Next.generation.wireless.networks

NEXTGENERATION

WIRELESSNETWORKS

THE KLUWER INTERNATIONAL SERIESIN ENGINEERING AND COMPUTER SCIENCE

NEXTGENERATION

WIRELESSNETWORKS

edited by

Sirin TekinayNew Jersey Institute of Technology

KLUWER ACADEMIC PUBLISHERS NEW YORK, BOSTON, DORDRECHT, LONDON, MOSCOW

eBook ISBN: 0-306-47310-0Print ISBN: 0-792-37240-9

©2002 Kluwer Academic PublishersNew York, Boston, Dordrecht, London, Moscow

All rights reserved

No part of this eBook may be reproduced or transmitted in any form or by any means, electronic,mechanical, recording, or otherwise, without written consent from the Publisher

Created in the United States of America

Visit Kluwer Online at: http://www.kluweronline.comand Kluwer's eBookstore at: http://www.ebooks.kluweronline.com

Contents

Introduction 1

1. Infostations: New Perspectives onWireless Data Networks 3Ana Lucia Iacono and Christopher Rose

2. Wireless Broadband Multimedia andIP Applications via Mobile ATM Satellites 65Abbas Jamalipour

3. Infocity: Providing QoS to Mobile Hosts 109Patricia Morreale

4. Assisted GPS for Wireless PhoneLocation- Technology and Standards 129Bob Richton, Giovanni Vanucci, and Stephen Wilkus

5. Evaluation of Location DeterminationTechnologies Towards Satisfying the FCCE-911 Ruling 157M. Oguz Sunay

6. A Series of GSM Positioning Trials 195Malcolm Macnaughten, Craig Scott, and Chris Drane

7. Enhancing Terminal Coverage and FaultRecovery in Configurable Cellular Networks UsingGeolocation Services 231Mostafa A. Bassiouni and Wei Cui

vi

8. UMTS Applications Development- DesigningA “Killer Application” 255Gunther Popischil, Ernst Bonek, andAlexander Schneider

Index 264

Introduction

This book is a collection of extended versions of the papers presented at theSymposium on Next Generation Wireless Networks, May 26, 2000, NewJersey Institute of Technology, Newark, NJ. Each chapter includes, inaddition to technical contributions, a tutorial of the corresponding area. It hasbeen a privilege to bring together these contributions from researchers on theleading edge of the field. The papers were submitted in response to a call forpapers aiming to concentrate on the applications and services for the “nextgeneration,” deliberately omitting the numeric reference so that the authors’vision of the future would not be limited by the definitive requirements of aparticular set of standards. The book, as a result, reflects the top-downapproach by focusing on enabling technologies for the applications andservices that are the defining essentials for future wireless networks. Thisapproach strikes a balance between the academia and the industry byaddressing new wireless network architectures enabling mobility andlocation enhanced applications and services that will give wireless systemsthe competitive edge over others.

The main theme of the book is the advent of wireless networks as anirreplaceable means of global communication as opposed to a mere substitutefor, or a competitor of, wireline networks. Geolocation emerges as thefacilitator of mobility and location sensitive services. The fields ofgeolocation and wireless communications have been forced to merge,following the Federal Commission of Communications’ (FCC) ruling thatobliges wireless providers with emergency caller geolocation. This initialdriving force has quickly been augmented by the already existing andincreasing popularity of positioning and navigation systems using the GlobalPositioning System (GPS), in addition to the wireless providers’ zeal to addvalue to the geolocation capability. The result is the currently experiencedevolution of wireless networks where mobile location is a natural aid tonetwork management, to a variety of applications and value added services.At this time, the path of the evolution is not clearly focused, neither is thewinning set of applications obvious. What we do know is that nextgeneration wireless networks will continue to change the way we live.

The first part of the book contains tutorials on three network architecturesthat aim to achieve this vision. The first chapter, by Ana Lúcia lacono andChristopher Rose of WINLAB, Rutgers University, presents the concept of“Infostations,” that arise from the tradeoff between the size of the radiocoverage area of a single transceiver and the feasible information rate.Infostations favor the latter, making use of mobility of users in redeeming theselective patchy coverage pattern. The “anytime, anywhere” motto of PCS isreplaced by “manytime, manywhere” access in Infostations. The secondchapter by Abbas Jamalipour of the University of Sydney describes the role

2 NEXT GENERATION WIRELESS NETWORKS

of satellites in broadband wireless access. In the third chapter, PatriciaMorreale of Stevens Institute of Technology describes the “Infocity”

in order to provide the envisaged high- speed ubiquitous access toinformation and communication.The second part of the book includes contributions that describe the state ofthe art wireless geolocation systems and trends. The first chapter is co-authored by scientists from Bell Labs of Lucent Technologies. Bob Richton,Giovanni Vanucci, and Steven Wilkus depict the widely accepted standardsolution to the wireless geolocation problem; i.e., “assisted GPS.” Thisconcept links wireless networks with GPS in order to reach the accuracy andavailability requirements for the user geolocation information. The wirelessnetwork assumes a supporting role in geolocation, in order to aid the enduser equipment through its prescribed communication with GPS. The secondcontribution, also authored by a Bell Labs scientist, Oguz Sunay, provides atutorial on all alternatives for wireless geolocation and details the evaluationprocedures that are currently under research by standards bodies and relevantwork groups. In the third contribution, Malcolm Macnaughton, Craig Scottand Chris Drane, researchers from the University of Technology, Sydney,present the chronicle of research efforts and empirical data collection on thegeolocation capability of the existing wireless infrastructure, during whichthey truly bring theory and practice together in efforts to sharpen thegeolocation capability of the wireless system.The third part presents contributions that demonstrate the use of locationinformation in next generation wireless network applications and services.The first contribution, by Mostafa Bassiouni and Wei Cui of the UniversityCentral Florida focuses on the use of real time geolocation measurements inimproving mobile connectivity and enhancing the effectiveness of faultrecovery in configurable cellular networks. The last contribution, byGuenther Popischil, Alexander Schneider, and Ernst Bonek of TechnischenUniversität Wien, portrays the creation of the “killer application,” for nextgeneration wireless networks.

I am proud to have put together this volume comprising of chapters bycontributors who are among the elite that are making the future happen. Ithank Dr. Oguz Sunay for his invaluable, tireless efforts in ensuring thetechnical flow and cohesiveness of the book. I would also like to thank AlexGreene, the Publisher of this book from Kluwer Academic Publishers, for hispatient, capable help. Finally, I’d like to express my gratitude to my brilliantresearch associate Mr. Amer Chatovich, whose careful, meticulouspursuance has made this project possible.

Sirin Tekinay

concept, which is based on the integration of wireless and wireline networks

Chapter 1

INFOSTATIONS: NEW PERSPECTIVES ONWIRELESS DATA NETWORKS

Ana Lúcia IaconoChristopher RoseWINLAB - Wireless Information Network Laboratory

Department of Electrical and Computer Engineering

Rutgers, The State University of New Jersey

Abstract We discuss file delivery issues for a new approach to inexpensive, highrate wireless data called Infostations. As opposed to ubiquitous cover-age, infostations offer geographically intermittent coverage at high speed(1Mbps to 1Gbps) since data, as compared to voice, can often toler-ate significant delay. The infostations paradigm flips the usual “slow-radio/fast-network” scenario upside down and offers intriguing new de-sign problems for wireless data networks. Collectively, we at WINLABbelieve that the infostations scenario, especially with the emergence ofthe World Wide Web as both a communications medium and defactostandard is one way to obtain low cost wireless data. And perhaps con-troversially, we offer arguments that currently proposed extensions tocellular systems (such as the coming Third Generation) will not be ableto offer data as inexpensively. In this chapter we describe the infosta-tions concept and then concentrate on issues above the physical layer.Specifically, we worry about delay bounds on information delivery forvariety of simple user mobility scenarios and infostation geometries. Wethen provide heuristic algorithms which closely approach these bounds.

Keywords: wireless communications, mobile computing, wireless data, wireless in-ternet, scheduling algorithms, delay bounds, mobility management


1. INTRODUCTIONOver the past 10 years, wireless voice communication has grown from

a rarity to a necessity. In contrast, wireless data services at rates andprice sufficient to generate equal excitement remain elusive. In response,the wireless industry has proposed third generation systems with ratesin the hundreds of kilobit per second range. However, the dominanttraffic on such systems will probably be voice at least initially, and herelies a “Catch-22” first observed by Roy Yates here at WINLAB [1].

Consider that the bit rate currently associated with voice communica-tions is on the order of 10 kbps and let us use this voice channel rate asour unit of measure. This channel costs v cents per minute. It thereforecosts approximately 13υ cents to transmit a one megabyte file – pro-hibitively expensive at current rates. In addition, what is particularlyinteresting is that this basic fact does not change with the introductionof higher rate services as long as voice is the dominant traffic. Onemegabyte of data always costs 13υ cents since the basic voice channelrate is unlikely to change drastically for both economic and legacy rea-sons. Thus, unless normal voice communications becomes essentiallyfree, it seems that wireless data will never be inexpensive when providedusing a cellular architecture.

This conundrum causes us to re-examine the cellular paradigm. Specif-ically, cellular wireless was built to carry voice traffic for people accus-tomed to the reliability and ubiquity of fixed telephone service. Thus,the goal of the cellular industry was coverage anytime and anywhere.However, to provide large coverage the system must be designed so thatusers both near and far from the access point (a base station) achievesome minimum quality of service. From a systems perspective however,it would be more efficient to serve users closer to the base station athigher rate, be done with them and then serve users farther away. How-ever, for voice systems the implication is intermittent coverage which isincompatible with continuous interactive traffic such as voice.

In contrast, data can tolerate delay and the system throughput couldbe increased by offering rates commensurate with achievable signal tointerference ratios. Add to this that customers are often in motion thebasic (somewhat surprising) infostations design precept emerges for sin-gle non-dispersive, non-directional channels:

Infostations should not be shared between users

That is, at any point in time, only one user should be attached to aninfostation. This basic idea has roots in information theory and water-filling of channels in space, time and frequency (see [2] for a developmenton multiple user dispersive channels). If we consider different frequency

Infostations: New Perspectives On Wireless Data Networks 5

or spatial sub-channels, then the precept still holds if each sub-channelis considered to be an infostation unto itself and users attempt to use the

might be co-located.A possibly non-obvious consequence of such spatio-temporal-frequency

water-filling results in another defining characteristic of the infostationsparadigm. For users in (ergodic) motion, the places at which transmis-sions should occur are where the channel quality is above some threshold– a result first shown by Joan Borras [3, 4] and based in part on work byAndrea Goldsmith [5] on fading channels. This implies that a user trav-eling with uniform velocity in an isotropic environment should transmitor receive only when it is close to an infostation, and from this the notionthat

Infostation coverage areas are spatially discontinuous

emerges naturally.Thus, we define infostations as a wireless communication system char-

acterized by sequential user access with discontinuous coverage areasand high data rate transmissions. As opposed to the moderate rateubiquitous coverage in cellular systems, infostations offer high speeddiscontinuous coverage which may be accessed by users in transientlyclose proximity to an infostation, and in fact can maximize system ca-pacity. Furthermore, the removal of the need to coordinate channelsamong multiple users and over the system as a whole should lead tosimple inexpensive realizations. And owing to the bursty nature of datacommunications and its tolerance of moderate delay, the infostation sce-nario with its inherently lower associated costs might be an attractivealternative to the classical concept of anytime anywhere communicationsnetworks.

1.1 EXAMPLESAlthough not specifically an infostation, consider a system introduced

by Apple, called Airport [6]: a base station that costs $299 and wirelessnetworking cards that cost $99 enable up to 10 computers to share a 11Mbits/second Internet connection at distances up to 150 feet. As PeterLewis describes in his article in The New York Times [7]:

“That is so important, and it has such potential to change the way weuse computers and information appliances around the house, that I’mcompelled to repeat it in a different way: I’m sitting outside the houseon the deck, with an iBook on my lap, enjoying a glorious autumn day,

reading the current e-news, checking e-mail … There are no wires,

“infostation(s)” with the best channel (s), even though these infostations


cables or extension cords in sight. As the stars come out, I simply strollback into the house and continue working from the sofa in the living

room.”

The salient feature of this narrative is a perceived desire for manytimemanywhere web access as opposed to the more traditional anytime any-where access we expect for voice services. Note in particular that Lewisdid not suggest he was using the computer during his journey from deckto sofa.

There are a variety of possible infostation system architectures. Forexample, many infostations may be owned by a single company and theymay be clustered and connected to cluster controllers according to theirlocation, creating a hierarchical architecture, as shown in Figure 1.1.This is somewhat analogous to the large telephone company cellularsystems where many base stations are connected to mobile telephoneswitching offices through dedicated high speed lines.

Another possible scenario might have small businesses such as conve-nience stores carry infostation service as a sideline – analogous to lottery


sales agents. This architecture is shown in Figure 1.2. To be econom-ically attractive, the start up cost to such a “Mom and Pop” operatorshould be low. There could also be a mixture of the two, as in a fran-chise setting where infostation operators leased the infrastructure fromthe founding company.

The network could also be isolated from the Internet and could beused for local communications, as in an office building or home. This isshown in Figure 1.3. Yet another architecture is to integrate infostationswith a ubiquitous, low data-rate system (e.g. CDPD or other [8]) anduse them as bandwidth boosters. Figure 1.4 shows one example of ahybrid infostation network. According to where infostations are placed,the user mobility can be characterized by three situations [9]: mobileusers moving with high speed, such as in a highway, characterize whatis called a “drive-through” scenario; users with medium speed, such asin a sidewalk or a mall, characterize a “walk-through” scenario; finallystationary users, such as in an airport lounge or a classroom, characterize

At WINLAB we have been studying several different problems relatedto an infostation network. A study of the infostation system performancein terms of capacity, throughput and delay was presented in [4] wherevarious models and different power allocation, symbol rate adaptation

“sit-through” scenario.


and modulation schemes are presented. A medium access scheme calledWINMAC has been proposed in [10] to support efficient packet commu-nications between an infostation and mobile terminals. This protocoladapts to the radio channel condition and achieves enhanced commu-nication reliability through packet retransmission and data rate adjust-ment. Due to the fact that one of the main services that infostation willprovide is Internet access, another area of interest is the design of a linklayer protocol to transmit IP packets efficiently via the wireless link. Anerror control scheme for the Radio Link Protocol is proposed in [11].The scheme uses multicopy and error threshold detection to improve thesystem performance. Infostation operation issues such as registration,authentication and billing are addressed in [12]. Some radio design is-sues are examined in [13]. There is also a variety of other work both atWINLAB and elsewhere ranging from physical layer issues up throughapplications [3, 12, 14, 15, 16, 4, 17, 18, 19, 1, 8].

1.2 USER MOBILITY AND INFOSTATIONSOne might wish to place an infostations system in an airport lounge,

in a conference room or in a small office at an affordable price. One


common characteristic of these situations is relatively low user mobility.Although the coverage area is small, the system is designed based on thefact that the user will stay in the coverage area during the time of theconnection and in fact, using beam steering techniques one might “movethe infostation” as opposed to moving the user. However, from theperspective of the fixed network, users are in relatively fixed locations.

However, when designing system where users roam, then user mobilitymust be considered since users may visit several infostations during asingle connection. For example, a user might roam over a shopping mallwith stores offering local infostation services and a user would not stayconnected to a single infostation while shopping. Likewise on a highwaywith infostations at regular intervals users might traverse great distances(from the fixed network perspective) between infostation contacts.

Now consider that data communication, such as messaging systemsor web applications, is inherently asymmetric with much greater volumeoccurring on the downlink from network to user. Under this scenario, ifthe information is available at the infostation, then the main issue is to


send it to the mobile as rapidly as possible. Since the data rate is high,as long as the user is in the coverage area, this can be done in a fewseconds. However, if the information is not available at the infostationand has to be transfered from a server, then the information has topass through the fixed network before reaching the mobile. Thus, inthe “drive-through” scenario, since the coverage area is small, the timewhich the mobile spends in coverage at a given infostation may not besufficient to transfer the information from the server to the infostation.

This is a situation peculiar to infostations where the radio rate isassumed much higher than the fixed network rate. Given an inexpensivehigh-speed radio, there are a number possible reasons for this inversionof the status quo. For economy, one could have a low cost relatively lowrate connection (i.e. commodity telephone modems) to each infostation.Alternately, even if one connects the infostations with high speed links,some types of services (i.e. HTTP request) have typical transmissionrates of the order of Kbits/second. The server transmission rate andnetwork congestion play an important role in determining the speedof the connection. Another scenario would have fixed network linksservicing some primary traffic with the infostation as an add-on servicesharing these links. Regardless, in all these cases although the radio rateis high, the user would have to restart a request at the next infostationin the path, and this process will increase the delivery delay, especiallyif only a small fraction of time is spent in coverage by any one user.

1.3 PROBLEM OVERVIEW, MOTIVATIONThe obvious solution to this radio/fixed-net mismatch is to cache or

prefetch information at the infostations. As an example, an intelligentprefetching algorithm which attempts to predict what the user will needwas proposed in [14] as a solution to a location dependent application(map request). The algorithm uses location and speed information to se-lect which of a set of maps should be prefetched. Based on location, timeor user dependency, different types of applications would need differentschemes for prefetching. However, suppose the information needed isknown and can be of any sort such as a web page, a map, or personale-mail. Then, the issue becomes how to partition the information, andthen when and where to send the packets over the fixed network so thatthey arrive at the user with minimal overall delay.

Thus, consider a system where the infostations are connected as acluster in a hierarchy where there is a higher level with a cluster con-troller, as shown in Figure 1.5. The cluster controller is the entity thathas information on all requests that were made in that cluster and how


many users are being served at every infostation in that cluster. Notethat a cluster would be a natural way of connecting different infostationsin the same geographical area, but the cluster controller does not haveto be necessarily in the same geographical area.

The cluster controller can coordinate the delivery of some packets tothe next infostation in the mobile path, so that they are locally availableat that infostation when the mobile user arrives in its coverage area. Ifthe path is not known then the cluster controller can send the packets toinfostations that are most likely to be in the mobile user path. Therefore,during the time the user is going between two infostations, the systemcan download the information to the next coverage area, reducing thedelivery delay, as shown in Figure 1.6.

The optimization problem is then, given some parameters and systemconfiguration, to deliver the information from its current location(s) tothe mobile user in a minimum amount of time. The important parame-ters that have to be taken into account are:

the overall amount of information that is requested, or file size;


the location of the file, which can be stored at the infostation, atthe cluster controller, at the home server (Internet) or distributedover a number of locations;

the data rate of the wired and wireless network;

the number of infostations at the cluster;

the infostations’ location;

the user mobility model.

To better understand the approach used here, consider the single usercase where there is a cluster with a given number of infostations, M.Let be the rate between the cluster controller and the Internet,the rate of each link between the cluster controller and the infostationsand the radio rate. Assume that the user requests a file, whichis then divided into packets, and each packet can be sent to differentinfostations.

Note that if the file is stored in “The Internet” then the networkwill be able to download the file to the user at the lowest link speed ofthe network. To take advantage of the fast radio, the cluster controllerwill prefetch file packets to infostations in the user path. Note that ifradio data rate were low, then every request should be re-initiated


at every infostation. That is, with a slow radio, there is little use inprefetching information to the infostations. Therefore we are interestedin the case where and . In this case the prefetchingapproach is helpful and only the radio rate will restrict the maximumamount of information that should be prefetched at a given infostationsince there is a limit on how much can be downloaded to the user in thecoverage area.

Given that the radio data rate is large, the specific delivery prob-lem ranges from the trivial to the difficult. Consider the case wheninformation is stored at some server on the Internet. Since we assume

, then the fixed network is the limiting factor. In the casewhen then the cluster controller can broadcast all the packetsreceived to all the infostations, as shown in Figure 1.7. All the infosta-tions will have the same information that the cluster controller has, asa copy network. As the user passes through the infostations the radiocan then download as many packets as possible to the user and discardpackets that were already received.


However if , then although the cluster controller cannot copyall the packets to all infostations, it can send different packets to differentinfostations, as shown in Figure 1.8. As the user passes through newinfostations, it can get new packets.

In the same way, if the requested file is locally stored at the clustercontroller, then again different file packets can be divided among all info-stations in the path, taking advantage of the parallelism of the network.The cluster controller has to decide which packets to put in which info-stations so that, when the user passes through the infostations, it obtainsthe most amount of information possible. The number of packets thatcan and should be prefetched is a function of the backbone rateand the radio rate . Note that if the Internet rate is very highthen the scenario is similar and we can assume that the file is locallystored at the cluster controller, as shown in Figure 1.9

In general, the cluster controller has a buffer where it queues all the filepackets. According to how large is, the size of the largest buffer (if thefile is locally stored at the cluster controller then the buffer contains allthe file packets). Thus, the cluster controller can coordinate the delivery


of these packets to different places and it can send copies and/or differentpackets to every infostation in the cluster. According to the radio rateand time spent in the coverage area, each infostation should only storesome maximum amount of information, since the user will not be able totransfer all the packets during its brief visit to the coverage area. Thismore general case is shown in Figure 1.10

Let us assume that there are N infostations in the user path beforecompletion of file delivery. Note that the value of N will be a functionof the file size, the delivery algorithm used, the user mobility model andthe rates and . In any case, we call this number N, althoughthe value could be different in each situation.

Assume that bits/second is the maximum data rate necessaryso that the radio is able to download all packets that are prefetched toa given infostation. Note that the value of M will be a function of thetotal file size, the delivery algorithm used and the user mobility model.

For a given value of , the problem space diagram is shown in Fig-ure 1.11. Region (1) is the case where and the cluster controller


should just broadcast every information received to all infostations. Re-gion (2) is where the radio is the bottleneck of the network and there isno reason for caching any information since it will not be delivered tothe user anyway. Therefore our study lies in region (3).

In region (3a) the rates and are as large as necessary to takethe most possible advantage of parallelism of the network for a given filesize. That implies that either the file is stored at the cluster controller or

. It also implies that the radio rate is very fast and all packetsprefetched to the infostations can be downloaded to the user during thetime it is in the coverage area.

If the rate is not as high then it is not possible to bring the thedesired amount of information to the cluster controller in order to spreadit over the many slow links. The cluster controller can then send differentpackets to some infostations and copy others in more than one place.Regions (3b) and (3d) represents this situation. In other words, thequeue at the cluster controller will have a small number of packets and


they can be copied to some infostations. In regions (3c) and (3d ) , theradio imposes a l imi t on the number of packets that should be prefetchedto the infostations since only some maximum number of packets can bedownloaded to the user during its passage through the coverage area.

Having exercised the various overall model parameters and identifieda number of scenarios – some trivial , some not, we can state the filedelivery problem for infostations simply. Let be a set of algorithmswhich transmit parcels of information to each infostation for delivery toa user. Let D be the del ivery delay seen by that user, defined as the


total time between the initiation of the request and delivery of the finalparcel. The optimization problem is then to

Find absolute lower bounds on delivery delay

Find algorithms which approach or meet these lower bounds

In this chapter, we will consider these problems for different mobilitymodels and infostation structures for single users. The multiple userscase is treated elsewhere [17] and will be the subject of near-future pub-lications .

2. TOURS IN ONE DIMENSION: THEHIGHWAY SCENARIO

Consider a one-dimensional model where users move along a line pop-ulated with equidistant infostations. Our objective is to derive lowerbounds on information delivery delay. In the first section we study thecase of constant velocity, where users move with a fixed velocity andfixed travel direction. In the second section we study the case whereusers travel with constant speed but random travel direction – a one-dimensional random walk. This constant velocity model, although sim-ple, covers important situations such as highway or railroad travel. Fur-thermore, it serves to illustrate some of the basic concepts of file deliveryunder the infostation model. In this section we assume that informationcan be delivered from the backbone to the clusters at or above the back-bone rate and that likewise, the radios are speedy enough that were anentire file available at the infostation, it could be downloaded during onepassage through the coverage area.

2.1 CONSTANT VELOCITYConsider a system with many infostations equally spaced at distance

d meters, as shown in Figure 1.12. Assume that the mobile travels atconstant velocity v m/s, the size of the coverage area is and that thewired backbone transmits at a rate of bits/sec.

The mobile will arrive in a given infostation and request a file of sizeF bits. If

then the mobile can receive only a part of the file at the first infosta-

the next infostation. Assuming a start/stop protocol where incrementalrequests must be initiated at each new infostation visited, the number

tion. Completion of the transaction must be deferred until it arrives at


of infostations, I, that the mobile has to pass to receive the whole file is

where represents the smallest integer greater than or equal to x.The time required for travel between two infostations is d/v and there-

fore the delay D in seconds, to transmit the file is

which we can rewrite as

Note that to decrease the delay one could decrease the distance be-tween infostations thereby increasing the infostations density and mov-ing toward a more ubiquitous coverage area scenario.

In our approach we assume that , which means that allinformation that is available at the infostation can be downloaded tothe mobile before it leaves the coverage area regardless of the amount.We also assume that , which means that the cluster controllerhas all the file pieces necessary to be able to prefetch any amount ofinformation necessary to the infostations. Note that if the informationis locally stored at the cluster controller, then the value of is irrelevantand the condition always holds.

In the case of constant velocity, given the initial position, the path isknown. Therefore, rather than initiating new transfers at each infosta-tion, the time the mobile is traveling between infostations may be usedto download part of the file to the other infostations along the mobiletour. Since the time spent traveling between two infostations is givenby , the system can download B (different) bits to each infostationin the tour, where


Likewise when the mobile leaves the second infostation the same amountcan be downloaded to the remaining infostations and so on. Therefore,to transmit the whole file we require

Thus, the smallest number of infostations, I, required for delivery ofthe file of size F is obtained by determining the smallest integer I suchthat

Therefore

and the delay , in seconds, is given by

which we rewrite as

Note that the delay now does not depend on the size of the coveragearea. In Figure 1.13 we can see the delay as a function of the file sizefor both situations: The delay without the prefetching approach, whichis given by equation 1.3 and the delay obtained when the prefetchingapproach is used, given by equation 1.9. We can see that there is alarge improvement when the prefetching approach is used. We also notethat the delay stays invariant for a larger range of file sizes when theprefetching approach is used.

Another interesting fact is that the velocity, υ, affects the delay. Equa-tions 1.3 and 1.9 show that the delay decreases as the velocity increases.Thus, it behooves the mobile to move rapidly, passing through a largenumber of infostations. As can be seen Figure 1.14, equation 1.9 ismore sensitive to the velocity. In this case, the network-to-infostation


bottleneck is essentially removed by spreading the communication overmany slow links. The ripples in the curve are due to the fact that theceil function remains constant for a range of values of υ, while thedenominator increases with υ

Certainly there is a limit to the improvement. It should first be notedthat we assumed that there were as many infostations as needed in theuser path. This may not be the case since the number of different infosta-

usually transmits packets and frames, there may be some minimum num-ber of bits, , that can be delivered during a transmission. Thus,deliveries could only be made by every infostation along the tour if v issuch that . However, even for infostation placement asclose as one city block (0.05 mile), a typical bytes, and aline transmission rate of 56Kbits/s, υ would have to exceed 2.7 miles persecond. For comparison, jetliner velocities are typically on the order of0.1 miles per second.

tions along a given tour may be limited. Furthermore, since the network


2.2 THE RANDOM WALKNow consider the scenario where the mobile moves with constant ve-

locity, but at each step the direction is random – it can go to the right orto the left with probabilities and , respectively. Note that inthis case the path is not known a priori. We would like to have boundsfor file delivery delay for the random walk scenario. Let a step be amotion between two infostations and, at every step, the mobile eithergoes to the left or to the right, it does not stay at the same infostation.

2.2.1 Delay Bounds. Assume that we have an optimum algo-rithm, in the sense that it minimizes the delay. That is, the algorithmsends file parts to each infostation as if it knew the path. Assume theinitial position is position 0. If the file size is F bits then it can bedivided into N segments, such that


If the optimum algorithm is used, then the maximum number of info-stations needed to the left and right of the mobile so that the whole filecan be downloaded is given by

where represents a boundary where the mobile stays until filedelivery completion. In other words, while the transaction is not com-pleted, the mobile will be restricted to the region , asshown in Figure 1.15. If the mobile actually goes on a straight line, the

delay is the minimum possible given by

If the mobile keeps hopping between two infostations then the maximumnumber of parts it can get is 1 in the first step and then 2 in all nextsteps. That will create a situation where the mobile passes through themaximum number of infostations, before file delivery completion.To find will be the smallest integer such that

Therefore, the maximum delay, in seconds, is given by

and thus,

Thus, equation 1.16 gives an upper an lower bound for the delay, inseconds, for file delivery completion using the prefetching approach.


2.2.2 Average Number of Segments. Equation 1.16 providesupper and lower bounds for the delay but it does not provide averagedelay for a given motion process. Therefore we will calculate the averagenumber of segments that can be downloaded to an infostation if anoptimum algorithm that minimizes the delay is used. From this wecan infer the approximate average number of steps, and thereby delay,necessary to deliver files of various sizes.

Assume that the mobile passes through position i for the first timeat time t and once again at time . Assuming the download processstarts at time 0, the maximum number of new file segments that can bedownloaded to that infostation at time t is t file segments. Furthermore,the maximum number of new file segments that can be delivered atposition i at time is n. Thus, it can be seen that for any optimalalgorithm, the cumulative number of file segments obtained by a visit tolocation i at time t is exactly t regardless of the path taken to locationi.

This result suggests that delay depends on the number of infostationsthat are revisited in a path, that is, if fewer infostations are revisitedthen more new file segments are obtained at each step and the timenecessary to completely transmit the file is reduced.

Let us assume that the path is limited to s steps. Let be theprobability that location x was last visited at time given that thepath is limited to s steps. If a location was visited at some time t, thenthe maximum number of segments that could have been downloaded,cumulatively, is t. If an optimum algorithm is used, then the number ofsegments downloaded would be exactly t. Given that, for an optimumalgorithm, the mean number of file segments, , picked up by step sis then

To calculate the value of it is necessary to calculate . Wewill do that using first passage times.

2.2.3 First Passage Times. Our goal is to calculate ,the probability that location x was last visited at time given thatthe path is limited to s steps. If one looks at the motion process in thereverse direction, then is simply the probability that, startingat some position n, the first passage through x is at time . Observethat for that to be possible it is necessary that


Without loss of generality, assuming , the number of paths fromthe origin that pass through position x for the first time at time ,

, is given by [20]

gives the number of possible paths. Each of these pathshas a probability which is a function of p and q. The sum of all theseprobabilities gives the probability that the first passage time throughposition x is given at time s – t.

In order to calculate this probability we divide the problem in twocases: the positions to the left and right of the mobile. We will assumethat the mobile at position 0. We first choose a position to the rightof the mobile, say position r. We want to calculate the probability ofa path that starts in 0 and passes through r for the first time insteps. We know that there are paths with this property. Butthe probability of each one of them is exactly the same. This statementwill become more clear with the discussion below.

The probability of a path that starts in 0 and passes through r for thefirst time in steps will be a function of the number of steps takento the right and to the left. Assuming all steps are independent of eachother, if the mobile take steps to the right and steps to the left,the probability, , of this path is given by

In order to satisfy the first passage time condition

In order to arrive to position r, we need that

which implies that

for all the paths. The number of steps to the right, , isgiven by


and then we conclude that the number of steps to the left and the numberof steps to the right is the same for all paths that start in 0and end in r after steps.

It is important to note that and must be integers and thereforemust be even or, equivalently, must be even.

The probability of a given path that starts in 0 and passes throughposition r for the first time in steps is then given by

And there are

of these paths.A similar analysis for a given position r to the left of the mobile gives

that the probability, , of a path that starts in 0 and passes throughposition for the first time in steps is given by

The total number of these path is also given by

Note that this case also requires to be even.Finally, the total average number of file segments picked by the mobile

after s steps, is given by

where

where the first summation represents the positions to the left of themobile and the second summation represents the positions to the rightof the mobile.


2.2.4 Special Cases. Two special cases of interest are the sym-metric case, where , and the straight line case, where

Assuming the symmetric case, the probability of each path is given. Then

and

where

For the case where we have the situation described inSection 2.1 (constant velocity), where the mobile comes from the left tothe right in a straight line. In this situation the second summation willbe zero, and the first summation is only non zero for , sincethere is only one possible path. Therefore we have

which is similar to the result in Section 2.1. In Equation 1.6 the variableT represents the number of infostations required for delivery of F bitsand plays the role of the variable s in Equation 1.33. The differencebetween Equations 1.6 and 1.33 is the factor used to transformfrom number of segments to number of bits.

Figure 1.16 shows the bounds on average number of pieces pickedafter a given number of steps, , as a function of s. We start with

and we increase the value of p. Because of the symmetryof the problem, the results increasing the value of q are similar, or, inother words. p and q can be exchanged.

As can be seen, the results for the straight line givehighest bound. As the probability p increases the bound approaches thelower bound, which is achieved with the symmetric caseThis suggests that the average number of pieces is closely related tothe number of revisits to a given place. In the straight line case the


mobile never revisits a given position, and therefore at every step a highernumber of pieces can be downloaded. As we mentioned in Section 2.2.1,the minimum delay is achieved when the mobile moves in a straight lineand the maximum delay is achieved when the mobile hops between twoinfostations. In the next section we will discuss this topic further.

3. EXTENDING TO HIGHER DIMENSIONS:THE GRID, THE CUBE, ETC.

Results for first passage times helped us to derive the bounds for theone-dimensional case. Since we were unable to find first passage timeresults for higher dimensions are not available in the literature, theseresults are provided here so that we may derive bounds on the averagenumber of pieces delivered. We assume the same radio and links ratescenario as in the single dimensional case.


3.1 THE TWO-DIMENSIONAL PROBLEMWe first extend the one-dimensional case to two-dimensions. Let us

consider that the infostations are equally spaced in a rectangular gridas shown in Figure 1.17. The mobile can go right, left, up or downwith probabilities and , respectively. At every step themobile chooses one direction and does not stay at the same position.The velocity is still assumed constant for internode moves.

We want to calculate the average number of file segments that can bepicked up by step s, . Let be the probability that location(x,y) was last visited at time given that the path is limited to ssteps. Similar to the one-dimensional case, we have

Observing the motion process backwards, as we did in the last section,then is simply the probability that, starting at some position

, the first passage through ( x , y ) is at time . Observe thatfor that to be possible it is necessary that

We need to calculate the first passage time through some position( x , y ) . Since we were unable to find this result in the literature we willderive it here, starting with the multinomial distribution. Assume x andy positive. Without loss of generality, assume .. We will firstcalculate the total number of paths that start in (0, 0) and pass through


(x,y) in s steps, . Note that a necessary condition is that

Let be the number of steps taken upwards, the number of stepsdownwards, the number of steps taken to the left and the numberof steps taken to the right. The total number of paths that start in theorigin and pass through (x, y) in step s is given by

where

In order that the mobile arrives at position y, it is necessary that

And since it will also have to get to position x ,

and

The number of steps upwards, needs to satisfy

Because the number of steps has to be an integer, we also need that(s + x + y) mod 2 = 0. Therefore, for and , we have

Note that the same holds for x or y non positive, depending only on theabsolute value of x and y. Therefore the number of paths with s steps,from the origin to (x, y) is given by


gives the total number of paths from the origin to someposition ( x , y ) in s steps. That includes paths that pass through ( x , y )at some time also. In order to obtain the number of paths thathave the first passage through (x, y) at step s, it is necessary to removethe paths that pass through (x, y) at s and before s also.

Let be the number of paths that start at some position andreturn to that same position for the first time in i steps. Note that imust be even. Then, the total number of paths starting at the origin thatpass for the first time through position (x, y) in s steps, , isgiven by

where we eliminated the paths that also pass through that position be-fore time s. These are the paths that pass through that position for thefirst time at time , and then return to that position in i steps, forall i even, and

To calculate we start with the total paths from some position toitself in i steps, . To calculate the total number of paths thatstart at some position and return to the same position for the first timein i steps it is necessary to eliminate the paths that also pass throughthat position before time i. Similarly to before, these are the paths thatpass through that position for the first time at time , and thenreturn to that position in p steps, for all p even, and

And we know that

therefore the value of can be calculated, using recursion, for any, i even.


The probability of the paths that start at the origin and pass throughposition ( x , y ) at time s, , will be a function of x, y and sand is given by

where

But calculating the probability of the paths that start at the originand pass for the first time at position ( x , y ) at time s is not trivial.For that reason we will consider here the symmetric case where

. Observing the motion process backwards, aswe did in the last section for the one-dimensional case, and consideringthe symmetric case, one can derive the average number of file segmentspicked up by step s is then given by

The condition is included because we assume that the useris at the origin at step s and therefore it got cumulatively s segments,which are added separately in equation 1.48.

In Figure 1.18 we present the average number of segments picked afters steps, for the symmetric two-dimensional case. We also present

for the one-dimensional case, for different values of p and q (pleaserefer to Section 2). One could expect that the curve for two-dimensionalcase would match with the case of . That is not thecase because in the one dimensional case the probability of revisits toa given position when is higher than for the two-dimensional case. As we can see from the figure, the value that mostapproaches the 2-d case is

3.2 THE N-DIMENSIONAL PROBLEMExtending the two-dimensional to the n-dimensional problem is straight-

forward. Of course, one might ask “why bother?” First, one could easily


imagine 3-dimensional infostations models. However, although the info-stations problem was couched in terms of moving matter (people, vehi-cles, etc.) and radios, a possibly quixotic generalization might includemigrant programs (mobile agents) operating over a computer networkin an almost arbitrary dimensional data space and other fixed programswhich need to pass large data files to the agents as they move over thenetwork. In addition, practical utility aside, the same machinery toconsider three dimensions allows consideration of N dimensions. Thus,since the incremental effort necessary for generalization is minimal, it istherefore provided here.

Equation 1.44 is still valid for the n-dimensional case, where nowis the number of paths with s steps from the ori-

gin to position . This number is easily calculated usingthe multinomial distribution as in Equation 1.42, but now with n – 1summations, as


and

But we know from condition 1.50 that

which implies that

We also know that the maximum value of is obtained whenand and therefore from condition 1.50 the maximum value of

is obtained when

which implies that

and similarly, the maximum value of is obtained when

which implies that

For the case where n = 3, for example, we would have


and finally we can write

where

and

Returning to the general case, the number of paths that start atthe origin and pass through for the first time after s steps,

, is given by

where

and

In general, for the symmetric n-dimensional case, we can write theaverage number of file segments picked at step s as


Figure 1.19 shows the average number of file pieces picked in s stepsfor one-, two- and three-dimensional problems, for the symmetric cases.

increases as n increases since the number of infostations revisited ina path decreases. Also shown in Figure 1.19 is the case where the mobilenever revisits an infostation. Figure 1.20 shows the average number oftimes the mobile revisits a given position in a motion process along apath of length s. This is simply the number of paths that start in theorigin and return to the origin after s steps times the probabilities thesepaths. For the two-dimensional symmetric case, for example, the averagenumber of revisits in s steps is given by . As can be seenfrom the figure, as the number of dimensions increases the number ofrevisits decreases.

4. A NEAR-OPTIMUM ALGORITHMIn the last two sections we presented bounds on file delivery for a gen-

eral n-dimensional model. Those results give bounds on the maximumnumber of file segments that can be downloaded for the user in the casewhere and and were derived assuming that an op-


timum algorithm is used. But it does not tell us how such an algorithmwould work. Therefore, in this section we will provide an algorithm forthe one-dimensional case. Given that the file can be divided in manydifferent smaller segments, and we label each segment, the role of thealgorithm is to decide after every step which segments should be sentto which infostations. As always, the goal is to minimize the overall filetransfer delay.

4.1 OVERVIEWWe will concentrate in the one-dimensional scenario, as shown in Fig-

ure 1.21, where infostations are equally spaced at distance d.


Assume a file of size F, and this file is divided in N segments of sizeB, where

and

Given that the mobile is at some position x, an algorithm that de-livers file segments to infostations around that user will have to deliversegments to a given number of infostation to the right and left of themobile. Therefore our algorithm will work in a range of Infostations.The first important characteristic of the algorithm is the calculation ofthe boundaries in the range. These boundaries are the maximum num-ber of infostations necessary to the left and right of the mobile so thatthe mobile is able to receive the whole file. Note that they should berecalculated at every step of the algorithm.

In order to maximize the number of file segments received at each step,the algorithm schedules different segments to each infostation, so thatthere are no repetitions and no segments wasted. After all the segmentsare spread over the range of infostations, then one can no longer avoidrepetitions. From that point on the algorithm avoids sending repeatedsegments to positions that are in a possible path for the mobile. Detailsof the algorithm are described in the following sections.

4.2 THE RANGE OF INFOSTATIONSAs seen in Section 2.2, as long as the mobile picks the maximum num-

ber of file segments at every step, the maximum number of infostationsneeded to the left and right of the starting position is given bywhere

If the mobile is at position i then an optimum algorithm does not haveto consider infostations that are outside the range

After the mobile moves to the left or right new boundaries must becalculated. To do so the history of the movements is considered. Allsegments that were already obtained by the mobile and all the segmentsalready delivered to infostations are considered. In addition, undeliverednew segments still to be scheduled must also be taken into consideration.The number of new segments that will be scheduled is calculated usingthe assumption that an optimum algorithm will be used. Note that the


boundaries will be given by one infostation to the left of the mobile andone infostation to he right of the mobile. We define

infostation at which the mobile user will receive the last filesegment if it moves from infostation i to in a straight line,taking only steps to the right. We assume that, when this path istaken, the mobile receives all the segments already delivered to theinfostations in the path plus new segments that will be deliveredusing an optimum algorithm (new segments would not be copiesof segments delivered before).

infostation at which the mobile user will receive the last file seg-ment if it moves from infostation i to in a straight line, takingonly steps to the left. We assume that, if this path is take, the mo-bile receives all the segments already delivered to the infostationsin the path plus new segments that will be delivered using an op-timum algorithm (new segments would not be copies of segmentsdelivered before).

For example, consider the situation where the file can be divided in 10segments, . In this case the maximum number of infostationsneeded to each side of the mobile is

Assume that the mobile is at position 5. Then the range of infostationsis given by [1,9], as shown in Figure 1.22.

position: 1 2 3 4 5 6 7 8 9M

Figure 1.22 Example for a file with 10 segments: calculating boundaries, and

After the boundaries are calculated, then the algorithm must decidewhich segments to prefetch in which infostations during the first step.The algorithm will deliver different segments to every infostation in therange, as shown in Figure 1.23

After the mobile moves, say to position 4, one segment is pickedup, as shown in Figure 1.24.


It is now necessary to calculate the new boundaries. The left boundaryis still the same, and the right boundary is calculated using the followingfacts:

Mobile has already received . There are nine segments missingto complete the file.

If the mobile goes in a straight line to the right it will pick allthe segments that are scheduled plus new segments that will bescheduled. If we assume that the next segments will be scheduledusing an optimum algorithm, then the mobile would pick two seg-ments in position 5, three segments in position 6 and finally foursegments in position 7, which gives a total of nine segments andtherefore the mobile would finish the transmission at position 7(please see Figure 1.25.


Therefore the new boundaries are and and positions 8and 9 do not have to be considered anymore in our scheduling process.

Now the algorithm has to schedule segments to be delivered to therange [1,7]. Note that the only segment delivered to the mobile is

and therefore should not be scheduled. Note also that segmentsand are already delivered to the infostations in the

range. Therefore segments and should be scheduled. Theother five segments must be a copy of segments already delivered to theinfostations. One way of scheduling segments is shown in Figure 1.26

Now suppose the mobile moves to position 5 and picked segmentsand . The boundaries must be re-calculated.

The mobile has already received 3 segments . To the leftthe mobile can receive two segments at infostation 4, four segmentsat infostation 3 and five segments at infostation 2, a total of elevensegments (please see Figure 1.27) Since only seven segments are missing,the left boundary is . To the right, since segments andwere already delivered to the user, they are wasted at infostation 6.Therefore at infostation 6 the mobile can receive only one new segment,four segments at infostation 7 and four segments at infostation 8, a totalof nine segments (please see Figure 1.28. Thus the right boundary is

Note that the scheduling choice was not very smart and some segmentswere wasted. Choosing where to copy which segments is not an easy task,and it is the objective of the algorithm. We will discuss the schedulingof segments in the next sections.

