Intelligent location tracking strategy in PCS

(1)

Intelligent

location tracking strategy in

PCS

K.Wang, J.-M.Liao and J.-M.Chen

Abstract: Ai1 alternative stntegy (AS) for reducing location update signalling load has hc11 yroposcd 10 improve the classical strategy (CS) used in the Global System for Mobile Communications (GSM). Bowevet, the AS is suitablc only Ibr long-term events and fixed movement tracks. Location tracking (paging) costs more when tlic rnobilc host (MH) changes iiiovcment habits or cncoiinlcrs short-term events. A novel method is proposcd that aims to reduce signalling load resulting from location tracking in ihc above situations. The key idea is to takc iiscr rccciit inoveinent information (called the paging i d o r m i o I 1 rccordj into account to detenninc which location area to page first. Performance evaluation has been conducted using ii city area zone modcl that can model H

redistic city area environment. Expcrimcnlal rcsults show that the method can rcducc location munagement signillling costs by 1 6 5 5 % and 6 3 9 % conipared to CS and AS, rcslxctivcly. The overhead of the :ipproach is the additional storage spacc rcquired (for storing MH procilcs and pdgitig infonnalion rccords) and the additionel processing time. Howcvcr, Lhc overhead is insignificant in

terms of loday's memoly ctipacity and processor speed, and Ihc r'cduction in signalIing load makcs tnorc bandwidth available.

1 Introduction

Wircless commuiiication is becoming cost-effective and is convenient for modern lifcslylcs. As il rcsiilt, the population oT mobile hosts (MH) has grown vciy rapidly in rcccnl ycars. The location of an MB must be updated (registered) when the MH moves into a new coverage area so that the location management system can roittc a call lo an M H at a n y h e . Third-generation mobile telecornniunication s p - kms [I], such 11s Universal Personal Tclcconiinunicalioi~s

(UI'Tj and Universal Mobjle Telecommunication Systems

(UMTS), will come into servicx in the next centuiy. How- ever, this will bring very large MH popul:itions and cmse

spectral congestion. To casc these problems, telecornmutii-

cation systems need to employ small radio cells lo reuse channels. Howcvcr, this will cause ti high rate of MH regis-

tration and is a heavy burden on thc nclwork.

In sccond genertition mobile cornmunicatiotl systems, such as thc Global System for Mobile Communications

(GSM)

[Z],

location trarcking (paging) [3] is achieved as Tal- lows. A network is divided into geographical arcas, callcd 'location ~ m s ' (LA) and the location managenmi system kceps track of the current LA of an MH. An LA may con- tain one or 1 1 1 0 ~ cclls. TIE locijtion information is stored in the netwoi-k dalahasc Tor localion managcincnl. To main- rain thc coiisistency of lociition information, an update process is triggered whenever an MH crosscs LA bounda- ries. When a call anivcs, a query to the lociition database is conducted 10 obtain thc localion informiition of the called MH. Then the location inanagcincnt systcin pages all cclls

in the corresponding LA siinultancoasly.

In third generation mobilc coinmiiiiicalion systciiis the same location manngemcnl ttpproilchfi inay not be appli- B IEE,

m

LEE Ancec~l~ttgs onlinc 110. 3HHS!27 UOt IO. l 0 4 ' l / i p a " 7

Papcl fila imcivcd 1st Octakx 1998 and in rcvised fatm 23td July 1999

'I'hc authors an: with the Dep;irlment or Compiiki md Infoin~alion Sciena,

National Chiao Tung Uniwrsity, Hsinchu, Taiwan 3IX150, I~cpuhlic of Chiria

cable. Owing to lhe hst growing population of MHs, lhc signalling load will become too heavy to Iiandlc location management, especially in dcnscly populated areas. Several maliods have k e n proposd Tor location management [ M I . They are classified into two lypcs. One type is based fin dgorithms and nctwork architectures (memorylcss), the cther type is bascd 011 slatistics to reflect the mobility

behaviour of an MH (memory-based) [C]. Thc mcinory- hised type relies on the information collection capsbility of

the network. The proposed mcthod klongs to the ineni- ury-based type.

