applications A Priority based Cross Layer Routing Protocol for healthcare Ad Hoc Networks

(1)

ContentslistsavailableatScienceDirect

Ad Hoc Networks

journalhomepage:www.elsevier.com/locate/adhoc

A Priority based Cross Layer Routing Protocol for healthcare applications

Hadda Ben Elhadj

^a^,^∗

, Jocelyne Elias

^b

, Lamia Chaari

^a

, Lotﬁ Kamoun

^a

aLETI Laboratory, Sfax University, Tunisia

bLIPADE Laboratory, Université Paris Descartes – Sorbonne Paris Cité, 75006 Paris, France

a r t i c l e i n f o

Article history:

Received 28 February 2015 Revised 9 October 2015 Accepted 20 October 2015 Available online 2 November 2015 Keywords:

Wireless Body Area Networks Healthcare

QoS MAC Cross layer Routing

a b s t r a c t

Wirelessbodyareanetworks(WBANs)representoneofthemostpromisingapproachesfor improvingthequalityoflife,allowingremotepatientmonitoringandotherhealthcareap- plications.DatadisseminationandmediumaccessinaWBANarecriticalissuesthatimpact thenetworkreliability,theeﬃciencyandthetotalenergyconsumedbythenetwork.Inthis paper,weproposeaPriority-basedCrossLayerRoutingProtocol(PCLRP)alongwithaPriority CrossLayerMediumAccessChannelprotocol(PCLMAC)forhealthcareapplications.

PCLRPcombinedwithPCLMACensuresreliabletraﬃcdisseminationandcustomizedchannel accessforintra-andinter-bodycommunications.Simulationresultsshowthattheproposed protocolachievescustomizedqualityofservicesandoutperformsstateoftheartexistingpro- tocolsintermsofpowerconsumption,packetdeliveryratioanddelay.

1. Introduction

Theincreaseinaveragelifespanandhealthcostsalong withtheadvancesinminiaturizationofelectronicdevices, sensing,batteryandwirelesscommunicationtechnologies haveledtothedevelopmentofwirelessbodyareanetworks (WBANs).Inthe healthfield,a WBANconsistsofa setof medicalsensors(i.e.,ECG,EEG)andacoordinator(personal digitalassistant(PDA)orasmartphone)implantedinoron theuser’sbody[1–4].Thesedevicesaimtocollect,storeand processpatient’sphysiologicalparametersandprovidehim ubiquitoushealthcareservices.Duetotheirspecificproper- tiessuchassmallsize,datarate,reliability,security,mobility,powerconstraint,QoSrequirements,andheterogeneous traffic,WBANsrequirespecialprotocolsdesigntomeettheir particularneeds. In other words, although WBANsderive

∗ Corresponding author. Tel.: +21625369105.

E-mail addresses: [email protected] (H. Ben Elhadj), [email protected] (J. Elias), [email protected]

(L. Chaari), [email protected] (L. Kamoun).

somehowfromWSNs,thereareintrinsicdifferencesbetween thesetwonetworks(whicharesummarizedinTable1).

EversinceWBANshaveemerged,differentoptimization schemeshavebeenproposed toovercometheabovechal- lengingissues.

Cross-layer approaches have proven to provide better WBANoptimizationresultsthantheirlayeredcounterparts [7].Indeed,layercooperationincross-layerbasedschemes wellenhancestheoverallWBANperformance.Forinstance, inacross-layerscheme,theQoSrequirementsattheapplica- tionlayercanbecommunicatedtotheMAClayerinorderto achievebetterresourceallocationfortherunninghealthcare application.Furthermore,thechannelstateinformationand batterylevelcanbefedtothenetworklayertoavoidpaths includingchannelsinabadstateordepletednodes.

