Enabling vehicular mobility in city-wide IEEE 802.11 networks through predictive handovers

Contents lists available atScienceDirect

Vehicular

Communications

www.elsevier.com/locate/vehcom

Enabling

vehicular

mobility

in

city-wide

IEEE

802.11

networks

through predictive

handovers

M. Mouton

∗

,

G. Castignani,

R. Frank,

T. Engel

InterdisciplinaryCentreforSecurity,ReliabilityandTrust,UniversityofLuxembourg,4rueAlphonseWeicker,L-2721Luxembourg

a

r

t

i

c

l

e

i

n

f

o

a

b

s

t

r

a

c

t

Articlehistory:

Received22October2014

Receivedinrevisedform10February2015 Accepted13February2015

Availableonline5March2015 Keywords: IEEE802.11 Vehicularcommunications Handoveroptimization Predictivehandover Context-awareness Experimentalevaluation

TheincreasingnumberofIEEE802.11networksdeployedworldwidegivesmobileusersthepossibilityof experiencinghigh-speedwirelessaccessonthemove.Moreover,thehighdensityofthesedeployments inurban areasmakeIEEE802.11 asuitable accesstechnology formoving vehicles.However, inorder toprovideaseamlessaccesstovehicles,thetransitionbetweenAccessPoints(APs)mustbequickand reliable.ThemainbottleneckofexistinghandovermechanismsisthelongAPscanningprocess,which onlyprovidesasnapshotoftheavailablenetworksatagivenlocation,impactingthehandoverdecision onmovingvehicles.Toovercomethislimitation,wepropose coper,a context-basedpredictivehandover mechanism thatconsiders vehicle’s trajectory,road topology, and network deploymentinformation to decidethebesthandoverlocationandcandidateaccesspoints.Wevalidatewithrealexperimentsina city-wide802.11networkandshowthat coper canprovidebetteraveragesignalstrength,datarate,and connectedtimethanotherexistinghandoverapproaches.

©2015ElsevierInc.All rights reserved.

1. Introduction

In the recent years, the number of mobile users connected

through802.11networkshavesignificantlyincreased.Atthesame

time, the increasing number of wireless network deployments

openthe waytowardsubiquitous networkconnectivity. However, undervehicular mobility,theselectionoftheAPinprovidingthe most sustainable network connectivity is a challenging task. In thiscontext, the standard IEEE 802.11 handover comprises three phases:scanning,authenticationand,association,andisinefficient inensuringaseamless transitiontothe bestAP.Itiswell known thatthescanningprocessinwhichtheMobileStation(MS)probes nearbyAPsoneachchannel andwaits forproberesponses,isthe majorbottleneck[1–3].Thisinefficiencyismainlycaused(i)bythe longdelay inthe handover and, (ii) by thelack of indication on short-termchanging inthenearbyAPs’ReceivedSignal Strengths (RSS).The firstimplieslongdisconnection periods,whilethe sec-ondleadstopotentiallyinconsistentAPselection.

Many solutions have been proposed to mitigate these issues withsome success, mostly by reducing the disconnection period duringthehandover.ThefutureemergenceofIEEE802.11p/WAVE

*

Correspondingauthor.Tel.:+3524666445877.

E-mailaddresses:maximilien.mouton@uni.lu(M. Mouton),

german.castignani@uni.lu(G. Castignani),raphael.frank@uni.lu(R. Frank), thomas.engel@uni.lu(T. Engel).

networks [4], which have been specifically designed for

vehicu-larcommunications, opens theway towardsseamless handovers.

One of the major modifications proposed in thisstandard is the suppression ofthe authentication and association phase,and the transmissionofWirelessAccessinVehicularEnvironments(WAVE) beaconscontaining theservicetransmissioncharacteristicsovera dedicated channel.However, this standard still providesno guid-anceastohowAPselectionmaybeoptimized.

As aresult, incurrentandfuture 802.11 deployments,the in-consistency oftheAP selectionremains an open issue.Currently, the majorityofthe proposed handoveroptimizationonly rely on the instantaneous sensing of the RSS. However, thismetric only provides asnapshot ofthe currentstate ofthenetworkanddoes not provide any trend for its evolution, which is critical in the context of vehicularcommunications. Gustafssonand Jonsson [5]

distinguishsessioncontinuityasoneofthemobilitymanagement enhancement that arenecessary toachieve theAlways Best

Con-nected (ABC)paradigm. The MS needs toaugmentits knowledge

on nearby networks and vehicle’s mobility such that it can an-ticipate their evolution. Thisanticipation allows the MS to avoid abruptconnectiondisruptionduetoinappropriateAPselectionand toprovideaseamlessconnectiontotheAP.

In this paper, we propose coper, a COntext-aware Predictive handovERtechnique conceived withtheaimofoptimizing vehic-ularcommunications througha metropolitanIEEE 802.11 hotspot network. The AP selection process of coper uses a prediction of

MS connectivity over the nearfuture such that the roaming de-cisionisperformedatthemostappropriate location.Inthiscase, a priorknowledge oftheroute takenby thevehiclewouldallow computingoptimalhandoverlocations.However,mostofthetime, peopledriveonpreviouslyknownroutes,thus,donotneedtobe assistedby anyembedded system.As aresult,there isnowayto knowthedestination,northeroutethedriverplanstotakein ad-vance.Infact,thisinformationisnotreliablesincethedrivermay change theroute on impulse. Consequently, coper includes a di-rectiondetectionmodulethat providesashort-term predictionof thedirectionofthevehicle.Inordertoperformthe direction de-tectionandtheAPselection, coper usestheknowledgeoftheroad topology,theAPs’locationsandtheirmodeled RSS,allstoredina ContextDataBase(CDB).

The remainder of this paper is organized as follows. In Sec-tion2 we presentrelatedwork. Section 3describesthe different modules composing coper andthe CDB construction. Theresults oftheevaluationstudyarepresentedinSection 4.Section 5 pro-posesa discussiononthepossibleenhancementsof coper.Finally, inSection6weconcludethepaper.

2. Relatedworks

During vehicular mobility, an MS connected through 802.11

networks faces frequent handovers and must select the next AP

andassociatewithintheshortestdelay.Priorstudiesonthe stan-dard IEEE802.11 handover mechanism[1–3] identified the scan-ning phase as the most costly in time. Indeed, during scanning, the MS probes nearby APs and waits for responses on all avail-able channels. Scanning implies a disconnection period because theMScannot exchangeframeswiththecurrentAPwhile listen-ingon otherchannels. Severalapproacheshavebeenproposedto reduce or even eliminate the impact of the scanning phase dur-ing handover. Montavont et al. [6] proposed an optimization to thisprocessbyperformingshortinterleavedscanningphaseswith on-going data communication. In SyncScan [7], Ramaniand Sav-age investigated a modification of the infrastructure where the APssynchronouslybroadcast beaconframesonthe samechannel so that the MS justneeds to switch the channel andwait for a shortperiodtoretrievetheavailableAPs.Anothersolutionconsists inrelocatingthe scanningtoa secondinterface. Theexperiments conductedbyRamachandranetal.[8]andBriketal.[9]showthat sucha solutioncanprovideasignificantgainintermsofthetime thattheMSisassociated.Inourpriorwork[10],weevaluatedthe impactofaddingasecondradioandshowedthattheMScanreach upto 98% of layer2connectiontime. Eventhoughtheimpact of scanningonhandoverdurationcanbesignificantlyreduced,itstill suffers from providing inaccurate results. Indeed, scanning is an instantaneoussensingofthenetworkthatdoesnotalwaysreflect reality.Becauseofbeaconloss,scanningcanmissan availableAP, especiallyindensedeployments, likethosediscusesin[11,12].To overcomethisissue,a solutionistouse repeatedmultiple scans, split into short interleaved scanning phases as described in [6]. However, thisapproach is not applicable to vehicular communi-cations,asthe sensedvaluesbecome rapidlyobsolete becauseof therelativelyhighvelocity.

