested by the user.

• For each application the User Profile shall contain the usage restriction • The User Profile shall incorporate also user's data such as address book,

telephone list, bookmark or favourite link list, etc.

Ubiquitous Access to Personalised Services 305

We propose a structure for the User ProfIle in UML (Unified Modeling Language) in Figure 2.

4. USER PROFILE AS AN XML WEB SERVICE

Several issues need to be addressed in the design of the User ProfIle Web Service (UPWS):

• Storage and retrieval of user profIle data • User access to and modification of user profile data • 3rd party application access to user profIle data

To manage the data retrieval and providing the UPWS we need a small application, and we will call it the User Profile Manager (UPM). This application will act as a middle layer between the data itself and the user terminal or application.

4.1 Storage and retrieval of user prorIle data

When deciding how to store and manage XML data there are at least four alternative ways to do it: In a fIle system, in a native XML database, in a modified object database or in a relational database. File systems and native XML databases are appropriate when one is storing XML documents rather that XML data, the latter being more structured. Object databases have been identified by several observers as a natural storage for XML, but have yet to prove their usefulness in this context. When storing structured data, it is generally recommended to use a relational database and then extract the information from the database to XML when needed.

Structured data will benefit from the relational model when it comes to retrieval, searching and aggregation of data. While this forces a need to transform, or decompose the original XML document, decomposing an XML document to persist it to a relational database is not all that difficult. Also, many relational database vendors are implementing thin XML serialiser wrappers that enable them to generate XML documents on demand from relational data. So even if data will be coming in and going out as XML, e.g. in the form of Web services SOAP messages, they can be stored as relational data.

One obvious advantage with using relational databases is that the technology is mature, stable and Ubiquitous, and there exists a whole range of tools for working with relational databases.

The UPM will extract the relational data into XML documents before providing applications and users access to them. It will also have to extract the data from XML documents and update the database. Depending on the

306 Tore E. l¢nvik, Anne Marie Hartvigsen and Do van Thanh

implementation, more or less of this functionality can reside in the database itself.

4.2 User access to and modification of user prorde data

Data Store

Figure 3. User updating the User Profile through a dynamically created user interface

The user must be able to access his/her profile to read and modify settings for existing terminals and add new terminals. This should be done through any kind of terminal, e.g. Web browser, mobile phone, PDA, etc. This means that functionality for reading from and updating the database information must be exposed to the user and a suitable user interface must be provided.

The UPM can apply XSLT style sheets to XML documents in order to transform the documents into HTML (XHTML) or WML (or any other desired mark up) according to what type of device is calling the service. See [6][7] for more details on possible architecture and implementation. Figure 3 shows a user accessing the user profile through a W AP interface.

4.3 3rd party application access to user prorde data and functions

The end user application provided by any 3rd party must get access to the UPWS in order to configure according to the settings defmed in the profile. The application should not have access to the whole profile, merely the parts relevant for the type of terminal and service in use, and in according to user

Ubiquitous Access to Personalised Services 307

specifications. Conforming to de facto Web service standards, the UPWS uses the SOAP protocol for wrapping data to be exchanged between the user profile and the application. Based on the information retrieved from the profile service, the application can provide the user with a personalised service, as shown in Figure 4.

User Profile Web Service (SOAP interface)

UPM Servlet

Data Store

Figure 4. Application using the UPWS for personalisation of its services to the end user

It might be desirable that the user be able to access the profile through the 3rd party application. The functionality required is the same as if the user accessed the profile directly, but the UPM does not need to provide any user interface, since that will be taken care of by the 3rd party application. However, a SOAP interface must be provided so that the 3rd party application can interact with the services.

4.4 Use Scenarios

1. The user accesses the profde through a mobile phone. After the user logs in through the UPM, the UPM extracts the correct user profile data from the database and wraps them in XML. Since the terminal is a WML browser, a WML style sheet is used to create WML pages that displays the desired parts of the XML document that constitutes the user's profile. Through the UPM the user gets access to the data and can update the User Profile. All needed functionality is disclosed to the user through the W AP interface.

2. The user uses an application that utilises the UPWS. In order to provide a personalised service to the user, the 3n1 party application contacts the UPWS and fetches the nessesary user data. These are then used to configure the interaction with the user. The user never

308 Tore E. If/Jnvik, Anne Marie Hartvigsen and Do van Thanh

has to worry about the profile, since all communication with the UPWS is between the UPWS and the 3rd party application.

3. The user modifies hislher User Profile while using a 3n1 party application. In this scenario the application must be able to update the database, using the same functions as in scenario 1, but this time the interface is not WML, but SOAP. The user performs the same functions, but through the user interface provided by the application, instead of the interface provided by the UPM.

5. CONCLUSION

We have suggested that it can be useful for telecom operators to offer User Profiles as XML Web services in order to provide 3rd party applications access to user data, preferences and settings. This way the application providers can offer customised services to the user anytime, anywhere and on any device, without the user having to maintain the same data in several applications.

Our work is in an early stage, and further work must be done in order to decide on implementation details. Choice of platform (J2EE based or .NET) is one of the significant decisions to be made. When considering platforms and other implementation details, interoperability issues inevitably become an important aspect, as does issues concerning security and reliability.

REFERENCES

[1] IBM Web Services Architecture Team. Web Services Architecture Overview. [Online]. Accessible: http://www-l06.ibm.comldeveloperworksllibrary/w-ovr/. 2000 [2001,18. December]

[2] Hagel III, J., & Brown, J. S. Your Next IT Strategy, Harward Business Review Vol. 79, Issue 9, 2001 p.l05-113.

[3] ITU-TS Universal Personal Telecommunication (UPl) Service Description, version 10 editio, January 1994. Draft recommendation F851, version 10

[4] TINA Consortium - Service Architecture - version 5.0 - 16 June 1997 -http://www.tinac.com

[5] ETSII3GPP - GSM- Digital cellular telecommunications system (Phase 2+) Organization of subscriber data (Release 1998) TS 03.08 V7.4.0 (2000-09)

[6] Biancheri, C., Pazzaglia, J.-C., Peddeu, G. EIHA?!? : deploying Web and WAP services using XML Technology. Sigmond Record. Vol. 30, Issue 1,2001, p. 5-12

[7] Coyle, F.P. Breathing Life into legacy. IT Professional. Vol. 3, Issue 5,2001 p. 17-24