Optimal Geolocation Updating for Location Aware Service Provisioning in Wireless Networks

Transcription

1 Optimal Geolocatio Updatig for Locatio Aware Service Provisioig i Wireless Networks Siri Tekiay Amer Catovic tekiay@adm.jit.edu axc4466@jit.edu New Jersey Istitute of Techology Uiversity Heights, Newark, NJ, USA 1 Abstract I this paper we preset a study of differet geolocatio updatig schemes for locatio aware service provisioig ad aalyze the ways to their optimizatio. We propose a hybrid periodic/predictive updatig method, i which adaptive update itervals derived from the geolocatio iformatio received from the mobile are used for MSs with domiat directio of movemet. Algorithm for detectig MSs with a domiat directio of movemet is proposed. This geolocatio update scheme reduces the sigalig load by a factor of two, ad opes the way for optimal geolocatio updatig. 1. Itroductio A defiig characteristic of ext geeratio wireless etworks will be the locatio aware services ad applicatios. These services ad applicatios truly take advatage of the tetherless ature of the users by providig custom tailored services to the user based o his/her locatio ad aidig i etwork fuctios such as routig, ad pre-fetchig. The mai cocept associated with locatio aware services is the cocept of target area. A target area is the arbitrary area specified by each locatio aware sessio or message, which ecloses the mobile or fixed odes destied to receive the service or message. A target area may be fixed for the duratio of a locatio aware sessio or it may be mobile with the MS at its ceter or edge. The major issue i providig locatio aware services for mobile cellular etworks is the mappig of the boudaries of the target area oto the cellular layout i a cost effective ad timely maer. Oe way of doig this is itroducig a cetral geolocatioaware compoet ito the etwork that has a view of the cellular layout, i.e. cellular boudaries, ad the capability of mappig the target area boudary specificatio oto the cells, ad cosequetly oto the Base Statios (BSs) that eed to be ivolved. After the BSs that will potetially trasmit for the locatio aware service sessio are determied, i tur, the geocast group is formed of the MSs withi the target area, based o the geolocatio iformatio submitted by MSs, i a maer that is trasparet to the seder. Cosequetly, the issue of target area mappig becomes a issue of maitaiig accurate geocast groups. The iaccuracy of the geocast group defiitios results from the fact that a MS ca be icluded i the geocast group, while it is outside the target area at the time of the arrival of a locatio aware service, Ad vice-versa. We deote such error as error i geocast group membership. This error is a measure of the quality of service offered by the locatio aware service-capable etwork. The ultimate quality of service parameter could be defied as the probability of havig oe error i geocast group membership, deoted as P c. Alteratively, P c could be defied as the ratio of the umber of errors i geocast group membership to the total umber of MSs that are sedig their geolocatio updates to the etwork. I this paper, we adopt the latter defiitio of P c. The accuracy of the geolocatio iformatio o MSs depeds o the frequecy at which the geolocatio iformatio is beig updated. O the other had, the amout of sigalig, measuremets, computatio, updates, ad lookups due to geolocatio ca become a sigificat cotributor to etwork maagemet costs. I the light of the above stated facts, the optimizatio of the update iterval is crucial for tradeoff betwee the miimizatio of sigalig load ad miimizatio of total probability of error i geocast group membership. While satisfyig the requiremets of providig accurate geocast groups ad miimizig the sigalig load, a successful geolocatio updatig scheme should be simple from the poit of view of computatioal complexity ad data storage requiremets, because of the limitatios imposed by the real-time eviromet ad umber of MSs, respectively. I the case whe geolocatio update itervals are determied by MS, the computatioal complexity ad data storage limitatios are further amplified by power cosumptio limitatios of mobile hadsets. I the subsequet sectios, we will examie differet geolocatio updatig schemes, ad show that a 1 This work has bee supported i part by grats from New Jersey Ceter for Wireless Telecommuicatios ad NTT DoCoMo, Ic.