InordertoprovideadynamicknowledgeofnearbyAPs,Mhatre andPapagiannaki [13] propose performing RSS predictions based onsmoothedRSStrends.A similarapproachisproposedbySadiq etal.[14].Nevertheless, thiskindof predictioncan only be per-formed over a very short period, since it doesnot consider any informationonthevehicle’smotion.Forinstance,theMSwillnot beabletopredictasuddensignaldecreasewhenthevehicleleaves thelineofsightoftheAP.

In the literature, multiple 802.11p network architectures for

Vehicle-to-Infrastructure (V2I) communications have been

pro-posed,includingmultiplehandovermechanisms.The network ar-chitecture described in [15] consists ofIEEE 802.11f Inter Access Point Protocol (IAPP) compatible RoadSide Units (RSU) intercon-nectedbyasetoflayer2switchesconnectedtoagatewayrouter. This makes, the network appear as a single distribution system andmobilitymanagement ishandledatthelayer2,avoiding the useofInternetProtocol(IP)mobilityprotocolsuchasMobileIPv6 (MIPv6)orProxyMobile IPv6(PMIPv6).Thehandoveristriggered

when the On-Board Unit(OBU) sends an IEEE802.11

disassocia-tionmessagetothecurrentRSU.TheRSU,then,communicatesthe OBUaddresstothenextRSUbyforwardingthepacketsaddressed toOBU.Themajordrawbackofthisapproachisthatitisdesigned

to workina highwayenvironment wherethenext RSU selection

istrivial. Notethatinan urbanscenario,theAPselectionprocess maynotbeastrivialasinthehighwayscenario.Anextended ver-sionofthisapproachpresentedin[16],isintendedtoworkinan urban environment.The authorspropose forwardingdatapackets toallthecandidateRSUs.WhenanRSU detectsthattheOBUhas entered inits coverage range, it isselected asthe next RSU and

sends a Move–Notify message to the remaining RSUs. However

suchanapproachimpliessignificantpacketoverhead.

In order to allow the MS to anticipate the short-term evolu-tionofthenetwork,itiscriticaltoprovideinformationaboutthe vehicle’s mobility.Based onthe fact that peopleusually drive on previously known routes, Deshpande et al. [17] propose replac-ingthescanningwithanAPselectionusinghistoricalinformation. Whiledriving,theMSlearnsandcachesinformationaboutnearby APs (ESSID,BSSID,channel) byfrequently sensing during inactive periods.Thisinformationis,then,usedtoscriptthehandover loca-tioninadvancesuch thattheMSisalways connectedtothebest candidate AP. Anotherapproach is proposed by Kwak etal.[18], who suggest includingthe trajectory ofthe vehicleandneighbor information in order to select APs along the vehicle’s path and predict an optimal handoverlocation. This approachwas

investi-gated by Montavont and Noel [19], who propose an augmented

versionoftheMobileIPv6(MIP6)architecturewithanew compo-nent,a GPSserverwhichisperiodicallyupdatedwiththelocation of the MS. Based on the evolution of the MS location and prior

knowledge of AP locations, channels, BSSIDs and IPv6 prefixes,

the GPS server triggershandover to theclosest APwhen theMS is about to leave the coverage range of thecurrent AP by send-ing a handoverindication packet tothe MS.In [20] and[21] the authors use large datasets gathering mobility history of cellular network users [20] andlarge indoor 802.11 network [21]. Zhang etal.[20] propose a cloud-basedmobilitypredictionsystem that consistsindetecting periodicityinthemobilitypatternusingthe Kullback–Leibler divergence (KLD) asthemetric andevaluate so-cial interplayinorderto identifysome pairsof callsthat willbe co-located. In[21]Abu-GhazalehandAlfa modelthemobility be-havior ofusers ina 802.11 networkasMarkov RenewalProcesse (MRP). This model intends to predict handovers andthe sojourn duration of a user based on the current location and the prior locationoftheuserimmediatelybeforethetransitionintothe cur-rentlocation. Inthe sameway,DeRango etal.[22] proposedan efficient prediction-basedbandwidth allocation schemeusing the MobileReSerVationProtocol(MRSVP)andananalysisofusers’ mo-bilitytoensurequality-of-service(QoS).Theseapproachespropose long-termpredictionsthatallowtheMStoensureservice continu-ityincasetheselectedAPisunavailable(switchoffornotableto provideenoughbandwidth).Inourprevious contribution(mroad)

Fig. 1. Description of the three main modules of coper.

Inthispaper weintend toselect thebest APandpredict the handoverlocationwithoutknowingtherouteofthevehiclein ad-vance.The twomain improvementsprovidedby thissolutionare theadditionofadirectiondetectionmodule,andan APselection module both relying on both the current location of the vehicle andaCDBoftheroadtopologyandthemodeledAPRSS.

3. COPER

In order to mitigate the limitations of the standard 802.11 handoverina vehicularscenario, wepropose a roamingdecision techniquethat intends toprovidealways bestconnected wireless access by avoiding the MS scanning process. The choice of the APprovidingthebest connectivityhastraditionallybeendone by selectingtheAPwiththehighestRSSdiscoveredthroughthe scan-ningprocess. In this work, we propose to base the AP selection processonprior knowledgeoftheroadtopology,the locationsof theAPsandamodeloftheirRSSs.Usingthisinformation,theMS isabletodetermineitslocationontheroadnetworkandusethe modeledRSSsofnearbyAPstochoosetheonepotentially provid-ingthebestsignalinthenewvehiclelocation.

AsillustratedinFig. 1, coper iscomposed ofthreemain mod-ules intended to provide seamless transitions between the best candidate APs regardless of the vehicle’s route. To this end, the DirectionDetectionModule(DDM)anticipatesthedirectionofthe vehicle usinga fuzzylogic based analysis trajectory that consid-erstheroadcharacteristicsstoredinapre-computedCDB.Forthis direction prediction, the AP Selection Module (ASM) makes use of the simulated AP RSS stored in the CDB in order to look for thebest APs andthe bestlocations to triggerhandovers. Finally, theConnectionModule(CM)detectswhenthevehicleisentering a handoverareaandattemptsa connectionbyprobingtheselected APandassociatingwithit.Thefollowingsectionsdescribethe con-structionoftheCDBandthe coper modules.