Note that if the algorithm is optimum then the new boundaries maydecrease, but never increase.

We then define, at any given step:


N: total number of segments in the file.

i: mobile user position.

total number of segments already delivered to the mobile user.

set of segments that were scheduled to infostation j but notdelivered to the mobile user.

set with segments that were scheduled to infostationsin the range [i, j]. Thus

And with these definitions we can write and as follows.

The right boundary is the smallest integer such that:


The left boundary is the largest integer such that:

4.3 PARTIAL PATHS

To maximize the number of file segments received at each step, thealgorithm should schedule different segments to each infostation, so thatthere are no repetitions and no segments wasted. After all the segmentsare spread over the range of infostations, then one can no longeravoid repetitions. The problem is then to decide which copies of seg-ments should be sent to each infostation.

It is important to note that although we can calculate the boundaries,that does not imply that the mobile will pass through all the infostationsin that range. For example, if the mobile goes in a straight line to theleft, it will never visit infostations to the right of its initial position. Thisfact is very important and will be used in the algorithm.

To be able to describe the algorithm we will define:

maximum number of file segments that can be delivered tothe mobile if it takes a given path P.

Partial Path from infostation i through infostation j: a path thatstarts at infostation i, passes through infostation j, and the mo-bile has the potential to receive at most the total number of filesegments still needed. Thus a path P is a partial path if and onlyif

set of all infostations that belong to all partial paths fromi through j.

set with all segments scheduled to infostations that belongto

To better understand the idea of partial paths, we consider the exam-ple given before. The mobile started at position 5, moved to position 4and received one segment , as shown in Figure 1.29.


Let us obtain the partial paths from infostation 4 through infostation5. The paths and

are all partial paths from 4 through 5. In path A the mobilecan receive at most two segments at infostation ; in path B themobile can receive two segments at infostation 5 and three segments atinfostation 6, a total of five segments and in path C the mobilecan receive two segments at infostation 5, three segments at infostation6 and four segments at infostation 7, a total of nine segments

Note that the path is also a partial path from 4 through5, but it includes the same infostations.

The paths and arealso a partial paths from infostation 4 through infostation 5.

In path D (please refer to Figure 1.30) the maximum number of seg-ments that can be delivered is segments (at infostation 3)segments (at infostation 4) segments (at infostation 5)

In path E (please refer to Figure 1.31) the maximum number of seg-ments that can be delivered is segments (at infostation 5)segments (at infostation 4) segments (at infostation 3)


We then know that infostations [3,7] belong to partial paths from 4through 5, or belong to . In order to find if infostation 2 belongsto we observe the paths (Figure 1.32)and (Figure 1.33). Sincesegments (at infostation 5) segments (at infostation 4) segments(at infostation 3) segments (at infostation 2) and segments(at infostation 3) segments (at infostation 2) segments (at info-station 3) segments (at infostation 4) segments (at infostation5) then

and

and thus G and F are not partial paths from 4 through 5. Thereforeinfostations 1 and 2 do not belong to partial paths from infostation 4through 5, and

It is important to note that after the boundaries are calculated, allthe infostations to the left of the mobile, except from , always belongto a partial path between the mobile and any infostation to the left ofthe mobile. The infostation may or may not belong to a partial path.The same argument applies to infostations to the right of the mobile.

For example, consider the case where the file divided in a total ofsegments. Assume the mobile is at position 3. The boundaries


in this case are and . but infostation 5 does not belong toa partial path from 3 through 4, since

4.3.1 Calculating . As seen in the last section, in order toverify if a path P is a partial path one need to find . Therefore itis important to have a general equation for

Consider the scheme shown in Figure 1.34.


Consider the path A as

can be written as (please see Figure 1.35):

It is interesting to observe the example in Figure 1.35 where we canclearly see that at step s the mobile can pick cumulatively at most ssegments.

Consider now a path B given by

Similarly

And therefore, in general, for a path where xand y are at opposite sides of i and x is visited before y, will be afunction of (i, x, y) and can be written as


where if then and if then and x is visited before

y.For example let us consider again the example given in Figure 1.29

and the paths and. Then, for path

. From Equation 1.80

For pathfrom equation 1.78

and the results match with the one presented in the last section, givenby Equations 1.74 and 1.75.

4.3.2 Infostations in Partial Paths. As we mentioned before,after the boundaries are calculated, all the infostations to the left ofthe mobile, except from , always belong to a partial path between themobile and any infostation to the left of the mobile. The infostationmay or may not belong to a partial path. The same argument appliesto infostations to the right of the mobile.

In order to guarantee the inclusion of and we define

Our goal now is to obtain the set or . We want to findall the infostations that belong to all partial paths from i through j. Aswe saw in the example in the last section, it is not necessary to obtainall partial paths from i through j in order to obtain

Consider again the scheme shown in Figure 1.34. The mobile is atposition i and one wants to find , where , or j is an infostationto the right of the mobile. We know that he set is constituted of all


infostations in , where k is the left-most infostation that belongs toa partial path. In order to check if an infostation k belongs to a partialpath one could check if the paths

and

are partial paths. Therefore it is necessary to verify, for each path, if thenumber of segments that the mobile can receive if the path is taken isless than or equal to the total number of file segments still needed. Weneed to verify if

and

Note that both paths contain the same infostations, but they may beof different sizes, or different number of steps. Define

the number of steps in a given path P.

But we know that

and therefore, it is necessary and sufficient to check if the shortest paththat starts at i and passes through j and k is a partial path.

So we conclude that, in general, to check if an infostation k is in apartial path from i through j, where i is in the middle of k and j

if , to check if an infostation k is in a partial path from ithrough j it is necessary and sufficient to check

if then to check if an infostation k is in a partial path fromi through i it is necessary and sufficient to check if

Therefore:

if then


if min then

* if thenbelongs to a partial path from i through j.

if min then

* if thenbelongs to a partial path from i through j.

if then

if then

* k belongs to a partial path from i through i.

4.3.3 Obtaining . As we saw in the last section, we canwrite

or,

if then , where is theleft-most infostation that belongs to a partial path from i through

if then , where is the right-most infostation that belongs to a partial path from i through j.


if then , where is the right-most infostation that belongs to a partial path from i through iand is the left-most infostation that belongs to a partial pathfrom i through i.

In other words, it is necessary to find the first infostation to the right(or left) to the mobile that belongs to a partial path from i through j.

4.3.4 The set v(i, j). Once the set is obtained, obtainingv(i,j) is straightforward. If the importance of this set in the algorithmis not clear yet, it should be after this section.

Assume the mobile is at position i and all the segments that were notdelivered to the mobile are already scheduled to the infostations. Thealgorithm has to decide what to schedule to a given infostation j.

Once we calculate the set of infostations that belong to partial paths,, the best solution is to schedule a segment that is not scheduled

to any infostation in the set , or a segment that do not belong toIf a segment that belongs to is scheduled to an infostation

that belongs to then there is a probability (different than zero)that this segment will be wasted if the mobile takes one of the partialpaths. On the other hand, if the segment does not belong to thenthis segment could be scheduled to any infostation in .

Assume . As we mention before the set may or may notinclude the the last infostation in the range, . But we also do not wantto copy a segment that is scheduled to in j since there is a probability(different than zero) that this segment will be wasted. The same appliesto infostation , if . For this reason we will work with the set

and v(i, j).In order to clarify the importance of these sets, please refer to the

example shown in Figure 1.24. The file is divided in 10 segments, themobile is at position 4 and picked one segment . The algorithm hasto schedule new segments for the next step. Since segments and

are not scheduled, they can be scheduled to positions 2, 3 and 4, forexample, as shown in Figure 1.39.


In order to decide what to schedule at the other infostations it isnecessary to consider each infostation separately. Let us consider info-station 5. Starting at the user position (infostation 4) there are severalpartial paths from 4 through 5. The set

includes all the infostations in all partial paths. This set creates a box,as shown in Figure 1.40.

All segments that are already scheduled to infostations in the box, orwere delivered to infostations in the box in previous steps, should notbe scheduled to infostation 5. The set with these segments is given by

Thus, we could schedule segment to infostation 5, as shown inFigure 1.41.

It is clear from the figure that the segment would never be deliveredtwice to the mobile, since the file delivery would end before the mobilepasses through both infostations, 5 and 1, no matter which path is taken.

The configuration shown in Figure 1.42 would also be possible, wheresegment is copied in two infostations. Note that in this case segment


could be delivered twice if path is taken.But note that this would not increase the delivery delay in terms ofnumber of infostations, since at infostation 2 only one segment need tobe delivered (assuming that infostations 3, 4 and 5 always deliver newsegments).

4.4 THE ALGORITHMLet N be the total number of file segments. Let i be the infostation

where the mobile is located at a given step. At every step the algorithmwill do:

1. Calculate boundaries and2. Obtain3. For each infostation do:

Find and v(i,j)

If then

– schedule a segment to infostation j;

– add the segment to

Else

– If then


* schedule a segment to infostation j;

* add the segment to

– If then

* schedule a segment and:

• If then schedule from the range start-ing from the set

• If then schedule from the range start-ing from the set

• If then schedule from the set or

• add the segment to

The algorithm loses its optimality if, at a given step,and, for some j, At this point the choice of what to send toj is heuristic. Using the idea that the boundaries may decrease, we bringsomething from the boundaries since that infostation may be removedfrom the range considered. This choice is not proven to improve theperformance, but it is a good heuristic choice. Other methods could beused.

4.5 AN EXAMPLE

In this section we will present an example of the algorithm applicationto a file that is divided in 10 segments. The number of infostations thathave to be considered is nine (four on each side of the mobile). Weassume that the mobile starts at position 5, and the partsare scheduled at every step. Below is the output of the algorithm:

step 0: mobile starts at position 5:

step 0:mobile: Mposition: 1 2 3 4 5 6 7 8 9

step 1: mobile goes to position 4:new boundaries are

and are repeated out of partial paths:


step 1:step 0:position: 1 2 3 4 5 6 7 8 9mobile: M

step 2: mobile goes to position 3:new boundaries are ;nothing is scheduled to position 6 since all parts belong to v(3, 6):

step 2:step 1:step 0:position: 1 2 3 4 5 6 7 8 9mobile: M

step 3: mobile goes to position 4:new boundaries are ;nothing is scheduled to position 6 since all parts belong tov(4,6);

contains all the segments and therefore is broughtfrom position 6:

step 3:step 2:step 1:step 0:position: 1 2 3 4 5 6 7 8 9mobile: M

step 4: mobile goes to position 3:new boundaries are ;nothing is scheduled to position 2 since all parts belong tov(3,2);nothing is scheduled to position 5 since all parts belong tov(3,5);

contains all the segments and therefore is broughtfrom position 5;

is repeated in position 3 since it is not in v (3,3):


step 4:step 3:step 2:step 1:step 0:position: 1 2 3 4 5 6 7 8 9mobile: M

step 5: mobile goes to position 2;file transfer is complete.

4.6 PERFORMANCE OF THE ALGORITHM

The average number of file segments picked up by step s, , wasobtained through simulations. We performed 900 trials for each value ofs, meaning that we have 900 independent random walks for each value ofs. Assuming a large enough number of file segments that it is impossiblefor them all to be picked up by step s, we take s steps and count thetotal number of segments that are picked up by the mobile. This resultis compared with the theoretical result given by Equation 1.31 and asimulation assuming an optimum algorithm. The results are shown inFigure 1.43 with confidence intervals about each point smaller than thesymbol size. As can be seen, the three curves are hard to differentiate.

Because the number of segments picked even with the optimal algo-rithm depends on the path chosen, there is considerable variance in thenumber of segments picked up at any given step. Shown in Figure 1.44is the corresponding standard deviation for our algorithm and for theoptimum which again suggests the near-optimality of our algorithm.

In Figure 1.45 we provide the complementary CDF of the differencebetween the delay for an optimal algorithm applied to a file of size 200pieces and the delay associated with our algorithm for the same randommobile user path. We did 1000 trials, and it can be seen that the heuristicalgorithm fails by more than 3% only 16% of the time and by more than10% only 0.4% of the time.

The excellent performance of the algorithm is in part due to the factthat we are transmitting in all infostations that will possibly be in thepath. In the single user case this is not a problem since all the linkscan be used for one user. If we increase the number of users it may notbe possible to coordinate all transmissions to all users if the boundariesoverlap.

Therefore, to evaluate the algorithm efficiency we change the numberof infostations to where the system transmits to. We consider the case


where the system transmits to C infostations to the right and to the leftof the mobile ( total number of infostations).

Figure 1.46 shows the total number of file pieces transmitted beforecompletion of file delivery, as a function of the file size, for different valuesof C. The total transmitted minus the file size represents the wastageof the system. Figure 1.47 shows the average delivery delay. As can beseen, reducing the value of C to 3, for example, increases considerably theefficiency and does not affect as much the delay performance. That showsthat it may be possible to accommodate more users without affectingvery much the system performance.

5. CONCLUSIONS AND WORK INPROGRESS

In this work we considered the problem of delivering a file in systemwhich features high rate discontinuous coverage. The collection of ac-cess points and the algorithms which support file delivery we call an


Infostation System. Assuming that there are several infostations in themobile path, the file is divided into segments and different segments canbe transmitted to different infostations along the path.

The constant velocity case was studied and the most interesting resultis that higher mobile velocity reduces delay if different data can be deliv-ered to multiple infostations in parallel over the fixed network. Then arandom walk mobility model was introduced with constant velocity butwith randomly chosen travel direction at each step. Results for boundson the average number of file segments picked after a given number ofsteps for a general infostations topology were obtained. It was shownthat the fewer infostations are revisited in a path, the larger the averagenumber of segments obtained at each step.

An algorithm for the one-dimensional case was proposed. The algo-rithm simply tries to avoid repetitions of segments in places where themobile is likely to visit along a path. The algorithm is not optimum inthe sense that it does not achieve the absolute bound associated withforeknowledge of the user path, and it fails when it cannot avoid the


repetitions. However, it is an open question whether this algorithm isindeed optimal among all algorithms which do not know the user pathbeforehand. Regardless, simulation results showed that its delay perfor-mance is extremely close to the known-path optimum.

Currently the authors are working on the scheduling problem for themultiple user case. This problem is much more complex than the singleuser case since each infostation has to decide not only which segmentsto transmit but also which user to serve.

Some other suggestions for future work are:

We have assumed a constant velocity between infostations. Itwould be interesting to consider the situation where the time spenttraveling between two infostations is a random variable. Yet an-other important factor is that in our analysis we assumed a gridscenario. When considering discontinuous coverage area, though,the system can be modeled as a complete graph, where every noderepresents the infostations and the weight of every edge is the tran-sition probability between the two infostations.


We assumed that the bottleneck of the system was the wired back-bone and that the radio was always capable of transmitting all thesegments to the mobile during the time the mobile is in the cover-age area. It is necessary to consider imperfections over the wirelesschannel, such as errors and retransmissions, which may cause somefile segments to be probabilistically missed at any given infostation.



References

[1] R.D. Yates and N.B. Mandayam. Issues in wireless data. In IEEE SignalProcessing Magazine, May 2000. To appear.

[2] R.S. Cheng and S. Verdu. Gaussian Multicast Channels with ISI:Capacity Region and Multiuser Water-Filling. IEEE Transactions onInformation Theory, 39(3), May 1993.

[3] D.J. Goodman, J. Borras, N.B. Mandayam, and R.Y. Yates.INFOSTATIONS : A New System Model for Data and MessagingServices. In Proceedings of IEEE VTC'97, volume 2, pages 969-973,May 1997, Phoenix, AZ.

[4] J. Borras. Capacity of an Infostation System. PhD thesis, RuthersUniversity, January 2000.

[5] A. Goldsmith and P. Varaiya. Capacity of fading channels with channelside information. IEEE Trans. Inform. Theory, pages 1218-1230, Oct1997.

[6] Apple, 1999. Apple Computer Inc., URL=http://www.apple.com/airport

[7] P.H. Lewis. Not born to be wired. The New York Times, CircuitsSection, November 25, 1999.

[8] J. Borras and R.D. Yates. Infostations overlays in cellular systems.In Proceedings of the Wireless Communications and NetworkingConference, WCNC, Volume 1, pages 495-499, 1999.

[9] R.H. Frenkiel and T. Imelinski. Infostations: The joy of"manytime, many-where" communications. Technical Report TR-119, WINLAB, Rutgers, The State University of New Jersey,April 1996.


[10] G. Wu, C.-W. Chu, K. Wine, J. Evans, and R. Frenkiel. Winmac: Anovel transmission protocol for infostations. Vehicular TechnologyConference, 1999.

[11] H.Mao, G. Wu, C.-W.Chu, J. Evans, and M. Carggiano. Performanceevaluation of radio link protocol for infostations. VehicularTechnology Conference, 1999.

[12] J. Irvine, D. Pesh, D. Robertson, and D. Girma. Efficient umts dataservice provision using infostations. Vehicular TechnologyConference, 3:2119-2123, 1998.

[13] J.G. Evans. A low cost asymmetric radio for infostations. TechnicalReport TR-130, WINLAB, Rutgers, The State University of NewJersey, September 1996.

[14] T. Ye, H.-A. Jacobsen, and R. Katz. Mobile awarness in a wide areawireless network of info-stations. ACM Mobicom, pages 109-102,1998. Dallas.

[15] J. Irvine and D. Pesh. Potential of dect terminal technology forproviding low-cost internet access through infostations. In IEEColloquium on UMTS Terminal and Software Radio, pages 12/1-6,1999.

[16] A.L. Iacono and C. Rose. Minimizing file delivery delay in aninfostations system. Technical Report TR-167, WINLAB, Rutgers,The State University of New Jersey , August 1998.

[17] A.L. Iacono. File Delivery Delay in and Infostations System. PhDthesis, Rutgers, The State University of New Jersey, June 2000

[18] A.L. Iacono and C. Rose. Bounds on File delivery delay in aninformation system. In Proceedings of the IEEE VehicularTechnology Conference, 2000. To appear.

[19] R.H. Frenkiel, B.R. Badrinath, J. Borras, and R.D. Yates. Theinfostations challenge: Balancing cost and ubiquity in deliveringwireless data. Submitted to IEEE Personal Communications, 1999.

[20] W. Feller. An Introduction to Probability Theory and Its Applications,Volume I, Chapters IV through XIV. Wiley, third edition, 1968.

This page intentionally left blank.

Chapter 2

WIRELESS BROADBAND MULTIMEDIA AND IPAPPLICATIONS VIA MOBILE ATM SATELLITES

Abbas JamalipourUniversity of Sydney, Australia

Abstract ATM is the promising technology for supporting high-speed data transferpotentially suitable for all varieties of private and public telecommunicationsnetworks. IP, on the other hand, is the fast-growing internet layer protocol thatis potentially applicable over any data link layer. New IP-based multimediaapplications require much higher bandwidth compared to the traditionalapplications run over the Internet which ignited the usage of the ATM in IPnetworks. With the revolutionary development in wireless cellular network inrecent years and the requirements of broadband data applications over thewireless channel, ATM and IP networks find their way of contribution in thisunderlying network. Mobility in wireless environment could however take itsultimate freedom when an integrated cellular-satellite supports the physicallayer of the network. In order to provide a global mobility for the futuremultimedia personal terminals, thus, there is a requirement of integration of allthese telecommunications technologies. Wireless ATM came to integrate thecell-switched ATM facilities in wireless environment and mobile IP has asimilar goal for IP networks. IP over ATM also proposed to merge the twoleading technologies of IP and ATM into a fast and efficient way ofmultimedia data transmission. Broadband ATM satellite systems also havebeen proposed in order to make the satellite channel a high-speed link forfuture networks. In this chapter, we will explain all these technologies andtheir mutual integration and then look into the issues of mobile satellites,ATM, and IP in a novel way in order to introduce the integration of the threetechnologies for future high-speed, global mobility-supporting, Internet-compatible wireless communication networks. We will discuss theapplications and their traffic and quality of service requirements. These arecrucial issues that need careful considerations for future multimediaapplications over the Internet.

Keywords: mobile satellite networks, cellular systems, wireless ATM, IP networks,routing, teletraffic, quality of service, mobility management.


1. INTRODUCTION

The idea of establishing personal telecommunication services viasatellites on non-geostationary constellations for commercial purposes wasfirst proposed in the early 90’s [1-8]. The proposal suggested that withsatellites on low earth orbit (LEO) or medium earth orbit (MEO), it ispossible to get rid of the highly restrictive long propagation delay and powerloss characteristics of the traditional geostationary earth orbit (GEO)satellites. Long propagation delay has always been a strict parameter inestablishing long-distance real-time communications such as voice and videotelephony via satellites. Long propagation loss, on the other hand, hasalways put a lower bound on the size of mobile terminals directly connectingto satellites. This is mainly because of the requirement of large batterycapacity for transmitting signal on the uplinks. By having satellites on orbitsmuch lower than the geostationary orbit, it is now possible to reduce thetransmission delay and the power of the transmitters so that satellite hand-held terminals could become a reality. These small satellite mobile phonesalso could provide users to have a unique and international network accessidentification number (NAIN) regardless of their location on the globe andthe availability of the terrestrial telecommunications infrastructures. Thesecharacteristics were so important and attractive for moderntelecommunications era that several satellite systems of this type wereproposed one after the other in a short period of time [1, 3, 9]. Although themajority of these systems have been proposed by the US companies, theywere highly supported internationally soon after so that the first system ofthis type started its service in the late 1998. An architectural example of thefuture mobile satellite systems for providing personal communications isshown in Fig. 1 [8].

Mobile satellite systems for commercial purposes were developed inparallel with the development of the second generation of the terrestrialcellular systems. Both systems look somehow to the same goal; that isachieving the issue of terminal mobility in telecommunications services.Some of the second generation terrestrial cellular systems such as GSM(Global System for Mobile communications), however, went further toprovide additional personal mobility. The differentiation between these twocomes from the way the moving object is defined. In the first proposal ofmobile satellite systems for commercial applications, the main purpose wasto provide basic telecommunication services1 (which were the dominantservices at that time) such as voice, telemessage, and paging regardless ofthe location of the user, specifically in remote areas. The user in such asystem can buy a specific terminal and subscribe to specific service(s)

1 For this reason, we may call this generation of mobile satellites as narrowband systems.

Wireless Broadband Multimedia and IP Applications 67

available within that terminal and the subscribed network. In order tosubscribe to other network services however, the user needs to buy adifferent terminal compatible with the new services. In GSM system on theother hand, the mobility is given to a user regardless of his terminal, and theuser is free to purchase and use any GSM-compatible terminals andsubscribe to new services after inserting his personalised SIM (subscriberidentity module) card in the new terminal. In both systems, however, themain idea of the mobility, that is, the ability of accessing telecommunicationservices from different locations by a terminal and the capacity of thenetwork to identify and locate the terminal are kept.

Another difference in these two parallel developing mobile systems is inthe range of mobility. Mobile satellite systems provide mobility in a muchbroader concept compared to the terrestrial systems. For example, if weconsider the mobility in the coverage area of a single base station (BS), itwould be in the order of a few kilometres in radius for a cellular system andseveral hundreds to a few thousands of kilometres for a LEO satellite (basedon the altitude of the satellite). Erection of a BS tower in a terrestrial systemwill also be limited to areas where the network service provider (NSP)expects to have some manhood population, such as cities, towns, and majorroads. This limitation is completely removed in the case of a mobile satellitewhich will cover anywhere on the globe including areas with no population.Therefore, the total coverage of the satellite systems (which is based ongeographical coverage and not on population coverage as in terrestrial

68 NEXT GENERA T1ON WIRELESS NETWORKS

cellular system) could be global. In this sense, we may consider the relationof satellite mobile systems and different terrestrial wireless systems in ahierarchical order, as shown in Fig. 2.

Mobile object also needs to be defined clearly when comparing terrestrialcellular systems and mobile satellite systems. The mobile object in aterrestrial system is the subscriber terminal, usually called a mobile station(MS), with a linear speed between zero and a few hundred km/h. In a mobilesatellite system, the moving object (or the mobile) is the satellite which has amuch higher speed; for example, in a LEO satellite system with a satellitealtitude of 1,500 km, the speed of the mobile comes to around 7.1 km/s or25,200 km/h! The changes in mobility characteristics, both the movingobject and speed, make the mobility management issues in mobile satellitesmore complicated when compared to that in terrestrial mobile systems.

Mobile satellite systems have a unique ability to establish a mobiletelecommunications network with or without their terrestrial counterpart. Inthe regions with no terrestrial wireless infrastructure, because of eithereconomical or technical reasons, mobile satellites can provide almost fullrange of telecommunications services. In the regions with developedwireless facilities, such as capital cities, the satellite can complement theservice or assist the terrestrial network in hotspot teletraffic handling.

1.1 Mobile satellites and cellular networks

As discussed above, the main reason for the success of the mobilesatellite system proposals was their ability to provide a ubiquitous means oftelecommunications. This ability has shown its importance in comparison


with GEO satellite systems because of the lower propagation delay and lossaccessible in lower orbit constellations and also service to the polar regionsusing inclination angles close to 90°. This latter characteristic of mobilesatellites (which is not possible in GEO constellation), secures a one-hundred percent global coverage possibility in the telecommunicationsystem. The idea was strong enough to support launching a very expensiveinfrastructure, both the initial and the consequent running and maintenancecosts, to the space. However, after the development of the second generationof terrestrial cellular systems in which at least a majority of countries use thesame standard, such as GSM in European and Asian countries (exceptJapan), roaming has put the usefulness of mobile satellite systems under abig question.

Roaming is an internetwork service in which a user or a terminal who issubscribed to a particular network, can ask to use temporarily a differentnetwork with the same standard which is not his home network (HN). Thesecond network, which we may call it a foreign network (FN), might havethe similar regional coverage as the HN or a completely different coverage.An example for the former roaming between networks could happen whenthe subscriber requires a network service which is not available in his HNbut can be supplied (maybe with a charge) from a FN in the same region. Forthe latter roaming, a good example would be the case when a cellularsubscriber travels abroad and wants to use his phone during his stay in theforeign country. This latter roaming clashes with the idea of the single globalnumber merit of the mobile satellite systems. Though the cellular phone userstill cannot use his cellular phone on the way to the other country and evenin the country of FN without a prior arrangement, the advantage of having asatellite phone for ordinary people who only travel to major cities ofcountries and not “deserts” ceases significantly.

The issue becomes even more apparent when we consider the fast growthof popularity and simultaneously decrease in price of cellular services in allparts of the world and even in developing countries where the satellitephones had targeted telecommunications services to those areas. As weapproach the third generation wireless mobile systems and the IMT-2000(International Mobile Telecommunications in the year 2000) [10], in whichthe internetwork connectivity is even considered between differentstandards, the advantages of the mobile satellite systems to terrestrial cellularnetworks are losing their importance gradually.

1.2 Future position of mobile satellites networks

Network service providers of the mobile satellite systems can still declarethat there is no terrestrial wireless system which can provide personal


telecommunication access to all parts of the globe including remote areasand polar regions. Also, there is no other wireless network infrastructure (oreven fixed networks) which could be reliable in the case of major naturaldisasters, whereas the mobile satellite systems have this ability to configurea complete network-in-the-space through satellites which will beindependent of any land system, and hence, reliable in the case of any typeof disaster on the earth. In addition, mobile satellite networks will providetheir subscribers a unique NAIN which they can use at any time without apre-arrangement before each travel, as in GSM networks. In addition to thebasic telecommunications services, these satellite-based networks canprovide other services such as vehicle navigation and GPS (GlobalPositioning System) on a personal basis.

But the question that remains is that whether these services are sufficientto promote the satellite personal communication networks so that we can seea rapid increase in number of subscribers to these systems as in the case ofterrestrial cellular systems. A quick answer would be “no” as it isexperienced in the recent financial failure of the first LEO satellite system.The above mentioned services and advantages could be attractive forcorporate and governmental subscribers but not for ordinary users as thecosts of satellite handset phones and subscription and call tariffs are too highand there is no optimistic expectation of significant reduction in the nearfuture. In order to achieve the goal in number of subscribers, the NSP of themobile satellite systems should focus on commercial applications that areattractive to ordinary people. These applications include broadbandmultimedia and especially Internet applications with lower costs that couldcompensate other expensive applications. The satellite NSPs have no choiceother than to compete the terrestrial cellular systems with their additionalservices and to integrate with them whenever this competition is not possiblein order to provide compatible service charges. In order to achieve this goal,the mobile satellite systems have to modify and adapt base on the newmultimedia applications. In this chapter, we will discuss some of these issuesrelated to the usage of the asynchronous transfer mode (ATM) to achievehigher data rates required in broadband networks and the Internet protocol(IP) applications. Such new mobile satellite systems, thus, will be referred toas broadband satellite systems.

1.3 Outline of the chapter

In the following section, we will review the characteristics of the mobilesatellite networks with emphasis on LEO constellations. In Section 3, wewill explain the ATM network originally developed for wired networks asthe most significant contribution to B-ISDN (broadband integrated services


digital network) and how it has been involved in wireless environment,namely the wireless ATM. Section 4 gives an overview to IP networks andhow they also came into the wireless world after the invention of mobile IP(MIP) in l996.

In Section 5, we will discuss specific issues related on the way ofintegration of ATM and IP networks with wireless and particularly with themobile satellite systems. The most important issues necessary to beconsidered are the quality of service (QoS) and traffic management. Theseissues are important in the sense that if considered carefully and thensophisticated techniques are designed and implemented, it would be possibleto provide more and better services to users and hence achieve highernumber of satellite mobile users. The quality of service has been vastlyconsidered for wired network but needs to be redefined when mobile andwireless channels are being involved. Different types of traffic and theirmanagement techniques are also required when considering multimedia andbroadband applications over the wireless link. After exploring these issues,we will introduce perspectives and applications of an integrated wireless IP-ATM network via mobile satellites. Finally, we will conclude our results anddiscussions in Section 6.

2. MOBILE SATELLITE NETWORKS

Mobile satellite networks refer to systems in which thetelecommunications satellites are on orbits other than the geostationary orbit.According the Kepler’s third law, the geostationary orbit is a uniqueequatorial orbit at a distance around 35,800 km from the earth surface [8]. Asatellite on the geostationary orbit can cover almost one-third of the earth,and hence, three satellites would be sufficient to cover almost all part of theglobe. This coverage excludes polar regions and other high latitudinal areas.The reason is simple if considering spherical shape of the globe and theposition of satellite over the equator. Since the satellites are stationary inrelation to the movement of the earth, antenna tracking and control would beminimal and the satellite gateways to the terrestrial public switchingtelephone networks (PSTNs) can always be faced to satellites for maximumsignal reception and transmission. This type of satellites, then, can be easilyused for long-distance telecommunications and broadcasting purposes.Because the length of the satellite transmission link is independent of theactual land distance of any given pair of the hosts on the earth, the long-distance communications cost will only depend on whether or not a satellitelink is used.


GEO satellite systems were successful in providing commercial services,both in telecommunications and broadcasting, since the establishment of thefirst system, the INTELSAT, in 1965. The key characteristic of the GEOsatellite systems was that they could be considered as a part of the fixedpublic switching network. In 1982, INMARSAT, another key-pioneered insatellite systems especially for mobile purposes, introduced its mobilesatellite services, or MSS, using GEO satellites in order to providetelecommunications services to ships and other large mobile vehicles. Thiscould be considered as the starting point of mobile communications viasatellites. However, there was always the problem of the round-trip distancebetween the earth and a GEO satellite which makes it impractical to havesmall-size terminals other than those mounted on vehicles. This restrictiveissue has become more visible when the people started to think aboutcommunications based on personal perspectives, or the personalcommunications services (PCS). It was clear that with GEO satellites it isdifficult, if it not impossible, to provide personal communications with smallhandheld terminals and phones.

Requirements of lower propagation delay and propagation loss togetherwith the coverage of high latitude regions for personal communicationservices have started a vast research on employment of satellites on lowerorbits, which in nature will have non-geostationary characteristics. Due tothe existence of the two Van Alien radiation belts, these mobile satelliteswere categorised into low earth orbit with a altitude of 500-2000 km andmedium earth orbit at around 10,000 km height. Generally, the lower theorbit the lower the propagation delay and loss and the higher the number ofsatellites (and the orbital planes) to cover the entire globe is resulted. Figures3 and 4 show the relationship between the altitude of satellites and thenumber of satellites and the number of orbits, respectively [8]. Besides, thefigures spot the actual constellation of some PCS non-geostationary satellitesystems.



As discussed in the previous section, the idea behind these mobilesatellite system which could provide a single and worldwide access numberwas so attractive that in a short period of time many of these systems havebeen proposed and found multi-national support [3]. Table 1 summarisessome of these PCS-based (narrowband) mobile satellite systems. In additionto these systems, there is another LEO proposal for global coverage, namedTeledesic, as a multimedia satellite system using Ka-band. The system isplanned to have data and Internet services at high data rates and in itsoriginal proposal 840 LEO satellites were considered. With a compromise onthe data rate actually required for the subscribers and the integration with theInternet service providers (ISPs), Teledesic has now changed its design to ahigher orbit height at 1,400 km which reduces the total number of satellitesinto 288 and may change further. Teledesic will use 1-Gbps links and 13.3-Gbps capacity satellites which state the potential applications of the systemto be broadband multimedia. For this reason, we may put the newly designedTeledesic as a broadband satellite system, why it is missed from Table 1.

Among the systems shown in Table 1, the Iridium [9], the first completedLEO satellite PCS system, has a unique design to achieve essential coveragewith minimal requirements of land-based gateways that connect to thePSTN. This is achieved by employing links between satellites, orintersatellite-links (ISLs) working at 23 GHz, which enable the system to


route the traffic from one satellite to another, forming a network in the space.A general overview of the Iridium system is shown in Fig. 5.

Each Iridium satellite has powerful on-board processing and routingfacilities. Traffic arrived in a TDMA (time division multiple access)timeslot, will be processed by the satellite and the routing decision will bemade. The next destination could be a ground gateway station via 20-GHzlinks or one of the four nearest satellites via ISLs. This type of user-satellite-gateway connectivity is shown in Fig. 6. The Iridium system employscircular polar orbits (86.5° inclination) which guarantee the servicecoverage to high latitude regions. The global coverage of the Iridium is oneof the main characteristics that distinguish this system from other mobilesatellite proposals. The footprint of each Iridium satellite is divided into 48cells via three L-band antennas forming a total of 3,168 cells on the earthsurface-A cellular-type satellite system. From those cells only 2,150 cellswould be enough for a global coverage, but with this plenty of cells, it ispossible for any given user to be in two or more cells simultaneously at mostof the times, providing a highly reliable communication.


The next Big-LEO satellite system to be in service soon is the Globalstar.This system does not claim offering a global coverage; instead it willprovide coverage to its partners in different countries with sufficientpopulation. This fact, together with higher altitude of the satellites, results ina less number of satellites than the Iridium system. Since the satellite orbitshave a 52° inclination, little or no coverage is provided beyond latitude.At most of the times, two or more Globalstar satellites will be visible fromthe designated areas on the earth. Another difference between the Iridiumand the Globalstar is that the latter does not employ ISLs, and as a result, asubscriber can access to the system on a bent-pipe fashion through a gatewaystation, as shown in Fig. 7. For a typical service area of about 1,600 kmaround a gateway station, global coverage requires more than 200 earthstations which is not planned in the system. Therefore, Globalstar will likelyserve national roamers in general. A satellite that is working as a repeater(e.g., the one used in bent-pipe scenario), is sometimes referred to as atransparent satellite.


The above discussion and explanation on mobile satellite networkproposals should make it clear that these systems will have a significant, ifnot dominant, role in the next generation wireless communications. Therecent financial failure of the Iridium however, does not change this role. Onthe contrary, it states the fact that the future trend in wirelesscommunications is the Internet and broadband services and any systemoptimised for voice-only communications subjects to failure, regardlesswhether it is terrestrial or satellite. In the following sections, we will explorethe ATM and IP networks and the potential integrity of the mobile satelliteswith these networks. This will be the most important issue for the mobilesatellite systems in order to compete or complement the next generationterrestrial cellular networks.

3. WIRELESS ATM NETWORKS

In this section, we will briefly explain the concept of ATM-basednetworks and how asynchronous mode of transfer provides extremely highdata rates in digital communication networks. The ATM switching is thepromising backbone technology for any data communication network,including telephony systems and the Internet. We will then develop the newtopic of wireless ATM and discuss the new elements added to the traditionalATM protocol stack. The wireless ATM will then be developed inapplications using mobile satellite networks in order to let those satellites bepracticable for the transmission of multimedia and broadband traffic.


3.1 ATM Networks

With the introduction of modern digital and high-speedtelecommunications with relatively low error rates, the requirements of longoverheads on the packets of the traditional packet switching networks(PSNs) became unnecessary. Since those overheads contain no userinformation, reduction in the amount of overhead bits could result in moreefficient utilisation of the channel capacity and higher data rates than whatcan be achieved in traditional PSNs. Frame relay networks make use of thisfact to increase the data rate from 64 kbps of the PSN up to 2 Mbps. ATMnetworks, on the other hand, reduce the overheads further by employingfixed-size packets, called cells, and increase the data rate to 10s and 100s ofMbps. As an analogy to frame relay, the ATM service sometimes is referredto as cell relay.

ATM has the most significant contribution in standardisation of B-ISDN[11-12]. In ATM, user information is split into 53-byte fixed-sized cells, asshown in Fig. 8, and then switched using fast hardware-based cell switching.Cell header, a 5-byte label, carries the minimum of overhead to supportmultiplexing and switching of the ATM cells. ATM leaves most of the errordetection and error correction and also out-of-sequence cell detection tasksto the higher layers of the network protocol stack, above the ATM layer andthe ATM adaptation layer (AAL). Asynchronous feature of the ATM mayseem conflicting to the periodic nature of the existing traffic from analoguesources, such as voice or video. However, the apparent periodicity is aproperty of the channel coding process and not of the information sourcesthemselves. With the powerful source coding mechanisms available now, itis possible to exploit the ability of ATM to absorb the essential burstinessthat characterises the analogue sources. ATM has the ability to multiplex andswitch data from various sources with varying rates and informationstatistics, and thus, is the most promising transfer mode for multimedia datawhether originating in B-ISDN or the Internet and intarnet segments.

In ATM, logical connections are referred to as virtual channelconnections (VCCs). A VCC is the basic unit of switching in B-ISDN and isset up between end user pairs through the network. A variable-rate, full-duplex flow of ATM cells is exchanged over the connections. Theseconnections are also used for control signalling between user and thenetwork and for network management and routing between one network and


another. All VCCs with the same endpoints are bundled in a virtual pathconnection (VPC) and switched along the same route. By groupingconnections sharing common paths through the network into a single unit, itis possible to control the cost of the high-speed networks significantly.Relation between the above connections is shown in Fig. 9. Virtual pathlevel and virtual channel level form the two sublayers of the ATM layer.

The cell header consists of 5 bytes. The format of the ATM cell headerfor the user-network interface (UN1) is shown in Fig. 10. For the network-network interface (NNI), there is no generic flow control (GFC) field and thevirtual path identifier (VPI) field fills the whole first byte of the header. Ann-bit label will support separate channels in the aggregate cell stream andas we will see in Section 3.3, it would be sufficient to support ISL routing inmobile ATM satellite systems. Resilience in the presence of errors isachieved by means of the header error check (HEC) mechanism. HEC is alsoused as a mechanism to control out-of-sequence cell arrival errors in ATMnetworks.

Payload type identifier (PTI), a 3-bit field, distinguishes particular classesof information flow and a single-bit cell loss priority (CLP) signals the cellto be discarded in the case of congestion in the network, similar to thetechnique used in frame relay networks. The GFC field is used to control the


traffic flow at the UNI in order to alleviate short-term overload conditions.This flow control is part of a controlled cell transfer (CCT) capabilityintended to meet the requirements of non-ATM local area networks (LANs)connected to a wide area ATM network.

As a final comment to the ATM networks, it is worthwhile to state thatthe ATM is intended to transfer many different types of trafficsimultaneously, including real-time flows such as voice, video, and burstyTCP (transmission control protocol) flows of the Internet. Trafficmanagement techniques have been developed for ATM in order to handlethese different types of traffic in an efficient manner based on thecharacteristics of the traffic flow and the requirements of the applications.All these issues are important in development and operation of a network,regardless of the medium whether it is wireless or wired, that is designed tohandle multimedia traffic and broadband applications. These issues will bediscussed in Section 5.

3.2 Extension of ATM into Wireless Environment

The ATM indeed can be considered as the main standard technology forbroadband communications in wireline infrastructure. However, the recentdevelopment in wireless networks which supports the mobility of users andthe strong requirement of supporting multimedia and specifically theInternet-based applications have opened new researches toward theintegration of ATM with the wireless, namely the wireless ATM (WATM).ATM has the advantages of high efficiency and QoS support for users and ifintegrates with wireless networks, can provide mobility-supported high-efficiency multimedia services.

WATM can be considered as an extension of a wired backbone networkwith the flexibility of wireless access and mobility support [13-17]. Thestandardisation of the WATM has been started within the ATM Forum andETSI (European Telecommunications Standards Institute) with contributionfrom other standardisation institute such as the IETF (Internet EngineeringTask Force). The first draft specification related to the WATM has beenreleased in December 1998 [13].

The WATM network has the traditional wired ATM network as itsbackbone. Therefore, we may consider a WATM network as a modifiedversion of a wired ATM network with new wireless links and equipment. Toinclude mobility, the traditional ATM switches are now complemented withmobility-supporting ATM switches connecting through enhancedpublic/private network node interface (PNNI). These new switches areconnecting the wireless access points (APs) or BSs to the wired network.Mobile terminals (MTs) which can be for example laptop or palmtop


computers and mobile phones are connecting via new wireless UNI usingradio channels. Connectivity between mobile and fixed hosts in the networkthus will be through wireless UNI, mobility supporting ATM switches,traditional ATM switches, and traditional wired UNI. The BS in thisconfiguration is sometimes named as the radio access unit (RAU) whichcontains all the link layer functional elements, including radio resourcemanagement and medium access control (MAC) functions, necessary tooperate over a shared radio frequency (RF) medium. Figure 11 shows ageneric configuration of such a WATM network.

The protocol architecture of the WATM also needs modifications formobility support [14]. Figure 12 shows such a modified protocol in whichnew mobility-related layers are shown in grey colour. As seen in the figure,both the user plane and control plane should be modified to support themobility in the network. Much attention should be given to inclusion ofproper MAC and wireless control protocols. More details on this architecturecan be found in [13-14].



In the case of mobile networks, including WATM, the main mobilityfunctions are location management and handover management. Since inthese networks users have not committed to be in any specific location, thereis the requirement of finding the real location of the MT from time-to-timeand also specifying its nearest point of attachment to the wired network. Thisissue will be required during the process of routing the information packetsfrom a MT to another MT or to a fixed host and vice versa. An efficient,reliable and quick handover technique is also necessary in order to maintainand reroute an ongoing session while the MT moves from the coverage areaof a BS to the next one. The location management and handovermanagement together is usually referred to as mobility management inmobile networks.

3.3 Wireless ATM and Mobile Satellite Networks

In the discussion on WATM given in the previous section, we have notspecified any type of physical channel used for transmitting the radiosignals. Consequently, it is possible in general to consider any type ofwireless media including satellite channels. Indeed, this is actually the ideabehind the new generation of mobile satellite systems based on the ATMarchitecture as their mode of transfer [18-22]. The satellites in these systemsare usually multispot beam with onboard processing capabilities. These


systems will provide services at high data rates in the order of 2 Mbps orhigher usually at Ka-band (30/20 up/down GHz) where the requiredbandwidth is available. Table 2 summarises some of satellite systemproposals for broadband applications [19]. Among these systems, SkyBridgeis the only one that will use Ku-band (14/11 up/down GHz). This band hasalready been used by the fixed satellite service (FSS).