Based

on thc inovcinciit records of a n MH during the last obserwtion period, thc localion managcment systeiii gcncralcs s n LA list. When there is R call for thc M 1-1, the locrttion tnan>igement system lakes the roccnl short-term events and thc LA l i s l into awount to generate a iicw list d e d ti 'paging list'. TIE LA lisl and the paging

list are the two priinary clcmcnls or the proposed location managcment method,

2 Existing methods

Existing location management srratcgics [9, 101 can k clas- sificd into two types, the memorylcss and the memoly- lxmxj. The m d n difference bctwccn thcsc two types is that the memory-biised type fakes tlic MI-I movement history inlo account. Hislory data of the MH is collected aver sev- cral wccks [7l. The decisions to perform l o d o n uflatc

and lwttion tracking arc closcly related to the mobility khaviour of a n MH. The mcmorylcss type is not con- ccmcd wilh MH movement history. In Ihc rollowing, we

dcscribc two representative location rnanagcment methods, tlic classical slratcgy (CS) [7] and the altcrnalivc stmlcgy (AS) [7, 81, 0111: of cach type.

2. I

Classical

strategy

(CS)

In CS

[I

(used in such systems tis GSM), thc radio covcr- age area is partitioned inlo LAs, as shown in Fig. 1. Each LA coiisists of a group of cells (I, 2 or more cells). MH ti-acking i s rcaliscd by updating the location whcncvcr an MH leaves iL3 current LA. Tlic location management

(2)

systcm always knows which LA an MH is in, and therefore when R ail1 w u r s Tor it, the systcm pages it over its currcnt LA. The loution arca identifier (LAI) is rccorded in a loca-

tion pointer stored in a visitor louition rcgistcr (VLR). The address of the currcnt VLR of an MI-I is kept in tlie hoiiic location register (HLR) of the MH. Thc major drawback or lhis method is the vcry high traffic gcncrated by location updtttes for an M H that frcqucntly changes its LA. If >in MH imivcs vcry few culls~ lhcii tlie sigiialling trariic gener- ated to track it is wasteful, and the ratio uf thc sig~ilalling

cost to the number of proccsscd calls is very high. This situation bccoines severe with thc growth of customer population and the very limikd bandwidth capacily.

-

- - -

+ user movement path

Fig. 1 Chwird .s/wtqy wrhitcctrrrt,

2.2 Alternative

strategy

(AS)

Tlie main goal of the AS [7, 81 is to reducc lhc location

update (rcgisti-atioii) cost by prcdicting users' inovcment pattei-ns. 11 is based on user iiiovcment habits. Most pcnple movc dong R 'usual' palh. I n this situation, an MH's inovenicnt path can be predictd l o avoid most location updatcs. I n this stmtcgy, tlic system handlcs a profile for cach uscr, where e x h u s d s mobility pattcrn is recorded. The structure nnd contents of this prolilc [7, X] for iin MH are as follows:

1, Over B period or time [ti,

$1,

it crosscs k Lhs.

2. The profile consists of two elemenls: (cy, p,] with 1 s f s k , where k is the number of LAs. q i s thc identifier of thc ,fill LA that nn MI-I can be 1oc;ited in and pJ is the rcsi-

d e n e probability that the MI-l is located in U+ wit11 pl 2

p? 2 ... k p k .