3.1.CDBconstruction

coperdecisions arebasedonthe roadtopologyandthe mod-eledRSSsthatare storedinaCDB.Tobuild thisCDB,weneedto gather multiple publicly-available sources of information and or-ganizetheminthe proposed datastructure.In thisworkwe use twonotionstocharacterizetheroadtopology.ThefirstistheRoad

Portion (RP)definedastheportionofroadboundedbytwo inter-sectionsoradeadend.AnRPischaracterizedbyanazimuth

α

and thecoordinatesofthestartingpointD andtheendingpoint E.The secondnotionistheRoadSegment (RS),whichisafragmentofthe RP.EachRPiscomposedofasetofRSshavingequalsize(around 5 m).NotethatbothRPandRSareoriented,meaningthata two-waystreetiscomposedoftwoRPs. TheCDBstructureconsistsof fivetables.Threedescribetheroadtopology:thefirstliststheRPs includingtheirboundaries(intermsofGPS coordinates),sizeand theirRSs;thesecondliststheRSsincludingtheirboundaries,size andtheRPtheybelongto;andthethirdliststherelationsbetween RPs.TheAPtablestorestheAPs’locations, SSIDs,frequenciesand theRPeachbelongsto.Finally,thelasttablecontains theRSSs of thelistedAPsoneachRSobtainedthroughsimulation.

The road topology is extracted from the OpenStreetMap1

database. The OpenStreetMap web interface provides links for

downloading the detailed map of the area selected by the user. This map not only includes the road topology but also a list of Points of Interest (PoI) and a description of the urban topology (e.g.,buildings,parks,pedestrianareas).Theroadtopologyandthe urbantopologyareextractedandstoredintwodifferentfiles.The roadtopologyisconvertedintoasetofRSsandRPsandstoredin theCDB.Inaddition,we obtainthelistoftheAPsandtheir loca-tions froman opendatabaseprovidedbythelocaladministration whichweincludeintheCDB.Thedatabasedcontainsanameand the GPS coordinates for each AP of the municipal network. It is usedtofeedawebsitemappingdifferentfacilitiesprovidedbythe municipality.The completelistofSSIDs wasprovidedby the net-workoperator.Theroadtopology,APlocationandurbantopology areprovidedasaninputforthesimulationframework.

First,we use sumo[25],an urbanmobilitysimulator, tobuild a simple mobility scenario where a vehicle drives along all the RPsoftheCDB.Thisscenarioisexported totheQualnet[26] net-work simulatorusing theVergilius framework [27]. TheRSSs are computedusing corner [24],an urban signal propagationmodel basedonknowledgeoftheroadtopology. corner computessignal attenuationconsidering three situations: (i) theAPis intheLine ofSight (LoS) oftheMS; (ii)the APis out oftheLoS ofthe MS withone-cornerseparation;or(iii)theAPisoutoftheLoSofthe MS withtwo-cornerseparation. corner provides agoodtrade-off betweenaccuracyandcomputational cost.In additionto corner, Qualnet issettoconsiderthe urbantopology inthecomputation ofthesignalpropagation,inordertoprovidemuchmorerealistic RSSsimulations.

A partial visualization of the CDB is shown in Fig. 2. Based onsignal propagationmodeling,the CDBcanprovideinformation

aboutthenumberof APsbroadcastingwithaminimum required

RSS(

−

85 dBm inFig. 2). 3.2. Directiondetectionmodule

The DDM attempts to predict the RP the vehicle is about to enter. This module relies on GPS location updates and the road topologyprovidedbytheCDB.Thisinformationiscriticaltotrigger the handover at theright location, in particular, when the vehi-cleturnsatan intersectionandisaboutto leavethelineofsight of the current AP. This situation often leads to a significant de-crease ofthe RSS of the currentAP, causingpacket loss or even completedisconnection.TheDDMdistinguishesbetweenthecases of a vehicle continues straight on or turning in an intersection. The first case must correspond to only one candidate RP. There-fore,we considerthat thevehicle continuesstraight onfromthe current RP to a candidate RP, if the azimuth difference between

Fig. 2. Example of AP density heatmap based on the CDB.

RPs is lower than 10◦. Otherwise, the candidateRP impliesthat aturnhasbeenmade. Thisthresholdissuesfromtheobservation that intersectionsalways compriseonlyone adjacent RP withan azimuth difference lower than 10◦ which is not the caseif one considers higherthresholds. Thenext direction isfound by com-paringthecurrenttrajectoryofthevehiclewiththecharacteristics thecandidate RPs (i.e.,the RPs starting at thenext intersection). The currenttrajectory is definedby thetwo most recentvehicle locationslt and lt−1, andthe bearingof the vehicle

β

. The next

directiondetectionstartswhenthevehicleentersthedirection de-tection area(starting 15 mbefore theendofcurrentRP).During thisphase, the DDM computesthe angulardisplacement

γ

such that

γ

= |β −

α

|

. The DDM concludes the vehicle has entered anRP when

γ

islower than20◦ andthedistancebetween vehi-cleandthe RP entrypoint isdecreasing (Eq. (1)). This threshold has been empirically obtained during a preliminary study. It re-sultsfromatradeoffbetweentheriskofwrong detectionandthe objectiveofearlydetection.



γ

<

20◦

(

a

)

d

(

lt

,

D

) <

d

(

lt−1

,

D

) (

b

)

(1)

Ifthe DDM has not detected any turning 10 mafter the end ofthecurrentRP,itconsidersthatthevehicleisgoesstraight on. Ononehand,thisrulenegatively impactstheperformance ofthe DDMintermsofdetectionanticipationbutontheother,itallows theDDMtodetectagreaternumberofturningevents.Inaddition, theearlystraightoneventdetectionisconsideredlesscriticalthan the turning eventdetectiondue to thefact that theMS ismore likelytostaylongerinthelineofsightofthecurrentAPwhenthe vehiclecontinuesstraighton.

Note that when the vehicle turns, its bearing often matches the azimuth of the next RP when the vehicle is about to enter toit. Therefore,inorder toincrease the anticipationof the

vehi-cle’s turning, we augment the DDM with a Fuzzy-based Turning

Detector(FTD).

The FTD usesa fuzzy logicsystem toevaluate the probability that the vehicle is initiating a turn. The FTD relies on a simple modelaiming at characterizing a turnbased on the angular dis-placement

γ

.Inourmodel,

γ

ischaracterized byanevolution in two phases(phase I: initialization; phase II: stabilization) as de-picted in Fig. 3. At start both

γ

andthe angular velocity

γ

 are zero,i.e.thevehicle isinthesamedirectionthan thecurrentRP. InphaseI,

γ

 increasesfromzerountilreaching amaximum,the turning phase is initiated. In phase II, the angular velocity

de-Fig. 3. Model of angular displacement during a turn.

creases until there is no more angular displacement.Entering in phaseIIindicatestheinitiationoftrajectorystabilization.