Both the transparent satellite networks and the systems with onboardprocessing satellites can be integrated with ATM networks. In the formersatellite ATM network, all protocol processing is performed on the ground atuser terminal, gateway stations, and the network control centre (NCC), sincethere is no such onboard processing facilities in the satellite to perform therequired processing at the ATM layer or above. These systems, however, canprovide a quick deployment of ATM connectivity using exiting satellites,and hence, providing high-speed network access by user terminals and high-speed interconnection of remote ATM networks. We will not discuss thistype of satellite ATM networks here, but a detailed discussion on thenetwork architecture of these systems can be found in [18].

In the networks with onboard processing satellites, the control functionsperform proportionally in the onboard ATM switch and the NCC on ground.The ATM interfaces between the payload switch and ground terminals canbe either a UNI or a NNI [18]. If the satellite links are low speed, then theywill be used to connect remote ATM hosts to a terrestrial network. Here, theinterface between the ATM hosts and onboard switch is a UNI and the onebetween the onboard switch and the terrestrial ATM network is an NNI.With high-speed satellite links, onboard satellite will function as an ATMnode and the interfaces will be NNI type. In a satellite system whichemploys ISLs, each satellite in the space network will act as a completeATM node and the network provides both network access and networkinterconnectivity. Here the interfaces between satellites are NNI type.

Figure 13 shows simple end-to-end communications between twosatellite mobile terminals, and and between a mobile terminaland a fixed terminal connected to the PSTN, respectively. In this figure,it is assumed that the mobile terminals have direct access to LEO satellites


and that the satellites are networked together via ISLs. What this simplefigure illustrates is that in a mobile satellite system with ISL networking, it ispossible to achieve high data rate long-distance communications directlybetween terminals, both mobility supported and fixed ones. The directlyconnectable terminal in such a system contains a satellite adaptation unitwhich performs all the necessary user terminal protocol adaptations to thesatellite protocol platform. This unit also includes all physical layerfunctionality such as channel coding, modulation/demodulation, the radiofrequency, and the antenna parts. The satellite contains onboard signalregeneration, and performs multiplexing/demultiplexing, channelcoding/decoding, and ATM switching.

In the communication path between the mobile terminal and the fixedterminal, there is a gateway station which provides connectivity between thespace and the ground segments. An interworking unit (IWU) included in thegateway station performs all necessary translations between the space(satellite) segment and other ground-based networks. The ground networksinclude PSTN, narrowband and broadband types of ISDN, frame relaynetworks, the Internet, and private and public ATM networks. A fixed userterminal equipment could be belonged and connected to any of thesenetworks. A network control centre might be required for an overall controlof the satellite network resources and operations. This includes allocation ofradio resources to the gateway stations, call routing and call managementfunctions such as location update, handover, authentication, registration,deregistration, and billing. In a complete LEO satellite system employingISLs, however, all these tasks could be distributed among the satellites,providing a more reliable control and then no NCC will be required. Anillustrative architecture for a global ATM connectivity using mobilesatellites is shown in Fig. 14.



Another viewpoint that relates the mobile satellite systems employingISLs with ATM networks is that we can consider each satellite as an ATMnode, each ISL as a single VCC, and the routing path of a connection as aVPC of an ATM network. Therefore, we can build a complete high-speedATM network in the space using the LEO satellites as its nodes and thenapply similar ATM-based algorithms in that network. Specifically, applyingthe VPC and VCC concepts in a mobile satellite system will benefit us byutilising many advantages of the ATM routing and transmitting schemes. Itwould also be much more convenient for mobile satellite networks to accessfixed terrestrial ATM networks. As explained in Section 3.1, an ATM cellheader in the NNI format contains 12 bits for VPI. This allows a maximumof VPs for each single ISL step. This number is more than

adequate for a mobile satellite system since the number of ISLs for eachsatellite nodes of the system is only between two and four (two links tosatellites in the same orbital plane and two to satellite in the firstneighbouring planes). Actually, one VPC has impact on only one ISL if thenode is at the terminal point and on two if the node is middle transient one.Thus the maximum number of simultaneous VPCs equals to the total numberof the all pair nodes, which can be defined as N(N-1)/2, where N is the total


number of the satellites in the system. More discussion on applying ATMrouting concepts for mobile satellite systems can be found in [20-25].

As a conclusion, we can say that satellite ATM networks can be used toprovide broadband access to remote areas and also to serve as an alternativeto the wired backbone networks. These satellite networks can effectivelyprovide both real-time and non-real-time communication services in a globalbasis to remote areas and other regions where land-based facilities are notsufficient or not available.

4. IP NETWORKS

In this section, we will overview the traditional IP networks in order tobriefly explain the new concept of mobile IP networks. The mobile IP is oneof solutions in providing macro-mobility in IP networks. This concept,though originally developed based on terrestrial wireless infrastructure,could have no logical objection to be integrated in mobile satellite networks.This integration will be discussed shortly in Section 5.

4.1 Conventional IP Networks

The Internet can be defined as a connection of nodes on a global networkuse a DARPA-defined (Defence Research Projects Agency) Internet address.The protocol suite that consists of a large collection of protocols that havebeen issued as Internet standards, is referred to as TCP/IP (transmissioncontrol protocol/Internet protocol) [26]. TCP/IP was a result of protocolresearch and development conducted on the experimental packet-switchednetwork ARPANET funded by DARPA. In contrast to the OSI (open systeminterconnection) reference model which was developed by the InternationalOrganisation for Standardisation (ISO), TCP/IP has no official protocolmodel, but can be organised into five layers of application, transport,Internet, network access, and physical. The network access layer can furtherbe divided into two sublayers called logical link control (LLC) and mediumaccess control (MAC). Information data processed in each application on ahost computer should go through all these layers until it can be passedthrough the physical media on a LAN and through intermediate routing andswitching facilities on the wide area networks (WAN) and the Internet.Figure 15 illustrates the connections and the required protocol stack in asimple TCP/IP-based network.



The two main components of the Internet, which are shown in Fig. 15,are the hosts and the routers. Hosts include any type of computer such asPCs and workstations. Routers forward datagram packets between hosts andother routers when there is no same link (e.g., a bus) connecting them. Arouter operates at the network layer of the OSI model to route packetsbetween potentially different networks. Another component that could beconsidered here is a bridge, which operates at data link layer and acts as arelay of frames between similar networks.

In order the routers perform their task, they use special procedures calledrouting protocols. Routing tables are built using these procedures and then arouter can select a path (hopefully the optimum one) for any given packetfrom a source host to a destination host. In the case of many routers betweena source and a destination, routing will be performed on a hop-by-hop basis,in which each router finds the next node (router) for sending a given packetunt i l the packet is being reached at its requested destination.

The IP is the most widely used internetworking protocol at the Internetlayer. An IP datagram includes a header and the payload. Payload of the IPpacket contains all the higher layer headers such as TCP in addition to theapplication layer data. The header for the IPv4 (currently deployed versionof IP) contains 20 bytes in addition to a variable size options field requestedby the sending host, as shown in Fig. 16.

The version field shows the version of the protocol, which is 4 for thecurrently used protocol. IHL (Internet header length) shows the size of IPheader. Type of service field specifies QoS parameters such as reliability,precedence, delay, and throughput. The maximum time that a datagram isallowed to remain in the Internet is specified in the “time of live” field.Header checksum is an error-detecting code for the header only and theprotocol field indicates the next higher level protocol that is to receive thedata field at the destination. The identification field is a sequence number toidentify a datagram uniquely together with the source address, destinationaddress, and user protocol.


The most important parts of the header are the source address and thedestination address. These are 32-bit IP addresses, as shown in Fig. 17,assigned to each network interface of a node. A node with multipleinterfaces, such as routers, then has more than one IP address. Each IPaddress has a network-prefix portion and a host portion. A network-prefix isidentical for all nodes attached to the same link whereas the host portion isunique for each node on the same link. In the next generation IP, or IPv6,address fields are extended into 128 bits which increases more number ofhosts in the network. Moreover, in IPv6 options are placed in separateoptional headers that are located between the IPv6 header and the transport-layer header. This will speed up router processing of datagrams. In addition,other enhancements such as address autoconfiguration, increased addressingflexibility for scalable multicast routing, and resource allocation that allowslabelling of packets belong to a particular traffic flow for special handling,are included in this new version of IP.

The most important task to be performed by the IP layer is routing.Whenever a packet received by a node, a host or a router, for which the nodeis not its final destination (i.e., having different destination IP address as thereceiving node), the node must find where the packet should be route inorder to be closer to its final destination. Therefore, in the process of routinga packet, a forwarding decision must be made by each node. This decisioncan be made using an IP routing table, which is maintained in each node.

Each row of the routing table usually has four components, namely,target, prefix-length, next-hop, and interface. Whenever a node has a packetto forward, it checks for matching between the packet’s IP destinationaddress field and the left-most prefix-length bits of the target field within therows of the table. If such a match is found, the packet will be forwarded tothe node identified by the next-hop field via the link specified in theinterface field in that row. In the case of more than one matching, the packetforwards to the one which has the largest prefix-length. This will ensure thatthe next node is the closest node to the final destination. An entry in therouting table might be a host-specific route, with the prefix-length of 32which can match with only one IP destination address; a network-prefixroute, with a prefix-length between 1 and 31 bits which match all destinationIP addresses with the same network-prefix; or a default route, with a prefix-length of zero. This last route will match all IP addresses but will be usedonly when no other matching were found.


The routing tables might be created statically (manually) or dynamically.Usually, these routing tables are produced using one of common shortest-path or least-cost algorithms such as Dijkstra or Bellman-Ford algorithms[26], widely used in other packet-switched networks. Because the Internetrouting is based upon the network-prefix portion of the packet destinationaddress, it is greatly improves the scalability of the Internet.

4.2 Mobile IP Networks

Mobile IP (MIP) is an extension to the currently deployed (fixed) Internetprotocol in order to provide wireless access to the Internet users [27-30].MIP is described in a request for comments (RFC) published by IETF first inOctober 1996 [27]. The most important barrier in developing mobileinternetworking is the way IP operates. Conventional IP supportsinterconnection of multiple networking technologies into a single, logicalinternetwork and is the most widely used internetworking protocol. An IPaddress is used to identify a host and contains information used to route thepackets.

Generally, in a mobile Internet the two lower layer protocolfunctionalities, i.e. physical and data link, are provided by cellular networks.However, the next higher layer protocols, that is network and transportlayers, should be modified in order to enable them to route and deliverpackets correctly in a mobile environment. As explained in Section 4.1, anIP address is assigned uniquely to each machine in the network and is usedby the network layer to route the datagrams. The concept of network prefixas part of an IP address however, is contradictory with the idea of mobility.This is because of this fact that in the case of movement of a terminal in thenetwork, it is not possible to maintain a single point of attachment for theterminal to the network; i.e. no logical network prefix would be available.Thus, any solution for supporting mobility in the Internet is constrained bythe requirement of the existing IP function and networking applications. Asthe mobile user is roaming between foreign networks, it will acquire a newIP address causing the established connection of the node to be lost.

MIP provides a means of delivering the packets addressed to the mobilenode. By defining special entities, home agents (HAs) and foreign agents(FAs), a mobile node (MN) is able to cooperate in moving without changingits IP address. Inversely, it provides a means for MIP to deliver packetsaddressed to a particular MN in the network. Furthermore, the solution canbe appropriately expanded to accommodate an increasing population ofmobile users (i.e., supporting the scalability).

In MIP, each mobile node is given a virtual home network. This remainsunchanged, and is used to assign the mobile node a constant IP address in the


same manner that a standard IP address is given to a stationary host. On thehome network, a location information database is maintained for each of itsattached MNs (which are currently visiting other networks). The accuracy ofthis information become vital when routers are to deliver any MN addresseddatagrams.

The core operations involved in MIP include agent discovery,registration, and packets tunnelling. This is exactly what mobilitymanagement is defined to be; i.e. to detect MN’s change of location, registerthe new location with HA (either directly or via FA), and finally to performhandover as MN moves to a new network.

Upon detecting a change in location, the roaming MN acquires a new IPaddress, a care-of address (CoA), either from the received foreign agentadvertisement (FACoA), or from an external dynamic host configurationprotocol (DHCP) server, a co-located CoA (CCoA). MH then notifies HA ofthe new location through the process of registration. Figure 18 gives a briefillustration to how Mobile IP works.

Data packets from a corresponding node (CN) are generally routed bydefault to the MN’s home address. HA attracts packets destined for thosenodes that are away from their home network, and redelivers them accordingto the corresponding CoAs being registered by each roaming node.

After the registration with the HA is completed, the mobilitymanagement protocols should secure a way for packets to be routed to thecurrent point of attachment, namely the tunnelling. The method used toforward data to roaming MN is known as encapsulation. Though MIP


assumes an IP-within-IP encapsulation methodology, shown in Fig. 19, otherencapsulation mechanisms are applicable upon agreement made betweenrelevant networks.

In general, M1P provides a good framework for handling users’ mobilityin a way that was never possible within the conventional IP networks. Thereare many benefits associating with this particular mobility managementtechnique. Table 3 briefly summaries some of these characteristics.

In spite of the advantages of the MIP in providing a mobile computingenvironment, there are a few concerns about its efficiency. Basically, theinefficiencies in MIP can be classified into three main categories accordingto each step of the mobility management process, say location management,routing management, and handover management. Regarding the locationmanagement, a serious inefficiency is widely evident because a registration


process with HA is required at every handover when changing either thenetwork or the link within the same network. This results in wastingresources that are associated with the frequent location updates arising fromevery single MN’s movement.

For the routing management, one of the biggest concern in MIP is theinefficiency associated with the way packets are delivered to roaming MNs,namely the triangle routing; an asymmetric routing with respect to topology.Specific concerns in this aspect include packet losses during handovers, highdata latency, and inefficient use of the network resources due to tunnelling.Route optimisation techniques are being developed to cope with this issue.The handover management also would be necessary to develop in order tocontrol large number of handovers by MNs as the size of cells in cellularsystem becomes smaller.

MIP which uses the terrestrial cellular infrastructure could be a goodstarting point for the implementation of IP services over mobile satellitelinks. The inefficiencies discussed above, however, should be carefullyconsidered in long-delay satellite links.

5. INTEGRATION OF WIRELESS ATM AND IP INSATELLITE NETWORKS

The integration of IP in ATM networks requires considerations of bothservice and performance issues. This consideration becomes even moreimportant when we apply the two protocols in a mobile and wirelessenvironment. In principle, the quality of service is the major issue which issupported in ATM networks and the applicability of IP over any data linklayer is the main characteristic of the IP networks. The integration of thesetwo networks aims to take the advantages of both and to optimise theintegrated network. Another important topic which requires moreinvestigation in this integration, is the different types of traffic to betransmitted over the integrated network and the traffic management policies.Therefore, in this section we will first look over the issues of quality ofservice and traffic management and then discuss the perspectives andapplications of the integration of IP and ATM in wireless and satelliteenvironment.

5.1 Quality of Service Requirements

The commonly used metrics for QoS in the telecommunications networksinclude bandwidth, throughput, timeliness (including jitter), reliability,perceived quality, and cost [31-34]. The management of the system


components becomes more complicated as we move from simple voice ordata services into multimedia and broadband applications. In this sense,because of certain limitations in portable computers, such as the restrictionsof battery life, screen size, and connection cost, management of deliveringthe required QoS in a mobile environment becomes even more complicated.

Depends on the type of an application, we may define different QoScharacteristics. For example, in transferring an image file, the picture qualityand the response time could be considered as appropriate factors. In general,the main technology-based QoS parameters are [31]:

Timeliness, including several parameters such as:delay (transmission time for a message)response time (time between the transmission of a request and thereceiving a reply)

jitter (variation in delay time)

Bandwidth, which may be defined at:system level data rate (required or available bandwidth in bit persecond)application level data rate (application specific bandwidth in itsunit per second)transaction rate (processing rate or requested rate of theoperations)

Reliability, which can be measured by:mean time to failuremean time to repairmeantime between failuresloss or corruption rate (e.g., because of network errors)

From a user-level QoS requirements, the following categories might beconsidered:

Critically, i.e. priority among different flows in multimedia streamperceived QoS, which is based on the type of data transmissionapplication can be defined by:

picture detail (e.g. resolution)picture colour accuracyvideo rate (frame per second)video smoothness (frame rate jitter)audio quality (sampling rate)video/audio synchronisation

Cost (a significant parameter considered by users) which can be eitherof the two:


per-use cost (connection establishment and/or resource accesscost)per-unit cost (per second or per unit of data cost)

Security, required in most probable applications, including;confidentialityintegritydigital signaturesauthentication

Certain controls and supervision, namely QoS management techniques,are required to attain and sustain the desired quality of service properties[31]. These techniques are required not only at the initiation of an interaction(namely, the static functions) but also during that interaction (namely, thedynamic functions). Definition of QoS requirements, negotiation, admissioncontrol, and resource reservation are some of the static functions, whereasmeasuring the QoS actually provided, policing, maintenance, renegotiation,adaptation, and synchronisation (e.g. combining speech and video streamswith temporal QoS) are the examples of dynamic functions. In the case ofATM satellite networks with onboard processors in which multiple IP flowsare aggregated onto a single VC, a QoS manager classifies the flows of IPtraffics in order to utilise the bandwidth efficiently [35]. This QoS manageruses IP source and destination address pairs. The manager also can furtherclassify the IP datagrams based on the type of service field (see Fig. 16)requested and available in the IP header.

In a mobile environment, mobility results in significant changes in QoSand a mobile system has to be able to adapt such changes. For the first QoSmetric, i.e. the bandwidth, we have to accept that for some time the wirelessnetworks can provide only bandwidth in an order much lower than fixednetworks. The freedom in mobility of terminals will also be limited by thecoverage area of the wireless infrastructure that a user is subscribed.Obviously, here another issue of QoS, i.e. the cost, will arise. As we move toa larger coverage and higher mobility during connection, e.g. from wirelessLAN into cellular phone networks and to satellite systems, higher costs maybe required though they may not provide higher data rate supportsproportionally2. Table 4 summarises the relationship between area ofcoverage and bandwidth for several common wireless networks. As it can beseen from this table, all these wireless networks provide much lower datarates than typical Ethernet LAN networks of 10 Mbps-1 Gbps.

2 Note that here we consider real terminal mobility and not nomadic systems which canprovide acceptable data rates at relatively low cost by using dial-up connections.


Nevertheless, in a mobile environment with data traffic applications, QoSmanagement requires much more sophisticated techniques than fixednetworks, because:

A short loss of communication during a handover, which is usuallyacceptable in voice-application system, is not desirable in dataapplications.

New point of attachment after a handover requires to have similarfacilities and resources as the old one, thus renegotiation procedureswould be required.

Blind spots where the signal is very weak, and hence, low quality, areunavoidable in mobile and wireless systems.

Certain specifications of portable terminal such as laptop computers mayalso affect the end-user QoS requirements in mobile environment comparedto fixed networks. These limitations include the battery limits, processingpower with low power consumption, screen size and their screen resolution.

5.2 Traffic Considerations

As explained in Section 4, the main structure of an IP application is basedon TCP and thus the performance of TCP is crucial in running IPapplications efficiently. In principle, TCP should work anywhere regardlessof underlying network architecture, however, it is optimised for operating ina wired network with relatively low bit error rates (BER), say in the order of

or less [36]. Consequently, TCP assumes that the major cause ofproblems in packet handling in the network is the congestion. In the case awireless link is used for transmission of packets, however, this assumptionwould be no longer valid as the main cause will change to the packet lossbecause of high BER of the wireless link. In the case a satellite link is usedas the wireless channel, the situation becomes even worse. For GEOsatellites the long delay and for LEO/MEO satellites the rapid delayvariation causes the acknowledgement- and time-out-based TCP congestioncontrol mechanism performs weakly. This in turn results in large number ofretransmissions which degrades the performance of the TCP. Therefore, the


relation of bandwidth-delay product and round-trip delay variation to theperformance of TCP requires development of new congestion control andtraffic management mechanisms in the TCP layer [35-36].

One issue in integration of IP traffic into ATM mobile satellite is toaccommodate multiple IP traffic onto a single VC. The primary reason forthis to be important is that the IP traffic must be transmitted within the ATMcells and through ATM VCs and that the number of these VCs are limitedbecause of the limitation of the earth stations and onboard satellites. Theclassification of large number of IP datagrams into limited number ofavailable VCs is performed by a QoS manager, as discussed in previoussection.

ATM is intended to carry different types of traffic simultaneouslyincluding real-time flows such as voice and video streams and bursty TCPflows [11]. Therefore, the ATM Forum has defined real-time and non-real-time service categories to accommodate all applications that require eitherconstant or variable bit rates. In general, real-time services concern about theamount of delay and the variability of delay (jitter). These applicationstypically involve a flow of information to a user that is intended to reproducethat flow at a source (e.g., voice or audio transmission). On the other hand,non-real-time services are intended for applications that have bursty trafficcharacteristics and do not have tight constraints on delay and delay variation(more flexibility for the network to handle traffic and to use statisticalmultiplexing).

The real-time services of the ATM include constant bit rate (CBR), andreal-time variable bit rate (rt-VBR). Non-real-time services also include non-real-time VBR (nrt-VBR), unspecified bit rate (UBR), and available bit rate(ABR). Among these services, UBR is suitable for applications that cantolerate variable delays and some cell losses (such as TCP-based traffic).Thus, no initial commitment is made to a UBR source and no feedbackconcerning congestion is provided. This service is best suited for the IPapplications in which a best-effort QoS (i.e., the primary service of IP) issufficient. The ABR service has been defined to improve service provided tobursty sources. In this service, a peak cell rate (PCR) and a minimum cellrate (MCR) are specified and the network allocates at least MCR to an ABRsource. The leftover capacity or unused capacity is shared fairly among allABR and then UBR sources. A guaranteed frame rate (GFR) has recentlybeen proposed by the ATM Forum which provides a minimum rateguarantee to VCs at the frame level and could enhance the UBR service [37].

Considering unavoidable delay and delay variation in mobile satellitenetworks, UBR and ABR seem to be the most practicable options for theimplementation of TCP/IP over ATM satellites. In particular, with UBRrouters connecting through satellite ATM network can make the use of GFR


service to establish VCs between one another. In the case of the ABRservice, the network can maintain low cell loss ratio by changing the ACRthrough the usage of a rate-based closed-loop end-to-end feedbackcongestion control mechanism. In the case of satellite systems that sufferfrom long round-trip delay, the control loop can be segmented using virtualsource and virtual destination concept which results in less bufferrequirements. More discussions on the teletraffic issues and operation ofTCP over wireless channel can be found in [38-40].

Nevertheless, new algorithms are being developed in order to evolve thebest effort service of the Internet into a QoS-supported one. Among them,the RSVP (resource reservation protocol) is enabling reservation ofresources within IP network [41]. This protocol provides a way for everysender to establish paths for identified IP flows. With such protocols, itwould be possible to define guaranteed QoS services for the delivery of IPdatagrams within a fixed delay and no loss.

5.3 Perspectives and Applications

ATM and IP networks have several common structural viewpoints so thatthe concept of “classical IP over ATM” has been ignited in the IETFworking groups [42]. In particular, IP datagrams, IP addresses, IP routing, IPQoS, and IP multicast could be mapped onto the corresponding ATM cells,ATM addresses, ATM VC switching, ATM QoS, and ATM point-multipointfeatures [43]. Thus, in this model the IP layer is entirely mapped onto theATM layer in order to use the general applicability of the IP over any datalink layer. In spite of the disadvantage of having many task duplications inthe two layers in this approach, this example shows that IP and ATM havesufficient potentiality for the integration. Such an integration, however,could provide new services for the IP networks other than its traditional besteffort; i.e. the QoS-supported services. Nevertheless, some optimisation isrequired for an efficient integration of IP and ATM.

Since the ATM is based on cell switching and not the conventionalcircuit switching, the network resources can be utilised optimally. Theguaranteed-QoS, the variable-rate support, and low-cost ATM chips are alsoadditional advantages of ATM in implementation of high-speed broadbandwireless pipes within the base station and advanced mobile terminals. By theusage of wireless ATM technologies, including signalling, access control,and resource management, it is possible to achieve high data rate broadbandpersonal communication services in the order of 2 to 10 Mbps or more.Transmission of a number of IP flows on individual VCs according to theirsource and destination addresses for a better QoS has opened lots of researchactivity in the area of IP over ATM networks (e.g., [44-46]).


According to the discussions given already on QoS and trafficrequirements, broadband mobile satellites with ATM switching would besuitable for some applications but less appropriate for others [47]. Some ofpotential applications are shown in Fig. 20. Telnet, or remote computeraccess, which is categorised in interactive computing applications, is feasiblein satellite systems on low earth orbits. The LEO satellite systems canprovide a relatively prompt response to a telnet connection. Multicasting andbroadcasting of large data files, as in the case of information disseminationand video broadcast, could be efficiently supported by the mobile satellitenetworks. The primary reason for this is the global coverage and startopology of the satellite networks. Video broadcasting is usually sensitive todelay variation but not to the delay itself. The reason is that for a QoSplayback of video streams, it is necessary to have each frame being equallyspaced. Multicasting of image and video files through group mailing list andemail also could be feasible with the broadband satellite networks. Includedin multicast applications is the transmission of geographical positioninformation to be used in GPS and other navigation instruments.

Net conferencing and video conferencing are also ideal point-to-pointand multipoint applications of broadband satellites networks. The delaywould be a problem in transmission of high-speed and high-quality videoimages but the LEO satellite link can be comparable with other long-distancecommunication media. Applications which are not delay sensitive are themost promising services of the satellite networks. This includes bulk transferof data. In addition to the above applications, low bit rate voice and imagetransmissions, paging, short message services would be included in basicapplications of the broadband satellites.



6. CONCLUDING REMARKS

Next generation broadband satellite networks is being developed to carrybursty Internet and multimedia traffic in addition to the traditional circuit-switched traffic. These satellites provide direct network access for personalapplications as well as interconnectivity to the terrestrial remote networksegments. In a data transmission environment, the traditional circuitswitching method would be insufficient, as it cannot utilise the link capacityefficiently. ATM, on the other hand, can provide high quality of servicesupport at a high data rate and good channel utilisation. Moreover, becauseof the existence of the real time traffic such as voice and video transmission,satellites on non-geostationary orbits have found much attention in thedevelopment of broadband satellite networks. In this regard, the nextgeneration of broadband satellite networks would have the concept ofintegration of mobile satellites and the ATM networks.

With the exponential increase in the Internet and web-based applicationsin addition to the requirement of supporting mobility, wireless IP networkssuch as mobile IP and cellular IP, have been developed by the integration ofcellular networks and IP networks. ATM could provide the high-data raterequirement of the multimedia applications and thus, many works have beendone in integrating of the ATM and IP.

In this chapter, we examined mobile satellite, ATM, and IP technologiesas well as wireless ATM and wireless IP and their mutual integration forproviding high-speed wireless multimedia services. In addition, wediscussed the new concept of integration of the three technologies in order toprovide global mobility to the future multimedia terminals. Different typesof traffic to be handled through these networks and the quality of servicerequirements have been explained. The mutual integration and the new ideaof integration of the three technologies can be considered as a hierarchicalresearch activities, as shown in Fig. 21.



REFERENCES

[1] W. W. Wu, et al., “Mobile satellite communications,” Proceedings of the IEEE, vol. 82,no. 9, pp. 1431-1448, September 1994.

[2] R. L. Pickholtz, “Communications by means of low earth orbiting satellites,” 25th

General Assmebly of International Union Radio Science (URSI), Lille, France, 1996.[3] J. V. Evans, “Personal satellite communications systems,” The Radio Science Bulletin,

URSI, no. 290, pp. 8-15, September 1999.[4] F. Abrishamkar and Z. Siveski, “PCS global mobile satellites,” IEEE Communications

Magazine, vol. 34, no. 9, pp. 132-136, September 1996.[5] M. Werner, et al., “Analysis of system parameters for LEO/ICO-satellite

communication networks,” IEEE Jour. Selec. Areas Commun., vol. 13, no. 2, pp. 371-381, February 1995.

[6] E. Del Re, “A coordinated European effort for the definition of a satellite integratedenvironment for future mobile communications,” IEEE Communications Magazine,vol. 34, no. 2, pp. 98-104, February 1996.

[7] B. Miller, “Satellites free mobile phones,” IEEE Spectrum, vol. 35, no. 3, pp. 26-35,March 1998.

[8] A. Jamalipour, Low Earth Orbital Satellites for Personal Communication Networks,Norwood, MA: Artech House, 1998.

[9] S. R. Pratt, et al., “An operational performance overview of the IRIDIUM low earthorbit satellite system,” IEEE Communications Surveys, Second Quarter 1999.

[10] F. Adachi and M. Sawahashi, “Challenges in realizing the multimedia mobilecommunications era: IMT-2000 and beyond,” Personal, Indoor and Mobile RadioCommunications Conf. (PIMRC '99), Osaka, Japan, 1999.

[11] W. Stallings, ISDN and broadband ISDN with frame relay and ATM, 4th ed., UpperSaddle River, NJ: Prentice Hall, 1999.

[12] M. Sexton and A. Reid, Broadband networking-ATM, SDH and SONET, Norwood,MA: Artech House, 1997.

[13] R. R. Bhat, K. Rauhala, eds., “Draft baseline text for wireless ATM capability set 1specifications,” BTD-WATM-01, ATM Forum, December 1998.

[14] B. Kraimeche, “Wireless ATM: Current standards and issues,” IEEE WirelessCommunications and Networking Conference (WCNC '99), New Orleans, 1999.

[15] C. K. Toh, et al., “Emerging and future research directions for mobile wireless ATMnetworks,” IEEE Wireless Communications and Networking Conference (WCNC '99),New Orleans, 1999.

[16] H. Nakamura, et al., “Applying ATM to mobile infrastructure networks,” IEEECommunications Magazine, vol. 36, no. 1, pp. 66-73, January 1998.

[17] R. J. Sanchez, et al., “Design and evaluation of an adaptive data link control protocolfor wireless ATM networks,” IEEE Global Telecommunications Conference(Globecom '98), pp. 2239-2244, Sydney, Australia, 1998.

[18] P. Chitre and F. Yegenoglu, “Next-generation satellite networks: Architectures andimplementations,” IEEE Communications Magazine, vol. 37, no. 3, pp. 30-36, March1999.

[19] I. Mertzanis, et al., “Protocol architectures for satellite ATM broadband networks,”IEEE Communications Magazine, vol. 37, no. 3, pp. 46-54, March 1999.

[20] M. Werner, et al., “ATM-based routing in LEO/MEO satellite networks withintersatellite links,” IEEE Jour. Select. Areas Commun., vol. 15, no. 1, pp. 69-82,January 1997.


[21] M. Werner, “ATM concepts for satellite personal communication networks,”Proceedings European Conference on Networks and Optical Communications (NOC'96), pp. 247-254, Heidelberg, Germany, 1996.

[22] S. Ray, “Network segment mobility in ATM networks,” IEEE CommunicationsMagazine, vol. 37, no. 3, pp. 38-45, March 1999.

[23] G. Dommety, M. Veeraraghavan, and M. Singhal, “A route optimization algorithm andits application to mobile location management in ATM networks,” IEEE Jour. Select.Areas Commun., vol. 16, no.6, pp. 890-908, August 1998.

[24] H. Uzunalioglu, “Probabilistic routing protocol for low earth orbit satellite networks,”IEEE International Conference on Communications (ICC '98), pp. 89-93.

[25] J. Chen and A. Jamalipour, “An improved handoff scheme for ATM-based LEOsatellite systems,” Proceedings of the 18th AIAA International Communication SatelliteSystems Conference, Oakland, CA, April 2000.

[26] W. Stallings, Data and Computer Communications, 6th ed., Upper Saddle River, NJ:Prentice Hall, 2000.

[27] C. E. Perkins, “IP mobility support,” IETF RFC 2002, October 1996.[28] J. D. Solomon, Mobile IP–The Internet unplugged, Upper Saddle River, NJ: Prentice

Hall PTR, 1997.[29] C. E. Perkins, Mobile IP–Design, principles and practice, Reading, MA: Addison

Wesley Longman, 1998.[30] A. Seneviratne and B. Sarikaya, “Cellular networks and mobile Internet, ” Computer

Communications, vol. 21, Elsevier Publishers, pp. 1244-1255, 1998.[31] D. Chalmers and M. Sloman, “A survey of quality of service in mobile computing

environments,” IEEE Communications Surveys, Second Quarter 1999.[32] X. Xiao and L. M. Ni, “Internet QoS: A big picture,” IEEE Network, pp. 8-18,

March/April 1999.[33] R. Guerin and V. Peris, “Quality-of-service in packet networks: basic mechanisms and

directions,” Computer Networks, vol. 31, Elsevier Publishers, pp. 169-189, 1999.[34] A. Iera, A. Molinaro, and S. Marano, “Adaptive QoS for multimedia applications in

personal communication networks,” Personal, Indoor and Mobile RadioCommunications Conf. (PIMRC '99), Osaka, Japan, 1999.

[35] R. Goyal, et al., “Traffic management for TCP/IP over satellite ATM networks,” IEEECommunications Magazine, vol. 37, no. 3, pp. 56-61, March 1999.

[36] P. Tran-Gia and K. Leibnitz, “Teletraffic models and planning in wireless IP networks,”IEEE Wireless Communications and Networking Conference (WCNC '99), NewOrleans, 1999.

[37] I. Andrikopoulos, et al., “Providing rate guarantees for Internet application traffic acrossATM networks,” IEEE Communications Surveys, Third Quarter 1999.

[38] F. Anjum and L. Tassiulas, “An analytical model for the various TCP algorithmsoperating over a wireless channel,” IEEE Wireless Communications and NetworkingConference (WCNC '99), New Orleans, 1999.

[39] R. Prakash and M. Sahasrabudhe, “Modifications to TCP for improved performance andreliable end-to-end communications in wireless networks,” IEEE WirelessCommunications and Networking Conference (WCNC '99), New Orleans, 1999.

[40] D. Grillo, et al., “Teletraffic engineering for mobile personal communications in ITU-Twork: The need to match practice and theory,” IEEE Personal Communications, pp. 38-58, December 1998.

[41] L. Zhang, et al., “RSVP: A new resource reservation protocol,” IEEE Network, vol. 7,no.5, September 1993.

[42] M. Laubach, “Classical IP and ARP over ATM,” IETF RFC 1577, January 1994.


[43] E. Guarene, P. Fasano, and V. Vercellone, “IP and ATM integration perspectives,”IEEE Communications Magazine, vol. 36, no. 1, pp. 74-80, January 1998.

[44] J. Hu, “Applying IP over wmATM technology to third-generation wirelesscommunications,” IEEE Communications Magazine, vol. 37, no. 11, pp. 64-67,November 1999.

[45] J. Aracil, D. Morato, and M. Izal, “Analysis of Internet services in IP over ATMnetworks,” IEEE Communications Magazine, vol. 37, no. 12, pp. 92-97, December1999.

[46] M. A. Labrador and S. Banerjee, “Packet dropping policies for ATM and IP networks,”IEEE Communications Surveys, Third Quarter 1999.

[47] D. P. Connors, B. Ryu, and S. Dao, “Modeling and simulation of broadband satellitenetworks-Part I: Medium access control for QoS provisioning,” IEEE CommunicationsMagazine, vol. 37, no. 3, pp. 72-79, March 1999.


ABOUT THE AUTHOR

Abbas Jamalipour is a Senior Lecturer in the School of Electrical andInformation Engineering at the University of Sydney, Australia, where he isresponsible for teaching and research in data communication networks andsatellite systems. He received his Ph.D. in Electrical Engineering fromNagoya University, Japan, in 1996. He was an Assistant Professor at NagoyaUniversity before moving to Sydney. His current areas of research includedata communication and ATM networks, mobile IP networks, mobile andsatellite wireless communications, traffic and congestion control, switchingsystems, and switch design. He is the author of the first technical book onLEO satellites, entitled Low Earth Orbital Satellites for PersonalCommunication Networks published by Artech House, Norwood, MA, 1998.He has served as the Registration Chair at the 1998 IEEE GlobalTelecommunications Conference (GLOBECOM ‘98) held in Sydney. He is aSenior Member of the IEEE and an organizing committee member of thejoint IEEE NSW Communications and Signal Processing chapter. He is therecipient of a number of technology and paper awards and the author formany papers in IEEE and IEICE Transactions and Journals as well as ininternational conferences.

Chapter 3

INFOCITY: PROVIDING QOS TO MOBILEHOSTSMobile Multimedia on the Wireless Internet

PATRICIA MORREALEStevens Institute of Technology, Hoboken, NJ, USA

Abstract: Future wireless networks will be integrated with existingwired networks. Together, this environment will compose amultimedia network infrastructure, providing advanced data,voice, and video services, which is referred to here as“InfoCity”. In this chapter, several emerging technologies,which might be used to provide the mobile multimediaservices needed in the event of such a technology integrationand convergence are presented. Careful consideration isgiven as to how these new technologies could best be used tooffer a state-of-the-art, networked “InfoCity”, as a solutionfor next generation distributed multimedia applications.InfoCity, as presented here, is envisioned as a wired andwireless co-existence environment, with seamless servicedelivery of full multimedia applications, regardless of theuser’s location and receiving device. Frame relay and ATMare presented as facilitating high-speed connections to afuture broadband architecture. In order to support the diverseservice needs of multimedia, Quality of Service (QoS) mustbe assured. Resource reservation protocol (RSVP) isconsidered as an example of the type of service-arbitrationtechnique which could be used to provide users with suchQoS assurance based on user need, rather than fair allocation.Finally, mobile IP is included as one approach to providingmobile host support in this new environment.

Keywords: multimedia network infrastructure, frame relay, ATM, QoS,RSVP, mobile IP


1. INTRODUCTION

The computer market has been the compelling force behind Internetdevelopment and technological growth. The market focus has been toconnect countless computers together in large business, government oruniversity communities. The computer market will continue to grow infuture decades due to expansion in new areas but growth will not beendlessly exponential. Saturation in the PC market, as demand movestowards integrated services, will result in this deceleration.

New markets will emerge as mobile computing and networkedentertainment becomes more and more ubiquitous. The possibility that everyTV will become an Internet host is not far away. The device control marketwill play an important role in the future. The electronic network-control ofeveryday devices – such as lighting equipment, heating and cooling motors,home appliances, which are today controlled via analog switches consumingimportant amounts of electrical power – will bring enormous futureopportunities. The potential of these markets is huge and requires simple,robust and easy to use solutions.

In this context it is imperative to imagine “InfoCity” as a geographicallocation containing a state of the art multimedia network that facilitates theintegration of new interactive applications as e-shopping and video ondemand with classical data services as fax and email and TV. This InfoCityis approaching reality, as more and more residential users network theirhome environments. This complements the wiring of the officeenvironments, which has taken place in previous years. The interconnectionof these sites, both residential and industrial, either by wired or wirelessmeans, would result in the InfoCity, a vision of the network infrastructure ofthe future.

To make InfoCity a reality in the coming century, currently available andproposed technologies must be analyzed in order to design a cost-effectivesolution. Multiple WAN and LAN services should be managed usingcommon software tools, in order to deliver bundled and premium serviceswithout compromising performance or scalability. The InfoCity architecturemust be efficient for low bandwidth networks, such as wireless, whileincorporating high-performance networks as ATM.

InfoCity: Providing QoS to Mobile Hosts 111

The Internet Engineering Task Force (IETF), ATM forum and otherstandards organizations have addressed all these issues in detail buttechnology standards no longer drive the data and telecommunicationsmarket. Customers do. Service providers are struggling to determine a suiteof protocols that match the requirements of today’s applications and alsomeets the needs of new emerging markets.

The new markets will create either a immense interoperable world wideinformation infrastructure, the InfoCity depicted here, based on open


protocols, or an interconnection of disjointed networks with protocolscontrolled by individual vendors.

In this paper we will analyze first WAN technologies as Frame Relay andATM, as they promise to provide high-speed connections in a futurebroadband architecture. This would be needed for InfoCity in order toprovide a WAN infrastructure for communication.

Speed alone won’t be sufficient in a future InfoCity infrastructure. It hasbecome clear in the last decade that users should not be treated equally butaccording to their needs or willingness to pay for a specific service.Resource Reservation Protocol (RSVP) – a QoS protocol - is one exampleof a service arbitration technique which is representative of this “pay-as-you-go” need. Therefore, RSVP is also considered here.

There is currently significant industry interest in providing wireless dataservices. In a global economy people seek to have Internet access andnetworking resources whenever and wherever they are. This can be easilyachieved with laptops and pocket devices such as Windows CE, or PalmPilot. Mobility improves the quality of people’s lives and gives them newbusiness opportunities such as: on the move collaborative andcommunication tools, access to corporate databases on the field, andlocation-based services.

In wireless networks mobile hosts should be able to enjoy interruptiblenetwork connectivity and different quality of services just like wirednetworks. To ensure the mobile hosts with interruptible network accessmobile IP schemes were developed. Mobile IP is also considered here.

Rather than presenting a specific implementation, this paper analyzes allthe above protocols and discusses the merits of proposed modifications inorder to offer a cost-effective, robust solution, such as that which might beused in the InfoCity of the future.

2. MULTIMEDIA APPLICATIONS

Consider networking applications whose data contains continuouslyevolving content, such as audio and video content. These will be called“multimedia networking applications”. Multimedia networking applicationsare typically highly sensitive to delay; for a given multimedia networkapplication, packets that incur more than an x second delay are useless,where x can range from a 100 milliseconds to five seconds. On the otherhand, multimedia distributed applications are typically loss tolerant, asoccasional loss only causes occasional glitches in the audio/video playback.Often these losses, which are local, not global, can be partially or fullyconcealed. Thus, in terms of service requirements, multimedia applicationsare diametrically opposed to fixed-content applications: multimedia


applications are delay sensitive and loss tolerant whereas the fixed-contentapplications are delay tolerant and loss insensitive.

3. WHERE ARE THE CHALLENGES?

Demand for increased bandwidth is coming from both the residential andbusiness community. Gaming web sites and interactive multimedia e-shopping are gaining rapid acceptance by consumers. The number of peopleaccessing these sites and the volume and density of the offered applicationsis expanding faster than ever.

A new generation of high-performance low-cost PCs is coming out withfree Internet connectivity as an initial inducement to try the hardware. Thecore Internet is changing to a high-speed backbone to reduce congestion.This change has been so significant, an “Internet 2” structure has beenproposed and developed in the U.S. to, once again, develop researchcommunity isolation, from the more mundane and ordinary traffic on theexisting Internet. This is discussed in detail in a following section.

In an expanding global economy, as corporations became geographicallydisperse, users are increasingly less happy with WAN bottlenecks and theylook for near-LAN speed for all their connection within the enterprisenetwork.

The content of the information exchanged between major services of anenterprise is driving companies to consider high-speed solutions.

We see over the last years an increasing demand in Tl services for LANinterconnections and for high-speed Internet at 64kbps, 128 kbps orfractional Tl, as well as for DS-3 (44.636 Mbps). The majority of users needT-l and greater services for running multi-applications using a commonbandwidth.

ATM has been designed specifically for broadband distributedmultimedia applications; therefore it will support the needs of the wide-bandservices. Frame relay will remain in the future a sub-Tl data networkingservice.

4. FRAME RELAY OR ATM

Internet Service Providers use Frame Relay to provide high performancecost-effective solutions to their customers. For mow, Frame Relay is thechoice for networks of E1/T1 services and below. Currently the major FrameRelay switch vendors as Cisco or Nortel, support speeds up to DS3 andsupport for OC-3 has come into the market in the last year.

Frame Relay is a circuit switching technology designed from the necessityto have a dial-on-demand service to handle multiple connections using asingle physical line. It alleviates the scalability problem of the leased lines


and it makes a better use of the line. In Frame Relay a subscriber site leasesa permanent dedicated line that is connected to a Frame Relay switch in atelephone central office. So logical connections are established to one ormore remote Frame Relay sites. The logical connections are called virtualcircuits and all share the same physical port on a router and the same leasedline and data service unit (DSU).

The cost of a leased line is proportional with its length. The cost of FrameRelay service is proportional with the required bandwidth. The speed of aFrame Relay service does not need to be the same for all the subscriber sites.A company can have a DS-3 connection for the headquarters office andseveral DS-1 circuits for branch offices. In conclusion Frame Relaysolutions have access rates at a lesser cost as leased lines so they are a goodeconomical solution.