The great number of proposed WBAN cross-layer approaches (reviewedin Section2) proves thatthere isstill aneedforfurtheroptimizationofsuchnetworks,andthat cross-layeringiseﬃcienttoaccomplishthat.Fromthispoint of view, this paper presentsa Priority based Cross Layer RoutingProtocolforhealthcareapplications,namedPCLRP.

PCLRPisanadaptiveprotocolinthesenseofslotassignment http://dx.doi.org/10.1016/j.adhoc.2015.10.007

(2)

Table 1

Comparison between WSNs and WBANs [5,6] .

Challenges WSN WBAN

Scale Monitored environment (m/km) Human body (cm/m) Node number Many redundant nodes for wide

area coverage

Fewer, limited in space Node tasks A node performs a dedicated task A node performs multiple tasks Node size Preferred small, but not important To be small is essential Network topology Very likely to be ﬁxed and static More variable due to body

movement

Data rates Homogeneous Heterogeneous

Node replacement Performed easily, nodes may be even disposable

Diﬃcult (implanted nodes) Node lifetime Several months/years Several months/years Path loss medium (free space) important Energy scavenging

source

Most likely solar and wind power Most likely motion (vibration) and thermal (body heat) Biocompatibility Not a consideration in most

applications A must for implants and on-body sensors

Security level Lower Higher, to protect data of patient Impact of data loss May be compensated by

redundant nodes

More signiﬁcant, may require additional measures to ensure QoS and real-time data delivery Wireless technology Bluetooth, Zigbee, GPRS, WLAN, ... Low power technology (i.e.,

Bluetooth low energy)

techniques,sleepandwakeupmechanismsinfaceoftopol- ogychanges.Moreover,itcombinesTDMAandpriorityguar- anteedCSMA/CAapproachestoaccessthechannelandwell definesa synchronizationschemeto avoidcollisions,data lossandidlelistening.Furthermore,PCLRPhandlesWBAN trafficheterogeneitybydefiningthreetrafficclasses:

• GeneralMonitoringpacketsforordinaryMedicaldata;

• DelaySensitivepacketsforHigh-prioritymedicaldata;

• Emergencypacketsforcriticalmedicaldata.

PCLRPfurtherensuresresourceallocationandroutese- lectionincompliancewiththeheterogeneousQoSrequire- mentsofsuchtrafficclasses.Infact,asakeyinnovativefea- ture, inthisworkwe investigatethechannel accessissue bothforintraandinterbodycommunicationswithaclear differentiationbetweenmultipletraffic typeswithrespect totheirQoSrequirements.Thispaperisasignificantexten- sionofexistingWBANMACprotocolsinwhichtheTDMA slotsallocationisrestrictedtointrabodynodes.Morespecif- ically,toensuremorereliabilityandcollisionavoidance,the PCLMACsuperframecontainsaContentionFreePeriod(CFP) customizedforinterWBANcooperation.

In summary, our paper makes the following key contributions:

• Wedefineasetofhealthcaremonitoringapplications(or trafficcategories)torepresentgeneralmonitoringtraffic data,highpriorityandemergencydata.

• Togivemeaningtothetrafficclassification,wepropose aPriorityCrossLayerMediumAccessChannelprotocol, PCLMAC,whichoperatesincompliancewiththedefined trafficcategories.

• Wefurtherproposeanintra-bodyandextra-bodyrout- ingprotocolsthatoperateincooperationwiththedeﬁned PCLMACprotocol.

• We perform a thorough performance comparison between our proposed approach and the Wireless Au- tonomousSpanningtreeProtocol(WASP)formulti-hop

wirelessbodyareanetworks[8]andDatacentricMulti objectiveQoS-awareroutingprotocol(DMQoS)forbody sensornetworks[9].NumericalresultsshowthatPCLRP isindeedeffective,sinceitsigniﬁcantlysavesenergyand ensureshighreliability.

Thepaperisstructuredasfollows:Section2discussesre- latedwork.Section3introducesourWBANnetworkmodel andtraﬃc categories.In Section 4 we presentour PCLRP approach,whileweillustrateanddiscussnumericalresults that show the eﬃciencyofour proposalin Section 5.Fi- nally,Section6concludesthispaperandpresentssomefu- tureworks.