The input variablesofthe fuzzysystemarethevehicle speed, the acceleration(computed asthe speed first derivative),

γ

and theangularvelocity

γ

 asthederivateof

γ

.Inallcases,we con-sider trapezoidal membership functions. The output variable, the turning metric, is a value in the interval

[−

180

;

180

]

that is in-terpreted asthe mostlikely angle ofthe vehicle’s trajectory. We definesevenlinguisticvariablesfortheturningmetric:oneforthe non-turningcaseandthree(low,mediumandhigh)foreach turn-ingdirection(i.e.,negativevaluesimplyturningleftwhilepositive valuesimplyturning right).The FTDconsiders aset ofrulesthat can be summarized asfollows. (i) There is no turn (i.e., turning metricequalszero) if

γ

isloworthevehiclespeedishigh.(ii) If the vehicle speed is low and there is negative acceleration, the turning likelihoodishigh.(iii)Iftheangularrateisnon-zeroand

γ

isincreasing, the turninglikelihood ishigh(phaseI inFig. 3). (iv)Iftheangularrateisnullor

γ

isdecreasing,theturning like-lihood decreases(phaseIIinFig. 3).And finally,(v)Ifthevehicle is driving off (i.e. low speed andhigh acceleration), the turning likelihood increases.Note alsothat theangular rateis usedhere asafiltersinceitindicateswhetherornottheturningmovement keepsgoing(see thefourthandfifthrules).Inordertoselectthe candidateRPsweevaluatetherelativeanglebetweentheazimuth ofthecurrentAPandeachcandidate.Wedefinesevengroups:one fortheRPthatimpliesthevehicleiscontinuingstraighton (rela-tiveanglebelow10◦),andthreegroups(forturningrightandleft) eachforRPsthatarelinkedtoalow,mediumorhighturnfor rel-ativeanglesbetween10◦ and45◦,45◦ and110◦ andhigherthan 110◦ respectively.NotethatthecasewheretwocandidateRPsare classifiedinthesamegroupisnotfrequentandconsideredoutof the scopeofthisstudy. Incontrasttothe previous approach,the FTD isabletoanticipatethevehicle’sdirection.However, theFTD canonlybeusedtodetectturning;thus,theDDMcombinesthese two approaches andconsiders the one providing theearliest de-tection.

3.3. APselectionmodule

Fig. 4. Sample of simulated and sensed RSS.

Fig. 5. The different cases handled by the ASM.

thecurrentRP.It,then,evaluateshandoverstoAPslocatedatthe roadsideofnextRPs andmakes ananticipatedhandoverdecision ifpossible.TheselectedAPsarethenstoredinthecandidateAPlist thatisusedbytheconnectionmodule(cf.Fig. 1).

3.3.1. Handover evaluationandlocalization

ForeachcandidateAP,theASMbuildsahandoverquality met-ricthatisusedintheAPselection.Inaddition,theASMcomputes

the handover location as the location where the RSSs of both

thecurrentand thecandidate APare simultaneouslymaximized.

Fig. 4(a) showsa sample of the simulated andreal RSSs of two APscalledrespectivelyAPi(currentAP)andAPj(candidateAP).

Notethat theCDBonlyprovides themodeledRSStotheASM.

The sensed RSS is only used to evaluate ASM performance. The

handovercomputation consistsof three steps.First, the ASM de-finestheoverlappingarea(OvS),i.e.,thesetofRPswheretheMS

isabletoreceiveasignalfrombothAPiandAPj.

Second,theASMselectsforeachRSthelowestRSS.LetminS

(

P

)

inEq.(2)be thelowest signal functionthat isevaluated overan RSP

∈

OvS.

minS

(

P

)

=

min

(

S

(

APi

)

P

,

S

(

APj

)

P

)

(2)

InFig. 4(a),fortheexampleoftheRS P ,minS

(

P

)

is S

(

APj

)

P.The

set ofall theelements ofthe lowestsignal function isshownby thecurvedelimitingthepattern-filledarea.

Third, theASM selects the handoverlocation asthe RS maxi-mizingEq.(2).

InFig. 4(a)we can seethehandover signalquality QS asthe

maximumof thecurve delimiting thepattern-filledarea and the handoverlocation HS,theRScorrespondingto H SS.

3.3.2. APselectionprocess

anotherAPlocatedinthenextRPs (e.g.,AP2orAP3inFig. 5(d)). NotethatwhenanAPisselected,theASMcomputesthehandover location (HS) asdescribed beforeand provides the CM withthe

3-tuple

{

APfrom

,

APto

,

HS

}

whereAPfromisthecurrentlyselectedAP

andAPto isthenextselectedAP.

Note that since the AP selection process relies on offline

in-formation, the ASM cannot guarantee that the candidate AP is

available and can provide enough bandwidth to the MS. In case

the candidate AP is unavailable, the MS should be able to

at-tempta connectiontoadifferentcandidateAPwithoutextradelay. Therefore,foreachAP, theASM shouldprovidea long-term eval-uationofthe potentialcandidateAPsinthe currentRP andafter. Thatis, the ASM should consider all the possible transitions be-tweenthecurrentAPandnextcandidateAPs(insideandoutside currentRP) andclassify them. In casethe candidate APselected bythe regularAPselectionis unavailable,the MSdirectlychoose thenextAPintheclassifiedcandidateAPslist.Someclassification methodsarediscussedlateronthispaper.

Wecanidentifytwo situationswheretheMSisnotassociated withthebest AP. Thefirst isdueto an incorrectestimate ofthe RSS, leading to situations where the handoverlocation basedon thesimulatedRSS(HS)doesnotcorrespondwiththeoptimal

han-doverlocationinthereality (HR).Thespatialshiftbetweenthose

twohandoverlocationsisShS/R (seeFigs. 4(a)and 4(b)). ShS/R is

considered positiveif HR islocated after HS andnegative

other-wise.Fora givenperiod,theRSSofthe currentAPis lowerthan the RSS ofthe best AP. This RSS difference iscalled



S1−3.The

secondsuboptimalAPselectioncaseisduetotheshort-term pre-diction of vehicle’s direction. As theDDM often detects the next RPattheendofcurrentRPtheASMdelaystheAPselectionwhen thecurrentAPcanprovidegoodsignaluntiltheendofcurrentRP. ThissometimesleadstosuboptimalAPselection.Incase3ofthe illustration(Fig. 5(c)),AP2andAP3providehigherRSSattheend ofthecurrentRP.InthiscasetheAPselectionissuboptimalsince the ASM does not select the best available AP. This situation is showninFigs. 4(a) and 4(b)),wheretheactualhandoverlocation H doesnotcorrespond tobesthandoverlocation HS.The spatial

shiftbetweenthetwoisdenotedasShS andtheRSSdifference,as



S1−2. ShS is consideredalways positive since H couldneverbe

aheadof HS.

3.4. Connectionmodule

TheCMisinchargeoftriggeringanewassociationwiththe se-lectedAPsatthehandoverlocationgivenbytheASM.Thismodule onlytakesasinputtheselectedAPlistandthecurrentvehicle lo-cation without performing any AP scanning. Fig. 1 describesthe connectionprocedurefollowedbytheCM.First,theCMparsesthe selected APlist by comparing theselected AP handover location withthecurrentlocationofthevehicle.Ifthehandoverlocationof aselectedAPisreachedoroutdistanced,theCMstartsthe proce-dureforconnectiontothisAP.Theconnectionprocedure consists oftwo phases. First, since the CM doesnot perform anyscan, it doesnot knowifthe selected APis actuallyavailable or not. In-deed,the fact that theselected APis listed inthe CDB doesnot guaranteethatthisAPisactuallyavailable.Asaresult,inorderto checkthe availabilityoftheAP, theCMsends aProbeRequestto thisAP.IftheCMdoesnotreceiveanyProbeResponseafterthree attempts,theselectedAPisconsideredunavailable.TheASM,then, performsa newAPselectionignoring thisunavailableAP. In con-trast,ifaProbeResponseisreceivedfromtheselectedAP,theCM initiatestheassociationprocess.Thisprocessconsistsofthe stan-dardIEEE802.11authenticationandassociation,i.e.,theexchange offourmessages(authenticationrequest/response andassociation request/response)withtheselectedAP. Ifthisphasefails, theCM retriesa connectiontotheselectedAP.