*

user movement nalh

_ _ _ -

Fig. 2 /t/mwtivc. S W I ~ P ~

PmiiIc coiiuliiis: ((1, 121180), (9, 14/l80), (IO, IWl80), (19, l h i l X l ) ) , (20, 125/180),

(21. 18/180), (14, 103/180))

1.A list: (20. 14, 21, 10, 19, 9, I }

An MH needs to download its profilc (and its LA list) at initial regisIration. When an MH updates its prolilc, the

(14

location in~tnege~ncnt system generates ii iicw LA list. The

LA list is a sct of LAs that appcar in tlie profile and the LA sequencc of the list is in descending order of the probability that an MII inay reside in an LA. Thc profile ofmcli user is slorod in tlie HER. If an MI-I is called, tlzc systcin pages him/hcr ovcr LA n, , If he/slic is not there, the systcni

will page tlic MI1 over LA (12 up to LA ah. (the last LA in

thc LA list). If an MH iiiovcx to an LA that is not in the LA list, the MH ninkes a registration to add a iicw LA to tlic profile anti the L A lis(. As sin example, in Fig. 2, an M H moves from LA 1 to LA 20, and then to LA 14. After ~ l i c MH updates its profile, the system generales m i LA list in the order of the MI-I residence probiibilily. If there is a call Ibr tlic MH, the system lirst pages LA 20.

Thc AS reduces the sigmtlling load by decreasing the times or rcgistiation. An MI-l usually moves along the same track because the destination i s the s l i m (i.e. office, school, tiomc ctc.) and the routc is usually the same. Thc cost of IocRtioii management C R ~ be rcduced by ensuring

highly preedict:ibli: mobility patterns. Howcvcr, sometimes an M H may ctimgc its dcstiiiation due to soiiic short-term cvcnt. In this case we may have a different movcinent path and some new LAs which the MH has not visited in reecent h c s . Under these conditions, the AS will coiisuine mole

paging costs to lowtc llic M H because it will pagc all the

old LAs in tlie LA list beforc il pagcs the new LAs which are ;it the cnd or the LA list. To avoid coiisuining excessivc network imoums, ii new nicthod is proposed lhal lakes shori-tcrm cveizts into a ~ o ~ u n t .

3 Design approach

3.1

Shopt-term

event

scenarios

In the AS, when an MH is in an LA of the LA list, the

location managcmciit system receivcs ncw location infur- mation only when the MH gciicratcs a call or afier iL is paged. A n MI1 inny not interact with thc system very oficn

Ixcause it does not makc R call or because it rcceives vcry

kw

rills. The systcin docs lint have enough information about the MH's short-term events. Wc dcscribe two short- t m i went sceniirios to illustrate the shortcomings nf the AS as follows. First, considcr the situation (sccnario I) in which M H is called and the timc since the last coniiection is short. With this int'onna~ion, the probability [hat the MI-I stays in the s m w LA is high with the AS, Lhc M H is paged from nl up to in thc LA list, wasting loo much paging cost. Tn our a p p r o d i , the MH is paged over thc L A whcrc the MH last appcarcd. Secondly, iin MI-I inay

chingc its movement habit or mow around soinc LAs (sccnario 2). The LAs wliich the MH visits may not be fre- quently visited LAs (which are not in the first scvei-al LAs o f llic LA list). So the AS may need to page all tlic LAs in the LA list, again resulting in high paging costs.

For scenario I , as shown in Fig 3, wliccn an MH in LA 14 has a call, thc AS will p g e it according to the LA order

in the LA 1 L . That is, the locrd mamgcmcnt systern will page LA 20 lirst, then it pages thc ciarciit LA {LA 14) where the MH resides. But if wc lakc short-term events into a w u n t , the location inanageinelit systeni may only need to page thc last visited LA (LA 14) instcad of LAs 20 and 14.