2. Relatedwork

Severalworkshaveappearedintheliteraturewiththe purposeofensuringefficientroutingandenhancingtheQoS ofWSNs[10,11].Nevertheless,asmentionedinTable1the specificityof theoperating environment andtreateddata makeWBANsuniqueandrequirespecificprotocoldesign.

Inbrief,WSNsprotocolswillnotworkaseﬃcientlyasthe protocolsspeciﬁcallydesignedforWBANs.In thissection, wesurveysomerelevantonesthataretightlyrelatedtoour work.

WBAN cross-layer protocol design is an emergent re- search area that aims to deliver greater eﬃciencies than singlelayeradaptationschemes[7].Wehighlightthepro- posedcross-layerroutingprotocolsforintegrationinWBAN systems.

Generally, cross-layer schemes may be either loosely coupledortightly coupleddesigned.Looselycoupledpro- tocol designs focus on communicating the lower layers available parameters to upper layers and/or coupling the functionalities ofsome adjacentlayers in orderto ensure overall network performance. Accordingly, in the loosely coupledapproachtheindividuallayerswithintheprotocol

(3)

stackremain,butparametersexchangeisnotlimitedtoad- jacentlayers.Inthetightlycoupledapproach,differentlayers areoptimizedtogethertoformonecompleteoptimizedsolu- tion.Thislatterapproachmaybettergetridofstackcommu- nicationoverheadthanthelooselycoupledone,butthismay beattheexpenseofprotocoltransparencyandmaintenance.

Consequently,themostcommonlyproposed WBANcross- layerprotocolsarelooselycoupledschemes[12].Authorsin [13]haveproposedacross-layersolutionforcriticaldatade- liveryapplicationsthatcombinesAPTEEN[14]andGinMAC [15]protocolswithnewfeatures,suchasmulti-hopsandmo- bilitymodules.Hence,theclusterheadsselectiontechnique usedin[13]isthatofAPTEEN.

APTEENoperatesintwophases,clustersetupphaseand datatransferphase.Intheclustersetupphase,clusterheads electionandclustermemberformationaredoneusingthe samealgorithmusedinLEACH[17].Then,eachclusterhead broadcastsanadvertisementmessagetotheentirenetwork includingaHardThreshold(HT)andaSoftThreshold(ST) attributes.Thatis,datatransmissionstakeplaceonlywhen actualsenseddataisgreaterorequaltoHTorchangesby anamountgreaterorequaltoSTcomparedtotheprevious sentvalue.Mostoperationsthatneedtobemadein[13]are takenfromAPTEEN,suchasselectingforwardingroutesand connection,andmobilitymanagement.Therewith,GinMAC followstheinformationgivenbyAPTEENandthenjustcon- ﬁrmsthatdataisforwardedtoanexthopoverasinglehop communication.It usestheTDMA protocoltoensurethat dataisdeliveredtothesinkinatimelyandreliableman- ner.Infact,aTDMAscheduleforeachclusterisdetermined.

Moreover,clusters,formedbyAPTEEN,operateondifferent transmissionfrequenciestoaccommodatealargenumberof nodes.

Althoughtheprotocolensuresmobilityawareness,itsuf- fersfromcomplexityofclusterconstructionatdifferentlev- elsandoverlookingtraﬃcdifferentiation.

Ruzzellietal.[16]andLatreetal.[18]incorporateaclosely coupledinteractionbetweentheMACandthenetworklay- ers.It isa mechanismwhereintheMACslotallocation is customizedfortheunderlyingroutingtree,therebyprovid- ingrouting-speciﬁcenergyeconomyattheMAClayer.Also, theseproposalshandlebodymobilitybyadaptivelyrecon- structingandmaintainingthetreetopologyusedforpacket routing.However,traﬃcdifferentiationandschedulingare overlooked.