Fig. 6. Map of the experimental path. 4. Evaluation

4.1. Testbedsetup

In orderto evaluate the performance of coper,we performed a set of experiments using the HotCity network, a metropolitan

IEEE 802.11 network deployed over a large part of the city of

Luxembourg. HotCity provides Internet access to mobile users in multiple areas of the city. It is composed of around 500 Cisco Aironet APs foroutdoor usage, embedding an IEEE 802.11b/g ra-dio forclientwireless accessandan IEEE802.11aradio formesh interconnection betweenAPs. HotCity is an open WiFi that uses Hyper Text Transfer Protocol (HTTP)authentication for registered users wanting to access the network. A detailed connection per-formancestudyofHotCityappears in[28].Notethatthescenario consideredinbuildingtheCDBincludesasubsetof392APsfrom the HotCity network, for which information is available online.2

Fig. 6 showstheexperimental route usedto evaluate coper.This route islocated in Gasperich,a residential neighborhood of Lux-embourgcity.Thisrouteis2.6 kmlongandcomprises18roadside APs alongthepath. Theroadsignalsinclude one traffic lightand multiple stop signs and intersections. However, since the speed limit is low (i.e., 30km/h)on almost the completeexperimental route, theduration ofeachindividual experimentremains almost constant.Theaveragedurationofasingleexperimentis353 swith a standarddeviationof 14 s.Fig. 7 highlightsthe numberofAPs

andthe maximumRSSdetected inall the RSsofthe

experimen-tal path. The distribution of the number of APs detected shows thatthereisnodisconnectedareaontheexperimentalroute.Also, thehighestRSSsensedineachRSoftheexperimentalrouteis al-waysgreaterthan

−

80 dBm.Notethat,asshowninFig. 8,anRSS lowerthan

−

80 dBm significantlydegradestheQoS.Theseresults areconsistentwith[29].Asaresult,inthebestcasetheMSis as-sociated withaHotCityAPandhasInternetconnectivityallalong the experimentalroute. TheMS usedintheseexperimentswas a

Fig. 7. CDF of the number of AP per RS and mean data rate per RSS.

Fig. 8. Relationship between RSS and data rate and dropped packets.

NexcomVTC6200-NI-DK3 _{Linux-basedembeddedplatform}

imple-mentingaKernelv3.11.Theexperimentswereperformedwiththe bulkIEEE802.11b/g/nminiPCIwirelesscardusingRalinkRT3092 chipsetand a Pulse multiband antenna4 _{providing 6 dBi gain in}

the2.4 GHzband.

In order to provide meaningful results, the performance of coper_{iscomparedtoother solutions:(i) networkmanager}5

(net-Man), (ii) a modified version of WPA-supplicant having an RSS

thresholdof

−

75 dBm and (iii) mroad [23]. We consider in this comparative studynetMan and WPA-supplicantin order to eval-uatehowlegacyscanning-basedapproachesperformcomparedto coperand mroad thatarespecificallydesignedforvehicular

com-munications. The comparisonbetween coper and mroad intends

toinvestigatethe eventualimpact ofthe directiondetection

pro-cess on MS communication. The experiments are performed as

follows.Beforeanexperimentstarts,theMSobtainsanIPaddress through DHCP and keepsit forthe whole experiment, since the HotCitynetwork supportshandoversatlayer2.Whenthevehicle isatthedeparture point,theMS initializes a2.5 Mbpsdownlink

User DatagramProtocol (UDP)flow from a remote server in our

laboratory.During theexperiment, wecapturethe received pack-etsusingtcpdump andrecordalogattheMS,includingtheLinux kernellog,theGPS locationofthevehicleandthe network inter-facestatistics(e.g.,RSSofcurrentAP,amountofdatareceivedand transmitted,packetloss).Theexperimentsarerepeatedfivetimes foreach considered solutioninordertoobtain representative av-eragedvalues.

4.2.Evaluationofthedirectiondetection

First, we investigate the DDM performance. The experimental routeforthisexperimentis 6.5 kmlongandcomprises62

inter-3

http://www.nexcom.com/Products/mobile-computing-solutions/in-vehicle-pc/in-vehicle-pc/transportation-computer-vtc-6200.

4 _{http://productfinder.pulseeng.com/productSearch/gpsdm700.} 5 _{https://wiki.gnome.org/Projects/NetworkManager.}

Fig. 9. CDF of anticipation distance.

Table 1

Turndetectionperformanceof DDMand DDMwith FTD.

Median detection Detection<10 m

dmm 12 m 45%

dmm+ ftd −7 m 92%

sections where the vehicle performs 43 turns. This scenario in-cludes multiple types of intersection,implying different mobility patternsforthevehicle(i.e.,differentspeedsandtrajectories).We useda4 HzGPSreceiverinordertohaveafrequentupdateofthe vehicle’strajectoryanddecreasedetectionlatency.Theexperiment was repeated multiple times in order to obtain a representative overviewofthedetectionperformanceoftheDDM.Asmentioned inSection 3.2,the mostcriticalpart ofthedirection detectionis turndetection:whenthevehicle turnsatanintersection,theMS hasmorechancestoleavethelineofsightofthecurrentAP,thus facinga significantdecreaseofcurrentAPRSS.Table 1highlights theperformance interms ofturningdetectionforthestandalone

DDM andthe DDMcombinedwiththe FTD.Weusethe distance

betweenthe detectionlocation andtheendof thecurrentRP as performance metric.Thisdistanceisnegative whenthe detection occursbefore the vehiclereaches theend ofthe currentRP (an-ticipateddetection)andpositiveotherwise.Wealsoevaluated the falsenegativerate,whichcorrespondstothecasewheretheDDM doesnotdetect thatthe vehicleisturningorselectsan incorrect RP,andthefalsepositive casecorrespondsto thecasewherethe systemdetectsaturnbutthevehicleisnotturning.Weobtained sevenfalsepositivesandsevenfalsenegativesagainst129 success-fuldetections(95%)forbothstandaloneDDMandDDMaugmented withFTD.ThissuggeststhattheadditionoftheFTDdoesnothave animpactonthedetectionsuccessoftheDDM.However,FTD pro-videsan enhancement to mediandetectionanticipation of19 m. In addition, 92% ofthe detections occur below 10 mversus 45% for the standalone DDM. If we focus on overall direction detec-tion(straightRPs andturnings),asdepictedinFig. 9,we observe amediandetectionanticipationof10 mwhenDDMisaugmented through FTD(3 m vs.13.2 m). Thisanticipation performance has a significant impact on the quality of AP selection, in particular when the vehicle isabout to turn.An early detection ofthe ve-hicle’s direction allows the ASM to detect when the RSS of the currentAPcould decreaseintheshort-termandtoanticipatethe selectionofthenextAPsuchthatthehandoveristriggeredbefore thecurrentAP’sRSSfallsdramatically.