Frame Relay is a technology deployed to provide the end user with aVirtual Private Network capable of supporting high-speed transmissionrequirements. A lot of functions performed by the classical data networks -as error correction and retransmissions are eliminated by taken intoadvantage of relatively error free transmission systems as optical fibers.

As the intelligence is build more and more into the end hosts, the FrameRelay network does not need to perform many QoS functions and themanagement operations are few. It is so called "bare-bone" approach thatresults in fast networks but places more responsibility on the end-usersystems and management entities.

The typical applications for frame relay networks are bursty data withhigh capacity requirements, client server database queries, broadcast videoemail and file transfer.

Frame Relay is not topology dependent but is currently implemented aspoint to point in a star scheme similar to the leased lines network. A veryimportant advantage of this network is the low cost. Let's take an example.The computer at the headquarter site that receives the traffic from manypoint to point connections is set initially as a percentage of the aggregateremote sites traffic. If the congestion becomes a burden at the headquarterlocation, another DS1 facility should be installed so the access rate to themain site is increased from DS1 to DS3 in increments of Tl (1.544Mbps) orEl (2Mbps).

Frame Relay is used as a high-speed technology to interconnect LANs. Ithas explicit flow control and the sequencing of data is the userresponsibility. A node is informing the network of a problem and does nottake any action as ceasing the transmission. Congestion and flow control areoptional and some vendors have not implemented them yet.

ATM is a circuit-oriented technology that was designed to providedifferent QoS levels to transport any type of user traffic: data, voice, andvideo. ATM traffic is transported into fix-length cells identified by a virtual


circuit identifier (VCI) in the header of the cell. The VCI is used to route thecells through the network. ATM does not support error correctionoperations. Only the signaling traffic can be retransmitted.

ATM was designed to deliver at least OC-1 (155 Mbps) but went down tosupport T3 and Tl widely in demand. Tl ATM has too much of overheadand does not justify the cost when compared to Frame Relay for datanetworking applications. But for applications that mix different types of data- voice, video - the most important issue are the QoS guarantee and thebandwidth optimization. The overhead is irrelevant in this case.

ATM can be seen as a solution to a core transport mechanisminterconnecting different types on networks. There are 2 methods tointerconnect ATM and Frame Relay. The first one is tunneling the FrameRelay frames through the ATM network. The variable Frame Relay framesare segmented and encapsulated into the payload of the ATM cells withoutdisturbing the Frame Relay header. The increased overhead is paid back byhigher switching speeds. The second method would be a translation servicebetween ATM and Frame Relay. The frame header information is mappedinto an ATM header so an ATM device can talk to a frame relay device. ButATM promises much more than only core network interconnections.

Frame Relay has a place in InfoCity as a cheap solution for providing Tland up to T3 speeds in access networks. But this place could be challengedby ATM soon.

5. IP OVER ATM

Exploring the demand for data networking and Internet services,researchers reached the conclusion that the classical solution of deployingdifferent networks and trying to interconnect them is very costly andresource intensive.

What if someone can deliver all the required services from the sameinfrastructure? Many efforts were put in designing integrated IP and ATMsolutions - LAN emulation [1], classical IP over ATM [2], address resolution[3], next-hop resolution protocol [4] - with the hope that the results willdeliver well known and widely used IP application taking advantage of theATM speeds. When high - speed multimedia application are offered as an IPover ATM service the real network topology is hidden from the IP layer andwe face inefficiency and duplication of functionality.

In case of LAN emulation for example, the higher layer (IP) isencapsulated into the appropriate LAN MAC packet format and then is senton the ATM network. The advantage is that no modifications are necessaryto the higher protocols to cooperate with ATM.

In classical IP over ATM both IP and ATM need their own routingprotocols. There is duplication in the maintenance and management


functions. New management functions are required for address translationand data format conversion. The problems are hard to identify and locate [5].When one tries to interconnect IP router using ATM switches the number ofneighboring routers in a cloud can become very large and the nest-hoprouting table is expanding uncontrollably. For N routers the update processis increased with N2. If a failure occurs a complex protocol should beemployed to deal with the recovery [6].

It is desirable to keep the best part of both technologies and try to unifythem in a common service architecture than to try to superimpose them.ATM offers link bandwidth scalability and speed and switching capacity forvarious types of traffic. TCP/IP has been imposed as the widest spread dataprotocol because of its connectionless nature that brings simplicity,scalability and robustness to failures. IP makes no assumption of theunderlying network beyond the capacity of forwarding a data gram packet tothe destination. No state has to be maintained in the intermediary routers forindividual connections since the destination address in contained in theforwarded packet. From here the robustness to failures.

Some researchers have proposed to implement IP directly on top of theATM hardware using the flow concept [7]. A flow is a sequence ofdatagrams that follows the same route through the network and receivessame service policies at intermediary routers. Flow carrying real time trafficwould be mapped to the ATM connections. Short duration flows as databasequeries wil l be carried by classical IP forwarding between routersinterconnected using an ATM network. Establishing an ATM connection forevery IP flow would impose a big burden on the ATM signaling protocol.

The ideal solution will combine the easy to use and scalability of IP withthe high QoS, speed and performance of the ATM. It has to provide virtualprivate networks that let users run critically IP-based applications securely.

A variety of value-added IP and ATM services should be able to deployeasily to meet customers needs. The service providers networks must beintegrated with a range of already installed technologies to cut the cost ofconnecting new users.

If past attempts in offering simultaneous IP/ATM services resulted intunneling, most recent ones as multiprotocol label switching [8] enable amore integrated and expandable solution.

In multiprotocol label switching the core network switches provideautomatically set up calls and dynamically switch IP traffic over ATRMnetwork in real-time, and the multiservice switches provide subscriberinterfaces for multiple data network service type: frame relay, ATM cellrelay, etc. By adapting each service to a common protocol and applyingtraffic management functionality, a multiservice switch can significantlyincrease the network performance [8].


ATM could provide the core network for InfoCity. Even though the costof ATM is still high for small private networks, it is definitely the futurebackbone solution. As a core technology ATM will be integrated inmultiprotocol switches capable or interconnecting different types of networkin a cost –effective manner.

Beside using ATM as a core technology tor InfoCity, LAN emulationcould be used in its small business private networks for videoconferenceapplications or for on-demand video services. IP services and well known IPapplications could be run over a high-speed ATM network to provide forexample, distance learning applications before different university campusesof InfoCity.

6. HIGH SPEED INTERNET: BACKBONE NETWORK SERVICENEXT GENERATION INTERNET AND INTERNET2

An example of IP over ATM implementation is Backbone NetworkService (vBNS) [9].

MCI's very-high performance network is designed to serve the researchcommunities that require superior performance than available commercialnetwork. It runs on an OC-12 (622.08 Mbps) and OC-48 SONET pipes andaims to provide a test-bed to new Internet technologies and services. Withmore than 91 connections throughout United States, 4 supercomputercenters, vBNS is a promising infrastructure that provides research anddevelopment institutions with high-speed data connectivity. Within vBNScongestion is not so much of a problem comparing with Internet because alimited number of research institutes, each with a separate Internetconnection, use vBNS only to test advanced networking applications andexperiments.

Next Generation Internet [10] is implemented by DARPA to provide anetwork that is at least 100 or even 1000 times faster than today's Internet.Intenet2 is a collaborative project for more than 120 universities that aims tofacilitate the development of state of the art distributed applications usingnetworks as vBNS and NASA Research and Education Network.

With this infrastructure in place (Figure 2) the question is which are thecore services offered in order to be tested and improved by the researchcommunity?


Beside best effort IP, switched virtual circuit logical IP subnets and pointto point permanent virtual circuits, an ongoing effort is concentrated towardsreserved bandwidth services because the traffic pattern - some bursty flowsand delay sensitive applications - require a dynamic and efficient allocationof the bandwidth on a session per session basis.

Resource Reservation Protocol (RSVP) may be implemented to triggerthe reserved bandwidth service.

There is a growing impression in networking research world in the lastyears that connections to the Next Generation Internet, Internet 2, or vBNSis vital for a place in future networking research. Having a broadband pipefor classical “best-effort” IP traffic is not a suitable solution for InfoCity dueto the rate at which developed applications are growing and bandwidthrequirements are exploding in recent years. Rather, in InfoCity, it would bebetter to concentrate our efforts toward technologies that bring controlledquality and can differentiate between users in term of resource allocation andmanagement.


7. RSVP – BRINGING QoS TO AN IP NETWORK

The RSVP protocol allows applications to reserve bandwidth for theirdata flows (see Figure 2). It is used by a host, on the behalf of an applicationdata flow, to request a specific amount of bandwidth from the network.

The routers use RSVP to forward bandwidth reservation requests. Toimplement RSVP, RSVP software must be present in the receivers, senders,and routers. The two principle characteristics of RSVP are:

1. It provides reservations for bandwidth in multicast trees (unicast ishandled as a special case).

2. It is receiver-oriented, i.e., the receiver of a data flow initiates andmaintains the resource reservation used for that flow.

RSVP is sometimes referred to as a signaling protocol. By this it is meantthat RSVP is a protocol that allows hosts to establish and teardownreservations for data flows. The term signaling protocol comes from theterminology of the circuit-switched telephony community.

RSVP operates on top of IP, occupying the place of a transport protocol inthe protocol stack. RSVP does not transport application data and it iscomparable to a control protocol like ICMP or IGMP.

RSVP makes receivers responsible for requesting QoS control. A QoScontrol request from a receiver host application is passed to a local RSVPimplementation. The RSVP protocol caries the request to all the nodes onthe reverse path to the data source.


In truth, the RSVP reservation message does not simply reservebandwidth. Instead it contains a flowspec, which has three parts: a serviceclass, a Rspec and a Tspec. The service class identifies the type of QoS thereceiver desires from the network. Two service classes are in the process ofbecoming Internet standards: the Controlled Load service class [11], whichpromises applications that their packets will usually see no queuing delaysand minimal loss; and the Guaranteed QoS service class [12], whichprovides deterministic delay bounds to packets. The Rspec (R for reserved)defines the specific QoS, such as the fraction of lost packets the receiver isprepared to tolerate. The Tspec (T for traffic) describes the data flow,typically described in terms of leaky bucket parameters. RFC 2210 [13]describes the use of the RSVP resource reservation protocol with theControlled Load and Guaranteed QoS services. The RSVP protocol definesseveral data objects that carry resource reservation information but areopaque to RSVP itself. The usage and data format of those objects is givenin this RFC.

Path messages are another important RSVP message type; they originateat the senders location and flow downstream towards the receivers. Theprinciple purpose of the path message is to let the routers know on whichlinks they should forward the reservation messages. Specifically, a pathmessage sent within the multicast tree from a Router A to a Router Bcontains Router A's unicast IP address.

Router B puts this address in a path-state table, and when it receives areservation message from a downstream node it accesses the table and learnsthat it should send a reservation message up the multicast tree to Router A.

In the future some routing protocols may supply reverse path forwardinginformation directly, replacing the reverse-routing function of the path state.

Along with some other information, the path messages also contain asender Tspec, which defines the traffic characteristics of the data stream thatthe sender will generate. This Tspec can be used to prevent over reservation.

Through its reservation style, a reservation message specifies whether themerging of reservations from the same session is permissible. A reservationstyle also specifies from which senders in a session the receiver desires toreceive data. Recall that a router can identify the sender of a datagram fromthe datagram's source IP address.

There are currently three reservation styles defined: wildcard-filter style;fixed-filter style; and shared-explicit style.

Wildcard-Filter Style: When a receiver uses the wildcard-filter style in itsreservation message, it is telling the network that it wants to receive all flowsfrom all upstream senders in the session and that its bandwidth reservation isto be shared among the senders.

Fixed-Filter Style: When a receiver uses the fixed-filter style in itsreservation message, it specifies a list of senders from which it wants to


receive a data flow along with a bandwidth reservation for each of thesesenders. These reservations are distinct, i.e., they are not to be shared.

Shared-Explicit Style: When a receiver uses the shared-explicit style in itsreservation message, it specifies a list of senders from which it wants toreceive a data flow along with a single bandwidth reservation. Thisreservation is to be shared among all the senders in the list.

Shared reservations, created by the wildcard filter and the shared-explicitstyles, are appropriate for a multicast session whose sources are unlikely totransmit simultaneously. Packetized audio is an example of an applicationsuitable for shared reservations; because a limited number of people talk atonce, each receiver might issue a wildcard-filter or a shared-explicitreservation request for twice the bandwidth required for one sender (to allowfor over speaking). On the other hand, the fixed-filter reservation, whichcreates distinct reservations for the flows from different senders, isappropriate for video teleconferencing.

The reservation can be removed by senders with a PathTear message orby receivers with a ResvTear message. Alternatively they can stop sendingPATH and RESV messages and the reservation state times out.

RSVP will be an option in a future modernized IP networks. Right now itis supported by CISCO routers and some research versions of a RSVPdaemon are in tests. But few commercial applications really make use if it.So the protocol will have a limited use in InfoCity at the beginning.

8. MOBILE IP

In the traditional Internet Protocol the IP address identifies without anyambiguity the user’s location. This assumption simplifies the work in allrouters in the network since every time a router receives a packet with adestination host address, it only needs to look into its routing tables todetermine on which port the packet has to be sent to that destination. MobileIP allows a host to move and connect to different subnetworks in a waytransparent to higher layer protocols such TCP [14].

In mobile IP every user has two addresses. The first one is the permanenthome address and the second one is the temporary “care of” address. Thehome address is the normal IP address that points to the location where themobile user is found most of the time, this is also the subnetwork the user isregistered with. The care of address is an indication of the actual location ofthe user as it moves to different subnetworks. Following this mobile user wewill have a router called Home Agent (HA) in the home subnetwork that willkeep track of the mobile user temporary location. A router in the visitingsubnetwork is called the Foreign Agent and assigns the care of address to themobile user during as long as it visits the subnetwork. (Figure 4)


Mobile IP is a layer 3 technology that can be used with any link-layerdevice wired or wireless. The mobility support is provided using tunnelingfor data forwarding from permanent home network to the visiting network.

When a mobile host connects to a subnetwork it realizes that it is visitinga different subnetwork by listening to a periodic beacon signal transmittedby the foreign agent at that location. The mobile host initiates a registrationprocedure with the foreign agent that ends with an IP care of addressassigned to it. This IP address is also transmitted to its home agent. When apacket is sent to the mobile host it finally reaches the home agent, becausethe packet contains the permanent address as destination. The home agentencapsulates the original IP packet into another IP packet with the care ofaddress as destination and it transmits the packet again. The packet reachesthe foreign agent using normal routing since the intermediary routers seeonly the care of address. This is called tunneling. The foreign agentdecapsulates it and send the original packet to the mobile host. The time amobile user is registered with a foreign agent is volatile so the mobile hosthas to register periodically.

There are two operation modes defined by the protocol. In basic mode thehome agent responds to registration request of mobile nodes away fromhome network. After completing the registration process it forwards thereceiving packet to the foreign agent of the foreign network. In advancedmode, after performing basic mode operations, the home agent sends mobilebinding information to the network router of sender. The router keeps thisinformation during communication. After receiving binding informationfrom the home agent packets generated from the sender are directlyforwarded to the foreign agent of the mobile node without connection withhome agent.

The biggest issue facing Mobile IP is security. Strong authentication isneeded because the mobile node may be accessing corporate resources from


the Internet. Fortunately, standards are available for authentication. Forexample, FTP Software's Mobile IP implementation provides mutualauthentication of the mobile node and the home agent. The thornier securityproblem is one of traversing firewalls. This problem is twofold because itinvolves firewalls both at the home network and at the foreign network.Many firewalls rely on packet filtering to implement some or all of theirsecurity. If the mobile node is trying to communicate with the home agent orother hosts on its home network, the firewall there may reject datagramsfrom the mobile node because the datagrams have the source address of aninternal node, but appear on an external port.

A similar problem can occur at the foreign network. Many firewalls areconfigured so that they will not pass datagrams from inside a network to theInternet if the source address differs from what is expected. The intent is toprevent the internal network from being a haven for malicious usersspoofing source addresses and perpetrating mischief on the Internet.Unfortunately, the Mobile IP node's address does not belong on the internalnetwork, so its transmissions may be blocked.

Although these firewall problems usually wil l arise during Internetcommunication, they also may come up in corporate Intranets as companiesincrease their use of internal firewalls. For instance, you may need toconfigure the firewall protecting the home network to allow ICMPdatagrams addressed to the home agent, which will allow the mobile node toregister its new location.

9. INTEGRATING RSVP WITH MOBILE IP – WHERE ARE THECHALLENGES?

There are some enhancements to Mobile IP in the scope of combiningmobility with QoS. When one tries to put RSVP and mobile IP to worktogether some interesting questions appear. Is it enough to perform anenhancement on both protocols to guarantee QoS to mobile users or theaccent should be put on the coordination between the two protocols? Thewell-known characteristics of the wireless environment – high non-stationary BER, less bandwidth, etc - have a great impact on both protocols.

When a mobile host moves to a different location all the packets in transitwill reach the old location and get lost. This situation may represent a non-desired long period of time in which the mobile host will stop receivingaudio or video stream. Here the basic hand-off protocols are not enough tohandle the situation and RSVP should be invoked to help during therecovery process.

In the original IETF draft for mobile IP every packet sent to the mobilehost had to go through the home agent. This leads to a non-optimaltriangular routing. The situation increases not only the delay end-to-end but


also generates wasted bandwidth in non-optimal paths that cannot be usedfor other connections. For this reason optimal routes play a key point inproviding QoS to mobile hosts. Here the challenge is how to manage routeoptimization while supporting QoS as well. In Mobile IP route optimizationis done by transmission of short messages among source, home agent (HA),foreign agent (FA) using normal best-effort IP transmissions. This might befast enough for data connections but for real time voice and videoapplication it might not be satisfactory. Another important issue would be tosee when is better to call RSVP: before, during, or after optimization.

10. RSVP AND MOBILE IP ENHANCEMENTS

RSVP was designed to work in wired networks. When one tries to extendit to include mobile hosts, some issues have to be solved. First, there is thefast movement of the mobile host. Instead of making a fresh reservation onthe move, it is better to provide the mobile host with the capacity to make anactive reservation in one foreign subnetwork and many passive ones inneighboring subnetworks.

The challenge is that the mobile nodes change their location up to onceper second. Every time a mobile node changes its location a new reservationshould be made.

Another problem appears because of the Mobile IP encapsulation. RSVPmessages have a descriptor of the flow for which the reservation isrequested. The flow descriptor contains a list of packets headers fields that arouter can use to distinguish the packets of the real-time flow which hasrequested QoS from a other data flows. When the packets travelencapsulated in the tunnel between the home agent and the foreign agent, theintermediary routers will treat all the flows as best-effort because theycannot see that a reservation is being made for a specific real-time flow.There must be a way to inform intermediary routers which flow shouldaccept which kind of services.

IP in IP tunnels are a widespread mechanism to transport datagrams in theInternet. Tunnels are used to route packets through portions of the networkwhich do not implement a desired protocol (Ipv6 for example) or to enhancethe behavior of the deployed routing architecture (e.g. Mobile IP).

There are many IP in IP tunneling protocols. To deploy RSVP with themaximum flexibility it is desirable for tunnels to act as RSVP-controllablelinks within the network.

A tunnel can participate in an RSVP aware network in three ways: as alogical link that may not support resource reservation or QoS control at all,as a logical link that may be able to allocate some resources specifically toindividual data flows or as a logical link that may be able to make


reservation for individual end-to-end data flows. The first one is called thebest-effort link, the second one is a configured resource allocation over thetunnel. For the last one the tunnel reservations are created and torn downdynamically as end-to-end reservation come and go.

When the two end points of a tunnel are capable of supporting RSVP overtunnels the proper resources have to be reserved along the tunnel. Dependingon the requirements of the situation one might want to have one client’s dataflow placed into an aggregate reservation (as in second type of tunnel) or, ifpossible, to have a new separate reservation for the data flow.

Currently RSVP signaling over tunnels is not possible. RSVP packets getencapsulated with an outer IP header and do not carry Router Alert option,making them invisible to RSVP routers between the two points of the tunnel.It is impossible to distinguish between packets that use reservation and thosewho don’t, or to differentiate the packets belonging to different RSVPsessions while they are in the tunnel.

Some enhancement should be added to IP tunneling to allow RSVP tomake reservations across all IP in IP tunnels. If packets require reservationwithin the tunnel there has to be some attribute other than the IP addressvisible to the intermediate routers, so that the routers may map any packet toan appropriate reservation. The solution chosen was to encapsulate such datapackets with an UDP header and to use UDP port numbers to distinguishbetween packets of different RSVP reservations.

A procedure to map the end-to-end session to the tunnel session isdetailed in [15]. The tunneling introduces new security issues as the need tocontrol and authenticate access to enhanced quality of service. Thisrequirement is discussed in RFC 2205 [16].

The IP in IP encapsulation is not sufficient because it does not provide atransparent way to classify the tunnel packets. The IP in UDP solves theproblem but the security issues are not negligible. The EncapsulationSecurity Payload and an authentication header can be used but theencryption can change the location of the header.

RSVP can handle slow changes in the established paths due to variationsin the topology or congestion conditions. A mobile node could provide anexplicit indication to a receiver that it has changed its location and thereceiver should reserve resources along a new path. It could use mobile IPregistration (the home agent will perform reservation between itself and themobile host).

When a mobile host moves into a new foreign subnetwork there should besome resources already reserved for this host with a high probability.Although this reservation may not satisfy the largest resource reservationrequirement of this mobile host, a partial resource reservation should be ableto satisfy the basic service requirement of the mobile host. Also the latency


between requesting a service and getting served with some degree of serviceshould be minimized.

Mobility will be a key functionality in InfoCity not only in corporatebusiness networks but also in access networks, as the management functionswill benefit enormously from it. Mobile IP is a new technology and it is notwell integrated in commercial products. Only few available products asRomIn from IKV++ (http://www.ikv.de/products/index.html) haveintegrated Mobile IP. These products should be considered as they can beintegrated with classical IP applications to provide mobility in InfoCity’sdistributed business networks

11. CONCLUSIONS

We have presented and discussed different candidate technologies andprotocols that could be integrated into a future Infocity state-of-the-artnetwork, as part of a multimedia network infrastructure, supporting bothmobile and stationary users. No implementation solutions were proposed;rather, a general view into the problem was provided, to identify where theoutstanding issues are that remain to be solved.

REFERENCES

[1] LAN Emulation over ATM Version 2 - LNNI Specification, ATMforum af-lane.0l12.000, Feb 1999

[2] RFC 1577, “Classical IP and ARP over ATM,” 1/20/94. Update inhttp://www.internic.net/internet- drafts/draft-ietf-ion-classic2-01 .txt.11/26/1996 and "Classical IP and ARP over ATM", 04/22/1997,http://www.internic.net/ internet-drafts/draft-ietf-ion-classic2-02.txt

[3] "NHRP Protocol Applicability Statement", D. Cansever, 07/25/1997,draft-ietf-ion-nhrp-appl-02.txt

[4] Cisco Implementation of NARP at http://www.cisco.com/univercd/cc/td/doc/product/ software/ios111 /mods/4mod/ 4c book/4cip.htm#xtocid 1050321

[5] Newman, P. “ATM local area networks” IEEE Communications,March 1994

[6] Liping An, N.Ansari, “Traffic over ATM networks with ABR Flow andcongestion control”, IEEE Selected Areas in Communications, August1997

[7] P.Newman, “ IP switching – ATM under IP”, IEEE/ACM Transactionson Networking, vol.6 no.2, April 1998

[8] E.Roberts “Getting a Handle on Switching and Routing”, IBM whitepaper Oct 1997


[9] K.Thompson, G.J. Miller, and R. Wilder “Performance Measurement onthe vBNS”, Interop ’98 Engineering conference

[10] http://www.ngi.gov/pub[11] RFC 2211 J. Wroclawski, “Specification of the controlled load network

element service”, IETF Network Working Group, September 1997[12] RFC 2212 Shenker, et al, “Specification of Guaranteed Quality of

Service”, IETF Network Working Group, September 1997[13] RFC 2210 J. Wroclawski, “The use of RSVP with IETF Integrated

Services”, IETF Network Working Group, September 1997[14] C. Perkins, “IP Mobility Support”, RFC2002, October 1996[15] A. Terzis, J. Krawczyk, J. Wroclawski, and L. Zhang, “RSVPOperation

over IP Tunnels”, Internet Draft, draft-ietf-rsvp-tunnel-02.txt, February1999.

[16] R. Braden, L. Zhang, S. Berson, S. Herzog and S. Jamin, “ResourceReservation Protocol (RSVP) – Version 1 Functional Specification”,

ABOUT THE AUTHOR

Patricia Morreale is an Associate Professor in the Department ofComputer Science and Director of the Advanced TelecommunicationsInstitute (ATI) at Stevens Institute of Technology, Hoboken, NJ. Herresearch interests include network management and performance, wirelesssystem design and mobile agents.

She received her Ph.D. in Computer Science from Illinois Institute ofTechnology, Chicago, IL in 1991. She is co-editor of theTelecommunications Handbook (1999) and the AdvancedTelecommunications Handbook (2000), both published by IEEE Press, andholds a patent in the area of real-time information processing. She has morethan 25 journal and conference publications. She is an editorial boardmember of the Journal of Multimedia Tools and Applications (KluwerAcademic).

She has served on the technical program committees of severalworkshops and conferences, and has organized and chaired sessions at IEEEconferences. She was a Guest Editor of the IEEE Communications SpecialIssue on Active, Programmable, and Mobile Networks. She will be Vice-Chair, IEEE INFOCOM 2002. Morreale is a member of Association forComputing Machinery and Senior Member of Institute of Electrical andElectronic Engineers.

RFC2205, September 1997


Chapter 4

ASSISTED GPS FOR WIRELESS PHONELOCATION – TECHNOLOGY AND STANDARDS

BOB RICHTON, GIOVANNI VANNUCCI, AND STEPHEN WILKUSLucent Technologies/Bell Laboratories, Whippany, NJ, USA

Abstract: Many approaches have been advanced for locating the geographic position ofwireless phones, both for emergency response purposes and for emerginglocation-based services. Depending mainly upon the services envisioned andthe particulars of the air interface, one or another approach appears appealing,.Increasingly, the assisted-GPS approach is gaining recognition as the approachthat can best meet all requirements. Among the important requirements arethose mandated in the USA by the Federal Communications Commission(FCC) for Enhanced 911. Assisted GPS provides the best accuracy for use inlocation services. It is rooted in the suitability of wireless networks to providedata over the air to enable fast acquisition and lower power consumption, aswell as provide indoor operation—capabilities that conventional GPS cannotprovide. The assisted-GPS approach promises to enable a new industry oflocation-based services and important new safety measures.

Keywords: cellular systems, geolocation, standards, wireless location, assisted-GPS, FLT,E-OTD, observed time difference of arrival, IS-801, WAG, FINDS.


1. INTRODUCTION

By now, the need for wireless geolocation* is well established. The needis driven in the USA largely by the Federal Communications Commission(FCC), which ruled that the location of a Mobile Station (MS) calling 911must be provided to the Public Safety Answering Point (PSAP). Besidesemergency services, many other geolocation-based applications have beendescribed[1]. Among these applications are:

– Location sensitive billing: Enabling price differentials based onthe caller location.

– Location-based information services: Providing directions to findrestaurants, hotels, cash machines, gas stations, etc.

– Network optimization: Used to improve daily operations of awireless network

– Fleet management and asset tracking: Giving ability to locate(FCC), which ruled that the location of a Mobile Station (MS)calling 911 vehicles, personnel, or property to more efficientlymanage operations.

Many more applications have been described[2]. Realizing this need,people unfamiliar with the details of wireless technology often think thatGlobal Positioning System (GPS)[3] could be combined with Mobile Stationsto support geolocation. However, combining GPS and mobile in astraightforward way turns out to be unsuitable for wireless applicationsbecause GPS:

– Does not work in buildings or shadowed environments, includingurban canyons

– Is too slow for some services, particularly for emergency use– Is too costly and too bulky to be included in a modern mobile

terminal– Drains common mobile station batteries at an unacceptably high

rate– Despite recent, dramatic advances in GPS technology, these

problems persist. However, “Assisted GPS” overcomes all theselimitations while providing better accuracy than any terrestrial-based approach, or conventional, stand-alone GPS.

This paper describes the basic technology of assisted GPS, whichcircumvents the problems listed above and achieves high accuracy atreasonable cost. The technique exploits the availability of the bi-directionalwireless link to divide the job of determining the mobile phone's location

* The term geolocation is used here to refer specifically to “location on the earth” (as latitudeand longitude) as opposed to “location within the network,” which is the more commonmeaning of the word “location” when used in the context of wireless communications.

Assisted GPS for Wireless Phone Location 131

between the phone itself and the wireless network. This results in a level ofperformance that exceeds that of a conventional GPS receiver, even ascomplexity is reduced. Assisted GPS is, therefore, neither purely a network-based, nor a handset-based solution, and is sometimes called a hybrid. Notethat for purposes of the FCC mandate, however, it would be consideredhandset-based because it requires new handsets.

In assisted GPS, the mobile includes a "partial" GPS receiver that iscontrolled (a better term might be "primed") by the network (which we'llrefer to as the "server") and the received GPS signal needs only minimalprocessing in the phone before being retransmitted to the server over thewireless link. The server can then perform the location computation takingadvantage of additional information (such as terrain and network data) notgenerally available to a GPS receiver.

There are two reasons why assisted GPS works so well: a) The assisted-GPS server has its own GPS receiver and, therefore, already knows withgreat accuracy what signals the mobile phone's GPS receiver is receiving;and b) the wireless system already has a reasonable estimate of the phone'slocation. These two elements enable the assisted-GPS receiver to detect areceived signal that is orders of magnitude weaker than is required byconventional GPS techniques, and to do so in a fraction of a second.

Note that while, on the surface, this technique sounds similar to the well-known Differential GPS (DGPS) technique, the underlying principles arecompletely different. The DGPS technique does not provide any improvedability to detect GPS signals under low Signal-to-Noise Ratio (SNR)conditions; it only improves the accuracy of the GPS location estimate. Ofcourse, the assisted GPS technique can (and should) also employ DGPSmethods in its location estimates, so that the expected accuracy of assistedGPS will be equivalent to that of DGPS.

1.1 Some Previous Works

Although assisted GPS has been widely described in standards bodiesand wireless industry meetings, few papers on the technology appear to havebeen published. NAVSYS Corporation’s description of TIDGET may havepresaged assisted GPS[4]. Another step towards GPS-mobile phoneintegration was described by DiEsposti et.al.[5] early in 1998, with perhapsthe first public description of assisted GPS coming from Moeglin andKrasner[6] later that year. A recent article by Norman Krasner provides avery readable account of assisted GPS and an interesting implementation ofit in the handset.[7]. Two other papers from the ION ’99 Conference, onefrom L. J. Garin et. al. [8] and one from A. J. Pratt,[9] cover architectureissues concerning assisted GPS.


2. FUNDAMENTAL CONCEPTS

2.1 The Global Positioning System (GPS)

A detailed discussion of GPS is beyond the scope of this document; wesimply summarize a few essential features that are needed for understandingassisted GPS. Additional details and parameters of GPS will be introducedas necessary. For a thorough description of GPS and additional referencessee [3,10].

The heart of GPS is a constellation of 24 satellites orbiting the earth at analtitude of about 20,000 km. The orbits are chosen to ensure that there arealways at least four satellites visible (i.e., sufficiently high above thehorizon) from any place on earth at any time. The satellites act like beacons,sending down radio signals that are carefully timed at the source in a pre-determined way. At a GPS receiver, signals from different satellites arrivewith different propagation delays, depending on the position of the receiver.The receiver makes an accurate measurement of the time of arrival of eachsignal, and the difference between the time of arrival and the (pre-determined and, therefore, known) time of departure yields the distance(range) to each satellite. Unfortunately, in most cases, the receiver does notinitially have a very accurate notion of time and the computed ranges willcontain a large error. Nonetheless, since the signals were perfectlysynchronized at transmission, the differences in their times of arrival arevery accurate and carry information about the mobile's position. Because ofthe inaccuracy of the measured satellite ranges, they are commonly referredto as “pseudo-ranges,” and the number of satellites required for a locationfix is four instead of three, as one must solve for four variables of latitude,longitude, altitude, and time.

To compute latitude, longitude, and altitude based on the pseudo-rangingmeasurements, the mobile must detect signals from at least four satellites†

and it must know the exact position of those satellites at the time the signalswere transmitted. For this purpose, each satellite also transmits a digital bitstream (at 50 bps) with precise information on the satellite's orbitalparameters (called ephemeris). An important feature of the GPS system isthat the same signal is used for both pseudo-ranging and for transmitting thedata. As we shall see, the fact that the signal carries data bits that are apriori unknown, limits the detectability of the signal. A key feature ofAssisted-GPS techniques is that this uncertainty is removed, resulting inmuch improved signal detectability.

† If altitude is already known, as is the case for a receiver that is known to be at sea level,three satellites are sufficient.


The satellites travel at high speed. The distance between satellite andreceiver may vary by more than one half mile per second. This results in adoppler shift in the signal carrier that may be as large as ±4500 Hz.

Each GPS satellite transmits several signals. The signal most commonlyused by civilian GPS receivers is the so-called C/A signal (C/A stands for"Coarse Acquisition") The C/A signal is a Direct-Sequence Spread-Spectrum signal with a chip rate of 1.023 MHz. Chip modulation is BinaryPhase-Shift Keying (BPSK) and the underlying (unspread) 50-bps data isalso modulated using BPSK. The spreading code is a shift-register sequence(or PRN sequence, for "Pseudo-Random Noise") with a repetition period of1023 chips. Each satellite uses a different PRN code and all the satellitesignals are transmitted in the same band centered around 1.57542 GHz.Satellite orbits and antenna radiation patterns are such that there is littlevariation in the received signal level on the ground as the satellite moves inits orbit. GPS specifications for the C/A signal call for a minimum user-received power of -130 dBm at a linearly polarized antenna with 3-dB gain[11] although, in practice, the system routinely delivers –125 dBm.

2.2 Assisted GPS

Figure 1 shows a diagram of a typical assisted GPS system: The mobilephone has a "partial" GPS receiver (or GPS “sensor”) that picks up thesignals from the GPS satellites. At the same time, the assisted-GPS servermonitors the same satellite signals through a reference GPS receiver.‡ Weexpect each GPS server to support many base stations (for example, theassisted-GPS server might be co-located with the Mobile Switching Center,or MSC). The assisted-GPS server should have exact knowledge of the GPSsignal being transmitted by the satellites.

Through its connection with the MSC, the assisted-GPS server knows thecell and sector where the mobile is located (which defines its position towithin a couple of miles or so). In more refined systems, the server mayhave even better knowledge of the mobile’s coarse location; the better thisinitial “guess” at the mobile’s location, the better the overall system willoperate. Through the wireless link, the assisted-GPS server will exchangeinformation with the assisted-GPS receiver, essentially asking it to makespecific measurements and collecting the results of those measurements.

‡ The GPS server does not actually need a GPS receiver in physical proximity, it can, in fact,use a service such as differential-GPS (DGPS) available through the internet or lowfrequency broadcast media.


The basic idea behind assisted GPS is to reduce the workload on themobile's GPS receiver as much as possible, at the expense of the assisted-GPS server. To this end, all complex calculations are done by the assisted-GPS server and, since the assisted-GPS server has its own source of GPSdata, there is no need to demodulate the ephemeris information from thesignal received by the assisted GPS receiver at the mobile, which is neededonly for the pseudo-ranging measurements. The signal processing requiredto obtain those measurements is divided between the assisted-GPS serverand the assisted-GPS receiver.

The real power of assisted GPS is that we can go beyond simply dividingthe labor between the assisted-GPS receiver and the assisted-GPS server: Asmentioned above, the assisted-GPS server already knows a coarse locationfor the phone, and it sees the signal coming down from the GPS satellites. Itcan, therefore, predict what signals the assisted-GPS receiver will bereceiving at any given time with great accuracy. Specifically, it can easilypredict the doppler shift experienced by the signal due to satellite motion,and it can also accurately predict other signal parameters that depend morestrongly on the mobile’s exact location. For example, the typical size of acell sector is about 2 miles or less, which corresponds to an uncertainty ofabout ±5 µs in the predicted time of arrival of a satellite signal at the mobile.


This, in turn, corresponds to an uncertainty of only ±5 chips of the spreadingcode of the C/A signal. Thus, the assisted-GPS server can predict the PRNsequence that the mobile receiver should use to de-spread the C/A signalfrom a particular satellite to within ±5 chips; and can communicate thatprediction to the mobile.

Let's say that the assisted-GPS server conveys to the mobile the correctdoppler shift and PRN synchronization for a nominal position in the centerof the sector. Then, if the mobile happens to be near the center of the sector,it can immediately begin to de-spread the corresponding satellite signal atthe doppler-shifted carrier frequency; but, even if the mobile is not exactlyin the center of the sector, a small amount of trial and error will be sufficientto hit the correct PRN. This is because, as we observed, PRN phaseuncertainty is only ±5 µs, and the doppler shift is virtually the same over theentire sector.

After de-spreading, the bandwidth of the signal is that of the underlying50-bps navigation data bits (the bits containing mainly ephemerisinformation). This bandwidth is so small that the assisted-GPS receivercould, in principle, digitize the signal and convey it to the assisted-GPSserver over the wireless phone link. In practice, there is a wide range ofpossibilities for how to split the job of obtaining a GPS fix between themobile and the server. As will be described later, one solution involvesadditional processing in the mobile; indeed, the recently-issued IS-801standard and the TIA/EIA-136 Rev. C draft standard allow for thepossibility of the mobile completing the location fix locally, with or withoutadditional information from the server.

2.3 Assisted GPS Advantages

The above procedure for locating a mobile phone through assisted GPSshould be contrasted to what a conventional GPS receiver needs to do.When it is first turned on, a conventional receiver has no idea where in theworld it is, which satellites are visible, where they are in the sky, what theirDoppler frequency offsets are, and what the timing is of the associated PRNsequences. It has to start a lengthy search over a vast parameter space (allpossible satellite PRN sequences, all possible PRN synchronizations, allpossible doppler shifts) to find the satellite signals. When it hits the correctPRN sequence with the correct timing and the correct doppler for one of thesatellites, it knows that it's good because a 50-bps signal emerges; however,since it doesn't know a priori what the bit modulation is supposed to be, thatsignal has to be fairly strong to both avoid false positives and allow reliabledemodulation of the bits. By contrast, the search space for an assisted-GPSmobile receiver is much smaller. Furthermore, the mobile can learn what


the 50-bps bit stream is supposed to be from the server and, therefore, it candetermine the presence or absence of the signal with a small fraction of thesignal strength otherwise required for full demodulation. The first advantagereduces acquisition time from minutes to less than a second; the secondadvantage allows operation in severely faded conditions such as indoors.

2.4 The Need For Accurate Timing

In this high-level description of the assisted-GPS technique we have, bynecessity, glossed over many details that must be dealt with in practicalimplementations. One of these, however, deserves special attention. It is theassumption that the assisted-GPS server can communicate to the mobile thecorrect timing of the PRN sequence with an accuracy of a fewmicroseconds. In practice, the assisted-GPS server communicates with themobile through a slow channel (e.g., 8 kbps) that goes through severalinterfaces and buffers before reaching the mobile. In general, there will bean unknown delay much larger than a few µs in that connection. For theaccurate PRN timing specification to be meaningful, the mobile must have anotion of time that matches that of the assisted-GPS server to an accuracybetter than the hoped-for PRN timing accuracy. Otherwise, the signal searchspace will have to be widened to include the larger timing uncertainty.

Through its GPS reference receiver, the assisted-GPS server cansynchronize itself to the GPS system which, in turn, is synchronized withUniversal Time (within a few nanoseconds). In the case of the ANSI-95wireless standard, all base stations are also synchronized to GPS Time, andthe mobiles derive their timing from the forward link, so that they, too, aresynchronized to GPS Time. The ANSI-95 standard specifies asynchronization accuracy of a few µs, which meets the needs of assistedGPS well. Other communication standards (e.g., GSM, TIA/EIA-136 orAMPS) do not have a similarly stringent synchronization specification, andthe design of an assisted-GPS system based on such standards musttherefore either include a solution to the timing requirement or endure alarger search space in the time domain. A possible solution for systemsbased on TIA/EIA-136 or GSM involves adding calibration receivers in thefield to monitor both wireless signals and GPS (or equivalent) timing signalsused as a time reference. It is worth noting that a “Time Calibrator” of thissort does not require synchronous timing in the exchange of messages withthe assisted-GPS server as long as both the calibrator and the server canunambiguously identify the same specific reference event in the wirelesssignal.

Even in the absence of a very accurate timing reference, the assisted-GPStechnique offers improved performance at lower cost compared to


conventional GPS. For example, a timing uncertainty of several milliseconds(which perhaps might occur in AMPS) implies a much larger search spacefor PRN sequence synchronization (up to about a thousand chip periods)which will increase acquisition time and require more signal processing inthe assisted-GPS receiver; but this is still a lot less than for a stand-aloneGPS receiver. More importantly, the availability of assisting GPS dataprovided by the server still allows the detection of the GPS signal at signallevels that are much lower than required by a stand-alone GPS receiver.

3. PRACTICAL IMPLEMENTATION

This section describes the two parts of assisted GPS: thereceiver/terminal and the server/network part. Figure 2 shows more detailfor the mobile and the server parts of the overall system, and each isdescribed in greater detail here.


3.1 The Terminal

As shown in Figure 2, and like most any receiver, the assisted-GPSmobile has an RF, an IF, and a digital section, although the IF may be a socalled, “low-IF” type. Both the GPS and ANSI-95 signals use spreadspectrum signaling with comparable carrier frequencies and chip rates.Therefore, acquiring a GPS signal requires functions somewhat like thoseused to acquire the pilot of a cellular or PCS COMA signal. This presentsopportunities to leverage commonalties in the IF and digital sections of anassisted-GPS mobile following the ANSI-95 standards.

This has been pointed out by Qualcomm, Inc. in their GPSOne product.12

3.2 The Assisting GPS Server

The main roles of the server are:1. To interface with the network entities that will request and/or

consume location data (note these are widely expected to be eitherin the private Wireless Intelligent Network [WIN] or the Internet)

2. to provide the assisting GPS data to the mobile3. to calculate the location of mobiles4. to interface to wireless network entities that may help the server to

improve the assisting data it will generate and/or provide data tothe server for more robust and accurate solutions for a mobile’slocation

The third function may not always be needed, since some high-endmobiles may conclude the location calculation themselves. Similarly, thefourth function may not always be needed—these are refinements of serverfunctions that improve performance but may not be essential to assistedGPS.

Starting with the left-hand side of Figure 2, we show input from areference GPS. The reference GPS’ function is to maintain current data forall visible GPS satellites. It is beneficial, and in some cases necessary, forthe GPS data to be project a short time into the future. The GPS data will beused to construct a Navigation Data Message for assisted GPS. A table ordatabase within the PDE (server) will maintain a record for each GPSsatellite, with information such as the PRN codes and observed or calculateddoppler shifts for all visible satellites. The reference GPS could be aconventional, high-quality GPS deployed at each server or could be aservice derived from existing, commercial Differential GPS serviceproviders. The Reference GPS could be thought of as maintaining GPS dataon expected satellite visibility within the server area, on the basis of cell andsectors covered by the assisted-GPS system, as well as the ephermeris of


each satellite that is expected to be useable and other orbital correctionparameters may be useful for the location calculation done at the server.Constellations of GPS satellites are visible over areas extending forhundreds of kilometers, so large networks may be supported with only a fewReference GPS sites.

The Assisting Message Constructor may formulate the assisting messagebased on network data such as Round Trip Delay (RTD), pilot phase offset,etc. These data can greatly help by reducing the size of the search windowthat the mobile will have to use in looking for GPS signals, and can be madeavailable in IS95 networks. Search window size is a parameter called for inthe IS801 standard, which will be described later in this paper.