2 simple liear predictive scheme performs sigificatly better tha the other schemes that do ot rely o previous geolocatio updates.. Novel Geolocatio Update Scheme for Locatio Aware Service Provisioig We cosider the referece framework i Figure 1. I this particular architecture, the geolocatio-aware compoet occurs i the etwork, ad is reached via the Mobile Switchig Ceter (MSC). I our referece model, all of the BSs belog to the same MSC. MSs are capable of obtaiig measuremets ad computig their ow geolocatios via cotiuous GPS. Figure. Sigalig flow for per-sessio updatig scheme I the periodic geolocatio update scheme, Figure 3, MSs are sedig their geolocatio updates regularly, regardless of whether a service alert was received or ot. The last available geolocatio is used to decide o the geocast group membership at the time of a ew sessio arrival. The update itervals are determied by the etwork s geo-compoet ad broadcasted to all mobiles i the area. The update iterval selectio will set the tradeoff betwee the sigalig load ad the probability of geocast group membership error. Clearly, periodic updates are oly applicable if the update frequecy is smaller tha the locatio-aware service arrival rate. Figure 1. Referece etwork architecture I the focus of our study is the compariso betwee three differet gelocatio update schemes. The itet is to determie if the trajectory predictio ad adaptive update iterval selectio ca help reduce the sigalig load ad reduce the probability of icorrect geocast group memberships. Our primary goal is to compare differet geolocatio update schemes i terms of sigalig load ad error probability for locatio-aware service delivery. Update iterval optimizatio for the purpose of sessio maiteace is a differet issue, which will ot be treated here. Therefore, the sessios are assumed to be oemessage log. I the first scheme (Figure ), MSs sed their geolocatio updates upo arrival of a ew locatio aware service. By comparig mobiles geolocatio with the target area boudaries, the geo-compoet decides whether they belog to the geocast group or ot. This scheme is iheretly delay geeratig, as geolocatio updates must be completed i the time betwee the locatio-aware service arrival ad the time whe geocast groups have bee formed ad the sessio ca actually begi. Also, if the locatio-aware service arrival rate is high, this scheme ca geerate sigificat sigalig load. I additio, sice the updates from all MSs are sychroized by sessio arrivals, update messages geerate bursts ad strogly bottleeck the system capacity. From the other had, this scheme provides very low error probability (moreover, if the delay is eglected, the scheme is error-free). Figure 3. Sigalig flow for periodic updatig scheme The predictive geolocatio update scheme (Figure 4) cosists i predictig the geolocatio of a mobile, based o the speed ad directio of movemet mobile had at the time of the last update. The update itervals are determied adaptively, ad they are equal to estimated time i which the MSs will cross a pre-defied distace step. The size of the step determies the average geolocatio update frequecy. The update itervals ca be determied by the geo-compoet ad paged to every MS; alteratively they ca be computed by MSs. Regardless of where the geolocatio update itervals are computed, the geo-compoet uses the last available geolocatio to determie the positio of each MS. Update itervals are beig adjusted adaptively to the displacemet of the MS i the previous update iterval. Therefore, MSs havig smaller average displacemet will sed their updates less frequetly, ad vice versa. The MS s geolocatio at the time of the sessio request arrival is predicted based o the average speed ad directio of

3 displacemet i the last update iterval, ad the time elapsed sice the last geolocatio update, as show i Figure 5. Figure 4. Sigalig flow for predictive updatig scheme Figure 5. Predictive geolocatio updatig Assume that the last geolocatio update iterval for the MS from Figure 5 was τ = t B - t A, where t A ad t B are the times at which the MS visited locatios A = (x A, y A ) ad B = (x B, y B ), respectively. The ext geolocatio update iterval, τ +1, will be determied as: s τ =, +1 d( A, B) / τ where d(a,b) deoted the distace betwee locatios A ad B ad s is a expected displacemet per update iterval. The expected displacemet is set by the etwork admiistrator ad represets the distace a MS is expected to pass betwee two geolocatio updates. It is directly related to the size of the target area. Assumig a sessio request arrival occurs at time t S before the ext geolocatio update, so t S < t B + τ +1, the predicted geolocatio of the MS at time t S will be D = (x D, y D ): d ( A, B) xd = xb + ( ts t B ) cosα τ d ( A, B) yd = yb + ( ts t B )siα. τ The exact geolocatio of the MS at the time t S will be locatio C. The predictio error will therefore be P e = d(c, D). This error ca result i the error i the geocast group membership for the MS. 3. Simulatio Model We assume 3 km wide square etwork coverage area with 100 per cet radio coverage. 