4.3. EvaluationofAPselection

Fig. 10. CDF of ShSand ShS/R. Table 2 DistributionofS whenS>0. Overall % of RS Median value(dB) Relative%ofRS whereS<5 dB Relative%ofRS RSS<−80 dBm 19.3 2.6 70.1 2.8

directionistakenorwhenthehandoverismisplacedduetoan in-correctRSSestimation.Thehandoverspatialshiftscausedbythese twoissuesareanalyzedinFig. 10,whichshowsthatthehandover location shifts dueto thelack of informationaboutthe vehicle’s directionaremuchshorterthanthosecausedbytheerroneous es-timationoftheRSS.WecanseethatShS isequaltozeroinalmost

79% ofthecases against21% forShS/R. Moreover,ShS is,in

gen-eral,below20 mwhereastheShS/R isratheruniformlydistributed

intherange

[−

40 m

;

40 m

]

.Wecanconcludethatthehandoveris misplacedmainlyduetothedifferencebetweenthesimulatedRSS (storedintheCDB)andactualRSS.ThedistributionoftheRSS dif-ferencebetweenthecurrentandbestAP(seeTable 2)revealsthat theRSS ofthecurrentAPisthe bestavailable inmore than80% oftheRSsoftherun.Themisplacementofhandovercomparedto theideallocationhasa negligibleimpactonthe RSS.Indeed,the medianRSS differencerelative to the bestAPis only 2.8 dBand in70.1% ofthe cases thisvalue is under5 dB.In addition, cases where suboptimal AP selection impliesa critically low RSS (i.e., below

−

80 dBm)arenot frequentlyobserved(i.e.,inonly2.8%of theRSweobserved



S

>

0).Asaresult,weconcludethat corner modeledRSSsarerealisticenoughtobeusedforAPselection.

4.4. ConnectionperformanceoftheMS

In thissection we analyzeseveral metrics relatedto MS con-nectivity during the experiments: the data rateof the downlink UDPflowattheMS,theMSdisconnectiontimeandtheRSSofthe currentAP.

4.4.1. Network performancecomparison

Fig. 11(a) showsthedistribution ofthemeandata rateofthe downlinkUDPflow.Fromthisfigurewecanobservethatthetotal timewheretheMSdoesnotreceiveanydatafromthenetworkis

muchlongerforbothnetManandWPA-supplicantbasedhandover

(around20%)thanfor mroad and coper (lessthan1%).

Table 3showsthat mroad and coper havebothhighassociated time(above99%)andhighpacketdeliveryratio(above95%).Inthe

same way,netMan andWPA-supplicanthaveapproximativelythe

samepacketdeliveryratio(around80%). Notethat thepacket de-liveryratioisanestimatesinceinourexperiments,theMSa flow of2.5 Mbpswhichishigherthanthecapacityofthenetwork.

Fig. 11(b)showsthedistribution ofthe disconnectedtime for all the considered approaches. The medianvalue forthe

discon-nected time is around 1200 ms (maximum 5385 ms) for

WPA-supplicant and1300 ms(maximum 7684 ms) fornetMan versus

around 100 ms (maximum 2277 ms and 1325 ms) for mroad

and coper respectively. This is explained by the fact that WPA-supplicantandnetMan triggerAPscanningateachhandoverand couldpartiallyexplainwhytheMSreceivesnodataduringaround 20%ofthetuneusingthesesolutions(seeFig. 11(a)).Weconclude thatavoidingthescansignificantlyreducesdisconnectionperiods.

Further, the study of the distribution of the RSS (Fig. 11(c)) showsagainalarge differencebetweenthetwosets ofsolutions.

We observedanRSSlower than

−

80 dBm in21%and33%ofthe

casesforWPA-supplicant andnetMan respectively, whileasignal levelthislow wasobservedinonly3.3% and2.8% ofthe casefor mroad _{and coper respectively. Also, the medianobserved RSS is}

around

−

64 dBm for mroad and coper versus

−

67 dBm for

WPA-supplicantand

−

71 dBm fornetMan.Again,thedistributionofthe RSS of mroad and coper aresimilar andindeedcloseto the dis-tributionofthemaximumRSSsensedovertheexperimentalpath (Fig. 7). Fig. 8shows that when the RSS drops below

−

80 dBm, thedataratedecreasessignificantly.

Fig. 11. Comparative results.

Table 3

Comparativeperformanceofdifferenthandoversolutionswithconfidenceintervala_.

Estimatedpacket deliveryratio

Associated time NumberofAP associationperrun

DistinctAPassociated (outsideroadside) Meanconnected time NetMan 0.79±0.05 88.2±0.68 15.4±1.17 15 (3) 23.79±13.25 WPA sup−75 0.80±0.02 78.8±0.47 39.7±3.12 25 (6) 7.15±0.87 mroad 0.97±0.03 99.06±0.17 19±0.00 18 (0) 15.25±1.43 coper 0.95±0.02 99.08±0.14 19±0.00 18 (0) 17.17±1.33

In contrast, the distributions for netMan and WPA-supplicant differsignificantly.NetManismoreaffectedbylowRSSthan WPA-supplicantbutthisseems tohavenoimpactonthedataratenor on the packet delivery ratio. We observe that the RSS

distribu-tion for valuesgreater than

−

65 dBm for coper and mroad are

notequal.This couldbe because APselectionfor coper is some-timesdelayedcomparedto mroad (cf.Section 4.3).Nevertheless, thishasnosignificantimpactonthedataratebecause,duetoISP bandwidthcapping,alltheRSSabove

−

77 dBm provideequivalent datarates(cf.Fig. 8).

Nowwe considerthe associatedtime, thenumberof success-fulassociations, themeanassociated time showninTable 3.The

associated time starts when the MS associates with an AP and

ends at the beginning of the scan that precedes the next

han-dover. These results show that, using mroad and coper, the MS is almost always associated withan AP (99%of the tune) while thisisthecaseforonly88.2%and78.8%ofthetimewithnetMan andWPA-supplicantrespectively. We observe a significant differ-enceof the meanconnected time with thesetwo solutions. The

MS remains connected with an AP on average 23.79 s for

Net-Manand7.15 sforWPA-supplicantversus15.25 sand17.17 sfor mroad_{and coper respectively. InthecaseofWPA-supplicant,the} higherre-connectionfrequency leads toshorterconnection dura-tion,whichincreasestheinstabilityoftheMSconnectioncontrary tonetManwhichisevenmoredurablyconnectedthan mroad and coper.WeobservedinFig. 11(b)thatWPA-supplicantcauseslong disconnectionduetoscanning,thus,thissolutionismoreimpacted bydisconnectionsthannetManbecauseittriggershandoversmuch moreoften.HerewerealizethatbothnetManandWPA-supplicant areaffectedbytwodifferentfactors,poorlinkqualityduringlong periods for the first andfrequent disconnections for the second, that seemsto havea similar impact asboth havethe samedata ratedistributionandpacketdeliveryratio.Theseresultsalsoshow that mroad and coper allowslongandstableassociations(the se-quenceofselectedAPsremainsthesameovertheruns).Wenote instability ofthe APselection beingbasedonly on instantaneous scanningresults,fromwhichbothnetManandWPA-supplicant se-lectbetweenthreeandeightAPsoutsidetherouteofthevehicle, whichareunabletoprovideasustainableRSStotheMS.Also,the APslocatedattheroadside arenotalways selectedintheirorder ofappearance.Incontrast, mroad and coper selectonlythe18APs providingthebestsignalallalongthepath.