NOW, wc dcscribc scenario 2 where an MI-I changes its usual inovcmcnt habit, as shown in Fig. 4. After the MH moves to LA 20 it then nioves to its destination (LA 13) which deviates froin its usual moveiiient path. In this silua- tion, the MH will pcrlbnn a locatioii update whcn cntering LAs 12 and 13 sincc thcsc two new LAs are not. in Ihc original LA list. Tlie system will add tlicsc two LAs to the end I I X ~ r ~ ~ I ' . - ~ , ' ~ i i i ~ } i : ~ i i , , Vol. 167. Nu. 1. F d w i w } J Zilffil

(3)

of the LA list. If there is a call for the MH in this silualion, thc AS will necd to spend signtilling cost and time paging all the LAs in the LA list to find ihc MI-I. But if we take short-term events into account, lhc system may only need

to page EAs 12 a n d 13. Scclion 3.3 describes how shorl- term events are taken into account.

@m

inmmlng call

_ - -

-b user mnvement path

incoming call

_I- -+ user mavement path

Fig. 4 ' h iconiiccrion p i n t : srenrrrio .?

Original 1.A list: 120, 14, 21, 10, 19, 9, 1 ) NCW LAs.

L?I!:Lilen M H is iii LA 13:

1 A s thal rtiiist hc paged in AS:

120, 14, 21, 10, 19,9, 1. 12, 13)

(20, 14, 21, 14 w,9, I. 12, 131

time

I

tag

I

3.2

Data structure

definition

Bcforc dcscribiiig how our intelligcnl location tracking strategy works, the data structui'c or the paging informa-

tion record inusl bc defined first, A sct of paging infoiination rccords stores informtltion about an MH com- municaling with the location nianagcmcnt system. Fig. 5 shows the three fields of a paging inforination record, which are explained as follows:

(i) LAI: The l o c a h n area identifier (LAI) idciitiks the LA in which an MH communicated with the location manage- ineiit syslcm during the most recent tiinc pcriod. Coinmu-

JLE Proc.-Uoimirrm.. Vol. 141, No. 1. i'dwary Z W i l

nication events including the MH entcring a new LA that is not in the LA list, the MH having a call to dclivcr, and a call for the MH, niust be i~ecorded.

(ii) Time; This is the slarl linic (last conncclion time} when an M H had a coiinection with the location inonagcinent systctn. This is an important factor in deciding which LA 10 pagc first for call delivery.

(iii) Tog: This is a mark (I: successful; 0: unsiiccesshl) to let us know if an LA lias ever been successfiilly paged during thc observation period hefore an MH updates the profilc in the system. It is a factor to decide aii offset

(W.

If there are

i paging inrominlion r ~ o r c l s with 1 in the T(4g field, thc offset

(W,

is defined as:

T.IJ = 2 x ?U

where w is B weighting parameter. W is a11 offset that the

location inanagcinciit system. should add to the

/ps

of all LAs in tlic paging inforination records when it rearrttnges the LA scquciicc in the EA list. The purpose of W is to express thc most rcccnt history of the MH.

vislted LA

no

determine the offset, W no

and generate apaglng

llst

I

page an IA based on the

LA order in the paging list

or the last LA

3.3 Intelligent location

tracking

strates

y

To reduce the paging cost under the two sccnarios

described in Section 3.1, we pay cxlra allcndon to short- tcrin events. Based on the AS slralegy, we add a new data itcm and store it in the HLR. The data itcm is callcd 'paging information records' (1, 2 or morc rccords), described

i i i the previous Seclion. Thc flow chart of the inteltigenl

location tracking sli~~lcgy is shown in Fig. 6. When ;in MH

is callcd, using this strategy the MH is not paged according to the original LA sequence in the LA list. First, wc dcter- niinc thc time difference between the time of the last d l to or rrom Ihc MH and the time of a iiew urrivd call to the MH. If thcic is inore than one record in the p4ging inlbr- niation records, we compute the time difference from the last record becwse this tiine diffcrcncc will bc tlic smallest. Then, we check if the time difference is sinallcr than the time tl~reshold, which is defined as the maximum residencc tiinc that an MH is in an LA. If not, the system will rcar-

(4)

rangc thc LA sequence in the LA list firs1 by taking the paging infomiation rccords into account. Those LAs that appcar in the paging inforination records incrcase their res- idcncc probability by an offset, W. The syskin sorts the

LA list and dm sorted list is called R ‘paging list’. Tlic sys-

tem then pages thc LAs according to the new LA scqucncc in the paging list until paging is successfd or the last LA is

encountered. The offset is intended lo reflect events that happened during this observation pcriod. The larger the offset, the higher the probability that an MH is in the LAs

of the paging infoimation records.