300 MSs are uiformly distributed withi the etwork coverage area. Siglemessage locatio aware service sessios, destied to all users withi the target area, arrive with expoetial iterarrival times, with the rate of λ=30 sessios/hour. A differet target area is associated with each ew message arrival. Therefore, the target areas ad the correspodig geocast groups are deleted after the messages have bee broadcasted to the MSs from the group, ad the geocast groups are re-created for every ew message. Target areas are circular, with radius r = 500 m, placed radomly ad eclosed withi the etwork coverage area. Cotiuous GPS feed is assumed, i order to elimiate geolocatio errors. Therefore, give 100 per cet radio coverage assumed, errors i geocast group membership ca oly be due to iaccurate (obsolete) geocast groups. We use the mobility model based o the two-dimesioal Browia motio [1], []. This model is capable of mimickig vide rage of mobility patters, by varyig model s parameters. At each time step t, MS moves by oe space step x to the right, left, forward or backward with probability p 1, q 1, p ad q, respectively. I such motio, the average drift velocity i oe of the two dimesios is give by µ 1 = (p 1 -q 1 ) x/ t, ad the diffusio is σ 1 = ((1-p 1 )p 1 + (1-q 1 )q 1 + p 1 q 1 )( x 1 ) t 1, ad for the other dimesio accordigly. The average drift velocity is the measure of average displacemet per uit of time, while the diffusio costat is the measure of locatio ucertaity. Although the motio is twodimesioal, the domiat directio of the movemet ca have arbitrary orietatio. We assume three types of MS with respect to their mobility patter, amely pedestria, urba vehicular ad highway vehicular. Their mobility parameters are give i Table 1. Table 1. Mobility model parameters This model gives rise to Gaussia margial probability desity fuctios alog the two orthogoal directios of movemet, with parameters ( µ, ) ad ( µ, ). 1 σ 1 σ 4. Results ad the User Classificatio Algorithm Figure 6 shows the graphs of the probability of error i the geocast group membership as a fuctio of the umber of geolocatio updates set per MS per hour for periodic updatig scheme. Each curve represets a system populated by users with particular mobility type amog the types discussed i the previous sectio. I additio, a system with all three users mobility types geerated from poit probabilities of 1/3 is also represeted. The umber

4 of geolocatio updates is iversely proportioal to the geolocatio update iterval. The probability of error was defied as the ratio of the umber of MSs that were mistakely icluded i or left out of the geocast membership group to the total umber of MSs that were sedig geolocatio updates. withi the etwork coverage area. The user powers up defaultig to periodic geolocatio updates, with update iterval τ, broadcasted by the BS. The etwork the keeps track of the first N geolocatio updates of the MS, occurrig withi the time period T = Nτ. If the MS stays withi the circle of radius R aroud the iitial locatio A durig the time period T, as depicted i Figure 8a, the etwork cocludes that his et displacemet per uit time does ot justify predictive geolocatio updatig, ad the MS cotiues to sed its geolocatio updates periodically. If the MS crosses the circle boudary at ay time withi the time period T, as i Figure 8b, the etwork decides that the MS has a domiat displacemet directio, ad switches to predictive updatig for this particular MS. Figure 6. Performace of the periodic updatig scheme Figure 7 shows the same curves for the predictive updatig scheme. The umber of geolocatio updates is cotrolled by settig the expected displacemet per update iterval, s. These figures should be compared havig also i mid the first geolocatio updatig method described i the previous sectio, which geerates sigalig load per MS equal to the umber of locatio aware sessio arrivals. Therefore, the umber of updates per MS per hour is λ 60 mi = Sice this updatig method provides o error i geocast group membership, the other two methods are meaigful oly if they geerate less tha 1800 geolocatio updates per hour. The graphs idicate that sigificat gai i terms of sigalig load could be achieved by predictive geolocatio updatig for MSs havig substatial et average displacemet per uit time, which idetifies a