Notethat coper outperformsMROADintermsofdisconnection time. Three factors can be cited to explain this slight difference ofperformance.The firstis thehandoverspatial shifts causedby the lackof information aboutthe vehicle’s direction, highlighted inprevioussection.Thismayalsobeduetotheexperimental

con-ditions not being exactly the same when evaluating mroad and

coperandbecause theperformance of agiven solutioncan vary slightlyfromonerunto theother.Nevertheless,wecanconclude thatby using coper asapredictive handovermechanism,we can providesimilar performance to mroad butwithoutrequiring any priorknowledgeofthecompletepathofthevehicle,whichgreatly increasesflexibility.

4.4.2. Impact oftheAPselectionandhandoverprocess

Finally, we study the impact of the AP transition process on MS connectivity.We distinguish betweenthe APselection phase,

which involves the offline computation to find the best AP, and

thehandoverexecution,wheretheMSprobestheselectedAPand associates to it. Fig. 12 illustrates the duration of the these two phases.

Thisoffline computation impliesthat the APselection is per-formedby theASM andthe candidateAPlistparsing bythe CM beforeattemptingahandover.ThemediandurationoftheASMAP selectionandthecandidateAPlistparsingare340 msand27 ms

Fig. 12. CDF of the handover phases duration.

respectively.Thisisa longprocesscompared tothehandover

ex-ecution, that hasa medianof12 msforAP probingand100 ms

forauthenticationandassociation.However,APselectionin coper is performedoffline,preventing a disconnectionbetweenthe MS andthecurrentAP. Infact,thedurationoftheAPselectionphase reflects thereactiontime of coper whena newRPis discovered, thus,intheworstcase,APselectiononlyslightlydelayshandover. Duringthe probingphase in coper,the MS sends abroadcast Probe Request on the channel where the selected AP is operat-ingandwaitsforProbeResponsesfromallAPsinthischannelfor a fixed time whilein thestandard handover, theMS repeats the sameprocess onall channels.Thisresultsina muchshorter me-dian duration for coper (98 ms) than for the standard handover usedbynetMan(1143 ms).NotethattheAPprobingduration dis-tributionsshowsthreemodesthatcorrespondto(i)thecasewhere the MS retrieves the selected AP in recently-stored scan results (

∼

20 ms),(ii)theMSreceivedaProbeResponsefromtheselected APafterthefirstattempt(

∼

100 ms) and(iii)the MSneeds sev-eralattemptstoprobetheselectedAP(APprobingdurationgreater than200 ms).

The handover phase of coper uses probing and association

methods available in the existing Linux kernel module, which

could be optimized.Forinstance, theprobing could be improved byspecificallytargetingtheSSIDoftheselectedAP,sending multi-pleprobeswithinashorttimeintervalortimingoutsoonerwhile waitingforProbeResponses.

5. Discussion

Further,theever-growingnumberofresidentialandcommunity networksofferswidespreadcoverageinurbanandsuburban envi-ronments.Atthesametime,cellularnetworkshavebeendeployed over a great part of the populated regions of the world. coper couldbe extendedtoall theseheterogeneousnetworks. However, a highnumberofAPsandfrequentchangesinnetwork infrastruc-turedeploymentsleads toopenquestionsconcerning information collectiononthesenetworksandthedynamicevolutionoftheCDB tokeepitup-to-date.

Asaresult,solutionsareneededtoenablethedynamic collec-tionandincorporationofavailablenetworkandmobility informa-tion andin the database. A collaborativeapproach based on the actual experience of usersis one possible strategy. Here, the MS periodicallycollects contextualinformationon thenetworks(e.g., RSS,packetloss,datarate)throughaMobileCrowdSourcing(MCS) application[32].Thecollectedinformation,storedandanalyzedin thecloud, could be usedto detect out-of-orderAPs,improvethe modelusedbythesimulationframework,andtobuildatemporal profileofeachAPthatcouldbeconsideredintheAPselection.

In order to improve coper reliability,the AP selection should focusonthelong-termevolutionofthenetworkenvironment. To thisaim,an APclassificationmechanismhasbeenpreviously de-scribedthatintendstoallowtheMSselectingwithnodelayanew candidateAPincasetheselectedoneisnotavailable.Considering onlytheinformation storedin theCDB,theclassification maybe performedbasedonthehandoverqualityparameter.Inaddition,it couldbe improvedbyintegratingalong-termmobilityprediction basedontheanalysisofthehistoricaltracescollected,forinstance, by the mean of the MCSapplication mentioned before. This ap-proachwouldallowtheMStomakeapre-selectionoftheAPthat willbeprobablyvisited.

6. Conclusion

Inordertoprovidean ubiquitousnetworkconnectionthrough 802.11networkstomobilevehicles,thestandardIEEE802.11

han-dover procedure needs to be amended withthe aim ofavoiding

longdisconnectionsandpotentiallyinconsistentAPselections. Al-though it has been proven that the disconnected time resulting fromscanningcanbesignificantlyreduced,thelackofinformation ontheshort-termevolutionoftheAPs’RSSssuggeststhatotherof sourcesofinformationshouldbeconsidered inordertoprovidea moreaccurateviewofnetworkevolutionovertheshortterm.

In thiswork we havepresented coper, a predictive handover mechanism that replaces the scanning procedure by using prior knowledgeofthe roadtopologyandthemodeled APsRSS.Based on thisinformationand knowledgeof the vehicle’spredicted di-rection, coper allowstheMStoselectbestnextAPsinthelightof theshort-termevolutionoftheirRSSs.Thisapproachshortensthe handover to a single channel probing and the authentication/as-sociationphase,whichsignificantlyreducesthehandoverlatency.

Our field experiments conductedon a city-wide 802.11 network

show that this handover optimization materially decreases the

time when the MS is disconnected. The average data rate

us-ing coper remainscloseto themaximumachievabledatarateon nearlythewholedurationoftheexperiments.

Wehaveshowedthat coper cansignificantlyimprovethe net-workconnectivityofavehicleconnectedtolarge802.11networks, whicharecharacterizedbyauniformAPdensitycomparableto fu-ture802.11pdeployments.Howeverfurtherinvestigationisneeded toevaluatesuchasolutiononheterogeneous802.11networks(e.g., residential and community 802.11 networks) that have different characteristics(e.g.,highnumberofAPs,shortercoverage,frequent topologychanges).Inaddition,someresearchshouldbedoneinto thebestwaytorunsuchexperimentsonalargerscale.

Acknowledgements

The authorswouldliketothanktheLuxembourg National

Re-search Fund (FNR) for providing financial support through the

CORE2010MOVEproject(C10/IS/786097).Wealsowanttothank

IsmailYildizforhiscontribution.

References

[1]Sang-HeePark,Hye-SooKim,Chun-SuPark,Jae-WonKim,Sung-JeaKo, Selec-tivechannelscanningforfasthandoffinwirelessLANusingneighborgraph, in:PersonalWirelessCommunications,Springer,2004,pp. 194–203. [2]AruneshMishra,MinhoShin,WilliamArbaugh,Anempirical analysisofthe

IEEE802.11MAClayerhandoffprocess,Comput.Commun.Rev.33 (2)(2003) 93–102.