The functioning of the server in most cases will be initiated via a requestfrom a location application, such as E911, a location-based billing service,or a navigation service. This function is shown in the lower center portionof Figure 2. The application could run in or through a Service Control Point(SCP) and would likely communicate to a wireless network via the WirelessIntelligent Network (WIN) using standard IS-41 messaging. The work ofTIA’s committee TR45.2 on projects PN3890 and PN4288 are expected tostandardize such messages. The interface can be expected to alsocommunicate either directly or indirectly with an authentication server toensure privacy of user's location data, which must be assumed to besensitive. Further discussion of location applications and their interactionswith assisted-GPS system are beyond the scope of this document.

The above discussion covers how the assisted GPS navigation message isconstructed for transmission to the assisted-GPS terminal. All that remainsto describe within the assisted-GPS server is the Location Calculator, whosefunction is obvious from the name. The Location Calculator receives datacoming from the assisted-GPS receiver, it performs the necessarycalculations to determine location. These calculations will likely includesteps to:

– Begin, of course, by determining which terminal is being locatedand which satellites it has detected. We assume here that theterminal has sent the assisted-GPS terminal PRN synchinformation (as described previously) for each satellite that it wasable to lock onto. Of course, there are many variations and thelocation calculator could operate on "raw" data from the terminal.

– The Location Calculator would calculate pseudoranges and thendo typical GPS receiver functions of converting pseudoranges todistances and distances to specific locations such aslatitude/longitude. Commonly used GPS techniques such asKalman Filtering would be used as appropriate, depending onspecific designs.


– Options to calculate velocity and heading in addition to locationcould be added.

Many variations are possible. While most are beyond the scope of thisdocument, we must mention differential GPS, which would be expected tobe used. Since Selective Availability causes most of the inaccuracy in GPSusage today, differential GPS (DGPS) corrections would be applied toimprove the solution accuracy. The inclusion of DGPS to the LocationCalculator would likely make assisted GPS the most accurate way to locateterminals; DGPS is commonly able to provide 10 meter accuracy or better.

Other techniques, such as applying atmospheric corrections, knowledgeof local terrain, altitude aiding, etc. would likely be applied. Thesetechniques are not likely to be available if the location calculation isconcluded in the mobile.

4. PERFORMANCE

Performance of wireless location systems is a complex subject13 and notaddressed here. We describe only key concepts relating to performance ofassisted GPS technology, not system performance of either mobiles orsystems.

4.1 Link Budget For Assisted GPS

The nominal loss from a 1-2 story building at 1.5 GHz is ~20 dB. Thisimplies that most conventional (i.e., not assisted) GPS receivers will notwork (or not get enough satellites) in such a building. This is consistent witheveryday experience. For example, in a typical 1-2 story open frameconstruction-house, a GPS will work somewhat in 20-50% of the rooms,working more often on the top floor and near windows. In a large buildingsuch as a major commercial location, conventional GPS receivers will notwork at all. The SNR improvement of assisted GPS is enough to make GPSlocation work in most such environments; note that the details ofimplementation become most important in this aspect of assisted GPS: thepercentage of buildings (and other obstructed sites) that can be coveredvaries dramatically with implementation details: in particular the hybridapproaches that use network data with GPS data greatly enhance in-buildingcoverage.

Table 1 is a GPS link budget for two scenarios.


Conventional GPS receivers typically integrate coherently over onemillisecond (one code period) and incoherently for six milliseconds and,consequently, have an acquisition threshold of (typically) –34 dB-Hz. Thus,a conventional GPS receiver can only acquire signals above approximately-130 dBm. Weaker signals require more processing gain (longer integration)for successful acquisition. Knowing “true” GPS time at the mobile stationand the approximate range to the satellite will enable the sensor to integratecoherently over 20 milliseconds (one navigation bit period).

Furthermore, if the network can predict (or obtain advance copies of) thebit sequence for some parts of the navigation message, the bit polarity canbe sent to the MS to enable integrating coherently over multiple bits. Thistechnique is known as “modulation wipeoff.” In addition to the sensitivityenhancement, knowing “true” GPS time at the MS reduces the time requiredto acquire the GPS signal for a given satellite. The serving BS sendsinformation regarding the search window center and the search window sizeto the MS. Hence, the MS need only search a small window in the timedomain rather than the whole code space. Once the network makes a coarseestimate of the position of the MS, that estimate can be used to compute thesearch window. Even if the uncertainty in the coarse estimate is as large asfour miles, the search window size for a satellite at the horizon is only 20chips, even less for satellites more directly overhead. This reduces searchtime per satellite compared with the case of no knowledge by a factor of 50.Search window size can be tightened substantially further if network dataare used to estimate the position of the phone.

4.2 Time Required For A Location Fix

Typically, a conventional GPS receiver takes a long time (minutes)to provide a location fix when first turned on. This is because it has tosearch a large parameter space in order to find the signals from the


satellites that are currently visible and, once it has acquired the signals,it takes about 30s per satellite to obtain a full set of ephemeris fromthe 50-bps data. By contrast, in an assisted-GPS system all thatinformation is already available in the assisted-GPS server.

In situations where a conventional GPS receiver has enough signalto operate, the assisted-GPS receiver will take only a fraction of asecond to make the measurements and relay them to the assisted-GPSserver. This is because, under such conditions, the integration timeneeds to be no longer than the 20-ms bit period. Thus, from the pointof view of the human user, the location fix is nearly instantaneous.The actual time it takes to make the measurements depends on fourparameters:

– The integration time, which is determined by the available SNRand can be as long as 1s to achieve maximum sensitivity.

– The timing uncertainty, which determines how many differentPRN synchronizations must be tried.

– The frequency uncertainty, which determines the Dopplerfrequency shift space that must be evaluated.

– The number of correlation channels available in the assisted-GPSreceiver.

These parameters can be handled as follows:Integration Time Initially, there is no way to know what SNR can beexpected at the mobile receiver, so it makes sense to have the receiverfirst make a quick measurement with a short integration time, and thenincrease it progressively if no signal is detected.Timing Uncertainty If the receiver knows GPS time to within a fewmicroseconds, and it is given the navigation bits for “modulationwipeoff,” then it can coherently integrate beyond the 20 millisecondlimit for conventional GPS receivers.Frequency Uncertainty Through simple calculations in the server, theDoppler shifts for each satellite can be calculated for the center of thecell and provided to the GPS receivers. By locking the receiver clocksto the base station’s clocks (which in the case of TIA/EIA-95 are alsotied to GPS time) the mobile terminal frequency error can beminimized.Number of Channels The number of channels will determine the Timeto First Fix (TTFF). We discuss this in terms of the number ofcorrelators. To simultaneously detect, say, 5 satellites with 20correlators, where the search window, as discussed above might be ±5chips, one might use a half-chip search approach to find thecorrelation peak. Satellites would then be found sequentially. If 100correlators were employed, all 5 satellites could be simultaneouslylocated.


5. STANDARDS COVERING ASSISTED GPS

U.S.-based service providers mandated by the FCC to provide E911geolocation are free to deploy either proprietary or standards-basedtechnology. Angle of Arrival, Reverse Link Time of Arrival, and othernetwork-based systems are being considered and have the virtue ofpotentially working with legacy terminals. However, there is concern that tobe fully deployed, these approaches require extensive build out of theinfrastructure and the siting of new base stations to provide geometries goodfor triangulation. Considering cost, coverage, and accuracy needed toachieve FCC mandated performance, handset-based geolocation, andassisted GPS in particular, becomes very attractive. To ensure that manyhandsets and various networks work together, as well as to improvemanufacturing efficiencies, standards are needed.

Several standards bodies have been actively working to specify themessages, parameters, and procedures needed for interoperability amongvarious proposals for assisted GPS. One of the first standards completedthat was primarily aimed at supporting assisted GPS was TIA/EIA 1S801,which was published in final form in November, 1999 by the TR45.5standards body. IS801 addresses both ANSI-95B and IS2000 (cdma2000).However, many of the techniques are applicable to all CDMA systems andthe general concepts are applicable to other systems as well. TIA/EIA-136TDMA, GSM, AMPS, although sometimes starting from differentperspectives, have progressed to various stages of completeness, as shown inTable 2. Because GSM defines network and air interface standardstogether, while TIA’s TR committees for network-side and air interfaces areseparate, resulting documents may appear quite different, but importantsimilarities exist that underline the base technology.

Perhaps because creation of the standards have been somewhat rushed,with standards committees having been strongly admonished about the needto complete their work in time for the FCC mandate, assisted-GPS standardsmay offer more options than would ideally be the case. The only way toreach rapidly obtain consensus was to accept all options that werecontributed in a reasonably complete and timely way. The marketplace mayeventually determine which options from assisted-GPS standards work best.For now, we describe the basics of each standard in the following sections,emphasizing IS801 as the most mature of the assisted-GPS standards.

Note that Table 2 is labeled 2G/2G+; although there has beenconsiderable, additional work done in 3G standards, that 3G work is beyondthe scope of this document. Other standards are not covered here becausethey do not touch upon assisted GPS.


5.1 IS95 & IS2000 CDMA: The IS801 Standard

IS801 could be said to support three technologies: GPS (autonomous andassisted); Advanced Forward Link Triangulation (abbreviated AFLT; this isCDMA pilot phase measurement); and a Hybrid Technique combining GPSand AFLT. IS801 location reports use the specification shown in Table 3.Note all parameters except latitude and longitude are optional. We focushere on assisted GPS and review IS801 operations, which are shown in asimplified view in Figure 3.


Figure 3 shows a simple, three-step ping-pong diagram representingthe basic operations for what can be regarded as the most commoncase of IS-801. The process begins with the network sending a requestfor location and assisting data. Several points are worth noting here:

– IS801 specifies only over-the-air messages, leaving all networkprocessing to other standards, such as the work now being done asPN-3890. Fig. 3 shows an actual Base Station, but note thatIS801 uses the words “Base Station” to refer not only the BaseStation itself, but to all network entities, including the MSC, PDE(Position Determining Entity), MPC (Mobile Positioning Center),SCP, etc.

– Before the network sends a request for location as shown in Fig. 3,it may recognize that a call has been placed to 911, and resultingspecial handling calls for location to be determined. Alternatively,some other location-based application may have requested thismobile be located.

– The first message sent performs two functions: requesting locationand providing assistance: IS-801 calls for requests and responsesand allows compound messages; thus the request for location can


be combined with assisting data for efficiency—just as the mobilereceives the request to perform the location function, it alsoreceives data that may be helpful in acquiring GPS signals.

– The data returned may also include multiple parts, such as theGPS and pilot phase data shown in the second step here.

– The last step (labeled optional) shows that the location result, oroutput as shown in Table 3, may be returned to the mobile,reflecting IS801’s consideration of location-based applicationsthat are concluded at the mobile, rather than in the network.

5.1.1 Types of GPS Assistance in IS801

1S801 specifies three types of GPS assistance: acquisition assistance,location assistance, and sensitivity assistance. Each is described brieflyhere:

Acquisition assistance provides basic GPS information to enable mobilesto rapidly acquire GPS signals. Specifically, satellite IDs, Doppler shifts,and the timing of spreading codes is provided such that the acquisition ofGPS signals, which could take several minutes in conventional receivers, isreduced to seconds or fractions of seconds.

Location assistance provides enough detailed information so a (properlyequipped) mobile can compute its own location with the accuracy ofdifferential GPS. The sequence of steps to use Location Assistance couldbe:

– PDE estimates a rough location (e.g., cell/sector) for the mobile.– PDE predicts GPS signal at estimated location at a future time.– PDE conveys to MS a Location-Assistance Message containing:

(a) the location estimate, and (b) the predicted GPS signal.– MS measures the discrepancy between the predicted and the

observed GPS signal at the specified time.– MS computes its own location through simple, linear math.

The key data for location assistance include: Elevation and azimuthangles of visible satellites, high-precision satellite Doppler shifts, and high-precision timing of spreading codes.

An important alternative also supported by optional parameters oflocation assistance messages conveys GPS almanacs, almanac corrections,and ephemerides, thus enabling equivalent calculations to those implied bythe above list.

Sensitivity assistance enables better penetration into buildings andfaded environments, as well as certain lower-cost implementations, byconveying predicted GPS modulation bits and their associated timingto the mobile. The steps envisioned are:

– PDE monitors periodic 50-bps modulation pattern of GPS signal.


– PDE predicts (or obtains advance copies of) future modulationbits.

– PDE conveys to MS the predicted bits (with their associatedtiming) and Doppler shift estimates.

– MS applies predicted BPSK. modulation to received GPS signal(also called modulation wipeoff); received signal becomes purecarrier (BW=0 Hz).

– Bandwidth reduction from ~50 Hz to 0 Hz allows MS to detectGPS signal with substantial sensitivity gain (10-17 dB).

5.1.2 Advanced Forward Link Trilateration (A-FLT) in IS801

As mentioned, IS801 also supports location by better enablingtrilateration (also called triangulation) by having the mobile measure theoffsets of pilots received from several Base Stations. Forward LinkTrilateration (FLT) presents an attractive option to locate CDMA mobilesbecause only software changes would be needed—no new network hardwareand no modified mobiles would be needed. Unfortunately FLT does notwork well enough to satisfy accuracy and coverage requirements ofimportant applications including E911, because of the coarseness of the chipresolution in ANSI-95 and ANSI-95’s power control imposing a regimewhere pilots from multiple Base Stations are too seldom “heard” by mobiles.

AFLT is accomplished by sending:Base Station Almanacs – Giving the mobile base station locations and

reference time correction. This supports position computations made at themobile.

Pilot Phase Measurements - Provides BS with forward link pilot phasemeasurements made by the MS. Position computation made at BS.

The mobile also returns data on pilot offsets and pilot RMS errors, whichrelate to pilot strength. These data can be used to perform trilateration at theserver.

5.1.3 Items Not Addressed in IS801

Several important items were discussed in the committee that wroteIS801, but were considered outside the scope of IS801, either because theyare not part of the air interface or because they were not essential to basicfunctioning of assisted GPS. These items will, respectively, be consideredby other standards bodies or in later versions of IS801. They include:

– Location of idle-mode mobiles—services such as location-basedbilling require location of mobiles in the idle state, and someefficiencies can be gained over having mobiles on traffic channelsby proper design of idle-mode operation. However, because this is


not needed for E911 services, idle mode operations are not nowcovered in IS801.

– Network-Side Messaging— is outside the scope of IS801;however, message exchanges between PDE and Base Stations,MSC and location-based applications have been defined, andconsiderable work has been done on PN3890 in the TR45.2organization.[14] In addition, there is new work to go beyond theFCC mandated requirements in Project Number PN4288.[15]

5.2 GSM Standards for Geolocation

Considerable work as also gone into the creation of the ETSI GSMstandard for geolocation where several important new documents have justbeen published.[16,17,18] This work, conducted by SMG3, in a Europeancontext, was not driven by the FCC mandate on accuracy or coverage, buteven so, these standards provide for a type of assisted GPS. This paralleldevelopment of the same fundamental technology is an indication of thebroad technical appeal of assisted GPS.

Of course, the GSM standard has options for many other geolocationapproaches, namely:

– TOA – Time Of Arrival: Trilateralization by 3 or more basestations of the reverse link from the mobile terminal. Works withlegacy handsets but may require additional siting of infrastructure.

– AOA –Angle of Arrival: Trilateralization by 2 or more basestations. This approach requires installation of substantial antennaand processing capabilities at well sited towers, but it works withlegacy terminals.

– Mobile Assisted E-OTD: The Enhanced - Observed TimeDifference of Arrival method can be used by new terminals thatcan record the relative time of arrival of bursts from two BTSs or(in the case of nonsynchronized networks) three BTSs. The mobileprovides the location server with the time measurements and theserver computes the location of the MS with information about thelocation and timing advance setting of each BTS involved.

– Mobile Based E-OTD: E-OTD can be implemented with newmobiles collecting all information including the assistance dataconcerning the location and timing advance information for eachsignal source. This requires new software/firmware but no newhardware, provided geometries are good.

– Mobile Assisted GPS: This is fundamentally the same approachdescribed throughout this paper. The differences are highlightedbelow but are not fundamental.


– Mobile Based GPS: In case of a high-end mobile, sufficientassistance information can be provided the terminal that it cancalculate its location without providing the network anyinformation other than its request for assistance. This capabilitywould be particularly useful in a navigation application where adriver is given navigation information continuously, withoutburdening the network with a stream of assistance.

– Conventional GPS: There is a provision in GSM 09.31 for aconventional GPS receiver that would communicate it’s positionthrough the GSM network. While in contact with the network, itseems appropriate to use network resources to help improve thespeed and accuracy of the GPS network, but one can still conceiveof times in which having a high-end “backpacker’s” phone thatcan double as a GPS receiver even without contact to the basestation would be useful.

The primary differences between the assisted GPS approaches of GSMand CDMA are network difference including an network element, the LMU(Location Management Unit) that the GSM networks use for timingmeasurements and calibration through the coverage area. These LMUs mayuse the air interface to report these measurements to the location server sothat they do not need extensive backhauling, but they do need to be situatedwhere multiple LMUs are able to “see” most locations in the coverage area.They need to be situated where they provide good trilateralization over thefield.

Another difference between the GSM and CDMA approach is that thesensitivity assistance is not included per se. The expectation that it will beneeded to meet the FCC mandate in indoor (highly faded) environments hasled to some recent activity to include it in the North American version of theGSM standards effort (T1P 1.5). In that forum, some navigation bits areallowed to be used as part of the sensitivity assistance, particularly over thebroadcast channel. It is likely that harmonization between the ETSI andT1P1 bodies will address this deficiency in future releases.

5.3 TDMA Standards for Geolocation

The ANSI organization TR45.3 has recently begun the standardizationeffort for the TIA/EIA-136 TDMA air interface, in working group 6(TR45.3.6). As of this writing, TR45.3.6 has drafted a stage one documentdescribing the requirements for geolocation, but it seems clear from earlierwork done in the UWCC-corel36 organization that assisted GPS will be thesole standard method. Other methods might be used, but would not needstandards support.


– Mobile Based GPS: In case of a high-end mobile, sufficientassistance information can be provided the terminal that it cancalculate its location without providing the network anyinformation other than its request for assistance. This capabilitywould be particularly useful in a navigation application where adriver is given navigation information continuously, withoutburdening the network with a stream of assistance.

– Conventional GPS: There is a provision in GSM 09.31 for aconventional GPS receiver that would communicate it’s positionthrough the GSM network. While in contact with the network, itseems appropriate to use network resources to help improve thespeed and accuracy of the GPS network, but one can still conceiveof times in which having a high-end “backpacker’s” phone thatcan double as a GPS receiver even without contact to the basestation would be useful.

The primary differences between the assisted GPS approaches of GSMand CDMA are network difference including an network element, the LMU(Location Management Unit) that the GSM networks use for timingmeasurements and calibration through the coverage area. These LMUs mayuse the air interface to report these measurements to the location server sothat they do not need extensive backhauling, but they do need to be situatedwhere multiple LMUs are able to “see” most locations in the coverage area.They need to be situated where they provide good trilateralization over thefield.

Another difference between the GSM and CDMA approach is that thesensitivity assistance is not included per se. The expectation that it will beneeded to meet the FCC mandate in indoor (highly faded) environments hasled to some recent activity to include it in the North American version of theGSM standards effort (T1P 1.5). In that forum, some navigation bits areallowed to be used as part of the sensitivity assistance, particularly over thebroadcast channel. It is likely that harmonization between the ETSI andT1P1 bodies will address this deficiency in future releases.

5.3 TDMA Standards for Geolocation

The ANSI organization TR45.3 has recently begun the standardizationeffort for the TIA/EIA-136 TDMA air interface, in working group 6(TR45.3.6). As of this writing, TR45.3.6 has drafted a stage one documentdescribing the requirements for geolocation, but it seems clear from earlierwork done in the UWCC-corel36 organization that assisted GPS will be thesole standard method. Other methods might be used, but would not needstandards support.


The TIA/EIA-136 wireless standard has a relatively narrowband RFchannel of 30 kHz, which limits ones ability to transport good timinginformation. The Cramer-Rao formula gives the limit on the ability tomeasure a transition in time that is inversely proportional to bandwidth andSignal-to-Noise Ratio.

Unlike CDMA, the TDMA networks are often unsynchronized,which further complicates the task of calibrating the TDMA time.Consequently, the TDMA standards effort is calling for “timingcalibrator” a network element similar to the Location MeasurmentUnit (LMU) defined in GSM standards, but with the sole function ofreporting on the time relationship between GPS and TDMA time asmeasured in bits in time slots in frames and superframes. Figure 4below shows the expected block diagram of components in the TDMAgeolocation reference model. Unlike Figure 1, this includes the timecalibrator function shown. Just as the LMU can be operated throughthe air interface, so too, can the time calibrator in this model.


5.4 Analog/AMPS Standards for Geolocation

The ANSI organization TR45.1 in its Working Group 1 has endeavouredsince midyear 1999 to standardize messaging required for geolocation inanalog AMPS under project number PN4662. The work of that group isexpected to be published in the Spring of 2000 as IS817. Key points ofanalog support for assisted GPS are:

– Assistance information is conveyed through existing “blank-and-burst” messaging mode. This means no hardware changes will berequired to Base Stations.

– Low bit rates that AMPS can accommodate (200-300 BPS) allowfor only limited assistance—nothing like the SensitivityAssistance of IS801.

– “Hybrid” solutions, as described previously, are not possiblewithout additional infrastructure changes.

– Performance of assisted GPS in analog modes will likely beinferior to digital-mode by

Decreased yield (decreased GPS sensitivity)Longer TTFF

Nevertheless, assisted GPS in AMPS is expected to meet FCC mandate,when considered in conjunction with superior performance of digital-modelocations.

6. SUMMARY OF BENEFITS

To summarize the key benefits that have been mentioned throughout thisdocument, assisted GPS's advantages are:

– Inexpensive, particularly for ANSI-95-CDMA terminals—Assisted-GPS terminals are expected to incorporate geolocationfunctions at the chip level, particularly in IS-95 terminals, whichtherefore might be made for only a few dollars per handset morethan conventional terminals. The network-side equipmentpromises to be much less costly than alternative approaches.

– Applicable to all air interfaces—although more straightforward forCDMA, as is reflected by the fact that IS801 standard was the firstto be completed in support of this technology.

– Differential GPS level of accuracy—particularly notable versusterrestrial triangulation systems seems unlikely to improve beyond100 meter accuracy of so; assisted GPS should be an order ofmagnitude more accurate.


– Locations available in buildings and other heavily-fadedsituations—the major problem with using conventional GPS isovercome.

– Little or no new hardware needed in Base Stations; no newconnections between network elements

– Rapid acquisition time-this is essential for emergency calling (911calling).

7. CONCLUSION

Other techniques that have been proposed to locate wireless phones|19]donot require any additional hardware in the phone itself. While the assisted-GPS technique has the obvious disadvantage of requiring special-purposehardware in the phone, it should be noted that the added hardware cost ismodest while the overall cost to the service provider is significantly reducedbecause, assisted GPS does not require modifications of all base stations.Indeed, the assisted-GPS server may be part of the Mobile Switching Center(MSC) or attached at even higher concentration levels of the wirelessnetwork and thus be shared by a very large number of base stations.

Close scrutiny of the associated costs are necessary for a fair costcomparison; however, the main advantage of assisted GPS lies in itssuperior performance; other techniques are typically characterized bycomparatively poor accuracy and limited availability[3,6,20,21] which makesthem unsuitable for advanced location-based services.[1] The FCC currentlyrequires service providers to locate 67% of E911 calls to within 100 metersor 50 meters depending upon whether network based or handset basedtechnologies are used[22]. This recent rulemaking reflects a easing ofrequirements from the FCC's 1996 (NPRM)[23] that expressed a desire of theemergency services community to have wireless location provide anaccuracy of 40 feet on 90% of all 911 calls, including a determination ofaltitude. While we believe that assisted GPS is the only approach that canmeet the current FCC requirements indoors, we are even more convincedthat it is the only way to meet the intention of the original emergencyservices community request for 40 feet accuracy. We see few prospects thatthis goal can be economically achievable by any other proposed technique.

ACKNOWLEDGEMENT

Our sincere thanks to Dr. Samir Soliman of Qualcomm, Inc. for valuableinput to this paper.


1 Mark Flolid, “Wireless Location Services,” NENANews (published by the NENA, theNational Emergency Number Association), December 1998 page 17.

2 E. McCabe, “Start Now, Evolve to the Future,” Telephony, Volume 236, Number 22, May31, 1999, page 36.

3 Elliot D. Kaplan, ed., Understanding GPS - Principles and Applications, Artech House,Boston (1996).

4 A. Brown/NAVSYS Corporation, “GPS Phone: An Integrated GPS/Cellular Handset,”issued at ION-GPS-97, The Institute of Navigation GPS-97 Conference, Kansas City,Missouri, September 16-19, 1997. Navsys Corp., 14960 Woodcarver Road, ColoradoSprings, CO 80921.

5 Raymond DiEsposti, Steven Saks, Lubo Jocic, and Capt. Jordan Kayloe, “Of MutualBenefit: Merging GPS and Wireless Communications,” GPS World, Volume 9, Number 4,April, 1998, page 44.

6 Mark Moeglin and Norman Krasner, “An Introduction to SnapTrack™ Server-Aided GPSTechnology,” ION-GPS-98: Proceedings of the 11th International Technical Meeting ofthe Satellite Division of the Institute of Navigation, September 15-18. 1998, Nashville,Tennessee, page 333.

7 Norman Krasner, “Homing In On Wireless Location,” Jan, 2000, CommunicationsSystems Design.

8 L.J. Garin, M. Chansarkar, S. Miocinovic, C. Norman, D. Hilgenberg, “Wireless AssistedGPS—SiRF Architecture and Field Test Results,” ION-GPS-99: Proceedings of the 12lh

International Technical Meeting of the Satellite Division of the Institute of Navigation,September 14-17, 1999, Nashville, Tennessee, page 489.

9 A. R. Pratt, “Combining GPS and Cell Phone Handsets—The Intelligent Approach,” ION-GPS-99: Proceedings of the 12th International Technical Meeting of the Satellite Divisionof the Institute of Navigation, September 14-17, 1999, Nashville, Tennessee, page 529.

10 J . J. Spilker, "Signal Structure and Performance Characteristics," Navigation, The Journalof the Institute of Navigation, Vol 25, Number 2, page 121. See also other articles withinthat issue of Navigation.

11 Elliot D. Kaplan, ibid., page 97.12 Qualcomm, Inc. datasheets for MSM3300™ Mobile Station Mobile and gpsOne™ enhanced

by SnapTrack; seehttp://www.qualcomm.com/ProdTech/asic/products/documents/MSMS3300.pdfandhttp://www.qualcomm.com/ProdTech/asic/products/documents/gpsOneSnapTrack.pdf

13 S. Tekinay, E. Chao, and B. Richton, “Performance Benchmarking in Wireless LocationSystems,” IEEE Communications Magazine, Volume 36, Number 4, page 72, April, 1998.

14 TR45:PN-3890, “Enhanced Wireless 9-1-1 Phase 2,” Rev. 13, February 15, 2000,Preballot Version

15 TR45:PN4299, “Wireless Emergency Services Features Beyond FCC Mandates,” charteravailable online at: http://www.tiaonline.org/pubs/pulse/1998/pulse0998-7.cfm.

16 ETSI GSM 03.71: “Digital cellular telecommunications system (Phase 2+); LocationServices (LCS); (Functional description) - Stage 2 (GSM 03.71 version 7.2.1 Release1998),” ETSI TS 101 724 V7.2 1 (2000-01), Published January 2000. (Available on lineat: http://webapp.etsi.org/workprogram/Report_WorkItem.asp?WKI_ID=9269. )

17 ETSI GSM 04.31: “Digital cellular telecommunications system (Phase 2+); LocationServices (LCS); Mobile Station (MS) – Serving Mobile Location Centre (SMLC) RadioResource LCS Protocol (RRLP) (GSM 04.31 version 7.0.1 Release 1998),” ETSI TS 101


527 V7.0.1 (2000-01), Published January 2000. (Available on line at:http://webapp.etsi.org/workprogram/Report_ Workltem.asp?WKI_ID=9263).

18 ETSI GSM 09.31:“Technical Specification Digital cellular telecommunications system(Phase 2+); Location Services (LCS); Base Station System Application Part LCSExtension (BSSAP-LE) (GSM 09.31 version 7.0.0 Release 1998),” ETSI TS 101 530V7.00 (2000-01), Published January, 2000. (Available on line at:http.//wehapp.etsi.org/workprogram/Report_Workltem.asp?WKI_ID=9217).

19 L. Stilp, “Time Difference of Arrival Technology for Locating Narrowband WirelessSignals,” Proceedings of the SPIE, Vol. 2602, October 25-26, 1995, pp. 134-144.

20 M J. Meyer, T. Jacobson, M. E. Palamara, E. A. Kidwell, R. E. Richton, G. Vannucci,“Wireless Enhanced 9-1-1 Service - Making it a Reality,” Bell Labs Technical Journal vol.1, n. 2, Autumn 1996.

21 E. Benedetto, V. Biglieri, and V. Castellani, Digital Transmission Theory, Prentice Hall,Englewood Cliffs, NJ (1987).

22 Federal Communications Commission, CC Docket Number 94-102, Action by theCommission September 15, 1999, by Third Report and Order (FCC 99-245), releasedOctober 6, 1999. Further information available at: http://www.fcc.gov/e911.

23 Federal Communications Commission, CC Docket Number 94-102, Report and Order andFurther Notice of Proposed Rulemaking (FCC 96-264), adopted June 12, 1996; releasedJuly 26, 1996. Available at: http://www.fcc.gov/e911.

ABOUT THE AUTHORS

Bob Richton is a member of technical staff in the Wireless TechnologyApplications Department at Lucent Technologies’ Bell Labs in Whippany,New Jersey. Since 1996, his work has focused mainly on systemsengineering, architecture, and opportunity analysis for wireless E9-1-1 andother wireless geolocation applications. Mr. Richton has a B.S. degree inphysics from the University of Massachusetts in Amherst, and an M.S. inphysics and chemistry from the Stevens Institute of Technology in Hoboken,New Jersey.

Giovanni Vannucci is a member of technical staff at Bell Labs inHolmdel, New Jersey. His primary responsibility is research in the area ofwireless and portable communications. He also conducts research inmicrowave, satellite, and optical communications, light statistics, quantumelectrodynamics, and visual psychophysics. Mr. Vannucci, a member of theAmerican Association for the Advancement of Science and a senior memberof the IEEE, received M.S. and Ph.D. degrees in electrical engineering fromColumbia University in New York. He also has a doctor's degree in physicsfrom the University of Pisa in Italy


Stephen Wilkus is a Technical Manager in the Wireless TechnologyLaboratory of Lucent Technologies. He received his MSEE degree from theUniversity of Illinois, Urbana-Champaign, in 1981. After working as asenior design engineer developing SAW devices and automated designsoftware, he began work at AT&T Bell Labs in 1986. He has led thedevelopment of several low-cost radio systems for indoor wireless LAN andElectronic Shelf Labels a product currently being sold by NCR and hasspearheaded several initiatives such as the development of the spectraletiquette approach to frequency allocation. He has authored severaltechnical papers in IEEE publications and has presented invited talks in anumber of International conferences and symposia, most recently at the ICMconference Wireless Positioning and Location Services Conference inLondon, March 6, 2000.


Chapter 5

EVALUATION OF LOCATIONDETERMINATION TECHNOLOGIESTOWARDS SATISFYING THE FCCE-911 RULING

M. SunayBell Labs, Lucent Technologies

67 Whippany Road,

Whippany, NJ 07981, USA

[email protected]

Abstract The recent FCC ruling has prompted the emergence of significant re-search on location determination technologies for wireless systems. Var-ious technologies have been proposed in the literature. The wireless op-erators need to select a location determination technology that satisfiesthe FCC requirements and is most appropriate to their needs by Octo-ber 2000. For this reason, proper evaluation techniques are necessaryto make an extensive and fair comparison of the different technologies.This tutorial chapter gives an overview of the specifics of the FCC rul-ing and the different location determination technologies that are beingconsidered by the wireless operators. A brief synopsis on how the indi-vidual location determination technologies may be evaluated is given inthis chapter as well.

Keywords: E-911, Location Determination Technologies, Wireless Assisted GPS,Time Difference of Arrival, Angle of Arrival, Multipath Fingerprinting,Evaluation and Testing Criteria.

1. INTRODUCTIONThe concept of using a single emergency phone number to report

any kind of emergency to a centralized reporting agency first originatedin Britain. Other countries, including the United States, followed suitshortly thereafter. In the United States, interest for a nationwide emer-


gency number sparked in 1957 when the National Association of FireChiefs recommended the use of a nationwide single phone number forreporting fires [1]. In 1967, the President’s Commission of Law Enforce-ment and Administration of Justice recommended that a single numberbe established nationwide, solely for the purpose of reporting emergen-cies. With support from other Federal Government Agencies and vari-ous government officials, the President’s Commission on Civil Disordersturned to the Federal Communications Commission (FCC) for a solu-tion. Later that year, FCC asked AT&T to find a means of establishinga universal emergency number that could be implemented quickly. In1968, AT&T announced that it would use the three digits, 911, as theemergency code throughout the United States. 911 was selected becauseit was short and easy to remember. Furthermore, 911 had never beenused as an area or service code and also met the long range numberingplans and switching configurations of AT&T.

With recommendation from AT&T, FCC designated 911 as the only“Universal Emergency Number” for public use in the United States torequest emergency assistance. On February 16th, 1968, Alabama Sena-tor Rankin Fite became the first person to place a 911 call. In Marchof 1973, the Executive Office of the President issued a bulletin, endors-ing the concept of 911 and urging its implementation nationwide. Thebulletin also provided for the establishment of the Federal InformationCenter to assist units of government in the planning and implementa-tion of 911 systems. In 1976, approximately 17% of the United States’population had access to 911 services. In the early 1970’s, AT&T beganthe development of sophisticated features for the 911 system, paving theway for the Enhanced-911 (E-911) service. Today’s wireline E-911 ser-vice provides the PSAPs an automatic caller identification (ALI) in theform of caller’s name, phone number and address. Furthermore, ALIcan be used to selectively route the 911 call to the proper PSAP whichis normally closest to the scene. As of November 1999, nearly 92% of thepopulation in the United States is covered by some type of 911 system[2]. However, the service footprint covers only 50% of the country’s phys-ical landscape. The current 911 service coverage in the United States ismapped in Figure 5.1. Coverage percentages in the individual states areshown in the figure.

The 1990s saw a tremendous growth in the acceptance and use of wire-less systems throughout the world. In fact, in the United States alone,the number of wireless subscribers grew from 44 million to 67 millionfrom 1996 to 1998. Needless to say, provision of the E-911 service towireless users is now a necessity. Statistics show that as high as 25% ofall 911 calls made in the last year have originated from a mobile phone.

Evaluation of Location Determination Technologies 159

Furthermore, wireless subscribers identify the ability to be able to call911 from wherever they are as one of the very important factors influ-encing their decision in subscribing to a wireless service [3]. However,unlike most wireline phones in the United States, which have access tothe E-911 service that automatically reports the caller’s location, whena 911 call is placed using a mobile station, currently the dispatcher atthe 911 Public Safety Answering Point (PSAP) does not know wherethe caller is and the wireless users who dial 911 usually cannot describetheir exact location.

On June 12, 1996, the Federal Communications Commission (FCC)adopted a Report and Order which established performance goals andtimetables for the identification of the wireless caller’s phone numberand physical location when dialing the 911 emergency services telephonenumber [4]. The FCC requirements have boosted much research in Loca-tion Determination Technologies (LDTs). Various different technologieshave appeared in the literature. These technologies may be grouped intothree general categories: Mobile Station Based LDTs, Network BasedLDTs and Hybrid Methods.

This chapter is intended to give an overview of the LDTs that havebeen proposed in the literature as possible solutions to the FCC re-quirements. A detailed summary of the FCC ruling on wireless E-911is given next in this chapter. Overviews of the Mobile Station Based,


Network Based and Hybrid LDTs follow. Evaluation criteria and evalu-ation methods to compare these different technologies is briefly outlinedin this chapter as well.

2. FCC RULING ON E-911 FOR WIRELESSSYSTEMS

To improve public safety and extend ALI to wireless callers, the FCCestablished a ruling, subject to certain conditions, for deployment ofE911 features by wireless carriers [4]. In Phase I, which began on April1, 1998, the wireless operators were required to forward the 911 callsfrom mobile phones to a PSAP without any interception for any valida-tion procedures or credit checks. Additionally, analogous to the wirelineE-911 service, the wireless operators were required to relay the caller’stelephone number and the location of the base station or cell site re-ceiving the 911 call. According to the June 12 ,1996 ruling, in Phase II,scheduled for October 1, 2001, wireless operators were required to pro-vide a much more precise location identification, within 125 meters, ofthe caller’s location to the PSAPs in 67 percent of all cases. The need forthe Phase II requirement stemmed from how wireless systems operate.In a practical wireless system, due to the imperfections of the mobileradio transmission terrain or network congestion, the base station thatprocesses the 911 call need not necessarily be the closest to where thecall is actually placed. With the Phase II implementation, ALI may beapplied to route these calls immediately to the proper PSAP, normallyone that is nearest the mobile station, not nearest the serving base sta-tion. In wireless systems, once Phase II is implemented, ALI may alsohelp PSAPs deal with sudden bursts of calls, which often occur afterincidents such as highway accidents. Knowing the location of the in-coming calls, the PSAP can better distinguish redundant calls about aparticular accident from calls concerning a different emergency.

Since the 1996 ruling a variety of LDT proposals surfaced. While someof these proposals required hardware changes to the mobile stations, oth-ers did not. Clearly, though they might be more accurate, LDTs thatrequire hardware changes to the mobile stations will potentially delaythe full availability of the Phase II ruling. FCC, taking this into ac-count, recently revised its Phase II ruling [5]. According to the October6, 1999 ruling, FCC now requires that LDTs requiring mobile stationhardware modifications be held to a higher accuracy standard than theLDTs that do not require such modifications. Allowing a rapid phase-inimplementation, FCC requires that for such LDTs, the modified mobilestations be made available earlier than the current October 1, 2001 de-


ployment date. Additionally, wireless carriers employing mobile stationbased LDTs take additional steps to provide location information forroamers and callers with legacy mobile stations.

FCC also replaced the RMS reliability methodology with a simplerstatistical measure. The 1999 ruling sets levels of accuracy that must beachieved for 67 percent and 95 percent of all calls. The revised ruling nowallows the wireless carriers to reach a 50 percent LDT coverage withinsix months of a PSAP request for Phase II services and 100 percentcoverage eighteen months after a PSAP request.

Specifically, requirements are placed in three categories [5]:

1. Decision for Technology Adoption: Wireless carriers are re-quired to report the LDT (or LDTs) of their choice to FCC byOctober 1, 2000.

2. Deployment Requirements:

(a) LDTs Requiring New, Modified or Upgraded Mobile Stations:

Regardless of whether there is a PSAP request for PhaseII implementation,

– The ALI-capable mobile stations need to be madeavailable to the public by no later than March 1, 2001.

– By October 1, 2001, at least 50 percent of all newmobile stations activated need to be ALI-capable.

– By October 1, 2001, at least 95 percent of all new dig-ital mobile stations activated need to be ALI-capable.

Specifically, once a PSAP request is received for a PhaseII implementation,

– Within six months of the request or by October 1,2001, whichever is later,

* The wireless operator needs to ensure that 100 per-cent of all new mobile stations activated are ALI-capable.

* The wireless operator needs to implement any nec-essary network upgrades to ensure proper opera-tion.

* The wireless operator needs to begin delivering tothe PSAP the location information that satisfiesthe Phase II requirements.

— Within two years of the request or by December 31,2004, whichever is later, the wireless operator needs


to strive for 100 percent penetration of ALI-capablemobile stations in its total subscriber base.

The wireless operators need to support a minimum ofPhase I requirements for roamers and other callers with-out ALI-capable mobile stations.For users with modified mobile stations, roaming amongdifferent wireless operators employing the same LDT needsto be allowed.

(b) LDTs Not Requiring New, Modified or Upgraded Mobile Sta-tions:

The wireless operators need to deploy Phase II to 50 per-cent of callers within 6 months of a PSAP request.The wireless operators need to deploy Phase II to 100percent of callers within 18 months of a PSAP request.

3. Accuracy:

(a) LDTs Not Requiring New, Modified or Upgraded Mobile Sta-tions: An accuracy of 100 meters for 67 percent of calls and300 meters for 95 percent of calls needs to be maintained.

(b) LDTs Requiring New, Modified or Upgraded Mobile Stations:An accuracy of 50 meters for 67 percent of calls and 150meters for 95 percent of calls needs to be maintained.

3. LOCATION DETERMINATIONTECHNOLOGIES

A variety of technologies are available for accurate location determi-nation. These technologies may be grouped based on where the measure-ments towards location estimation are made in the system. LDTs thatuse radiolocation measurements performed only by the mobile stationcan be grouped under the class, Mobile Station Based Methods. Simi-larly, LDTs that use radiolocation measurements performed only by thebase stations can be grouped under the class, Network Based Methods.LDTs that utilize radiolocation measurements performed at both themobile station and the base stations can be grouped under the class,Hybrid Methods [6]. Note that even though this classification is madebased on where the radiolocation measurements are performed, it doesnot specify where the actual location estimation calculations are made.That is, a Mobile Station Based Method, where the measurements to-wards the location estimation are performed by the mobile station, mayhave the calculations done either at the mobile station, at the network


or both. Under the current classification all such alternatives fall underthe same category.

3.1 MOBILE STATION BASED METHODSBroadly defined, a mobile station based LDT is one that detects

and processes signal(s) transmitted from multiple base stations and/orsatellites. Specifically, such methods may be divided into three sub-categories:

3.1.1 MS Based Methods Using Wireless System Signals.The location determination technologies that use signals transmitted bybase stations serving the system to perform algorithms fall into thiscategory.

Theoretically, the position of a receiver can be estimated from themeasurements of the arrival times, directions of arrival, or Doppler shiftsof electromagnetic waves sent by various transmitters whose exact loca-tions are known. If the arrival times are known, the distances betweenthe individual transmitters and the receiver are also known. Supposethat we know the distance, d, between a single base station and the mo-bile station whose location is to be found. As can be seen from Figure5.2, this knowledge narrows down all possible locations the mobile sta-tion could be to the surface of a sphere that is centered around the basestation and has a radius of d.

Suppose next that the distance between a second base station and themobile station is also known. It is then possible to draw two spheres,each centered around one of the two base stations. The mobile stationhas to be somewhere on the circle where the two spheres intersect asseen in Figure 5.3.

From Figure 5.4, if the distance of the mobile station from a thirdbase station is known as well, the position estimation region narrowsdown to only two points where the three spheres intersect.

In order to decide between the two intersection points, a fourth mea-surement could be made. In practice however, usually one of the two


points is an improbable solution and can be rejected without a measure-ment.

Following this reasoning, forward link hyperbolic location systems,often called forward link time difference of arrival (TDOA) systems, lo-cate a mobile station by processing signal arrival-time measurementsfrom three or more base stations. The arrival time measurements fromtwo base stations are combined to produce a relative arrival time that,in the absence of noise and interference, restricts the possible mobilelocation to a hyperboloid with the two stations as foci. Mobile stationlocation is estimated from the intersection of two or more hyperboloidsdetermined from at least three base stations. In the current TIA/EIA-95based CDMA networks, all base stations continually transmit pilot sig-nals which amount to approximately 20% of the total transmitted power.Therefore, forward link time difference of arrival algorithms are readilyapplicable to TIA/EIA-95 systems, where the relative arrival times ofthree or more pilot signals emanating from different base stations areused for the mobile station location estimation.


The requirement for the forward link TDOA to work is for the mobilestation to detect signals from at least three base stations in a tightlysynchronized network. The relative arrival times of the signals from thevisible base stations are then used to form hyperboloids, the intersec-tion of which gives us the location estimates. If there is informationfrom more than base stations, it is possible to form more than two hy-perboloids and find the intersection of all of the hyperboloids.