domiat displacemet directio i MS s movemet. The amout of gai is proportioal to the magitude of the mea displacemet i the domiat directio. For users with small or zero et displacemet per uit time, the periodic updatig scheme performs better. It follows that a combiatio of periodic ad predictive geolocatio updatig could pave the way to the optimal updatig scheme If the etwork was capable of detectig MSs with a domiat displacemet directio, it could perform the predictive updatig o such users. O the other had, MSs with o domiat displacemet directio could sed their geolocatio updates periodically. A simple algorithm for detectig MSs with a domiat displacemet directio could execute as the MS becomes active at a locatio Figure 7. Performace of the predictive updatig scheme Figure 8. Illustratio of User Classificatio Algorithm Parameters R ad T reflect the policy etwork adopts to classify the MSs accordig to their et displacemet per time uit.. Parameter R determies the magitude of the et displacemet per time uit required to trigger predictive updatig. The amout of sigalig the etwork ca afford i for the purpose of user classificatio is set by the parameter T. The bigger the T, the loger will highly directioal users sed their updates periodically, ad thus create uecessary sigalig traffic. Icreasig R

5 ad T will allow more users to display the directioality i movemet ad thus be switched to predictive updatig. Figure 9 shows the result of the simulatio of the etwork employig user classificatio algorithm with parameters R = 00 meters, T = 60 sec, τ = 15 sec. As ca be see from the graph, the predictive geolocatio updatig with classificatio shows potetial for optimizatio of geolocatio updatig for locatio aware services. For example, for probability of error of 1%, which is a reasoable QoS level, selective update scheme requires 70 updates per hour i average, while periodical update scheme requires 600, which is improvemet of more tha 100%, yieldig proportioal reductio i sigalig load. For etworks with eve higher percetage of users havig domiat directio of movemet, the gai will be eve bigger. For etworks pre-domiatly populated by users without directioality i movemet, the selective update scheme will coverge to the periodic scheme, which gives the best performace i such case. The performace of the predictive updatig scheme with classificatio will greatly deped o the fie-tuig of the set of parameters R, T, s, τ }, where τ p is the { p update iterval for users who have bee classified for periodic geolocatio updatig. The impact of τ p grows with the populatio of MSs with o domiat displacemet directio. Figure 10 shows compariso betwee selective updatig with differet periodic update itervals, τ p, for a etwork where 1/3 of MSs have bee classified for periodic geolocatio updatig. Parameter optimizatio is tightly related to the uderlyig mobility model, ad subsequetly, o the mobility model s parameters. Estimatio of the mobility model parameters for each MS is a separate issue that is out of the scope of this work. O the other had, aalytical optimizatio of geolocatio update parameters uder the assumptio of kow mobility parameters is a importat issue that deserves close attetio i further research. It is clear that further examiatio of differet techiques for fast ad efficiet classificatio of users accordig to their optimal updatig methods would yield additioal gais i terms of sigalig load ad error i service delivery. Our future work will be focused o aalytical evaluatio of differet user classificatio algorithms ad geolocatio updatig schemes. Figure 9. Performace of predictive updatig with classificatio 5. Coclusio I this work, we itroduced the problem of optimal geolocatio updatig for locatio aware services i wireless etworks. We proposed ad aalyzed differet geolocatio updatig schemes for locatio aware services i wireless cellular etworks. We showed that trajectory predictio ad adaptive updatig itervals could sigificatly decrease the amout of sigalig load geerated by geolocatio updates set by mobile statios. Itelliget yet simple scheme for trackig ad predictig users'locatios ad computatio of geolocatio updatig itervals provides a multifold decrease i sigalig load i etworks where users with a domiat directio of movemet are preset. Figure 10. Impact of τ p o the geolocatio updatig performace Refereces [1] S. Tekiay, Modelig ad Aalysis of Mobile Cellular Networks with Highly Mobile Heterogeeous Traffic Sources, PhD Dissertatio, School of Iformatio Techology ad Egieerig, George Maso Uiversity, Virgiia, 1994 [] C. Rose, R. Yates, Locatio Ucertaity i Mobile Networks: A Theoretical Framework, IEEE Commuicatios Magazie, February 1997