This strategy can resolve the two sho~~t-term event sce- narios which were shown in Fig., 3 and 4, and that result in high paging cost. In scenario 1, since the time differelice is smaller tban tlie time threshold, thc last visited LA (LA

14) i s paged first. As a result, the p r o p o d method can reduce needless paging.

In scenario 2, since the

inovenicnt habit of the MH has chaiigcd, Ihe AS needs to pitge RH the

LAs in the LA list. However, using the proposed strfitegy,

the location iipdatc cvcnts that the MH moved to R iicw LA would be stored in thc paging itiformation records. When 21 call arrives, thc ncw LAs 12 ancl 13 i n the paging

information records increase their rcsidcncc probability by ail offsct W Then the system will page an LA according to thc LA order in the paging list. In this way, the paging costs can be reduced.

4 Evatuation

In this Section we describe a simulahn model and use it to gcneralc M I-I movement data. We first defiiic sonic parani- cters {like MH groups, MH distribution, ctc.) to simulaTe an cnvironnient which is close to the M H s movcincnt in a rcal city. Then we compute the location updalc cost and tlie location tricking (paging) cosi or the three approaches

(CS, AS and our proposed approach) based on ihc simulation model.

4.7

Simulation

model

The simulation modcl is bascd on a city zone area model [I, 111. Thc city area is assumed to have a radius of 20km. There are four area typcs: thc city centre, the urban area, the suburban areii, and lhc rural area. This city model consists of 32 area z o n a (cigllt per area type), four circular highways, and four radial highways. Note that each arm is covcrcd by one base station, which corresponds to an LA. In the simulation we simulate a sainplc of 100 MHs roam- ing within city arcas. Thcre are three MH groups based 011 their mobility; high mobility users, working people, and housekeepers. Now we dawibe some environment pari” ters of this modcl.

Table 1: Distribution of MH groups

MH group Probability

HiQh mobility 15% Working people 45% Housekeepers 40%

MH groups boxed on !heir mobility: Table 1 shows the distribution of the three MH groups. Each MH group has its own mobility charactcristics. An MH movement depends on which MH group thdt it bclong to. When an M B cntcrs an LA, a residence time is sssigned. Tlic high mobility group has a short residence time in each LA on its way

to a dcstination. An MH in this group will be assigned a ncw dcstination after it reaches the original dcslinalion. 66

This process will bc rcpcalcd during simulation. The working pcople group will be assigned

a

destinatioil as its work place. When an M H in this group reaches its destination it will stay there for a long pcriod of tiinc. The housekeeper group is the same as tlie working pcoplc group except that it has low movement speed.

iMvverrient iltdtwtiorz pnin!.~ We assume each MH gro:aup has its own inovcnient attraction points. Movement attrat; tion points arc places where the MH may reside for a paiod of timc. Thcy may be houses, workplaces, shopping centres, and parks In cach area zone, we define the probability that an MFI inay be attracted into this area [L], as shown in Tahlc 2. An MWs destination is selected according to which MH group that it helongs to and the MH’s movemenl attraction points, as shown in Tables L and 2.