As in Figure 5.5, assume that the coordinates of the three base sta-tions are known. Without any loss of generality, one can form localcoordinates where the first base station, BS# 1 is centered at the originand the second base station, BS# 2 is somewhere along the local y-axis.The local coordinates of the third base station, BS# 3 and the mobilestation, MS can then easily be defined relative to those of BS# 1 andBS# 2. In other words, assume that the coordinates of the three basestations are as follows:

Furthermore, assume that the mobile station is located at,

Then, the distances between the mobile station and each of the basestations can be calculated using,


where and are the time it takes for the signals (Pilot Signals inthe case of TIA/EIA-95) to travel from BS# 1, BS# 2 and BS# 3 tothe MS, respectively and c is the speed of light. Now, for the 1.2288MHz TIA/EIA-95 system, the arrival times can be described in termsof PN chip offsets using the following ratio,

where PN — OFFi is the offset between the actual PN chip of the i’thbase station (which is the base station identification) and its measuredcounterpart. The TDOA algorithm draws two hyperboloids using,

The above two equations have two unknowns, x and and areknown from the base station GPS coordinates and as well asis measured and thus and are known as well. The two equations in(5.5) and (5.6) can be solved in many different ways. They can be solvediteratively using the Steepest Descent Method, or visually by plottingthe hyperboloids, using Taylor series expansion etc. It is also possible tosolve the set of equations analytically (for a two-dimensional solution)since it is possible to reduce the problem to the solution of a quadraticequation [7, 8]. For a three-dimensional solution, the problem becomesa quartic equation whose analytical solution, though algebraically morecomplicated, is still available.

Taking the squares of both sides of the equalities in (5.5) and (5.6)yields,

Provided that is not equal to zero we can write,


One can re-write (5.9) as,

where

Now, substituting (5.10) into (5.7) results in

which results in

(5.14) is a quadratic equation whose roots give the two coordinatesof the intersection points of the hyperboloids. The corresponding xcoordinates may be found using (5.10).

In an ideal world, where there are no detection errors, no multipathand non-line-of-sight propagation and perfect synchronization amongstthe base stations, the TDOA algorithm will always converge to the truemobile position. In the wireless channel none of these conditions hold.Multipath propagation is prevalent and especially in urban areas, there isa very high probability that most of the received multipaths will be non-line-of-sight. Synchronization errors and detection errors (measurementerrors) are also present. All such impairments cause errors in the locationestimation algorithm. The existence of multipath causes errors in thetiming estimates even when there is a line-of-sight path between thebase station and the mobile station. Conventional delay estimators suchas the delay locked loop, are influenced by the presence of multipathespecially when the multipath signals arrive within a chip period of oneanother [9]. When the first arriving multipath is less powerful than thosearriving later, the delay estimators detect a delay in the vicinity of themore powerful multipath signals. The non-line-of-sight propagation, onthe other hand, introduces a bias in the TDOA measurement because thesignal arriving at the mobile station from the base station is reflectedand thus takes a longer path relative to the line-of-sight path. Thepresence of multiple access interference also influences the accuracy ofTDOA systems. Analogous to the multipath propagation effects, the


existence of multiple access interference deteriorates the performance ofthe delay estimators.

If the base station signals follow a direct line of path, and if the signalarrival times can be detected exactly, the TDOA approach always givesthe true mobile location. This ideal case is illustrated for a specificexample in Figure 5.6. In this case, the true mobile location lies on oneof the intersection points. Figure 5.7, on the other hand, represents asituation where all possible impairments are present. In both figures the

represents the locations of the base stations and the ’o’ representsthe calculated mobile station location. In Figure 5.7, the solid curvesare the hyperboloids drawn when impairments are present in the systemand the dotted curves are the hyperboloids if there were no impairments,i.e., if a genie were to tell us the true distances between the mobile andall three base stations. The ‘o’ represents the estimated mobile locationwhereas the represents the true mobile location.

The relative geometry of the base stations performing the TDOA mea-surements is critical in the performance of the TDOA algorithm. Poorgeometry can lead to high geometric dilution of precision (GDOP). Ifthe geometry of the three base stations performing the TDOA measure-ments is such that the intersecting hyperboloids intersect at a very small


angle, significant estimation errors are observed even with the slightestof impairments. If on the other hand, the hyperboloids intersect at al-most right angles, the impairments present in the system translate tosmall offsets from the true location.

3.1.2 MS Based Methods Using Satellite Signals. The loca-tion determination technologies that use signals transmitted by a numberof satellites to perform algorithms fall into this category. Note that here,technologies make use of only the satellite signals for location determi-nation. Each mobile station needs to be furnished with a stand alonesatellite receiver in this case.

The Global Positioning System (GPS) is a worldwide radio-navigationsystem formed from a constellation of 24 satellites, each 11,000 nauti-cal miles above the Earth, arranged in 6 orbital planes with 4 satellitesper plane as shown in Figure 5.8 [12]. The constellation is designed sothat signals from at least six satellites may be received nearly 100% ofthe time from any non-obstructed place on earth. The GPS satelliteseach take 12 hours to orbit the earth. Each satellite is furnished withan atomic clock that keeps accurate time to within three nanosecondsso it can broadcast its information signals coupled with a precise timing


message. In this regard, the GPS system can also be used as an accuratetiming reference. A global network of ground stations monitor the con-dition of the satellites. Five stations exist: Hawaii and Kwajelein in thePacific Ocean, Diego Garcia in the Indian Ocean, Ascension Island inthe Atlantic Ocean and Colorado Springs in the continental USA. Thisnetwork regularly uploads navigation information and other data to thesatellites without impacting the regular operation of the GPS system.GPS can provide service to infinitely many users as it is a broadcast only

system. GPS is managed for the US Government by the US Air Force.In an effort to make GPS beneficial to non-military applications as well,two GPS services are provided. The Precise Positioning Service (PPS)is available primarily to the US Military and its allies. The StandardPositioning Service (SPS) is designed intentionally to provide a less ac-curate positioning capability than the PPS for civil and all other usersthroughout the world.

As seen from Figure 5.9, the GPS system has 3 parts: the spacesegment, the user segment, and the control segment. The space segmentis made up of the 24 satellites orbiting the earth whereas the controlsegment is made up of the 5 ground stations. The user segment consistsof a GPS receiver placed with the user whose location is to be estimated.

The GPS system utilizes the concept of time difference of arrivalwhich was described in the previous section. As stated before, the GPSsatellites use precise atomic clocks on board to control the frequency


and modulation rate of two L-band carriers, andand are selected to be integer multiples of

a 10.23 MHz master clock [13]. Similarly, all of the signal clock ratesfor the codes, radio frequency carriers and the navigation data streamare coherently related to the master clock. Each satellite has a uniquespreading sequence so that users, upon detection of a satellite signal, candetermine from which satellite the received signals originated. All GPSsatellites transmit signals on the same carriers using DS CDMA. Thesatellites send time stamps of when their codes pass through a phasestate. Based on when the user’s receiver detects that phase state inthe received signal, the propagation delay and therefore distance fromeach visible satellite at time of the time stamp can be estimated. Thisestimate is commonly referred to as the “pseudo-range” in the GPS ter-minology. The satellites also transmit information about their orbits(ephemeris data). The information describing the satellite’s orbit, thecode phase state time stamps and clock offset corrections are providedin the GPS navigation message, D(t), on both and Using thisinformation as well as the pseudo-ranges from at least four satellites, theuser’s position can be determined. From TDOA analysis we know thatonly three satellites would be sufficient to estimate the user location. In


practice, however, the user’s timer is significantly less accurate than anatomic clock. For this reason, information from a fourth satellite is usedto correct the clock bias errors possibly present at the user receiver.

The signal is modulated by both a 10.23 MHz clock rate Precision(P) signal, P(t) and by a 1.023 MHz Coarse Acquisition (C/A) signal,G(t) using quadrature phase modulation. P(t) is used to provide PPSwhereas G(t) is used to provide SPS. The i’th satellite spreads D(t)using both P(t) and G(t) as follows,

)

where is the frequency, represents a small phase noise and oscil-lator drift component and and are the C/A and P signal powers,

respectively. and are the i’th satellite’s navigation,C/A and P signals, respectively. The GPS navigation message, D(t) isa 50 bps signal that has a 1500 bit long frame made up of five subframes.Each satellite begins to transmit a frame precisely on the minute and halfminute, according to its own clock [12]. Subframes 1, 2 and 3 containthe high accuracy ephemeris and clock offset data. The data content ofthese three subframes is the same for a given satellite for consecutiveframes for periods lasting as long as two hours. New subframe 1, 2 and3 data usually begin to be transmitted precisely on the hour. Subframe1 contains second degree polynomial coefficients used to calculate thesatellite clock offset. Subframes 2 and 3 contain the orbital parameters.Subframes 4 and 5 are subcommutated 25 times each, so that a com-plete data message requires the transmission of 25 frames. A satellitetransmits the same data content in subframes 4 and 5 until the next isuploaded by the ground stations, usually for about 24 hours. These sub-frames contain almanac data and some related health and configurationdata. The navigation message contents and format are summarized inFigure 5.10.

As seen from Figure 5.10, each subframe starts with a Telemetry(TLM) word and a Handover word (HOW) pair. The TLM word con-tains an 8-bit Barker word for synchronization. The HOW contains a17-bit Z-count for handover from the C/A code to the P code. Theremaining slots in the subframes are allocated for the clock correction,satellite health, ephemeris and almanac data depending on the subframenumber. Almanac data consists of course orbital parameters for all satel-lites. Each satellite broadcasts almanac data for all satellites. This datais not very precise and is considered valid for up to several months.Ephemeris data by comparison consists of very precise orbital and clock


correction for each satellite and is necessary for precise positioning. Eachsatellite transmits only its own ephemeris data. This data is consideredvalid only for about 30 minutes. Each set of ephemeris data gives a fit in-dication which tells how long the particular data is valid. The ephemerisdata is broadcast by each satellite every 30 seconds.

The C/A code, G(t) is a satellite unique Gold code of period 1023 bitsand has a clock rate of 1.023 Mcps. The P code, P(t) on the other hand,has a period that is slightly more than 38 weeks if allowed to continuewithout a reset and has a clock rate of 10.23 Mcps. On the C/Acode strength is nominally set to be 3 dB stronger than that of the Pcode.

The signal is bi-phase modulated normally by the P code but theC/A code may be selected by ground command as well. For the i’thsatellite, the same 50 bps navigation signal, is modulated by P ( t ) inthe normal operation as follows,

where is the signal power, is the P code for the i’th satellite,which is clocked in synchronism with the codes. Schematically, eachsatellite generates the and signals as shown in Figure 5.11.


A GPS receiver can be visualized as performing four primary func-tions:

1. Determine the code phases (pseudo-ranges) to various GPS satel-lites,

2. Determine the time-of-applicability for the pseudo-ranges,

3. Demodulate the satellite navigation message,

4. Compute the position of the receiving antenna using the pseudoranges, timing and navigation message data.

Most commercial GPS receivers perform all of these operations withoutany external assistance. In the conventional GPS receivers, the satellitenavigation message and its inherent synchronization bits are extractedfrom the GPS signal after it has been acquired and tracked. Such areceiver is illustrated in Figure 5.12. The GPS receivers use correlatorsto compute the pseudo-ranges. A classic hardware correlator based re-ceiver multiplies the received signal by a replica of the satellite’s C/Acode and then integrates the product to obtain a peak correlation signal.Initially, the search for the correlation peak is done over three dimen-sions: satellite, time and frequency.

Satellite: Since each satellite has its own C/A code, a GPS receiver,not knowing which satellites are visible, has to search through allpossible C/A codes to find a correlation peak.


Time: For each satellite, the signal structure consists of a 1023chip long pseudorandom sequence sent at a rate of 1.023 Mcps.To acquire in this dimension, the receiver needs to set an internalclock to the correct one of the 1023 possible time slots by tryingall possible values.

Frequency: The receiver must also correct for inaccuracies in theapparent doppler frequency of the satellite. The receiver’s crystaloscillator may be off by up to due to the doppler off-sets. If we assume that the frequencies are searched in steps of500 Hz, about 40 frequency cells have to be tested for each timeoffset of each satellite making the overall acquisition process quitelaborious.

If no ephemeris data is available from a previous search, the GPS receiveris said to go through a “old start” acquisition. If on the other hand,ephemeris data of three still visible satellites are available, the receiveris said to go through a “warm start” acquisition. Clearly, warm starttakes up considerable less time than cold start.

Once a signal is acquired, the process enters the tracking mode inwhich the C/A code is removed and the GPS navigation message isdespread. The navigation message can be reliably demodulated if thereceived signal strength is above approximately -135dB m for the du-ration of the message being received. As stated before, the navigationmessage structure has a 1500 bit message sent at a rate of 50 bps taking30 seconds. Conventional GPS receivers require the demodulation of a


complete, unbroken 1500 bit message block to use in location estimation.If the detection happens to start at the beginning of the 1500 bits, ittakes 30 seconds for the receiver to copy the entire message content. Ifon the other hand, the detection starts from the first bit of the message,the receiver has to wait for 30 seconds to the start of the next unbrokenblock of data resulting in a processing time of 60 seconds. The averagelatency in this case is 45 seconds.

The performance of the GPS system is affected due to the followingimpairments:

Atmospheric Conditions: As a GPS signal passes through thecharged particles of the ionosphere and then through the watervapor in the troposphere it no longer travels at the speed of light,and this creates the same kind of error as a bad clock would.

There are a couple of ways to minimize this kind of error. Forone thing one can predict what a typical delay might be on atypical day. This is called modeling and it helps but, of course,atmospheric conditions are rarely exactly typical.

Another way to get a handle on these atmosphere-induced errorsis to compare the relative speeds of two different signals. This dualfrequency measurement is very sophisticated and is only possiblewith advanced receivers.

Multipath: Once the GPS signal reaches the earth ground it maybounce off various local obstructions and travel via several pathsbefore it reaches the receiver in question. High end GPS receiversuse sophisticated signal rejection techniques to minimize this prob-lem.

Imperfections at the Satellites: Even though the satellites are verysophisticated they do account for some tiny errors in the system.Although the atomic clocks used in the satellites are very precise,they are not perfect. Minute discrepancies can occur, and thesetranslate into travel time measurement errors. Furthermore, eventhough the satellites’ positions are constantly monitored by theground stations and necessary adjustments are made to the satel-lite signals accordingly, they cannot be watched continuously. Soslight position or ephemeris errors can sneak in between monitor-ing times.

Geometric Dilution of Precision: As with TDOA, basic geometryitself can magnify the errors already present in the system with aprinciple called Geometric Dilution of Precision (GDOP) for the


GPS. If the satellites visible to a user are close together in the sky,the intersecting hyperboloids that define a position will intersectat very shallow angles, causing significant estimation errors evenwith the slightest of impairments. If on the other hand, the visiblesatellites are widely separated, the hyperboloids intersect at almostright angles and therefore the impairments present in the systemtranslate to small offsets from the true location.

Intentional Errors: For civilian use of the GPS system, the De-partment of Defense introduces some noise into the satellite’s clockdata which, in turn, adds noise (or inaccuracy) into position calcu-lations. The Department of Defense may also be sending slightlyerroneous orbital data to the satellites which they transmit backto receivers on the ground as part of a status message. Militaryreceivers use a decryption key to remove the intentional errors.

A method called Differential GPS can significantly reduce these prob-lems. Differential GPS involves the cooperation of a reference receiverwhose location is exactly known free of error. The reference receiver

ties all the satellite measurements into a solid local reference. The ba-sic Differential GPS scheme is shown in Figure 5.13. If the reference


receiver is fairly close to the one whose location is to be estimated, saywithin a few hundred kilometers, the signals that reach both of themwill have traveled through virtually the same slice of atmosphere, andso will have virtually the same errors. Then, the reference receiver whoseexact location is a priori known uses its coordinates to calculate whatthe true GPS signal timing values should be, and compares them withwhat they actually are. The difference is an error correction factor. Thereference receiver then transmits this error information to the receiverin question so it can be used to correct the measurements. Since thereference receiver has no way of knowing which of the many availablesatellites a receiver might be using to calculate its position, it quicklyruns through all the visible satellites and computes each of their errors.Then it encodes this information into a standard format and transmitsit to the receiver. The differential GPS algorithm enhances the accuracyof the GPS system significantly.

3.1.3 MS Based Methods Using Wireless System and Satel-lite Signals. The location determination technologies that fall intothis category use signals transmitted by a number of GPS satellites aswell as a number of wireless system base stations to estimate the mobilestation location.

The GPS system, even though an accurate means to estimate the lo-cation of a user, may be unsuitable for the E-911 application without anymodifications. This is because the GPS system, especially when it hasto go through a cold start, is too slow. It may take up to several minutesfor a GPS receiver to deliver the location estimate. Furthermore, theGPS system, due to the weakness of the received satellite signals onthe earth surface, does not work in buildings or shadowed environments,limiting the E-911 service coverage greatly. Last but not least, a GPSreceiver incorporated into the mobile station hardware may drain thebattery at a very high rate. A number of assisted GPS systems havebeen proposed to circumvent these problems [14, 15, 16]. Figure 5.14shows a diagram of a typical assisted GPS system. The cellular systemalready monitors the GPS signals continually to draw its timing. Theassisted GPS system adds a server into the cellular system architecture.The server needs to be placed in close proximity to the user whose loca-tion is to be estimated since both the server and the user have to see thesame satellites and the satellite signals need to experience similar im-pairments on route to both receivers. In that capacity, the server maybe co-located with the base stations or the switching centers. Throughits connection with the MSC, the assisted GPS server knows the servingcell and sector of the mobile station that is to be located which gives it


a rough idea of how far the mobile station is from the server. Takingthis into account, the server formulates aiding information to the mobilestation so that it can detect the satellites better and quicker. By send-ing the aiding information, the server practically converts the problemof detection of unknown satellite signals that the GPS receiver withinthe mobile station has to the problem of detection of known satellitesignals thereby increasing the probability of detection even when thesatellite signals are weak. Furthermore, the GPS receiver within themobile station no longer needs to go through every possible combina-tion within the three dimensional search for acquisition but only a smallfraction of them, cutting the acquisition delay significantly. Dependingon where one desires to perform the location estimation calculations,it is possible to place a fully functional or a partial GPS receiver intothe mobile station hardware. If a full receiver is placed into the mobilestation, obviously the entire location estimation process can take placewithin the mobile station and the mobile station has to transmit onlythe final latitude-longitude information back to the cellular system forE-911 purposes. If, on the other hand, only a partial GPS receiver isplaced in the mobile station receiver, calculation of some of the locationestimation procedures has to take place at the cellular network. In thiscase, the mobile station has to transmit the measured pseudo-range val-ues back to the cellular network and the assisted GPS server performsthe necessary calculations to calculate the location estimate. This is


done to ease the burden on the mobile station so that battery drainageor mobile station size does not become an issue. However, clearly, thisis done at the expense of a more complicated assisted GPS server andslightly more signaling load on the cellular network.

3.2 NETWORK BASED METHODSBroadly defined, a Network based LDT detects the signal transmitted

from a mobile station and uses that signal to determine the mobile sta-tion location. Within the category of Network based methods, there arethree techniques that are primarily employed: Time Difference of Ar-rival, Angle of Arrival and Location Fingerprinting. These techniquesmay be employed either individually or in combination. The followingis a brief description of each technique:

3.2.1 Time Difference of Arrival. The most commonly usedtechnology for network based location systems is time difference of ar-rival (TDOA), which computes the caller’s location by measuring thedifferences between the arrival times of mobile station transmissions atindividual base stations or cell sites. The TDOA concept has alreadybeen discussed in great detail in section 3.1.1 for the case where thebase station signals arriving at the mobile stations are used to form thetime difference equations. The theory given in that section applies hereas well. One potential concern on using TDOA using the mobile sta-tion signals is the need to ensure that at least three base stations detectthe mobile station’s signal. This may translate to situations where themobile station transmission power needs to be increased significantlyto provide location estimation, causing near-far problem to the othercellular users in the same cell.

3.2.2 Angle of Arrival. Another widely used technology for net-work based location systems is angle of arrival (AOA). The AOA tech-nique determines the direction of arrival of the mobile station’s emittedsignal at the LDT receiver antenna. The phase difference of the signal onelements of a calibrated antenna array mounted at the cell site providesa line of bearing to the mobile station. The intersection of the lines ofbearing from two or more receivers provides the location. As observedin Figure 5.15, there is no ambiguity here because two straight lines canonly intersect at one point. The AOA technique receivers usually eitherutilize the existing base station antennas or use their own antenna ele-ments that are typically co-located with the wireless network’s cell sitebase station.


Like TDOA, AOA is affected by the impairments in the wireless chan-nel as well. Scattering near or around the mobile station as well as thebase station will impact the AOA measurement. When non-line-of-sightsignal components exist, the antenna element may lock onto one of thereflected paths that may not be coming from the direction of the mo-bile station. This will impose a potential problem even if a line-of-sightsignal component is present as well.

The accuracy of the AOA method is inversely proportional to thedistance between the mobile station and the base station. This is dueto the fundamental limitations of the antenna elements used to measurethe arrival angles as well as the changing scattering characteristics of thewireless channel.

The geometry of the antenna elements used to draw the two straightlines affects the performance of AOA as well. If the antenna elementsare located such that the two lines of bearing intersect at a 90° angle,the error is at a minimum. If the two sites are not optimally placed, orif one site is unable to determine a line of bearing for some reason, athird site will prove valuable. The presence of a third sight will improvethe AOA performance in any case, however, the accuracy gain from theaddition of the third line of bearing is small [17].


3.2.3 Location Fingerprinting. A location fingerprinting tech-nique has been proposed as a network based LDT where distinct RFpatterns (multipath phase and amplitude characteristics) of the radiosignals arriving at a receiver antenna from a single mobile station areutilized. The proponents of this technology claim that unique chan-nel characteristics, including its multipath pattern can be linked to acertain geographical area [18]. In other words, the multipath and ampli-tude suppression characteristics for a given location may be regarded asa fingerprint. Thus, as illustrated in Figure 5.16, once a mobile stationtransmits a signal, an associated fingerprint can be calculated from thereceived signal characteristics at a number of base stations in the vicinityof the mobile station [18]. The so-called fingerprint is then compared toa database of previously fingerprinted locations, and a match is made.By matching the fingerprint of the caller’s signal with the database ofknown fingerprints, the caller’s geographic location is identified to oneof the surveyed areas.

3.3 HYBRID METHODSThe Hybrid methods make use of radiolocation measurements per-

formed by both the mobile station and the base stations in conjunction,to produce a more robust estimate of location in a single process. Twotechniques are primarily employed:

3.3.1 Hybrid MS Based Methods Using Wireless Systemand Satellite Signals Plus Network Based Methods. These tech-niques combine GPS satellite and wireless system assisted MS basedmethods with Network based methods. The mobile station collects ge-olocation measurements from the GPS satellite constellation as well assignals from the wireless network’s base stations. The mobile stationthen sends the information back to the PDE which combines these ge-olocation measurements together with geolocation measurements made


by the base station to produce an estimate of the mobile station’s loca-tion. In the absence of sufficient satellite visibility, hybrid methods stilloperate by using the knowledge of the mobile station’s reference timeand pilot phase measurements, as well as round trip delay measure-ments made by the base station. This clearly improves the availabilityof the location service.

3.3.2 Hybrid MS Based Methods Using Wireless Systemplus Network Based Methods. These techniques combine wirelesssystem assisted mobile station based methods with network based meth-ods. The mobile station collects measurements from a number of basestations to perform TDOA. These measurements are then sent back tothe network which combines them together with measurements madeby the network towards an AOA and/or round trip delay analysis toproduce an estimate of the mobile station’s location.

4. EVALUATION OF LOCATIONDETERMINATION TECHNOLOGIES

The E-911 ruling has prompted extensive research on mobile stationlocation determination technologies. As listed in the previous section,a number of technologies have been proposed. According to the FCCruling, the wireless system operators need to choose one (or a subgroup)of these technologies for implementation in their service areas by October1, 2000. Obviously, the operators would like to make sure that the LDTsthey choose at least satisfy the current FCC conditions for E-911. In thiscapacity, a rigorous evaluation criteria needs to be established so that theLDTs can be compared extensively and fairly. The CDMA DevelopmentGroup (CDG) recently published guidelines for testing and evaluatingLDTs that are applicable to the TIA/EIA-95 and TIA/EIA-2000 familyof systems [6].

The CDG guidelines require that field tests be conducted to evaluatethe LDTs using the vendor hardware and software [6]. The use of simu-lations for the evaluation is considered only as an additional option andis not seen as a replacement for the field tests.

4.1 TEST SCENARIOSMobile stations operate in a wide range of environments and condi-

tions. To characterize the LDT performance under a realistic range ofdistinct service areas and environments, aspects such as the type of ter-rain, presence of natural and man-made structures, speed, location ofthe mobile station and time of day, etc. should all be taken into account.


The concept of test scenarios provides a means to condense the rangeof typical operating conditions into a manageable number of test cases.This allows for a comprehensive evaluation of LDTs. The CDG classifiesthe scenarios as rural, suburban, urban, highway and water [6]. Withinthese classes, further distinction regarding type of terrain and foliage,indoor/outdoor location, types of man-made structures, speed and timeof day are made. For fair comparison, the same system operating pa-rameters should be used for all tests. Customization of LDT-specificoperating parameters for a given test scenario should not be permitted.We now define the individual test scenarios.

4.1.1 Rural Environments. Sparsely populated geographic ar-eas with isolated dwellings characterize the rural class of scenarios. Thisclass specifically excludes corridors along highways and freeways as thosescenarios constitute a separate class. Specifically, the following defini-tions apply for analog and digital systems:

AMPS Coverage Area: Isolated Single AMPS Rural CoverageCase. A single, large AMPS omni-directional (or sectorized) base sta-tion coverage area defines this case with no hand-off candidates. Themobile station can only detect a single AMPS FOCC. Only the servingAMPS base station can detect the mobile station.

AMPS Coverage Area: Nominal AMPS Rural Coverage Case.A single, large AMPS omni-directional (or sectorized) base station cov-erage area defines this case with limited hand-off candidates. The mobilestation detects and monitors the strongest AMPS FOCC, though addi-tional weaker control channels may be detectable. Multiple base stationsmay detect the mobile station, but the serving AMPS base station re-mains the same.

CDMA Coverage Area: Isolated Single CDMA Base StationRural Coverage Case. This case is defined by a single, large CDMAomni-directional (or sectorized) base station coverage area with no ad-ditional base stations, either above or below the CDMA T-ADD systemparameter. The mobile station can only detect a single base station pilot.Only the serving CDMA base station can detect the mobile station.

CDMA Coverage Area: Nominal CDMA Rural Coverage Case.This case is defined by a single, large CDMA omni-directional (or sec-torized) base station coverage area with no other base stations exceeding


the CDMA T-ADD system parameter. There is only one base stationin the mobile station’s Active Set.

4.1.2 Suburban Environments. Medium levels of populationdensity, where 1-2 story residential neighborhoods, 2-3 story office build-ings, and public spaces such as large shopping malls and multi-level park-ing garages characterize the suburban class of scenarios. Specifically, thefollowing definitions apply for analog and digital systems:

AMPS-only Coverage Area: Nominal AMPS Suburban Cov-erage Case. A single AMPS omni-directional (or sectorized) base sta-tion coverage area defines this case with hand-off candidates. The mobilestation may detect several AMPS FOCCs and occasionally change thecontrol channel it monitors. Multiple base stations may detect the mo-bile station, with the serving AMPS base station changing occasionally.

CDMA Coverage Area: Nominal CDMA Suburban CoverageCase. This case is defined for soft/softer handoff coverage areas wherethere are 1-3 CDMA omni-directional/sectorized base station(s) in theActive Set. 1-3 base station(s) detect the mobile station.

4.1.3 Urban Environments. High levels of population densitycharacterize the urban class of scenarios, multi-story/high rise apart-ment/office buildings as well as medium height and narrow streets aretypical. Specifically, the following definitions apply for analog and digitalsystems:

AMPS-only Coverage Area: Nominal AMPS Urban CoverageCase. This case is defined for the unlikely case of AMPS-only urbancoverage. The mobile station typically detects multiple AMPS FOCCsand reselects a different control channel with only minor movement ofthe mobile station. Multiple base stations typically detect the mobilestation.

CDMA Coverage Area: Nominal CDMA Urban Coverage Case.This case is defined for soft/softer handoff coverage areas where thereare 1-6 CDMA omni-directional/sectorized base station(s) in the ActiveSet. 1-6 base station(s) detect the mobile station.

4.1.4 Highways. Freeways, primary and secondary roads betweenmajor population centers characterize the highway class of scenarios.Excluded from these areas are heavily urbanized areas where build-


ings are over 2 stories. There is significant overlap in adjacent omni-directional/sectorized base station coverage for mobile station servicealong the driving corridor. Foliage may range from non-existent to adense canopy. Specifically, the following definitions apply for analog anddigital systems:

AMPS-only Coverage Area: Nominal AMPS Highway Cover-age Case. This case is defined for the unlikely case of AMPS-onlyhighway coverage. The mobile station typically detects multiple AMPSFOCCs and reselects a different control channel with only minor move-ment of the mobile station. Multiple base stations typically detect themobile station.

CDMA Coverage Area: Nominal CDMA Highway CoverageCase. This case is defined for soft/softer handoff coverage areas wherethere are 1-6 CDMA omni-directional/sectorized base station(s) in theActive Set. 1-6 base station(s) detect the mobile station.

4.1.5 Water and Waterfront Environments. Proximity to wa-ter bodies such as a lake, bay or ocean categorize the water class of sce-narios. There may be a significant RF delay profile due to over-waterpropagation effects. Specifically, the following definitions apply for ana-log and digital systems:

AMPS-only Coverage Area: Nominal AMPS Water CoverageCase. This case is defined for the unlikely case of AMPS-only watercoverage. The mobile station typically detects multiple AMPS FOCCsand reselects a different control channel with only minor movement ofthe mobile station. Multiple base stations typically detect the mobilestation.

CDMA Coverage Area: Nominal CDMA Water Coverage Case.This case is defined for soft/softer handoff coverage areas where thereare 1-6 CDMA omni-directional/sectorized base station(s) in the ActiveSet. 1-6 base station(s) detect the mobile station.

4.2 TESTING METHODOLOGYTo ensure variety in the test points that make up the statistics, the

CDG requires that three different locations that fit the environmentdefinition be used for each of the test scenarios considered. A totalof 120 test points make up the statistics for a given test scenario in theCDG guidelines document. For testing in CDMA coverage, the locations


selected should be sufficiently far apart so that the mobile stations atthese locations will have entirely different base stations in both theirActive Sets and Candidate Sets. For GPS-enabled LDTs, the locationsshould provide entirely different satellite constellations. To further addvariety to the test points making up the statistics, tests in one of thethree locations is required to be conducted at busy hour while tests forthe second location is required to be conducted at an off-peak hour.Tests for the third location is required to be conducted at night.

The Service Provider conducting the test may choose the scenarios tobe tested based on its network coverage area. For example, a wirelesssystem operator that operates in a landlocked geography need not con-duct water front tests. The outcome of every conducted test should notonly present results for the individual tested scenarios but also a cumu-lative result that is achieved by weighting the tested scenarios based onthe population density and the wireless E-911 calling patterns. It is thisweighted average that may be used to assess whether the LDT undertesting satisfies the FCC criteria in the region of operation.

To obtain the proper weights for a given service area, the wirelesssystem provider should identify the complete set of scenarios that arerepresentative of its service area and establish the expected fraction oftotal calls in each scenario. Additionally, the provider should establishthe fraction of total calls in each scenario during the peak, off-peak andnight hours. The actual sites selected for the tests should be repre-sentative of the expected traffic conditions, as well as, the propagationconditions associated with each scenario. Thus, test results of each sce-nario should have a weight reflecting its expected spatial and temporaldistribution of calls in the service area. Ideally, the sum of the weightsshould be equal to one, however, good results in the high traffic scenar-ios may reduce the need for tests in the very low traffic scenarios duringcompliance tests.

4.3 EVALUATION CRITERIA

The CDG guidelines document identifies 5 criteria to evaluate theLDTS [6]:

1. Accuracy

2. Latency

3. Capacity

4. Reliability

5. Impact on the Wireless Network


We now describe each of these criteria.

4.3.1 Accuracy. The accuracy of the geolocation technology is ameasure that defines how close the location measurements are to the truelocation of the mobile station being located. Of the 5 criteria identifiedby the CDG, accuracy is the only one that is explicitly stated in theFCC ruling.

The accuracy can only be determined when the LDT under testingcan actually provide an FCC Phase II compliant location report withcontents other than sector and cell information. In other words, theaccuracy figures should be composed of test points and times where theLDT is reliable. Reliability is another CDG criteria and is explainedbelow. As stated before, to achieve meaningful statistical results, suf-ficiently many measurement trials should be taken for a particular testscenario to assess the accuracy of the LDT. In this case, accuracy canbe defined as a distribution of the relative distance between the locationestimates and the true location as described by a ground truth algo-rithm. Therefore, the LDT accuracy can be presented graphically asa probability density function and a cumulative distribution functionfor the individual test scenarios. To assess whether the LDT satisfiesthe FCC ruling, the cumulative result achieved by weighting the testscenarios based on the Service Provider’s population density and E-911calling patterns shall can be presented using the probability density andcumulative distribution functions as shown in Figure 5.17. The 95%and the 67% circular error probability (CERP), as well as CERP valuescorresponding to 50 meter, 100 meter, 150 meter and 300 meter errorsas singled out on the resultant graphs since these specific numbers areexplicitly mentioned in the FCC ruling.

4.3.2 Latency. Latency is defined as the time needed from theinstant of mobile station call origination to the instant the location re-port record is sent from the PDE. Even though it is not explicitly statedin the FCC ruling, latency is a very important criteria for the LDTevaluation. In fact, an accuracy value without an associated latencyfigure is not very meaningful as many of the technologies may use post-processing techniques to improve their accuracy numbers at the expenseof increased latency. However, the very nature of the E-911 service re-quires that the position determination be completed as fast as possible.For this reason, the CDG guidelines document requires that each accu-racy number given as a result of an LDT testing in a given test scenariobe coupled with the associated latency figure.



4.3.3 Capacity. The capacity of an LDT is defined as the maxi-mum number of independent, simultaneous location determinations thetechnology can sustain for a given wireless systems load. Capacity mea-surements should be made for unloaded, lightly loaded, medium loadand heavily loaded systems. It should be noted that the capacity mayaffect the location accuracy as well.

Specifically, the capacity of an LDT can be expressed as

The maximum number of independent, simultaneous location es-timates an LDT can handle, expressed as the number of simulta-neous locations, for an unloaded system,

The maximum number of independent, simultaneous location es-timates an LDT can handle, expressed as the number of simulta-neous locations, for a lightly loaded system,

The maximum number of independent, simultaneous location es-timates an LDT can handle, expressed as the number of simulta-neous locations, for an average load system,

The maximum number of independent, simultaneous location es-timates an LDT can handle, expressed as the number of simulta-neous locations, for a heavily loaded system.

These values should be given for each test scenario. Also an averagecapacity value shall be presented for a weighted inclusive set of testscenarios. Clearly, the desired LDT should be able to sustain a largelocation determination capacity for all possible network loads.

4.3.4 Reliability. Reliability is defined as the total number of E-911 calls that result in a location report divided by the total number ofE-911 calls, for each test scenario and for the weighted inclusive set oftest scenarios. The reliability is a measure of the coverage of the LDTwithin the wireless network. Reliability figures should be given for eachtest scenario as well as for the weighted average.

4.3.5 Impact on the Wireless Network. No specific test ormeasurement is needed for this evaluation. However, the following issuesneed to be understood/observed and should be documented:

1. Configuration changes in the cellular network.

2. Software changes in the cellular network.

3. Physical footprint size of the LDT equipment, power requirementsand environmental conditions required, e.g., air conditioning, etc.


4. If the LDT product is evaluated with one wireless technology, e.g.,AMPS and N-AMPS, then how much of the hardware can be re-used (leveraged) for supporting another wireless technology, e.g.,CDMA, or for supporting two wireless carriers using the same ordifferent wireless technologies or for supporting multi-band, i.e.,800 and 1900 MHz, wireless technologies.

For evaluating the impact on the wireless network, the CDG guidelinesdocument suggests recording of the following for each of the LDTS:

1. Hardware additions

2. Software additions

3. Modifications to the communications link

4. Physical area needed to house various components of the LDTequipment

5. Power requirements of the LDT equipment

6. Air conditioning requirements of the LDT equipment

7. Amount of hardware sharing of LDT equipment working with twoor more air interfaces or wireless services, e.g., cellular and PCS.

A somewhat more difficult issue to test is the impact of the LDT onthe wireless system capacity. In other words, how much, if any, doesthe wireless system capacity go down if one or more users request E-911service assuming all other parameters in the wireless systems are keptunchanged. Obviously, the desired LDT should have little, if any, impacton the wireless system capacity.

5. CONCLUSIONSFCC’s mandate to accurately locate wireless 911 callers has acted

as a catalyst for an emerging industry that is focused on developinglocation determination technologies. As a result, many approaches tolocation determination have been introduced. These technologies may begrouped based on where the measurements towards location estimationare made in the system. This chapter gives an overview of the FCC rulingand the location determination technologies that are being consideredby the wireless operators for adoption. In this capacity, TDOA, GPS,Assisted GPS, AOA and Location Fingerprinting technologies have beensummarized.

The wireless operators need to select a location determination tech-nology that satisfies the FCC requirements and is most appropriate to


their needs by October 2000. For this reason, proper evaluation tech-niques are necessary to make an extensive and fair comparison of thedifferent technologies. The FCC ruling quantifies the LDT performanceonly through the location estimation accuracy. Other criteria are nec-essary as well for proper evaluation of different technologies. The CDGrecently established 5 criteria for evaluation purposes, namely, accuracy,latency, capacity, reliability and impact on the wireless network. In thischapter, we provide definitions for the evaluation criteria and describehow the individual LDTs can be tested for each of them.

AcknowledgmentsSection 4 of this chapter contains material from the CDMA Development Doc-

ument prepared by joint efforts of a number of companies and service providers.

Towards this end, Iftekhar Rahman of GTE Labs, Matthew Ward of TruePosition,Scott Fischel of Qualcomm, Len Sheynblat and Karin Watanabe of SnapTrack and

Eddie Hose of Signal Soft are acknowledged.


References

[1] National Emergency Number Association, “The Development of911,” http://www.nena9-l-l.org, Last viewed March 3, 2000.

[2] Minnesota Department of Administration, “911 Population Cov-erage,” http://www.admin.state.mn.us/ telecomm/911.html, Lastviewed March 3, 2000.

[3] Public Opinion Strategies, “National Survey Conductedbetween July 31 and August 4, 1997,” http://www.wow-com.com/consumer/highway/reference/e911poll.cfm, Last viewedFebruary 28, 2000.

[4] Federal Communications Commission, “FCC Adopts Rules to Im-plement Enhanced 911 for Wireless Systems,” FCC News Report,No. DC 96-52, CC Docket No. 94-102, June 12, 1996.

[5] Federal Communications Commission, “Third Report and Order:Revision of the Commission’s Rules to Ensure Compatibility withEnhanced 911 Emergency Calling Systems,” FCC Document, CCDocket No. 94-102 RM-8143, Document No. FCC 99-245, October6, 1999.

[6] M.O. Sunay, “CDG Test Plan Document for Location Determina-tion Technologies Evaluation,” CDG Document, February 17, 2000.

[7] M.O. Sunay and I.Tekin, “Mobile Location Tracking in DS CDMANetworks Using Forward Link Time Difference of Arrival and ItsApplication to Zone-Based Billing,” Proceedings of the IEEE Globe-com’99 Conference, Rio De Janeiro, December 3-6, 1999.

[8] B.T. Fang, “Simple Solutions for Hyperbolic and Related PositionFixes” IEEE Transactions on Aerospace and Electronic Systems,vol. AES-26, no. 5, pp. 748-753, September 1990.

[9] M.K. Simon, J.K. Omura, R.A. Scholtz and B.K. Levitt, SpreadSpectrum Communications Handbook. New York: McGraw Hill,1994.

[10] J.J. Caffrey, Jr. and G.L. Stüber, “Overview of RadiolocationCDMA Cellular Systems, IEEE Communications Magazine, vol. 36,no. 4, pp. 38-45, April 1998.

[11] D.J. Torrieri, “Statistical Theory of Passive Location Systems,”IEEE Transactions on Aerospace and Electronic Systems, vol. AES-20, no. 2, pp. 183-198, March 1984.

[12] L.F. Wiederholt, E.D. Kaplan, “GPS System Segments,” in theedited book, Understanding GPS: Principles and Applications.Boston:Artech House, 1996.


[13] J.J. Spilker, Jr., “GPS Signal Structure and Theoretical Perfor-mance,” in the edited book, Global Positioning System: Theory andApplications. Washington, DC: American Institute of Aeronauticsand Astronautics, 1996.

[14] M. Moeglein and N. Krasner, “An Introduction to SnapTrackServer-Aided GPS Technology,” http://www.snaptrack.com. Lastviewed March 16, 2000.

[15] B. Peterson, D. Bruckner, S. Heye, “Measuring GPS Signals In-doors,” Proceedings of the ION-GPS-97 Conference, Kansas City,September 1997.

[16] B. Richton, G. Vanucci and S. Wilkus, “Assisted GPS for WirelessPhone Location - Technology and Standards,” Chapter 4 in thisbook, May 2000.

[17] H.D. Kennedy and R.B. Woolsey, “Direction-Finding Antennas andSystems,” in the edited book Antenna Engineering Handbook, thirdedition. New York: McGraw Hill, 1993.

[18] US Wireless Corporation, “Location Pattern Matching andthe RadioCamera Network,” http://www.uswcorp.com/USWCMainPages/our. htm, Last viewed April 11, 2000.

About the Author

M. Sunay has been a Member of Technical Staf at Bell Laboratories, LucentTechnologies since 1998. From 1996 to 1998, he was a Research Enginner at NokiaResearch Center. He received his B.Sc. from METU, Ankara, Turkey and M.Sc. andPh.D. from Queen’s University, Kinston, Ontario, Canada, respectively. His currentresearch interests include third generation CDMA systems’ physical and MAC layers,wireless packet data, wireless ad hoc networks, and wireless geolocation systems. Hehas authored numerous articles on these areas in refereed journals and internationalconferances, and has over 10 issued and pending U.S. and European patents. He hasserved and contributed in various telecommunications standards bodies on cdma2000.He was a guest co-editor for the January 2000 special issue of the IEEE Communi-cations Magazine, titled “Telecommunications at the Start of the New Millenium.”His latest appointment involves chairing a task force at the CDMA DevelopmentGroup, which is responsible from developing test plans and criteria for the evaluationof wireless E911 location determination technologies.

Chapter 6

A SERIES OF GSM POSITIONING TRIALS

Malcolm D. MacnaughtanFaculty of Engineering, University of Technology, Sydney

[email protected]

Craig A. ScottFaculty of Engineering, University of Technology, Sydney

[email protected]

Christopher R. DraneFaculty of Engineering, University of Technology, Sydney

[email protected]

Abstract Researchers at UTS have developed a prototype positioning receiver toinvestigate the achievable performance of cellular positioning using theGSM network. Static positioning trials using this receiver have yieldedaccuracies of the order of 100 to 150 meters. The experimental setupfor these trials, the locations in which the trials were conducted and theresults achieved are discussed in detail. The trials have also yielded in-sights into a number of factors affecting the achievable positioning accu-racy including multipath, NLOS reception, interference and the physicalconfiguration of the cellular network. This paper also explores some ofthe complexities associated with establishing conformance regimes forthe FCC E911 mandate.

Keywords: GSM, E-911, positioning, geolocation, cellular systems, location services


1. INTRODUCTIONThe desire for increased subscriber safety [7], combined with growing

awareness of the commercial opportunities [14] has generated significantinterest in cellular geolocation technology. In the particular case of theGlobal System for Mobile Communications (GSM), work is in progressto develop a standard for so called GSM Location Services (LCS), coor-dinated by working group T1P1.5. [8, 9, 15]1. This enhancement will beincorporated in the next release of the GSM standard by the EuropeanTelecommunications Standards Institute (ETSI).