[3]GermanCastignani,AndrésArcia,NicolasMontavont,A studyofthediscovery processin802.11networks,Mob.Comput.Commun.Rev.15 (1)(2011)25–36. [4]IEEEStandardforInformationtechnology–localandmetropolitanarea net-works–specificrequirements–Part11:WirelessLANMediumAccessControl (MAC)andPhysicalLayer(PHY)specificationsamendment6:wirelessaccess invehicularenvironments,in:IEEEStd802.11p-2010,July2010,pp. 1–51. [5]EvaGustafsson,AnnikaJonsson,Alwaysbestconnected,IEEEWirel.Commun.

10 (1)(2003)49–55.

[6]JulienMontavont,NicolasMontavont,ThomasNoel,EnhancedschemesforL2 handoverinIEEE802.11networksandtheirevaluations,in:IEEE16th Inter-nationalSymposiumonPersonal,IndoorandMobileRadioCommunications, vol. 3,PIMRC2005,IEEE,2005,pp. 1429–1434.

[7]IshwarRamani,StefanSavage,SyncScan:practicalfasthandofffor802.11 in-frastructurenetworks,in:24thAnnualJointConferenceoftheIEEEComputer and CommunicationsSocieties,INFOCOM 2005, in: Proc. IEEE, vol. 1, IEEE, 2005,pp. 675–684.

[8]KishoreRamachandran,SampathRangarajan,JohnC.Lin,Make-before-break MACLayerHandoff,in:IEEEInternationalConferenceon802.11Wireless Net-worksCommunications,ICC’06,vol. 10,IEEE,2006,pp. 4818–4823.

[9]VladimirBrik,AruneshMishra,SumanBanerjee,Eliminatinghandofflatencies in802.11WLANsusingmultipleradios:applications,experience,and evalua-tion,in:Proceedingsofthe5thACMSIGCOMMConferenceonInternet Mea-surement,USENIXAssociation,2005,p. 27.

[10]MaximilienMouton,GermanCastignani,RaphaëlFrank,LaraCodeca,Thomas Engel,Ontheevaluationofmake-before-breakhandoversinurbanWiFi net-works for moving vehicles, in: 10th Annual Conference on Wireless On-demandNetworkSystemsandServices,WONS2013,IEEE,2013,pp. 170–177. [11]GermanCastignani,AlejandroLampropulos,AlbertoBlanc,NicolasMontavont,

Wi2me:a mobilesensingplatformforwirelessheterogeneousnetworks,in: 32ndInternationalConferenceonDistributedComputingSystemsWorkshops, ICDCSW2012,IEEE,2012,pp. 108–113.

[12]AndrésArcia-Moret, Laudin Molina,Nicolas Montavont, German Castignani, A. Blanc,Accesspointdiscoveryin802.11networks,in:6thInternational Con-ferenceonNetworkGames,ControlandOptimization,NetGCooP2012,Nov. 2014.

[13]Vivek Mhatre,KonstantinaPapagiannaki,Using smarttriggersfor improved userperformancein802.11wirelessnetworks,in:Proceedingsofthe4th Inter-nationalConferenceonMobileSystems,ApplicationsandServices,ACM,2006, pp. 246–259.

[14]AliSafaSadiq,KamalrulnizamAbuBakar,KayhanZrarGhafoor,JaimeLloret, SeyedAliMirjalili,A smarthandoverpredictionsystembasedoncurvefitting modelforFastMobileIPv6inwirelessnetworks,Int.J.Commun.Syst.27 (7) (2014)969–990.

[15]Jungwook Choi, Hyukjoon Lee, Supporting handover in an IEEE 802.11 p-basedwirelessaccesssystem,in:ProceedingsoftheSeventhACMInternational WorkshoponVehiculArInterNETworking,ACM,2010,pp. 75–80.

[16]Young-uk Chung, HyukjoonLee, Yong-Hoon Choi, Chungwon Lee, Proactive cachingand forwardingschemesforseamless handover inIEEE WAVE net-works,Int.J.Distrib.Sens.Netw.2013(2013)1–16.

[17]PralhadDeshpande,AnandKashyap,ChulSung,Samir R. Das,Predictive meth-odsforimproved vehicularWiFi access, in:Proceedingsofthe7th Interna-tionalConferenceonMobileSystems,Applications,andServices,ACM,2009, pp. 263–276.

[18]DaehanKwak,JeonghoonMo,MoonsooKang,InvestigationofhandoffsforIEEE 802.11networksinvehicularenvironment,in:FirstInternationalConference onUbiquitousandFutureNetworks,ICUFN2009,IEEE,2009,pp. 89–94. [19]JulienMontavont,ThomasNoel,IEEE802.11handoversassistedbyGPS

infor-mation,in:IEEEInternationalConferenceonWirelessandMobileComputing, NetworkingandCommunications,2006,WiMob2006,IEEE,2006,pp. 166–172. [20]DaqiangZhang,MinChen,MohsenGuizani,HaoyiXiong,Mobilitypredictionin telecomcloudusingmobilecalls,IEEEWirel.Commun.21 (1)(2014)26–32. [21]HaithamAbu-Ghazaleh,AttahiruSuleAlfa,Applicationofmobilityprediction

[22]FlorianoDe Rango,PeppinoFazio,SalvatoreMarano,Utility-basedpredictive servicesfor adaptivewireless networkswithmobile hosts,IEEE Trans.Veh. Technol.58 (3)(2009)1415–1428.

[23]MaximilienMouton,GermanCastignani,RaphaëlFrank,ThomasEngel,Mroad: scanning-freemodel-basedroamingfor802.11networksundervehicular mo-bility,in:WirelessDays(WD),IFIP2013,IEEE,2013,pp. 1–8.

[24]EugenioGiordano,RaphaelFrank,GiovanniPau,MarioGerla,Corner:aradio propagationmodelfor VANETs inurbanscenarios, Proc.IEEE99 (7)(2011) 1280–1294.

[25] Simulation ofurban mobility, [online]accessed January 2015, http://sumo. sourceforge.net/.

[26] Qualnet network simulator, [online] accessed January 2015, http://www. scalable-networks.com.

[27]EugenioGiordano,EnzoDeSena,GiovanniPau,MarioGerla,Vergilius:a sce-nariogeneratorforVANET,in:IEEE71stVehicularTechnologyConference,VTC 2010-Spring,IEEE,2010,pp. 1–5.

[28]German Castignani, Juan Monetti, Nicolas Montavont, Andrés Arcia-Moret, RaphaëlFrank,ThomasEngel,A studyofurbanIEEE802.11hotspotnetworks: towardsacommunityaccessnetwork,in:WirelessDays(WD),2013IFIP,IEEE, 2013,pp. 1–8.

[29]GermanCastignani,AlbertoBlanc,AlejandroLampropulos,NicolasMontavont, Urban802.11communitynetworksformobileusers:currentdeploymentsand prospectives,Mob.Netw.Appl.17 (6)(2012)796–807.

[30]JavierGozálvez,MiguelSepulcre,RamonBauza,IEEE802.11pvehicleto infras-tructurecommunicationsinurbanenvironments,IEEECommun.Mag.50 (5) (2012)176–183.

[31]K. PaulYoon,Ching-LaiHwang,MultipleAttributeDecisionMaking:An Intro-duction,SagePublications,1995.