Table 2: Distribution of movement attraction points

City Urban Suburban Rural area centre area area

MH group

High mobility 39% 34% 20% 7%

Workingpeople 40% 34% 16% 10%

Housekeepers 11% 30% 47% 12%

Time zunes.: The simulatioii time runs from 7:OO am. 10 1O:OO a.m. Thc total simulation time is three hours and our major goal is to simulate MHs moving from home to ofice. Tlic simulations o f MHs moving from office to

hoinc arc similar. The difference is that destinations will bc sources, and vice versa.

iMobility aud tru@ runditioiru: We follow tlie model in

[I L]

to determine mobility conditions (wemge pedcstrian speed and vehicle velocity). The movement algorithms and Iraffc conditions for high mobility users, working peaplc and housckccpci-s (residential users) are also based on Ihc imodel in [ll].

Initiui 1WI distribution over the dry are(i: Since the distribution of M H s is an important f x t o r in detcmiining MH

sources, we initialise the M H distribution ovcr the city mea by assuming that most of them live in the rural arcas (45‘1/0), as shown in Table 3.

Table 3: Initial MH distribution over a city area

~~

City centre Urban area Suburban area Rural area

10% 22% 23% 4 5%

4.2 Simulation

results

In the simulntioiis wc sct w = 201180 and tiwe thtwholcl =

20 minutes. These two values cdn be adjusted bascd on the history of the MH. Fig. 7 shows that the location update costs of the AS and the proposcd inclhod are much less

than that of the CS with the variance of repeat probability. The rcpcat probability is the probability of the M H moving over the samc path as in the last time period. If the repeat probability is 1000/, this mcans that thc MH is moving over the same path as in the last timc pcriod during simulations and the iimcs or location updatc are zero. If the repeat probability is Oo/, this ineans that the MH never inovcs over the same path as in the last time period during simulations. However, this docs not incan that all the EAs on the movement path of the MH are complctcly different from the last period. That is, mine LAs may bc the same as

in the last time period. As the repeat probability dccrcaTs, the location update most increases. This is becausc thc MH

(5)

changcs its movement path. It will stay in some LAs that

didn't appear in the last time period.

e 8 Boo0

6 g

7000 3

&

6000 U

3

5000

3

2

c La E 4000

.p

3000 ~ 2 0 0 0 1000

In Fig. 8 wc

can

SIX that the proposed approach per-

roorms bctlcr ilian the AS on paging cost. Wlicn ilic rcpcat probability decrcascs, Llic paging cos1 of the AS increases faster than that of the proposed inethod. According to the SS7 MAP incssage size in GSM from [RI, which i s shown in Tablc 4, wc can compare the total location nianagcnicnt cost among thcsc three slrategies. Fig. 9 shows that o w incthod can savc 14-%?'0 of location mEmngelllcnl's signalling cost compareci LO the CS, and save 1 ~ 3 9 % signalling

cost compwed to thc AS under diffeient repeat probabili- ties. Our incthod pcrronns signifimintly better than tlic AS

when the repeal probability is low. Low repeat probability

implies high occurrciicc probability of short-iciin cvcnts. The advaiitagc or our method is due to the two short-term event sccnarios incnlioned before. Since the AS did not actually handle thc short-(elm events, in these two sccnar-

P

- - ~ -

-

01 ' 1 1 ' ' 1 ' 1 ' 1 I 6000 r 8 1000

1

Table 4 SS7 MAP message size in GSM

Variable Oescription SS7 bytes

b,, location update 461

b,l mobile terminated call 858 mobile terminated without authentication 470 bP2

1

P

(6)

ios, it may nwd to page almost all the LAs in the LA list before locating an MIJ. As shown in Fig. 10, we use the call-mobility ralio

(CMR)

as a variable to cvaluate location tracking cost. CMR is the ratio of call incoming ratc divided by LA crossing ralc. The location tracking cost of our nicthod is smalkr than that of Ihc AS under diffcreut CMRs. As expected, our method and the AS havc more

location r “ n g cost than the CS. In Fig. I I it is shown that our method 1x1s the lowest total (tocation sigiidling) cost among tlie t h e e methods except whcn CMR is closc to 1 .