For the past 6 years, a research team at the University of Technology,Sydney (UTS), have been investigating the problem of positioning usingcellular mobile phone signals, with a particular focus on GSM. Thisresearch has led to the development of a prototype positioning receiversuitable for use in field trials. The prototype positioning receiver used inthese trials operates in essentially a self-positioning mode. That is, thereceiver measures its own position. This is in contrast to a network ofgeographically distributed receivers measuring the position of a mobilephone (usually referred to as remote-positioning) [4, 3]. The use of aself-positioning architecture, however, does not diminish the utility ofthe trial results since the majority of the factors which affect the resultshere are common to both self and remote positioning architectures.

The architecture of this positioning receiver and the results of fieldtrials conducted in Sydney using the Telstra GSM cellular network dur-ing December 1999 and January 2000 form the basis for this paper. Theequipment used for these trials, including the prototype positioning re-ceiver is described in section 2.. Section 3. describes the experimentalprocedure as well as the metrics that are used to report the positioningperformance. Section 4. summarises the trials carried out in 4 differentlocations in Sydney along with the results observed. A number of addi-tional observations and insights into the factors affecting the achievablepositioning performance are discussed in section 5.. Some of these obser-vations are of particular interest in relation to the mandate issued by theUS Federal Communications Commission (FCC) [10], as they illustratethe complexities that will be associated with establishing conformancetests to verify compliance with the regulation.

1Readers interested in the evolving GSM location standard should visit the T1P1.5 websiteat http://www.tl.org/index/0521.htm

A Series of GSM Positioning Trials 197

2. TRIAL EQUIPMENTThe test setup for the positioning trials consists of several items of

equipment as illustrated in figure 6.1. This includes a prototype posi-tioning receiver which is supplemented with other equipment to aid inperformance evaluation and to display the measurements on a map. TheUHF receiver together with the modules housed in the VME rack formthe core and are referred to as the GSM Positioning System (GSMPS).The GSMPS is responsible for the actual position measurements. Theposition measurements are displayed in text mode on a monitor and arealso output via an RS-232 interface to the Remote Mapping Terminal(RMT) which is implemented on a laptop PC. The Orbitel-901 field testGSM mobile phone shown in figure 6.1 is used to scan the GSM spectrumto confirm the frequency allocations of GSM cells in the test environ-ment and update the network information file used by the GSMPS. TheDGPS receiver is used for two main purposes. The first is to provide theGSMPS with a measure of its own position when calibrating the RealTime Differences (RTDs) between cells (discussed further in section 3.2).The DGPS is also used when making position measurements to providea reference position measurement for comparison with the GSMPS mea-sured fixes. The GSMPS and RMT are described in greater detail in thefollowing subsections.

2.1 GSMPSThe GSMPS is a flexible 4-channel digital receiver. The front end

is provided by a VHF/UHF communications receiver. This is followedby a high speed, 12-bit A/D converter, a 4-channel Digital Down Con-verter (DDC) board, a Quad DSP board hosting 4 Texas Instruments


TMS320-C40 DSPs and a 486-DX2 IBM-compatible PC. The receivercomponents are, in the main, off-the-shelf components hosted in a VMEequipment rack. Figure 6.2 is a photograph of the receiver. A block di-agram of the receiver showing the modules which comprise the receiverand the interconnections between them is presented in figure 6.3. Thefollowing paragraphs describe the important features of the modules andthe interconnections.

The VHF/UHF front end provides frequency coverage of up to 1200MHz. Received GSM signals are down converted to a 21.4 MHz IF witha bandwidth of 8 MHz. The IF signal is then amplified using an AGCto maintain the signal envelope close to full-scale at the A/D input.Receiver tuning and control is achieved from the PC via an RS-232interface.

The A/D clock frequency is set at 34.66667 MHz, a multiple of theGSM bit rate. The 12-bit output from the A/D converter is supplied viaa ribbon cable interface to the DDC board. This digitised IF signal isapplied to 4 identical but independent digital down converters in parallel.The parameters for each down-converter are able to be programmedeither by the PC (via VME), or by the respective DSPs (also via VME).In the present configuration, the DDCs are programmed to down-converta selected 200kHz GSM channel from the digitised IF signal to baseband,


yielding quadrature outputs at a rate of 541.6667 kHz, i.e. 2 samplesper GSM bit.

The output from each DDC is supplied to the corresponding DSPvia a ribbon cable interface. The interface is interrupt driven, with eachDSP accepting data in blocks of 256 complex samples. The DSPs processthe signals, measuring Time of Arrival (TOA) and other related signalcharacteristics. These measurements are stored in globally accessibleRAM on the DSP board from where they can be retrieved by the PCvia the VME bus.

The PC functions as the VME bus system controller and performs theoverall receiver control functions, coordinating the operation of the fourDDC/DSP channels to obtain signal measurements from the selectedGSM channels. These measurements are then combined at the PC andused to calculate a position measurement.

The 4 channel receiver described above provides significantly moreprocessing power than is required for a self-positioning mobile phone re-ceiver. This architecture, however, provides a great deal of flexibility forresearch and development. For example, four independent channels al-low the system to simultaneously process signals from multiple channels,enabling us to investigate practical issues such as the number of suitablecells for positioning that are available and how that number changesover time as the receiver moves in different localities. Another conve-nient feature of the receiver is the ability to test and compare differentsignal processing algorithms on identical channels by loading the differ-ent algorithms into separate DSPs and testing them on the same signal.Current developments include modifications to the receiver software touse only a single channel, sequentially scanning the selected cells to maketiming measurements and calculate position (which is more typical of apractical implementation in a commercial GSM handset). Depending


on the time and processing resources available in a commercial handset(which will be determined by several factors including the power budgetas well as the other tasks to be performed by the processor), the accu-racy of the single channel system should be only moderately poorer thanthe performance achievable using 4 channels.

2.1.1 Receiver Control and User Interface. The GSMPS canbe controlled either locally with a monitor and keyboard connected tothe embedded 486 PC or alternatively via the serial interface from theRMT. The device drivers supplied with the embedded PC only supportDOS and consequently the local GSMPS display operates in DOS 43line text mode (see figure 6.4). The GSMPS display is divided intoa number of sections. The top right hand corner shows the status ofthe various receiver components including the RF front-end, DDCs andDSPs. For each of the 4 receiver channels, the current operating modeis displayed together with the identifier of the GSM cell currently beingprocessed and parameters such as the received signal level and signal tonoise ratio. The upper left corner of the screen shows the most recentposition measurements together with the true location (using DGPSfixes supplied from the RMT via serial port or entered manually fromthe keyboard). The DGPS input is also used to automatically calculateaccuracy measures when making position measurements. The error inthe most recent measurement is shown in the middle of the upper portionof the screen while several accuracy measures are displayed at completionof a series of measurements in the main window. The screen shot infigure 6.4 shows the accuracy measures calculated after a typical run of100 position measurements.

2.2 REMOTE MAPPING TERMINALThe Remote Mapping Terminal (RMT) is a suite of software applica-

tions running on a separate laptop PC. These applications integrate toprovide a remote interface to the GSMPS as well as providing a real-timemap display of the GSMPS position estimates and DGPS ground-truthposition. The elements of the RMT and the information passed betweenthese elements are illustrated in figure 6.5.

2.2.1 Position Server. The position server is the central com-ponent of the RMT. The position server receives data from the OrbitelMobile Phone, DGPS receiver, and GSMPS via a multi serial port PCM-CIA card. This data is reformatted, filtered and then made available viaTCP/IP to the appropriate recipient(s) as well as being displayed on theposition server window (figure 6.6). The DGPS position information is


transmitted to TCP/IP clients such as the Map Client as well as tothe GSMPS. Position measurements received from the GSMPS are con-


verted from Australian Map Grid (AMG) Eastings and Northings toWorld Geodetic System 1984 (WGS84) latitude and longitude beforetransmission via TCP/IP.

The GSMPS can also send status information, identical to that whichis displayed on the GSMPS monitor, to the position server which isequipped to display this information in the position server window (seefigure 6.6). The position server also has a command window (bottomright-hand corner). Any command entered here is sent verbatim to theGSMPS where the command is buffered for execution at the first avail-able opportunity. Thus the position server can be used to remotelycontrol the GSMPS.

2.2.2 Map Client. The map client receives GSM and GSMPSpositions via a TCP/IP interface to the position server. These positionsare displayed in real-time on a rasterised map display as illustrated infigure 6.7. The map client is a Windows 95 application developed usingMapInfo’s MapX library. The Map Client application supports featuressuch as automatic scrolling/panning, zooming and displaying measure-ment history trails.

2.2.3 Telnet Client. Since the position server provides a TCP/IPinterface, it is possible to use the standard windows Telnet client to con-


nect to the position server. Once connected, a Telnet client will displayall messages sent from the position server. That is, the Telnet clientprovides a real-time listing of the GPS and GSMPS position fixes (fig-ure 6.8). Any data entered into the Telnet client is also sent to theposition server. This enables the GSMPS to be monitored and evencontrolled remotely for instance via a radio modem, enabling the resultsof positioning trials in the field to be displayed at some central site.

2.3 DIFFERENTIAL GPSThe Differential GPS (DGPS) receiver consists of two hardware com-

ponents: a Garmin 12XL GPS receiver and a DCI RDS 300 DGPS datareceiver. The RDS 3000 provides differential GPS corrections to the12XL via an RS-232 serial port. The DGPS measurements from the12XL are sent to the position server using the NMEA 0183 protocol viaa RS-232 serial port. These measurements are used as the ground truthposition for determining the accuracy of the GSMPS and are also usedin calculating the RTDs (discussed in section 3.2).


2.4 ORBITEL MOBILE PHONEThe Orbitel 901 is a special purpose GSM phone designed to report

on the operating status of a GSM network in the local area. Any termi-nal emulation program can be used to control the Orbitel and extractinformation on the current configuration and status of the network. Inthe context of the GSMPS project, the main function of the Orbitel isto report the 4 digit Cell ID and assigned GSM channel frequency ofthe cells in the local vicinity. This data is used to manually update thedatabase of cell sites and frequency allocations etc. in the GSMPS.

3. POSITIONING TRIAL PROCEDUREThe positioning trials involve three main tasks. First, after selecting a

test area, all the cells from a particular network operator in the vicinityare identified2. The second step is to pseudo-synchronise the network,as discussed in section 3.2. Finally, the GPSMS is used to make positionestimates. DGPS measurements are recorded simultaneously as ground-truth for accuracy estimation.

2The reason for not using cells from different operators simultaneously, is simply becausethe 8 MHz IF bandwidth of the prototype receiver will not accommodate any more thanone operator’s spectrum allocation at any one time. There are 3 GSM networks operatingin Australia at present, each using a separate block of approximately 8 MHz of the 25 MHzGSM uplink/downlink bands


3.1 SITE SURVEYSExperiments with the GSMPS can be conducted without the need

for cooperation from the network operator. The receiver is passive, us-ing only broadcast signals from GSM Base Transceiver Stations (BTSs).However, the GSM networks are dynamic as operators expand and re-configure their networks and as a result new cells are added and existingfrequency allocations change frequently. In a commercial positioningimplementation operated by, or in cooperation with the GSM carrier,these changes would be immediately reflected in the system database.During the trials described here, we did not have ready access to thistype of information and therefore each set of field trials required a sitesurvey to determine the locations and broadcast frequency used by eachcell. During these surveys, our DGPS was used to measure the locationof each BTS.

3.2 BASE STATION SYNCHRONISATIONPerforming hyperbolic self-positioning using Time Difference Of Ar-

rival (TDOA) measurements on signals from 3 or more transmittersusually requires the transmitters to be synchronised. This is not thecase with GSM, as the recommendations do not impose any require-ment for BTS synchronisation. However, since the BTS timebases arerelatively stable with respect to one another3 in the short term, it is pos-sible to subtract out the time differences between cells, thereby pseudo-synchronising the cells. These time differences can be calculated by plac-ing the prototype receiver at some known location and then measuringthe TDOA for signals from a pair of cells. These measured time differ-ences are commonly referred to as Observed Time Differences (OTDs).Knowing the location of the receiver as well as the location of the twoBTSs, it is possible to calculate the propagation times from each of theBTSs to the receiver and then subtract these propagation times from theOTDs, leaving only the actual time difference between the BTS clocks.These actual time differences are referred to as Real Time Differences(RTDs).

To measure the RTDs during these positioning trials, the GSMPS isplaced at a location which affords good reception from all of the cellsselected for a particular trial. OTDs between the selected cells are thenmeasured. A DGPS receiver is used to measure the location of thereceiver accurately, enabling the RTDs between all the cells to be com-

3Confirmed during earlier positioning experiments [5].


puted from the OTD measurements. Typically between 100 and 500OTD measurements are made for each pair of cells, in order to averageout noise and interference errors. The RTD measurement process is re-peated for each set of trials as the RTDs drift over time and are onlyaccurate for a period of a few hours (for the network we were using).(Note that in a practical system, the measurement of RTDs would bethe responsibility of the system infrastructure rather than the position-ing receiver).

3.3 POSITIONING MEASUREMENTSThe final stage in the experimental procedure involves the actual po-

sition measurements. The receiver is placed at a series of randomlyselected test sites where a series of position measurements (usually 100)are made. When making position measurements, the OTDs between3 or 4 cells are measured. The previously calculated RTDs were thensubtracted from these OTDs to yield the time differences due only topropagation time (essentially the reverse of the calculation described insection 3.2 above). These differences are then applied to the positioncalculation algorithms to derive a position estimate. A DGPS positionfix is also recorded at each test site for comparison with the calculatedpositions.

The position calculations are made in AMG, a Universal TransverseMercator projection of the Australian Geodetic Datum (AGD) (1966).The DGPS operates in WGS84, as does the map display software. Con-versions between the various datums are performed using functions de-veloped in-house.

As the aim of the trials was to measure the accuracy and performanceof the system and to understand the cause and magnitude of the factorsaffecting the performance, only detailed static position measurementswere conducted in this set of trials. This is not a limitation of the ex-perimental setup or the GSMPS however. Instead we limited ourselvesto static trials only because the already large number of factors affectingthe performance is significantly increased if the receiver is moving, mak-ing detailed performance analyses very difficult. We intend to proceedwith dynamic trials after addressing the issues identified in the statictrials. It should be noted that position measurements from a movingreceiver should provide more accurate results due to the decorrelationin multipath errors.


3.4 PERFORMANCE METRICSThe main performance metrics calculated during these trials are ac-

curacy metrics. It was not the our aim in the trials to date, to providequantified coverage results. The tables of results which follow reportthe 2D Standard Deviation 2DRMS Error, 90% 2DRMS, and the67th percentile error. The use of more than one metric is designed toprovide a clearer overall indication of the errors as no single metric pro-vides a comprehensive description of the magnitude and distribution ofthe errors. The standard deviation measures the variation of the posi-tion estimates about the mean position estimate. As such it indicatesthe repeatability of the measurements or how well the system would per-form if systematic biases and other effects such as static multipath biaseswere eliminated. The 2DRMS error provides an overall estimate of thesystem’s accuracy. However a small number of poor measurements in aset of otherwise accurate measurements can significantly distort the 2DRMS measure, and hence the 90% 2D RMS error metric compared tothe 2D RMS gives an indication of the accuracy after outliers have beensuppressed. The 67th and 90th percentiles provide further indicationsof the magnitude of the positioning error distribution.

4. POSITIONING TRIALS

The prototype positioning receiver and ancillary equipment was in-stalled in a van and powered from batteries via an inverter which pro-vides 240VAC (see figure 6.9). The van was driven to several locationsin Sydney where positioning trials were conducted. The main factor inselecting test sites was the need to start and stop the test vehicle re-peatedly without risking our safety and without interfering with traffic.A secondary factor affecting site selection was convenience. CentennialPark was chosen for its proximity to the University. The remaining siteswere close to the residence of one of the researchers. Four series of trialsare described in the following subsections.

4.1 POSITIONING TRIALS INCENTENNIAL PARK

The first series of positioning trials were conducted in CentennialPark. The park is located near the centre of Sydney and covers anarea of approximately two to three square kilometres. The trials de-scribed here were conducted along a virtually straight stretch of ParkesDrive at the southern end of the park, approximately 800m in length.The immediate surroundings for these trials consist of open grassy areas,


fringed with dense stands of trees (see figure 6.10). There is also a smalllake immediately to the left of the road in the upper photograph. Thisend of Centennial park is virtually ringed by low hills. The areas outsidethe park, approximately 500 meters from the sites used for these trialsare built up with 1 to 2 storey houses, see figure 6.11.

The experiments conducted in Centennial Park included 1,200 mea-surements at a dozen sites. The results are presented in table 6.1. Theaccuracy of the results can be summarised as follows :

Standard deviation of the measurements ranged between 26.2 mand 94.1 m. Across all sites, the average is 64.7 m.

2DRMS error of the measurements ranged between 71.1 m and273.5 m. Across all sites, the average is 156.3 m. The greatermagnitude of the 2DRMS errors compared to the standard devia-tion is the result of significant biases in the measurements at somesites, discussed later in this paper.

The 90 percent 2DRMS error of the measurements ranged between56.8 m and 256.2 m. Across all sites, the average is 143.4 m.


The 67th percentile of the errors at each test site ranged between60.1 m and 298.6 m. Across all sites, the average is 162.8 m.

4.2 POSITIONING TRIALS IN MONTEREYThe second series of positioning trials were conducted along a 1 km

stretch of O’Connell Street in the Sydney suburb of Monterey, located12km south of the CBD, just west of Botany Bay. The surroundingregion is relatively flat and would probably best be described as suburban(see figure 6.12). Both sides of the street are lined with medium density1-2 storey houses and there is a steady stream of light-vehicle traffic.

The experiments conducted in Monterey included 2,600 measurementsat 26 sites. The results are presented in table 6.2. The accuracy of theresults can be summarised as follows :



2DRMS error of the measurements ranged between 45.1 m and457.6 m. Across all sites, the average is 132.8 m.



4.3 POSITIONING TRIALS IN SANS SOUCIPOINT

The third series of positioning trials were conducted along a numberof streets in the suburb of Sans Souci 16km south of the CBD where theGeorges River runs into Botany Bay. The suburb is similar in terrain anddevelopment to Monterey but the streets on which we conducted testshad very little traffic (see figure 6.15). The most significant differencebetween the Monterey and Sans Souci trials was that the latter wereconducted over many different streets covering a much larger area.


The experiments conducted in Sans Souci included 1,800 measure-ments at 18 sites. The results are presented in table 6.3. The accuracyof the results can be summarised as follows :


2DRMS error of the measurements ranged between 54.0 m and405.2 m. Across all sites, the average is 141.3 m.



4.4 POSITIONING TRIALS IN ALLAWAHThe final series of positioning trials were conducted over a number

of streets in the suburb of Allawah 12km south of the CBD approxi-mately 4km west of Botany Bay. This area is much more hilly thanany of the other sites, the streets are in general narrower and tree-lined(see figure 6.15). However, the area is still probably best described assuburban.


The experiments conducted in Allawah included 1,600 measurementsat 16 sites. The results are presented in table 6.4. The accuracy of theresults can be summarised as follows :


2DRMS error of the measurements ranged between 45.5 m and378.5 m. Across all sites, the average is 120.7 m. The greatermagnitude of the 2DRMS errors compared to the standard devia-tion is the result of significant biases in the measurements at somesites, discussed later in this paper.




4.5 SUMMARY OF TRIAL RESULTS4.5.1 Accuracy achieved during trials. The primary aim ofthese trials was to assess the achievable accuracy in a range of locali-ties. The results tabulated above show RMS errors on average in theorder of 100 to 150 metres. While there are likely to be a number offactors contributing to these errors including noise, interference, trans-mitter/receiver clock drift etc. we believe that the major contributorsto the overall error are multipath and possibly NLOS reception. Thebasis for this view is that in all cases, the variation within a set of 100measurements is significantly smaller than the actual RMS error for thatset of measurements. In other words, the measurements are clusteredat some consistent offset from the true location. Noise and interference


effects will be uncorrelated from one measurement interval to the nextand can be expected to produce an unbiased elliptical 2D error distrib-ution [17]. By contrast the fact that each set of 100 measurements weremade while the receiver was stationary means that the multipath andany NLOS effects would be largely stationary, resulting in a consistentbias. (Some variation in multipath could be expected due to passingvehicles at some sites).

The multipath and NLOS errors can enter the position calculationsin these trials at two stages. The first is during the RTD measurements


described in section 3.2 which are made to pseudo-synchronise the BTSs.In this case the measured RTDs will be biased. In a series of positionmeasurements along a straight road, (such as made in these trials), abias of this type in the RTD measurements will manifest itself in theform of a varying bias at each of the sites that follows an approximatelylinear progression. We have observed such errors in previous trials wherethe RTD measurements were made at a single site. In these recent trials


however, we have attempted to reduce such errors by making and aver-aging RTD measurements at a number of sites. The sites were spacedsufficiently apart to decorrelate the multipath. The fact that the RTDmeasurement sites were spaced 50 to 100 m apart in an area where themain propagation obstacles are 2 storey houses should mean that theNLOS errors (if any) will also be reduced by averaging).

The second stage at which multipath and NLOS errors may enter theposition calculations is in measuring the individual TOAs to calculateposition. In this case, the multipath effects will be uncorrelated fromone trial site to another (the trial sites were separated by at least 50m), in which case we expect the positioning errors to exhibit a randombias from one site to the next. This matches the pattern of resultsobserved in these trials. This observation highlights the importance ofdeveloping robust techniques for dealing with the multipath errors inorder to achieve accurate positioning.

4.5.2 Coverage. Although these trials were not conducted withthe objective of assessing the coverage, we can make a few observations


about the likely coverage, based on the number of cells that we couldhear in the different trial areas. In the areas that we used, the numberof cells available for positioning ranged from 3 to 10. In most of thesecases however, there were one or more pairs of (sectorised) cells from thesame site meaning that the actual number of useful cells for positioningis lower,

5. EXPERIMENTAL OBSERVATIONSThe trials described in this section are only the first in an extended

series of trials planned in a range of environments. Although these tri-als were somewhat limited in scope, there are a number of points to benoted as well as a number of conclusions that can be drawn. In particu-lar, the trials have illustrated the difficulties associated with specifyingpositioning performance criteria for cellular mobile phone systems.

5.1 OBTAINING CELL SITE COORDINATESThe rate of expansion of the GSM networks in Sydney proved to be

a source of difficulty. Although we had been provided with a list of cellsites and coordinates some months earlier, we found several new cellsin each of the trial areas. While this is an issue for experimenters such


as ourselves, the close coupling of the positioning system with the GSMnetwork proposed in the forthcoming GSM LCS standard will make iteasier for this type of information to be delivered to the positioningprocessor.

Another factor affecting our initial experiments was the accuracy towhich the location of the cell sites had been measured. While the net-work operators know the location of their base stations, their accuracyrequirements are significantly lower than the requirements when usingthe cells for positioning. (We also observed this problem when conduct-ing trials in the UK). For the trials described here, we surveyed each of


the cell sites ourselves using a DGPS although limited access in somecases meant we could not get directly under the antennas and had to es-timate the location of the antenna using a DGPS fix at a perimeter fence.During our surveys, we also observed that whilst many co-located cellsare treated as having the same coordinates in the network database, theantennas can actually be some distance apart, especially when the anten-nas for sectorised sites are mounted on different faces of a building. Forthe positioning trials described here, this necessitated identifying whichcell was which and then surveying each set of antennas separately.

A further factor which has limited our trials, particularly in Urbanareas, is the fact that an increasing number of the cells in these areasare so-called microcells. Typical installation sites for microcell antennasinclude tops of traffic signals, on ledges above shop doors etc. In thesecases, the actual BTS equipment may be installed some distance fromthe antenna, with a long fiber-optic cable run to the antenna4. In suchcases, the list of cell sites with which we were provided, only lists the lo-cation of the BTS equipment, not the actual antenna site. As a side note,

4 Private correspondence with a Telstra engineer


if in the future base stations are synchronised, the separation of the mi-crocell from the BTS equipment will complicate the positioning processas the signals radiated from such separated antennas will be delayed bythe cable propagation delay from the BTS equipment to the antenna,introducing a bias into the position calculations if it is assumed thatthe transmitted signals are synchronised. To further complicate mat-ters, some microcells are simulcast (1 cell operating from two antennasat distinct locations), and some are time multiplexed (a busy simulcastcell is dynamically split into two cells on different frequencies).

The geographic distribution of cells is another factor affecting posi-tioning performance. Typically cell site locations are selected in an effortto provide adequate voice coverage. Often the cells will be installed onhigh ground and the distribution will, in general, not be homogeneousover the coverage area. This distribution affects the positioning accu-racy that can be achieved. For positioning, it is desirable (for smallHDOP) to use cells which form a polygon enclosing the receiver. In sev-eral cases however, a large building or a stand of trees blocked receptionfrom a particular side significantly increasing the HDOP. In addition, onseveral occasions, we observed that we could receive more signals thanare required for positioning but several of these signals originated fromco-located (sectorised) cells. The majority of cells in the areas used forthe trials are sectored with three cells installed at one location. Theoverall effect was that we did not have the superfluity of useable cellsthat we expected. One implication of this is that as the demand forcellular positioning grows in the future, network installations may alsohave to take into account the needs of positioning as well as the existingcapacity considerations in cell site placement.

5.2 MULTIPATH ERRORSMultipath is likely to be one of the major factors affecting the accuracy

of a GSM positioning system. Firstly, the so-called fast fading can resultin significant variations in the received signal quality. This can resultin strong channels becoming virtually unusable with movement over arelatively short distance. This may mean for instance, that very smallmovements make the difference between having 3 or more cells or havingtoo few cells to make position measurements. These dramatic variationscaused by movements over small distances is one issue that any FCCconformance regime must address, perhaps by evaluating a large numberof position measurements over a wide range of representative locations.

Multipath also results in positioning errors for TDOA based posi-tioning systems by distorting the shape of the correlation peak used to


measure the signal TOA [6]. Simulations using existing ETSI modelsfor multipath in GSM indicate that TOA errors ranging between a fewtens of meters in rural environments and several hundred meters in moredense urban environments can be expected [12]. This is confirmed by ourexperiments with relatively small variations in successive (static multi-path) measurements at a particular site but relatively large variationsbetween measurements at different sites separated by even small dis-tances (due to the decorrelated multipath effects at the different sites).The reduction of multipath errors is a key area of on-going endeavor.

5.3 NLOS ERRORSA further significant source of errors, particularly in more densely

built-up environments, is likely to be NLOS reception. The receiverused during these trials did not employ any techniques to explicitly tryand identify or correct for such errors. It is difficult to estimate whatinfluence such errors may have had on these trials, however it is likelythat further improvements would be gained with the incorporation oftechniques to deal with these potentially large errors. While some rel-atively primitive techniques have been proposed for dealing with theseerrors [13, 18], this remains an open area of research.

In one spot in Monterey, we did have the opportunity to experimentwith the effects of NLOS. We were using a DGPS to measure RTDs.Because of the relatively flat area we could visually identify several ofthe cells in neighboring suburbs on the horizon. In one spot on O’ConnelSt. by moving a few meters we could move from having LOS to a par-ticular cell to being hidden behind a large 2-storey house. Measuringthe RTDs before and after moving showed a consistent jump of approx-imately 1900m

5.4 PSEUDO-SYNCHRONISATION ERRORSAs described in section 3.2, it is necessary when implementing a

TDOA-based GSM positioning system to pseudo-synchronise the basestations. Errors in this process also contribute to the overall positioningerrors. In fact this has proved to be a major source of errors in ourtrials to date. This is due firstly to the difficulty in finding suitable sitesfor measuring RTDs which have LOS reception from a number of thecells to be used in the trials. A second problem is the potential for largestatic multipath errors in the RTD measurements. We have tried to re-duce these errors by repeating the RTD measurements at multiple sitesand averaging the results. In some cases however the variation in RTDsbetween observation sites was of the order of 200 to 300 meters. In such


cases, averaging at 4 different sites can still leave errors of around 150meters which manifest in the position estimates as biases. While thisis a serious hurdle for these types of positioning trials, there are severaloptions for limiting these errors in a practical implementation. Theseinclude carefully selecting reference receiver sites, to provide clear LOSto the cells of interest. The use of directional antennas would also signif-icantly reduce the problems caused by multipath. Another alternativewhich completely removes the need for measuring RTDs would be tosynchronise all the cells, perhaps using GPS time transfer receivers.

Further errors are introduced by the relative drifts between BTS clocksthat occur after our RTD measurements are completed. Our analysis in-dicates that at this stage however, these errors are negligible comparedwith the errors due to multipath, noise and interference. A commer-cial installation measuring RTDs would be able to install a network ofreceivers utilising directional antennas, averaging and other means in-cluding predictive models for the relative clock drifts to minimise theerrors in the RTDs. (Some of the T1P1.5 contributors have completedsimilar measurements, and used them as the basis for calculations of theminimum update rate for RTD measurements. See for instance, [2]).

5.5 CELL SELECTIONAs noted above, the number of cells available for a position measure-

ment varies significantly depending on the environment. The prototypepositioning receiver used for these trials is designed to track up to 4 cellsat any one time. In situations where there are more than 4 cells availablefor positioning this necessitates a decision on which 4 of the availablecells should be used. The most straightforward way to do this is to usethe carrier to interference ratios on each of the channels to estimate thelikely ranging error and then use these ranging errors to predict the po-sitioning error (which is a function of these ranging errors as well as therelative geometry, i.e. the HDOP). The set of 4 cells with the lowestpredicted positioning error is then the natural choice.

Experiments during these trials showed however that this decisionprocess does not always lead to the best solution, in fact it can resultin significantly larger errors than if some other method was used. Onereason for this is that with the dynamic nature of the traffic in the GSMnetwork (based on handovers, frequency hopping, DTX etc.), carrier tointerference ratio measurements for a particular cell are strictly onlyvalid for the timeslot in which they were measured. Although on av-erage a measurement in one TDMA time slot is a reasonable predictorfor the conditions in a subsequent time slot, we have frequently found


this not to be the case. As a result the decision about which cells touse (particularly where the receiver is tasked with making a series ofrepeated measurements) needs to be a flexible one which can be revisedfrequently5.

Another reason why the intuitive cell selection approach above maynot always be the best is that channel ranging error predictions basedon the measured carrier to noise and interference ratio do not take intoaccount multipath or NLOS errors which are likely to be larger on aver-age than the errors due to interference and or AGWN. There are somepossibilities for estimating the relative likelihood of significant multipatherrors including the relative locations (particularly elevations) of cells aswell as analysing the measured channel power delay profiles. Similarlythere are techniques available for detecting channels with large NLOSerrors. However these all require the receiver to actually make somemeasurement using the channel in question, before the NLOS errors canbe detected. Once again this means that the channel selection decisionshave to be flexible and able to be altered quickly.

5.6 VARIATIONS WITH TIME OF DAYExperiments in our laboratory as well as during these trials have

shown that the time-of-day has a significant effect on the accuracy ofthe system. At certain times of the day, especially the morning and af-ternoon peak transit times, the level of GSM subscriber traffic increasessignificantly and with it the levels of co-channel and adjacent channel in-terference. This produces a corresponding degradation in the achievableaccuracy. Again this is a factor that will have to be considered whentesting systems for compliance with the FCC regulations.

5.7 INTERFERENCE VARIATIONS WITHELEVATION

Another factor that affects the level of interference observed by thepositioning receiver is elevation. At ground level, the signal level ofdistant base stations, and hence the interference level, is likely to berelatively low as large buildings and in some cases hills will block thesignal. In tall buildings, however, particularly near external walls andwindows, there is often line-of-sight reception from neighbouring base

5The cost/benefit analysis for dropping one cell and picking up an alternative one is mademore complicated if the receiver uses averaging across multiple bursts as a technique forimproving the TOA measurement accuracy. In such cases, dropping an existing cell losesthe accumulated information from that cell and means that there will be a period of relativeinaccuracy with the new cell as the receiver gathers data to average


stations as well as a number of more distant cells resulting in greaterco-channel interference and adjacent channel interference.

5.8 VERTICAL ERRORSThe GSMPS operates in two dimensional space. This is because the

difference in elevation between the GSMPS and the neighbouring basestations is relatively small resulting in a very large Vertical Dilution ofPrecision (VDOP). As a result attempts to measure location in the ver-tical dimension are virtually pointless. Simple geometrical calculationsreveal however, that when a receiver at ground level is very close to a basestation whose antennas are elevated on a multi-storey building, there isa measurable increase in propagation time due to the height componentbetween the GSMPS and the base station. This will introduce errorsinto the 2D position estimation algorithm. We have developed a recur-sive algorithm to resolve this problem but have not yet implemented itin the GSMPS.

5.9 PROBLEMS ARISING FROMCOMPARABLE BASELINE LENGTHAND RANGING ERROR MAGNITUDES

Cellular mobile phone networks are different from many positioningsystems in that the baselines between transmitters can be of a similarmagnitude to the expected signal measurement errors. We have observedmany locations where the timing errors experienced are comparable tothe distances between the base stations6. This can result for instance ina noisy time-difference-of-arrival measurement which is larger than thedistance between the two base stations. This required modifications tothe “standard” positioning algorithms which in general do not have toaddress this problem as the baselines of such systems are many magni-tudes greater than the timing errors.

5.10 OTHER EXPERIMENTAL PROBLEMS

Finding areas where we can start and stop the test vehicle repeatedlywithout interfering with traffic and without affecting our safety, is an-other experimental problem we have had to address. This has affectedthe choice of areas in which we have conducted trials. In a moving trial

6 Compare GPS where typical ranging errors to a given satellite are of the order of a few tensof metres while the average satellite-to-user distance is of the order of


this would be less of a problem although the need to determine RTDsusing would still involve some stationary measurements.

In several cases, the dynamic range of our wideband receiver provedto be a significant limitation. This limitation means that on occasionsstrong signals from nearby cells may prevent reception of weak signalsfrom distant cells. This problem arises because of the particular ar-chitecture of our prototype receiver, being a common limitation withwideband digital receivers [1]. While this is less likely to be a problemfor a typical narrowband GSM Mobile Station (MS) receiver, it may wellbe a consideration for a remote positioning receiver where a widebanddigital architecture might be preferable to multiple narrowband chan-nels. One option in such cases is to use an A/D converter with greaterresolution and therefore greater instantaneous dynamic range. In factthis is presently an area of rapid technological advance and it appearslikely that A/D converters with the requisite bandwidth and dynamicrange will be commercially available shortly. Further improvements canbe made by using amplifiers with better linearity and mixers with betterintermodulation performance to reduce the likelihood of strong signalsswamping weaker signals.

6. SUMMARYGSM self-positioning trials were conducted in four different areas of

Sydney using the Telstra GSM network. Overall the accuracy observedduring these trials was of the order of 100 to 150 metres. We believe asignificant proportion of these errors were the result of systematic errorsin the test setup. Several directions have been identified for reducingthese systematic errors which we expect will lead to increased accuracyin subsequent trials. Comparing the results of these trials with expec-tations based on extensive simulations [12] shows a satisfactory level ofagreement after taking into account the biases introduced by inaccuratecell site coordinates and biased RTD measurements.

We have not undertaken any moving trials as yet. This has beenprimarily to limit the number of variables to enable a careful analysisof the factors affecting the positioning performance. In dynamic trials,the envelope fading caused by multipath can make it necessary for thereceiver to dynamically acquire and discard particular cells, which canobscure other more fundamental issues. Our intention is to commencedynamic trials once we are confident that we have addressed the sig-nificant issues arising from the static trials. Based on prior experience


with radio positioning systems however7, it is likely that the accuracywill improve with a moving receiver as the movement will decorrelatethe multipath induced errors enabling them to be reduced by averaging.

There are a range of measures for the performance of wireless loca-tion systems, covering aspects including accuracy, coverage, update rateetc. [3, pp. 25-46], [16]. In this report we have dealt primarily withthe achievable accuracy of a GSM positioning system in selected envi-ronments. It is too premature to discuss the other aspects since theseare only preliminary trials in a few areas, and cannot be considered tobe representative of the full range of practical environments in whichGSM positioning systems will be expected to operate. Clearly therefore,there is much scope for further practical trials in many more locations,at different times of day, both while moving and while stationary etc.

The trials described here have been conducted using what amountsto a self-positioning receiver. In general the results and the conclusionsdrawn from those results are also applicable to a network based archi-tecture, while taking account of a few differences. These differencesinclude the shorter training sequence lengths used on uplink bursts, thepower control and frequency hopping that may optionally be employedon uplink channels and the variation in interference levels in the uplinkbands compared to the downlink BCCH channels. In addition, a selfpositioning receiver is likely to have more opportunities for making andintegrating TOA measurements. This combined with the longer trainingsequence in the Synchronisation Bursts (SBs), is likely to lead to greateraccuracy than for a network-based solution. In other words, trials ofnetwork-based positioning, in similar circumstances to those describedabove, are likely to lead to moderately larger errors.

As a side note, the prototype positioning receiver is a powerful toolfor positioning research. Following the recent upgrades, it now enables alarge number of position measurements to be gathered easily along witha significant amount of other pertinent information including receivedsignal levels, signal quality estimates, frequency offset estimates, powerdelay profiles etc. The flexible architecture of the receiver also meansthat it can be easily adapted for positioning trials with other mobilephone systems. The interface with the RMT also supports a real-timemap display. All measurements made by the GSMPS are also written toa log file with a time stamp which enables trials to be replayed for furtheranalysis. The main problem with the receiver is a lack of dynamic range

7Prof. Drane’s prior experience with spread spectrum tracking systems.


which in particular limits the operation of the receiver when it is in veryclose proximity to a particular base station.

In the future we are planning to implement a single channel versionof the prototype positioning receiver to experimentally demonstrate theaccuracy that would be achievable using a standard mobile phone re-ceiver (i.e. without the advantages of a multi-channel receiver). Wehave been investigating multipath rejection algorithms for some timeusing both simulations as well as tests with the GSMPS. We have alsodeveloped many other ideas for investigation which should lead to an ongoing improvement in the performance of the prototype.

Acknowledgments

The authors wish to thank Mr. Miguel Miranda for his development efforts on the

position server and map client software as well as Mr. Brett van-Zuylen of Telstra forenlightening us on some of the practical aspects of GSM network configuration.

References

[1] B. Brannon. Wide dynamic range A/D converters pave the wayfor wideband digital-radio receivers. Analog Devices, Inc. technicalpaper, 1996.

[2] J. Clarke. T1P1.5/99-642: BTS synchronization requirements andLMU update for E-OTD. Submission to location standards workinggroup T1P1.5 by CPS, October 1999.

[3] C.R. Drane and C. Rizos. Positioning Systems in Intelligent Trans-portation Systems. Artech House, 1998.

[4] C.R. Drane. Positioning Systems, A Unified Approach. LectureNotes in Control and Information Sciences. Springer-Verlag, 1992.

[5] C.R. Drane, M.D. Macnaughtan, and C.A. Scott. The accurate lo-cation of GSM mobile telephones. In Proceedings of Third WorldCongress of Intelligent Transport Systems, FL, October 1996.

[6] C.R. Drane, M.D. Macnaughtan, and C.A. Scott. Positioning GSMtelephones. IEEE Communications Magazine, 36(4):46-59, April1998.

[7] C.J. Driscoll. Locating wireless 9-1-1 callers is there a solution tothe problem. 9-1-1 magazine, pages 38-42, July/August 1995.

[8] ETSI. GSM 02.71: “Digi ta l Cellular Telecommunication System;Stage 1 Services Description of LCS Phase Ver. 7.0.0, 1999.

[9] ETSI. GSM 03.71: “Digital Cellular Telecommunication System;Stage 2 Functional Description of LCS Phase Ver. 7.0.0, 1999.


[10] FCC. Revision of the commission’s rules to ensure the compatibilitywith the enhanced 911 emergency calling systems, June 1996. FCCDocket No. 94-102.

[11] B. Hoffman-Wellenhof, H. Lichtenegger and J. Collins. GPS Theoryand Practice, chapter 6, pages 124-128, Springer-Verlag, 3 edition,1994.

[12] M.D. Macnaughtan. Accurately Locating GSM Mobile Telephones.PhD thesis, University of Technology, Sydney, Australia, March2000.

[13] M.I. Silventoinen and T. Rantalainen. Mobile station emergencylocating in GSM. In Proceedings of IEEE International Conferenceon Personal Wireless Communications 1996, pages 232-238, 1996.

[14] Wireless location services: 1997. Industry survey, 1997.

[15] T1P1.5 GSM10.71: “Digital Cellular Telecommunication System(Phase Project Scheduling and Open Issues: Location Services(LCS).”

[16] S. Tekinay, E. Chao and R. Richton. Performance benchmarkingfor wireless location systems. IEEE Communications Magazine,36(4):72-76, April 1998.

[17] Don J. Torrieri. Statistical theory of passive location systems. IEEETransactions on Aerospace and Electronic Systems, AES-20(2):183-198, March 1984.

[18] Marilynn P. Wylie and Jack Holtzman. Non-line of sight problemin mobile location estimation. In Proceedings of 1996 5’th IEEEInternational Conference on Universal Personal Communications,ICUPC’96, 1996.


About the Authors

Dr. Malcolm Macnaughtan has recently completed his PhD at UTS. His thesisentitled, ”Accurately locating GSM mobile telephones”, examines a range of issues forGSM positioning, focusing particularly on reducing the errors caused by multipath.His doctoral research included simulations of the GSM radio channel as well as thedevelopment of two prototype receivers and an extensive series of practical positioningtrials. Since completing his PhD, Dr Macnaughtan has continued to work with theIntelligent Transportation Systems group at UTS, carrying out further research intocellular positioning. His other research interests include signal processing for wirelesscommunication and software radio receivers.

Dr. Craig Scott is a Senior Lecturer and Program Director for Computer SystemsEngineering at the University of Technology, Sydney. Dr. Scott has been involvedin positioning research for the last 10 years. His doctoral thesis examined meansfor improving the tracking of motor vehicles by incorporating extra sources of infor-mation, in particular maps. Since completing his PhD, he has concentrated on theGSM mobile phone positioning research project at UTS. In particular, for the past 8months, Dr. Scott has used his sabbatical to work on the project full time improv-ing the system’s software, and extending and improving the underlying positioningalgorithms.

Professor Chris Drane is Professor of Computer Systems Engineering at the Uni-versity of Technology, Sydney (UTS). His research group works in cellular positioning,positioning theory, and the application of positioning to Intelligent TransportationSystems. He received his BSc(Hons) from the University of Sydney in 1976 and hisPhD from the Physics School at the University of Sydney in 1981. He has been atUTS for nine years with sabbatical leaves at Cambridge University and ITS America.He is the author of many papers and two books on positioning.


Chapter 7

ENHANCING TERMINAL COVERAGE ANDFAULT RECOVERY IN CONFIGURABLECELLULAR NETWORKS USING GEOLOCATIONSERVICES

MOSTAFA A. BASSIOUNI and WEI CUISchool of Computer Science- University of Central Florida, Orlando, Florida, USA

Abstract: In this paper, we discuss the application of geolocation services in improvingmobile connectivity and enhancing the effectiveness of fault recovery inconfigurable cellular networks. Real-time location measurements (e.g., GPS)are used to guide the movement of the mobile base stations to provide bettercoverage of the different groups (swarms) of mobile terminals. Umbrellacoverage via a more powerful transceiver is used to enhance the overallterminal coverage and simplify the movement coordination strategies andimprove the efficiency of the channel allocation protocol. When a base stationbecomes immobilized or faulty, a recovery protocol is used to preventdiscontinuation of coverage for the mobile terminals that were being servicedby this base station. Real-time location measurements are crucially importantfor the proper execution of the recovery protocol.

Keywords: mobile positioning, GPS, cellular networks, mobile base stations, channelallocation, handoff blocking, new call admission.