5 Conclusions

The AS is suitable for situations

or

high mobility raLc and small LA size because it significantly reduccs the location update costs cornparcd to the CS. Our approach rurther itducas localion tracking cosb it1 comparison with the AS, without increasing location updatc COSB. It just nDcds extra storage spaw atid some simplc computations. Since the advances in memory capacity atid proccssor speed arc faslcr than improwinelits in the availability of radio channels, this ovcrhead is nol significant. Our method

cm

reflect an MB‘s movcment belluviour by generiiting a dynamic paging list, If an MH performs actions that it sel- dom or never did during tlie previous time period, 0111‘ approach will increase an offset for each LA in the paging inronnation records In this way, the proposcd location m a n a g c m c t i t system cmi significantly rcduce the totd signalling cost by responding lo the MH’s bchaviour quickly before an MH updates its profile.

6 Acknowledgment

This work was supportd in part by thc National Science Council of the Republic of Chitia under &ant NSCSS-

221 3 - W 9 - 0 3 9 . 7 I 2 3 4 S 6 I 8 3 I O 1 1 References

MARKOULIDAKIS, J.G., LYBEROPOULOS, G.L., ‘ISIRKAS: D E , wntl SYKAS, E.D.: ‘Mobility modcling in third-gcncrtition

mohilc tclccummunication systems’, I i X K Pcrsoiitd Catnniun., 1997,

nn dI..M

,, >,

, , , I _

REILLY, P.1. ~ ‘Signal traffic in tlic pan European digital ccllulat radio (GSM) systcm’. Prucedings of IUEB Vehicular ‘lcchnology

Socictv 42nd VTS confcittm. Mev 1992. Vol. 2. m. 721-726 BAlt-NOY, A., KESSI,E,H, i,, a i d SIDI, M.: ’ f k k i n g strakgies in

wirclcss ne~works’. Proocodings or [EEE 8th Convention of Elcctricd and Elcctronicu Engneers in Israel cunferencz, 1995, pp. 5.3.4!1-3 JALN, K., LlN, Y.-B., LO, C., ond MOHAN. S.: ‘A caching stnregy to 1-educe nctrvork iinpticls or PCS’, 1E.M J. Sd. Arcm Catnnw.,

1994, 12, (S), pp. 1434-1444

JAIN, R., LIN, Y.-H., LO, C., and MOHAN, S.: ‘A forwanliiig strat- egy to rcduce network impacts of PCS’. Procecdings ul‘ IEEE [ N I 0 COM ‘95, April 1995, Vol. 2, pp. 481-489

TABBANE:, S.: ‘LuMtion managcincnt methods faor third-gccnernlion mobilc systems’, iEiX Comwrt~i, M c i ~ . , August 1997, pp, 72-84

‘I’AIIMNE, S.: ‘An altcmativt strattgy for locatiun tracking’, IiiLE

J. S(d A r m C~~nmiun., 1995, 13, (3, pp. 880 ,892

POLIJNI, G,, and TABBANE, S.: ‘The intelligent nctwork signaling

and switching~wosts of an altcri&ive localion s h t c g y iising miniory: Promdings of IEEE ‘93 Vehicdnr ‘t’cchnology confcrencc, 18 .-20

May 1993, pp. 931-931

AKYOL, i3.A.: m l COX, U.C.: ‘Htintlliiig mobility in a wirclcrs ATM network’. Proceedings of ICBE INFOCOM’96, May lW6, Vol. 3. nn. 1405-1413 ~~ ~ ~~~

1 LIN. Y.-D.. wnrl HWANG. S.-Y.: ‘CoInoarinc thc PCS tomtion track-

ing slratcgics’, IEEE TwH,< Vdt, 7?chd., 1936, IS, ( I ) , pp. 114-I21 MEROS. D.V.. ‘ISIRKAS. D.E.. and SYKAS. E.D.! ‘lntclliecot ~ I B -

LY IXI~OPOULOS, CLI-., MARKOULIDAK IS, J.G,, POI ,Y-