1. INTRODUCTION

Cellular wireless systems and their related algorithms have beenproposed and evaluated for the purpose of achieving better utilization of theradio spectrum and improving the QoS of wireless connections [CHO98,CHI00]. In traditional cellular systems, the service area is divided intoregions called cells. Each cell is served by a stationary base station (BS).Base stations are connected via wirelines to mobile switching centers which


provide the interface to the wired backbone. When the mobile crosses theboundary of its current cell and enters into a new cell, the base station of thenew cell must assume the responsibility of servicing the ongoing call(connection) of that mobile. This process is called handoff and it is themechanism that transfers an ongoing call from the current cell to the nextcell. It is possible that the new base station does not have a free channel(e.g., frequency band in FDMA cellular systems) to service the incomingmobile and the connection of that mobile gets blocked, i.e., is forced toterminate. The handoff blocking probability is an important quality ofservice (QoS) parameter in cellular systems; careful design schemes andarchitectures must be used to minimize the handoff blocking rate. Anotherimportant parameter is new call blocking probability which is the fraction ofnew calls that get turned down because of channel insufficiency in the cellwhere the new call is generated. A successful handoff provides continuationof the call which is vital for the perceived quality of service (QoS) and asuccessful establishment of a new call helps improve the throughput of thesystem. In general, the blocking of a handoff request (i.e., dropping of anongoing call) is less desirable than the blocking of a new call. Minimizinghandoff blocking has therefore received considerable attention; thechallenging design issue is how to reduce this probability without muchdegradation to new call acceptance rates.

Recently, there has been increasing interest in cellular networks withmobile base stations [GEL99, NES99, CUI00]. These networks have beenreferred to as "fully" or "totally" wireless networks. In these networks, themobile base station (MBS) moves from one place to the other in order tostay close to its group of moving users (called mobile terminals or hosts).Totally wireless networks are advantageous in many applications, e.g.,combat and military operations, emergency evacuation of disaster areas,rapid deployment of dynamic networking capabilities, the temporaryreplacement of destroyed infrastructure, etc. In this paper, the termconfigurable cellular network (CCN) will be used to represent the generalclass of totally mobile wireless networks, i.e., cellular networks in which thebase station can dynamically move in order to stay close to the group ofmobile terminals being serviced by this base station.

2. THE USE OF GEOLOCATION SERVICES INCCN ENVIRONMENTS

The capability to perform real-time location measurement [FCC96,DRA98, TEK98]of mobile terminals and mobile base stations in CCNenvironments is crucially important. This capability is needed in two aspects:

Enhancing Terminal Coverage and Fault Recovery 233

a) Movement co-ordination strategies of the mobile base stations.b) Recovery protocols when a base station becomes faulty or is

immobilized.

Below we discuss these two aspects in the context of hierarchical cellulararchitectures.

2.1 Hierarchical Architectures for CCN

Most previous studies on configurable cellular networks (CCNs) havebeen limited to non-hierarchical architectures. In [NES99], a distributedalgorithm for channel allocation is presented using techniques inspired bysolutions of the well-known mutual exclusion problem. A mobile terminal(MT) cannot directly communicate with another MT. Rather, the connectionfrom/to a MT must go through its mobile base station (MBS). A set ofchannels, called backbone channels, is dedicated for communications amongthe MBSs while another set of channels, called short-hop channels, is used tosupport communications between MTs and MBSs. A different aspect oftotally mobile wireless networks, namely movement strategies for MBSs,has been investigated in [GEL99] and algorithms that can allow MBSs tofollow a swarm of MTs were proposed and evaluated. The movementalgorithms investigated in [GEL99] include center of gravity (COG), SocialPotential Fields (SPF) and movement with power control. The model used todevelop these movement strategies was based on a one-tier architecture,fixed channel allocation for the spectrum available to MTs, and a separatewireless resource (e.g., satellite links) for communications among MBSs.

2.1.1 Proposed CCN Architecture

In [CUI00], we advocate the use of hierarchical cellular schemes forCCN. Our proposed hierarchical design borrows some basic ideas from themacro/micro cellular architecture used in wireless networks with stationarybase stations [BER96, HU95]. For the purpose of illustration, we shall use atwo-tier hierarchy, i.e., the mobile base stations will be divided into twocategories:

– mobile base stations with larger range of coverage; these will be calledlarge MBSs or LMBSs and

– mobile base stations with smaller range of coverage; these will bedenoted SMBSs.


The cellular coverage areas of SMBSs could overlap with or could betotally overlaid inside those of LMBSs. Fig. 1 shows an example of a two-tier configuration with one LMBS and four SMBSs. Communicationsamong the mobile base stations (SMBS and LMBS) is achieved by anexclusive set of radio channels, called backbone channles, while another setof channels, called short-hop channels, is used to support communicationsbetween base stations and their mobile terminals [NES99]. Alternatively,communications among the mobile base stations can be achieved by satellitelinks [GEL99].

In a two-tier CCN, the large cell areas of LMBSs act as an overflowbuffer that can cover the mobile terminals that drift away from the coveragearea of their current SMBSs. Among SMBSs (and similarly among LMBSs),there should be sufficient separation of location to avoid excessive celloverlap. The cell areas of SMBSs, however, can overlap with or can beentirely overlaid inside an LMBS cell. With the availability of the umbrellacoverage from LMBSs, the movement strategy for SMBS should not overlyworry about losing few mobile terminals near the boundary of theirtransmission range; these terminals can be handed over to the overlayingLMBS.


Our initial comparisons between the two-tier and one-tier systems usingsimple Center of Gravity (COG) movement and fixed channel allocationhave given us very good insights into many performance issues and havealso generated a number of interesting problems. An example of theseproblems concerns power consumption in relation with the way we splitchannels among large and small MBSs.

2.1.2 Movement Coordination Policies

The first aspect in the design of fully configurable cellular networksconcerns the strategies for coordinating the movement of the mobile basestations (MBSs). The aim of these strategies is to maximize the percentageof covered mobile terminals and reduce the blocking probability duringhandoffs. Gelenbe et al [GEL99] examined algorithms that can allow MBSsto effectively follow a swarm of mobile terminals. In the center of gravity(COG) algorithm, each MBS periodically calculates the gravimetric centerof its swarm and then moves toward this location. Another algorithm, calledthe Social Potential Fields (SPF), uses a distributed control paradigm inwhich the mobile base stations decide their behavior based on a set of fourforces: a repulsive force that tries to avoid excessive and useless cell overlapamong MBSs and three attraction forces that try to prevent scattering ofresources, allow MBSs to intelligently follow the mobile terminals, and alsore-arrange the positions of the moving cells based on their load in order torelieve hot spots.

The purpose of the COG algorithm is to maintain the location of theSMBS and LMBS at the center of swarm to better serve the mobilesubscribers. Evidently there are different strategies for SMBS and LMBS.

SMBS interacts directly with MTs, two possible approaches toimplement COG are

a) Try to maintain the location at the center of all MTs within the servicerange of the SMBS.

b) Try to maintain the location at the center of the MTs that are beingserved.

Approach a) requires all MTs to register themselves once they enter anew cell and then report their locations periodically to the SMBS. Theselocation updates could be sent through some shared control channels.

To implement the COG algorithm for SMBS using approach a), we startthree concurrent threads to collect location updates from MTs, calculate


COG and update SMBS location. Lock/unlock operations for sharedvariables are omitted for clarity.

Thread MT_dr should be running at each mobile.

thread MT_dr(){

while (true){

cur_loc = get_location(); // Get current location from GPS.t = get_time(); // Get current time from GPS.

if (previous update has not been sent)clear it from sending buffer;

put location update to sending buffer;

set GPS timer and sleep until next update epoch Eu;// Eu occurs after Tu seconds.// The constant Tu is the location update period.

// Since GPS timers are accurate and synchronized,// all MTs are actually waken up at exactly the same// time. Therefore, all the location updates are// synchronized, i.e., they represent a snapshot of// all MTs at an epoch. This is desired for COG calculation.// However, since the location updates must be sent// through some shared uplink control channel, the MBSs// will not get all the location updates immediately, they// must listen for a small period to collect all location// updates. When the MBS finishes the calculation of COG,// the real COG has drifted away. Our simulation shows// that this discrepancy is not significant when Tu is small enough.

}}

The following variables are shared among the three threads running oneach SMBS:

cur_loc is a 2D vector which is initialized to the starting location ofSMBS,


new_loc is a 2D vector which is the calculated COG which SMBS shouldmove to,

V is a normalized 2D vector which specifies the direction of SMBSmovement,

MT_locs is a list of 2D vectors which is initialized to nil,N is an integer to count the length of MT_locs and is initialized to 0.

Thread listen_updates, COG and update_loc are running on each SMBS:

thread listen_updates(){

while (true){

listen to location updates from MTs;

insert the received location update to MT_locs,replace existing entries if necessary;

N++;}

}

thread COG(){

while (true){

// Give enough time to thread listen_updates().

set GPS timer and sleep until next COG calculation epoch EC;

// The relationship between next update epoch and next COG// calculation epoch is:// Ec = Eu-Tp// Tp is the maximum time to do the following calculation.

// Copy MT_locs and N to local variables.myMT_locs = MT_locs;myN = N;

// Reset shared variables MT_locs and N.MT_locs = nil;N = 0;


II sum is a 2D vector.sum.x = 0.0;sum.y = 0.0for (i = 0; i < myN; i++){

sum.x = sum.x + myMT_locs[i].x;sum.y = sum.y + myMT_locs[i].y;

}new_loc.x = sum.x / myN;new_loc.y = sum.y / myN;

V = new_loc - cur_loc;normalize V;

}}

thread update_loc(){

while (true){

if (cur_loc != new_loc){

move toward direction V;update cur_loc;

}else{

keep current location;}

}}

Approach b) has no major difference from a) except that only MTs withongoing calls need to send location udpate. Evidently approach b) savesbandwidth and processing power for both MTs and SMBSs. However, oursimulation shows that approach b) does not maintain the position of SMBSto the COG of the MTs very well when there are not enough calling MTs inthe cell. If the number of calling MTs is increased, the trace of approach b)will converge with that of approach a).


LMBS only need to communicate with the SMBSs within its cell tocalculate the COG. The code is similar to what is given above and isomitted.

To reduce the number of location updates from MTs to the SMBS, deadreckoning model could be introduced. The pseudo code is given below:

We assume the dead reckoning model object has 2 member functions:

update(loc, t)set the starting location of the DR model to loc;set the starting time of the DR model to t;

predict(t)return predicted location at time t;

This thread should be running at each active mobile.

thread MT_dr(){

cur_loc = get_location(); // Get current location from GPS.t = get_time(); // Get current time from GPSdr.update(cur_loc, t); // Initialize DR model.

send dr to SMBS;

// Tdr is a predetermined constant. If the MT has not reported its DR// model to the SMBS for Tdr seconds, then it must report its DR model// even if the DR is still accurate.

dr_timer = Tdr;

while (true){

sleep for T seconds; // T is a predetermined constant// and T < Tdr.

cur_loc = get_location();t = get_time();dr_loc = dr.predict(t); // Update dr model and save

// predicted location.D = dr_loc - cur_loc; // D is a 2D vectordr_time = dr_timer - T;


// Dmax is a predetermined constant. When the difference between// current location and the predicted location of DR is greater than// Dmax, the DR must be updated.

if (|D| > Dmax || dr_timer < 0.0){

dr.update(cur_loc, t);send dr to SMBS;dr_timer = Tdr;

}}

}

For SMBS, in addition to the shared variable in approach a), thefollowing variables are added:

MT_drs is a list to record the DR models of the MTs.MT_timers is a list to record the timers for the DR models in MT_drs.

thread listen_drs(){

while (true){

wait for DR models from MTs;

if (there is a DR for the MT in MT_drs){

replace the entry in MT_drs by the new DR;replace the entry in MT_locs for the new DR;reset the corresponding timer in MT_timers to Tdr;

}else{

insert a new entry in MT_drs for the new DR;insert a new entry in MT_timers for the new DR;insert a new entry in MT_locs for the new DR;set the corresponding timer in MT_timers to Tdr;N++;

}}

}


thread scan_drs(){

while (true){

sleep for T seconds;t = get_time();

for (i = 0; i < N; i++){

MT_timers[i] = MT_timers[i] - T;if(MT_timers[i]<0){

mark entry i in MT_timers for delete;mark entry i in MT_drs for delete;mark entry i in MT_locs for delete;

}else{

MT_locs[i] = MT_drs[i].predict(t);}

}

remove all marked entries from MT_timers, MT_drs and MT_locs;update N correspondingly;

}}

thread COG(){

while (true){

sleep for Tcog seconds; // Tcog is a predetermined constant which// is the period for COG calculation.

// Copy MT_locs and N to local variables for convenience.myMT_locs = MT_locs;myN = N;

// No need to reset shared variables MT_locs and N.

// sum is a 2D vector.sum.x = 0.0;


sum.y = 0.0for (i = 0; i < myN; i++){

sum.x = sum.x + myMT_locs[i].x;sum.y = sum.y + myMT_locs[i].y;

}new_loc.x = sum.x / myN;new_loc.y = sum.y / myN;

calculate a vector V which point from cur_loc to new_loc;normalize V;

}}

thread update_loc(){

// the same as approach a)}

2.2 Simulation Model

Several attempts [BER96, HU95] have been done to evaluate models forstatic multi-tier wireless networks with stationary base stations. So far, therehas been very little work on hierarchical networks with mobile base stations.This section addresses aspects of these latter networks that have not beeninvestigated before. The problem we are tackling, however, is extremelycomplex and involves many factors, e.g., moving base stations, powerconsumption, cells of different sizes, different mobility patterns such asswarm and random motions, overlaid and overlapping cells with time-varying positions, etc. Analytical models for the hierarchicalmacro/microcellular architecture with fixed base stations have manylimitations and cannot be easily adapted to the case of mobile base stations.We have therefore performed our evaluation using extensive simulationbased on a flexible software model; the simulation results gave very goodinsights into many performance issues.

We have developed a detailed and flexible simulation program to test andevaluate the proposed hierarchical scheme for CCN. The simulationprogram has a visualization module that displays a real-time 2-dimensionalanimation of the movement of the base stations and mobile hosts as well asthe wireless connections between mobiles and base stations. The code is


written in C++ and executes under Linux and Solaris. The visualizationmodule is written in C and runs on top of Xlib. The simulation program cansupport one or more hierarchical levels of mobile base stations. All the testsreported in this paper were executed for both single level (one-tier) and two-level (two-tier) systems. Below, we describe the various features andassumptions of our simulation model.

– The range of coverage of each mobile base station is circular with acontrollable radius. Our tests covered cases ranging from relatively smallcoverage areas (radius of 500 meters for SMBS and 1000 meters forLMBS) to larger coverage areas (5 and 10 kilometers for SMBS andLMBS, respectively). In all the tests reported here, we have made theradius of the umbrella coverage (LMBS) double that of the regular cell(SMBS).

– The number of mobile terminals (users) and the number of mobile basestations are controllable parameters. Unless otherwise stated, the testsreported in this paper use a total of 600 mobile users supported by 4 or 5mobile base stations.

– The duration of each call is exponentially distributed with a mean of 180seconds. New calls arrive according to a Poisson process and arehomogeneous among all users. The number of channels allocated to eachmobile base station (i.e., SMBS or LMBS) is a controlled parameter. Inmost of our tests, the number of channels per SMBS was 64 while thenumber of channels per LMBS was varied as will be stated in thedescription of the different experiments.

We have simulated two cases of mobility: i) Swarm motion and ii)Random motion. In the Swarm Mobility model, the mobile terminals areinitially divided into swarms (groups). In each swarm, mobile terminals haverandom movements but the whole group exhibits a general heading. Ideally,each swarm is supported by a mobile base station (SMBS). However, as theMTs move, they cross boundaries of coverage and drift from one swarm tothe other. The program continually updates the position of each MT. TheMT moves with an average speed of 25 meters/sec. Mobile base stationsmove based on the various attraction and repulsion forces of the movementstrategy. The maximum speed for a mobile base station has been restricted to30 meters/sec. In general, each SMBS tries to stay at the center of its ownswarm and each LMBS tries to stay at the center of all neighboring swarms.Whenever the COG of the mobiles hits the boundary of the simulated(rectangular) area, a new general heading vector is generated to restrict themovement of the swarms within the simulated boundaries. Attractive andrepulsive force vectors between swarms (MBSs) are also generated to


maintain appropriate distance and prevent too much overlapping. Thesimulation makes the mobile terminal follow the general heading of itsswarm but allows it random motion within this framework. If the mobiledrifts away from the center of the swarm by certain controllable threshold,an attraction vector towards the center is enabled and is added to the randommovement component. We have adopted the COG movement methodbecause of its elegance and simplicity. In the COG method, every SMBScalculates the gravimetric center of its MTs and then moves towards thiscenter. In real-life situations, a COG-like movement may ensue from thenatural behavior of the SMBS’s driver who usually tries to stay centeredamong the MTs served by his vehicle. Furthermore, the rapidly evolvingpositioning technology (e.g., GPS, GSM) [DRA98, TEK98] and the E-911ruling for locating mobile callers [FCC96] are expected to result in an arrayof low-cost positioning tools that can enable base stations to determine thelocation of active mobile terminals in their cells. With the help of these tools,automated guidance to help implement COG movements will becomefeasible.

2.3 Performance Results

To evaluate the proposed scheme, we compared the new call success rateand handoff blocking rate of a two-tier CCN with those of a one-tiercounterpart. In both schemes, location measurements have been assumed tobe provided in real-time to each mobile base station in order to compute thegravimetric center of the mobile terminals and then subsequently movetoward this center. In the graphs shown in Figures 2 and 3, the one-tiersystem has four SMBSs with 64 channels per SMBS. The total number ofmobile terminals is 600. The movement strategy used in the simulation isCOG (Center of Gravity). The simulation assumed fixed channel allocation(FCA). Each point in these graphs is the average of 10 test runs, and thelength of each run was sufficiently long (4 hours of simulated time) to ensurestability.

One consideration used in our simulation is the signal fading rate. Theaverage received signal strength, P, at a distance r from the transmitter isusually modeled as

where c and the path loss exponent are propagation constants and isthe transmitter power. The value of can range from 2 to 5; the value 4 is


commonly accepted as a typical path loss exponent and will therefore beused in our tests.

Fig. 2 compares the new call success rate for four differentconfigurations. The notation means a two-tier CCN with S smallcells whose base stations have channels each and one large (umbrella)cell whose base station has channels. If the term is omitted, thecorresponding CCN is a one-tier system.

Fig.2 plots the new call success rate (which is the complement of thenew call blocking rate) for four configurations at different load levels. Wehave opted to represent the load as a percentage of the service capacity of theone-tier system. This is explained as follows. Suppose the total averagearrival rate of calls from all MTs is the total number of channels in thesystem is C, and the average call duration is d. Then the total load requestedper second is and the total capacity that the base stations can provide inone second is C. Therefore the load percentage (used in the horizontal axisof Fig. 2) is given by

As shown in Fig. 2, the one-tier system and the two-tier systemwith equivalent power consumption have close performance results in termsof new calls; the two-tier system is actually slightly better but it should benoted that the configuration slightly improved both the acceptancerate of new calls (Fig. 2) and in the same time slightly reduced the handoffblocking probability (Fig. 3). The other two-tier configurations withincreased power consumption gave significant improvement commensuratewith the extra number of channels used by LMBS.


If we assume that the large cell (corresponding to LMBS) has double thediameter of the small cell (SMBS) and that has the typical value of 4, thenassigning one channel to LMBS is equivalent, as far as power consumptionis concerned, to assigning 16 channels to SMBS. Consider now aconfiguration consisting of four SMBSs each with 64 channels. Thisconfiguration is a one-tier system and is denoted by <4x64>. If we convertone SMBS to LMBS by doubling its cell diameter, then we must reduce thenumber of its channels from 64 to 4 so that the total power consumption byall four base stations remains the same. The new configuration is ahierarchical two-tier system and is denoted <3x64+4>, meaning three small


base stations with 64 channels each and one large base station with 4channels. We could also construct other two-tier configurations having thesame total power consumption, e.g., <3x48+7>, <3x32+10> or <3xl6+13>.Figure 4 compares the scaled throughput of these different two-tierconfigurations. The Y-axis in this figure is the scaled throughput obtained bydividing the throughput by the average call arrival rate and the x-axis isproportional to the load level. Notice that the throughput is negativelyimpacted by blocked new calls and broken connections (due to failedhandoffs or non-coverage). The configuration <3x32+10> performsreasonably well for the entire load range, is much better than thecorresponding one-tier system and seems our best choice. These resultsimply that some power-preserving adjustments in channel splitting can leadto significant performance improvement. They also imply that the optimalstrategy would be to use the splitting method that optimizes coverage andminimizes broken connections, combined with a rotating assignment policythat allows the base stations to switch the role of LMBS and SMBS in orderto balance power consumption among all base stations.

3. FAULT-TOLERANT DESIGN ISSUES

In this section, we identify the major types of faults that can seriouslydegrade or totally disable the communication services of a mobile basestation and discuss in high level details the possible approaches to handle


these faults. We shall divide the faults that affect the operation of a MBSinto two categories.

Immobilization of MBS:In this case, the mobility of the base station is eliminated or is severely

limited but its transceiver continues to function properly. Scenarios for thistype of fault include mechanical failures, road obstruction, flat tires, andmalicious attacks. The immobilized unit becomes a stationary base stationand cannot therefore follow the group of moving terminals.

Destruction of MBS:The transceiver of the mobile breaks down and the fault entirely halts the

communication services of the base station.

To deal with the above faults, the protocols for CCN need to be enhancedby adding “fault-tolerance” design features. Additional hardware andsoftware resources are needed to achieve this task. Below we outlinepossible approaches for handling the immobilization and destruction faultsof mobile base stations in CCNs.

Our solution uses the basic principles of fault-tolerant computing designs.In order to be able to handle these faults when they occur, the normaloperations of the various protocols need to be changed by adding redundanthardware and incurring some extra overhead (e.g., additional messageexchanges). Additional vehicles in each moving swarm can be equipped withspare transceiver hardware and can become an operational MBS byactivating and properly configuring this hardware. Understandably, there is atradeoff between the cost of the extra hardware and the level of effectivenessand gracefulness by which the faults can be handled.

The immobilization fault is easier to handle. When this fault occurs, theimmobilized base station is still functioning and can participate in therecovery protocol, i.e., the protocol that aims at finding a new base stationfor each affected mobile terminal and possibly adjusting the movement andtransmission power of adjacent operational base stations in order toguarantee continuous coverage of the newly transferred terminals. Theimpact of the immobilization fault on a mobile terminal is similar to anormal handoff.

Unlike immobilization faults, destruction faults affect the QoS of theongoing connection for some period of time, i.e., until that connection ispicked up by an alternate base station. With proper recovery protocol and


redundant hardware, the down period after the occurrence of a destructionfault should be quite short. The impact of this short disruption on real-timeaudio connections should not be too bothersome and there is no additionalretransmission delay after re-establishing the audio connection. Forconnections intolerant to data loss (e.g., transmission of textual messages orfax), additional overhead is incurred to retransmit the packets that may havebeen already sent to the destroyed base station. This is basically done by aprotocol that reconfigures the sender’s side of the TCP connection.

Below, we discuss two possible approaches for the fault-tolerantprotocols.

3.1 Backup Base Stations

This approach is similar to the resilient execution of database transactionupdates (e.g., resilient two-phase commit protocol). For each operationalbase station, a backup base station (e.g., a spare transceiver mounted on anearby vehicle in the same cell) is identified. The primary base stationcommunicates with its backup to update its information about establishedconnections as well as to exchange periodic “I am still alive” signals. Thiscommunication represents a low-level demand on power consumption.Destruction of the primary base station can be easily detected by its backupand the recovery protocol can then be invoked. The backup base stationperiodically collects the location information about the different mobiles andapplies the COG strategy to direct its future movement. It should be notedthat switching the affected terminals to the backup base station does notrequire additional channel resources and does not therefore generate theconnection-admission problems that occur when an operational base stationis faced with multiple new calls or handoff requests (i.e., bulk arrivals). Inthe latter case, some or all of the multiple requests may be denied because ofthe lack of channel resources (whereas a backup base station will have noproblem reusing the channels that were used by the destroyed base station).Increased resiliency to handle multiple concurrent destruction faults can beobtained by using multiple backups and extending the protocolappropriately.

3.2 Decentralized Recovery

Rather than relying on a backup for each operational base station, adecentralized recovery approach can be used. At the time of establishing aconnection during normal operation, each mobile terminal stores appropriateinformation that helps in the re-establishment of the connection if it gets


broken later by a destruction fault. When the mobile detects prolonged lossof communication with the current base station, it submits a handoff requestto the base station having the strongest detected beacon signal (as in anormal handoff procedure). The connection re-establishment information isalso submitted to the new base station to offset the fact that the destroyedbase station will not participate in the handoff process. The key conditionfor making this approach successful is that base stations should operate atlow or medium loads so that they can absorb multiple handoff requests fromterminals affected by a destroyed neighboring base station. In other words,the redundant hardware for fault-tolerance is used to increase the number ofoperational base stations relative to the number of supported terminals. Theincreased number of operational base stations requires adjusting the celldiameters and the channel allocation procedures, similar to the process ofmigrating from a macro- to a micro-cellular architecture.

To implement decentralized recovery, mobiles must be able to detect theloss of the serving base stations. When the loss of the communication withthe current base station is detected, there are 3 possible reasons:

a) The base station is still running but the mobile has moved out of theservice area. Usually the received signal strength (RSS) decreases graduallyas the mobile moves out of the cell.

b) The RSS could decrease sharply to cause the loss of communication ifthe mobile is roaming in special terrain, say, maintained area.

c) The base station is not functioning.

Therefore, a mobile cannot blindly assume the loss of a base station whencommunication is lost. We need to consider RSS as well as the distancebetween the mobile and base station. Below is the pseudo code:

Procedure handoff is called by a MT when its RSS is below operationthreshold.


procedure handoff(){

// This procedure is triggered by the loss of communication.

struct handoff_request r;

locate the base station B which has strongest beacon;

if (no strong enough beacon detected){

drop call;return;

}

r.MT_id = my_id; // my_id is the permanent// id of the mobile.

r.MT_loc = get_location(); // Current location of the mobile.

r.prev_base_id = base_id; // base_id is available from the// beginning of the session

r.prev_base_loc = base_loc; // We assumes each base station// will periodically notify the mobile// of its location, and the mobile will// store it in local variable base_loc.

r.prev_base_TxPwr = base_TxPwr;// Base station should notify mobile// the actual transmitting power.

// Alternatively, prev_base_loc and prev_base_TxPwr could be// omitted from handoff request if each base station knows the// location and transmitting power of its neighbor bases.

r.rss = RSS; // RSS of last detected signal from// old base.

send r to B;wait for reply;

if (reply is positive)switch to new channel;


elsedrop call;

return;}

Procedure process_handoff_request is called by a SMBS when it receivesa handoff request.

procedure process_handoff_request(struct handoff_request r){

// This procedure is running on base stations// and is triggered by incoming handoff requests.

if (r.prev_base_id is NOT registered as dead)

{V = r.MT_loc - r.prev_base_loc;

// |V| is the distance between r.MT_loc and r.prev_base_loc,// f is constant and 0 < f < 1,// R(r.prev_base_TxPwr, |V|) is a function call to calculate the// ideal RSS at distance |V|, given transmitting power of// r.prev_base_TxPwr.if (r.rss < f * R(r.prev_base_TxPwr, |V|)){

// r.prev_base_id is possibly dead,// try to verify r.prev_base_id.start thread verify_base(r.prev_base_id);

}}

try to find a channel for r.MT_id;

if (channel is available){

assign the channel to r.MT_id;send positive reply to r.MT_id;

}else{

send negative reply to r.MT_id;}


}

Thread verify_base will start when a SMBS feels necessary to check thevitality of another base station.

thread verify_base(base_id){

contact base_id;wait for reply;if (time out){

// r.prev_base_id is considered lost,// try to reclaim its channel resources.

negotiate with neighboring base stations to re-allocate thechannel resources previously owned by r.prev_base_id;

negotiate with neighboring base stations to increasetransmitting power;

register r.prev_base_id as dead;}

}

It is not necessary for the surviving base stations to change movementstrategies. When the power is properly adjusted, either COG or SPFalgorithm will automatically adjust the position of the surviving base stationsto serve the mobiles previously owned by the dead base station.

4. CONCLUSIONS

In this paper, we discussed the use of geolocation services in improvingmobile connectivity and enhancing the effectiveness of fault recovery inconfigurable cellular networks. Example performance results werepresented showing the tradeoffs and performance implications of using ahierarchical CCN architecture augmented with a capability for real-timelocation measurement.


Acknowledgment

This work has been supported by ARO Grant No. DAAH04-95-1-0250 and by an I-4 Corridor grant from Harris Corporation and theState of Florida.

References

[BER96] R. Beraldi, S. Marano and C. Mastroianni, "A Reversible Hierarchical Scheme forMicrocellular Systems with Overlaying Macrocells", Proc. IEEE INFOCOMM '96, pp. 51-58, 1996.

[CHI00] M. Chiu and M. Bassiouni “Predictive Schemes for Handoff Prioritization inCellular Networks Based on Mobile Positioning” to appear in IEEE Journal on SelectedAreas in Communications (JSAC), Vol. 18, No. 3, March 2000.

[CHO98] Sunghyun Choi and Kang G. Shin, "Predictive and Adaptive BandwidthReservation for Hand-offs in QoS-sensitive Cellular Networks", Proc. ACM SIGCOMM'98., pp. 155-166, 1995.

[CUI00] W. Cui and M. Bassiouni “Two-Tier Channel Assignment for Wireless Networkswith Dynamic Cellular Architecture” to appear in Proceedings of IEEE InternationalConference on Third Generation Wireless Communications, Silicon Valley, June 2000.

[DRA98] C. R. Drane, M. Macnaughtan and C. Scott, “Positioning GSM Telephones,” IEEEComm. Mag. Vol. 36, No. 4, pp. 46-59, Apr. 1998.

[FCC96] “FCC Adopts Rules to Implement Enhanced 911 for Wireless Services, ” FCCNews, CC docket no. 94-102, 1996.

[GEL99] E. Gelenbe, P. Kammerman, T. Lam "Performance considerations in totally mobilewireless", Performance Evaluation, Vol. 28 & 29, pp. 387-399, 1999.

[HU95] Lon-Rong Hu and Stephen S. Rappaport, "Personal Communication System UsingMultiple Hierarchical Cellular Overlays", IEEE JSAC, Vol. 13, No. 2, pp. 406-415, Feb.1995.

[TEK98] S. Tekinay, E. Chao and R. Richton,“Performance Benchmarking for WirelessLocation Systems, ” IEEE Comm. Mag. Vo. 36, No. 10, Apr. 1998, pp. 72-76.

[NES99] S. Nesargi and R. Prakash “Distributed wireless channel allocation in networks withmobile base stations” Proc. INFOCOM, 1999, pp. 592-6000.

Chapter 8

UMTS APPLICATIONS DEVELOPMENT –DESIGNING A “KILLER APPLICATION”

Günther Pospischil and Ernst BonekForschungszentrum Telekommunikation Wien (FTW),

1040 Vienna/Austriaand

Institut für Nachrichtentechnik und Hochfrequenztechnik der Technischen Universität Wien,

1040 Vienna/[email protected]@nt.tuwien.ac.at

Alexander SchneiderMobilkom Austriaand previouslyInstitut für Nachrichtentechnik und Hochfrequenztechnik der Technischen Universität Wien,

1040 Vienna/Austria.

[email protected]

Abstract The upcoming generation mobile communications system UMTS/IMT-2000will provide high data rates and advanced means for service generation. In con-junction with so-called smart phones a cass of entirely new services, exploitingmobility and multimedia will become possible.

We present a framework for providing such services, as well as their keyrequirements and concepts to meet these criteria.

Keywords: UMTS, IMT-2000, Mobile Multimedia, Service Module, Navigation, LocationBased Service


1. INTRODUCTIONCurrent mobile communications systems, like GSM or IS-136, offer circuit

switched data at 9.6 kbps. The next steps of evolution are packet switched trans-mission and higher data rates of up to 40 kbps (theoretically up to 150kbps) [1].

On the terminal side, smart phones, using WAP (Wireless Application Pro-tocol) to display simplified web-pages [2], are becoming available. In the verynear future devices with bigger (touch screen) displays, an open operating sys-tem (EPOC) and advanced programming capabilities, for example utilizingJAVA [3] [4] [5], are expected.

On the internet side, web-portals for information retrieval, e-commerce, com-munity actions (chat, mail, private sales) have become popular recently. Theupcoming 3rd generation mobile communications system UMTS/IMT-2000 isexpected to integrate mobile communications and internet in a totally new way,enhanced by localized information and data rates up to several hundred kbps.

Therefore we propose a web portal architecture, specially designed for loca-tion based mobile applications. It is called “City Life” [Fig. 8.1] and providestourist information (sight seeing, restaurants, hotels), shopping facilities, med-ical care, and a private area with address book, appointments, notes, etc. Thus,unified messaging [6] is also integrated. Another important part is the personalprofile, which can be seen as extension to the popular personal cards includedin electronic mail or short messages.

UMTS Applications Development 257

2. ARCHITECTUREOn one hand, the number of different devices, offering very different capabil-

ities in terms of display, memory, and processing power, increases steadily. Onthe other side, ubiquitous access of data can evolve into the "killer application".Users want to access their data anytime and anywhere on any device. Fromthe service providers point of view it is necessary to keep data management assimple as possible, for example without the necessity of manual content adap-tation. Taken all this together, a location based mobile multimedia service hasto meet the following requirements:

Any Device:

– Notebook, Palmtop,Smart Phone

Any Network:

– GSM/GPRS, UMTS/IMT-2000, generic TCP/IP, IS-136, etc.

– Circuit switched or packet switched

– Different QoS, even no QoS guarantees

Any User:

– Service subscription (on demand)

– Service personalization

– Public data and private data


To achieve an efficient service implementation, we suggest to use several“service modules”, for example a location module, a customization module, aQoS module, etc. [Fig. 8.2]. Modules can be realized either in software, byexternal hardware, or be integrated into the transport network. This makes itpossible to provide the service in an efficient manner to simple PDAs as wellas to high performance PCs. The User Interface module, for example, could bean extended Web-browser with JAVA and some plug-ins. On the other hand, itcould make use of a clipping technology [7], where the client only displays animage which “looks like the actual web page”, the processing is done at a proxy.

The modular concept means that upgrades of specific modules do not effectthe functionality of the others. Also it is possible to use “emulation modules” ifsome functions are not (fully) supported by the infrastructure. If, for example,a PC connected to a TCP/IP network is used, position information might not beavailable. Thus, there could be a “Position emulation module”, simply askingthe user to provide his or her position, e.g. street name, building and roomnumber.

3. CONTENTIn principle, all information that is available on the internet, should be usable.

But, due to the limited resources for mobile devices (screen size, memory, pro-cessing power) and transmission bandwidth, specially designed Web-Portalsare advantageous.

A portal provides a homogeneous environment for all services a specific usercommunity needs. In our case, the users are visitors in Vienna (or any othercity). Therefore the service “City Life” consists of route finding, tourist infor-mation, and a private area (which can be synchronized with the user’s homePC) [Fig. 8.3, 8.4].

The concept of Web-Portals also offers a Win-Win situation for network op-erators and content providers. Operators can differentiate against each other byspecialized access mechanisms and system services. Also they can add valueto their transport infrastructure, for example by delivering location informationto a navigation service. Content providers can use the operator with the bestinfra structure for a specific service, still allowing limited access to the servicefrom other operator’s networks. Furthermore, they can directly benefit fromthe operator’s customer data base [8].

Access everywhere is ensured by using only basic transport mechanisms,e.g. a TCP/IP functionality. Additionally needed features, for example QoS or


location information, can be provided by the transport network or be emulatedwithin terminal, proxy and/or server (using appropriate emulation modules).To ensure a specific QoS, for example, the service may use buffering. If thenetwork does not deliver location information, the user might supply his/herposition manually.


4. SERVICE “CITY LIFE”

City Life uses a horizontal structure (Client/Network/Server) [Fig. 8.2] anda vertical decomposition [Fig. 8.5]. The vertical components are distinguishedaccording to their range (Network/Cell/Building/ Room/Object), building a hi-erarchical service. The upper 2 layers use a Network-to-User data transmission(broadcast), while the lower 3 layers employ a Object-to-User data transfer(network assisted specific transmission).

On Network Level, the “service directory” is provided (a service home page)which includes the information overview and user profile management (servicesubscriptions, interesting information categories, device capabilities).

On Cell Level, the key feature is location and direction specific information(local broadcast with push or pull characteristics) and navigation to a select-ed destination (needing real time position calculation and information transfer).

On Building Level, location is again the key issue. In this case other al-gorithms are used, for example manual selection of current position (using asimple menu, voice input/output, or a 3D-map). Also a detailed user profile ismaintained to supply the user with interesting information (ranging from build-ing history to people working in the building).

On Room level and Object Level, very local broadcast mechanisms (e.g.Bluetooth) are used to transfer user specific information.


It is obvious that different user interfaces (voice, menu, graphics), position-ing methods (Cell ID, signal run time, manual positioning), and multimediafeatures are necessary. Methods for location/direction determination and userinteraction are presented in the remaining sections of this paper. The methodsfor automatic content adaptation are beyond the scope of this work.

5. LOCATION AND DIRECTIONWe require three different functions: location determination, direction find-

ing, and navigation. Therefore an iterative algorithm is used, successivelyrefining location.

First, an initial location estimate is achieved by the Cell ID. Then an iterativeupdate process is performed to cope with user movements and to achieve ahigher position estimation accuracy. The refinement algorithms are based onsignal run time (OTD, TOA) [9], user tracking (using a history of Cell- IDsand/or smart antennas with spatial reference algorithms) and direction findingwith a compass and a map (for plausibility checks and navigation display).To achieve sufficient reliability and accuracy, some user interaction may berequired. This is done via a combination of guiding and asking for information.Guiding is done by presenting additional information for navigation (e.g. “passstraight ahead until the next street corner”). If the system needs additionalinformation, it presents a graphical menu with icons of (possibly) surroundingbuildings. The user selects the nearest one, thus specifying his/her position andimplicitly also the speed of movements (calculated from the time differencebetween two location estimates).

6. PRESENTATIONSince system acceptance is determined by the usability, special considera-

tions for the user interface have to be undertaken. Especially for the interactiveposition refinement, the information transfer must be quick and easy. Sincemaps are difficult to understand, especially in the case of a very limited displaysize and resolution, other concepts, like images or direction arrows, are mainlyused.

For information input, the user is required to select information items insteadof giving full textual input as far as possible. An example is the location esti-mate. The current position is determined by selecting a picture of a buildingfrom a menu instead of typing a buildings name. Also street names are nottyped, but rather selected from a list, which is generated according to the CellID, the last position estimate (using an internal map to find surrounding streets)and the first character of the street. Also voice recognition (and voice synthesis)


is heavily used. This has the advantage that the user is not required to look atthe screen and that small screens are sufficient. To achieve high recognitionrates, the location server considers the user profile to find the user's languageand to activate the appropriate grammar.

The future work is mainly dedicated to location/ direction estimation and userinteraction (speech recognition), therefore no quality measures can be given yet.

7. CONCLUSIONWe feel that information access independent from specific networks and ter-

minals is the key issue: users need their information anytime and anywhere,therefore the integration of new services with existing networks is essential.This means that missing system features have to be emulated.

Another important result is that value added services are possible with rel-atively modest system requirements; the critical factors are content and userinterface.

It cannot be expected to have a universal device or a universal service. In-stead, a universal portal can be created allowing for access to different cus-tomizable services. To cope with different devices, automatic adaptation, de-vice profiles, and low device requirements (e.g. voice output instead of largegraphics) are used.

“City Life” implementation will be completed within the near future, thenfield trials and development of advanced algorithms for navigation and userinteraction will continue the project.

References

[1] 3GPP (http://www.3gpp.org and ftp://ftp.3gpp.org). GPRS Ser-vice description Stage 2. 3G TS 23.060.

[2] WAP-Forum (http://www.wapforum.org). WAP Architecture.SPEC-WAPArch-19980430.

[3] Symbian/EPOC (http://www.symbian.com). EPOC’s JAVA imple-mentation. http://www.symbian.com/epoc/papers/java/java.html.

[4] Symbian/EPOC (http://www.symbian.com). EPOC hardware.http://www.symbian.com/epoc/hardware/hardware.html.

[5] 3GPP (http://www.3gpp.org and ftp://ftp.3gpp.org). MExE Func-tional Description Stage 2. 3G TS 23.057.

[6] WebForUs (http://www.webforus.com).


[7] Eric A. Brewer, et al. A Network Architecture for HeterogeneousMobile Computing, pp. 8-24. New York, USA: IEEE PersonalCommunications, October 1998.

[8] Takeshi Natsuno (NTT DoCoMo). NTT DoCoMo’s I-mode ser-vice. London, UK: Proc. Mobile Software Forum, September1999.

[9] ETSI (http://www.etsi.org). Location Services Stage 2. ETSI TS101 724, GSM 03.71.


Index

3rd generation mobile, 255Accuracy, 187, 188Acquisition assistance, 146Advanced Forward LinkTriangulation, 144Aires(ECCO),74Almanac, 172Angle of Arrival, 148, 180AOA, 148Assisted GPS, 129, 133, 135,178ASTROLINK, 84ATM, 109, 114

networks, 78Authentication, 8Billing, 8Capacity, 187, 190CCN, 232Cell selection, 222Center of gravity, 235Cluster controller, 13Configurable cellular network,232Copy network, 13CYBERSTAR, 84E-911,158Ellipse, 74Ephemeris, 172ETSI, 80EUROSKYWAY, 84Fault-tolerant design, 247FCC ruling, 157, 160Forward Link Trilateration, 147Frame Relay, 109, 113Frequency, 175

Uncertainty, 142Geolocation, 129, 148, 149, 151Geometric Dilusion ofPrecision, 176Geostationary earth orbit, 66

Globalstar, 74GPS, 130, 132

assistance, 146Gravimetric center, 235GSM, 67, 148, 196, 256GSMPS, 197Handover management, 83Highway Scenario, 18Highways, 185Hybrid methods, 182ICMP, 119ICO-Global, 74IETF, 80IGMP, 119Infocity, 109Infostations, 3Integration Time, 142Interference, 195IP, 65

networks, 88over ATM, 115

Iridium, 74IS-801, 129, 139, 144Latency, 187, 188LDTs, 159Link budget, 140Location, 146

assistance, 146determination

technologies, 162lingerprinting, 182management, 83

Low earth orbit, 66Medium earth orbit, 66Mobile Assisted E-OTD, 148Mobile Assisted GPS, 148Mobile Based E-OTD, 148Mobile Based GPS, 148Mobile IP, 109, 121

Networks, 92


Mobility, 8management, 83

Multimedia, 112Multipath, 176, 195, 220Network access identificationnumber, 66Next Geberation Internet, 117Next Geberation Internet2, 117NLOS, 195, 221N-STAR, 84Number of Channels, 142OTDs, 205Personal mobility, 66Prefetching, 10QoS, 109Quality of service, 95Radio Link Protocol, 8Random walk, 22Range, 38Real Time Difference, 197Registration, 8Reliability, 187, 190Remote Mapping Terminal, 197RMT, 197Roam, 9RSVP, 109, 119RTD, 197, 205Rural environments, 184Satellite, 65Self-positioning, 196Sensitivity assistance, 146SIM, 67SKYBRIDGE, 84Social Potential Fields, 235SPACEWAY, 84SPF, 235Suburban environments, 185Synchronization, 205TDOA, 167, 205TELEDESIC, 84Telemetry, 172Terminal mobility, 66Time, 175

Difference of Arrival, 180Of Arrival, 148

Timing, 136Uncertainty, 142

TLM, 172TOA, 148

Two-dimensions, 29UMTS/IMT-2000, 255Urban environments, 185Water and waterfrontenvironments, 186WATM, 80WEST, 84WINMAC, 8Wireless, 65

ATM, 65location, 129

Next.generation.wireless.networks

Documents

upper saddle

intelligent

generation

federal communications

configurable

wireless broadband

generation

test vehicle