WASP[8]isaslottedspanningtreebasedcross-layerpro- tocol.WASPtimeaxisisdividedintocontentionfreeandcon- tentionbasedslots,groupedincycles,whichareassignedto nodesina distributedway.InaWASP-cycle,eachnode is allowedtosenditsdataand/ortoforwarddatareceivedin thepreviouscycletothenextnode.Atthebeginningofeach cycle,thesinkbroadcastsaschememessagecalledWASP- schemetoinformitschildrenwhentheycansendtheirdata.

Thesechildren respondby sending out their own WASP- schemeintheir designatedtimeslots[8].Whereas,WASP overlookstraﬃcheterogeneity,nodemobilityandnodesyn- chronization.

Although, authors in [19,20] have treated WBAN heterogeneity,these proposals,which are based on the IEEE 802.15.6standard,sufferfromhighenergyconsumptiondue toidlelistening.

DMQoS [9]isamodularQoSbasedprotocol. Itsetsup a modulararchitecture whereindifferentmodules coordi- nate witheach other toprovide QoSpreferment services.

DMQoSclassiﬁesdatapacketsintofourmaincategories:or- dinary,critical,delaydrivenandreliabilitydrivenpackets.

Theroutingsofdelaycriticalandreliabilitycriticalpackets arehandledseparatelybyemployingindependentmodules foreach,whereasforthemostcriticalpacketshavingboth stringentdelayandreliabilityconstraints,thecorrespond- ing modules operatein coordination toguarantee there- quiredservice.Whilein[9]thedelaycontrolmodulechooses the next-hop router node offering highervelocity ofdata packets, thereliabilitycontrol moduleinjects minimalre- dundantinformationbyexploitinghighreliabilitylinks.To ensurereliability,DMQoSduplicatestransmitteddatapack- ets, which leads to energy consumption and interference increase.

In order to mitigate packet losses, Gaudadri and Ba- heti[21]proposeacross-layerschemebasedonapplication andMAClayerinteractions.Theyfocusontreating theis- suesofend-to-endpacketlossesduetolinkdeterioration, interference, congestionandsystem load. The packetloss mitigation approachis basedontheemerging signal pro- cessing concept,the compressed sensing,wherein signifi- cantlyfewsensormeasurementscanbeusedtorecoversig- nalswitharbitrarilyfineresolution.Lostpacketsareiden- tifiedattheapplicationlayerviaasequencenumberfield in the packet header of the lower layers. However, au- thorshavenotconsideredneithertrafficheterogeneitynor node synchronization. The main characteristics of WBAN cross-layerprotocols,previouslyreviewed,aresummarized inTable2.

Despitethefactthatmanyoftheseproposedtechniques look promising,there are still manychallengesthat need to be solved. From thispoint of view,we propose inthe next sectionacross-layerprotocolthattakesadvantageof strengthsandcopewithweaknessesofpreviouscitedpro- posals. In other terms,a protocolthat ensures energyef- ﬁciency, traﬃc differentiation at the channel access and routing levels,reliability, nodesynchronizationandmobil- ityawareness.Thismaybeachievedviadifferentlayercoop- erationthatlimitsidlelistening,overhearing,collisionsand interference.

3. Networkmodelandtraﬃccategories

Thisworkfocusesonremotehealthcaremonitoringsys- tems,whichareoneofthemostpromisinghealthcareappli- cations.Inthissection,weintroduceourhealthcarenetwork modelaswellastheproposedtraﬃccategorizationinthis latterdomain.

3.1. Networkmodel

Inournetworkmodel,weconsiderasetofmobileper- sonal WBANs(e-healthusers) denotedby P anda set of WiFi gateways(G)^.^Each ^personal ^WBAN (^p∈P) ^aims ^to

ubiquitouslydisseminateits collecteddatatothee-health staff (doctor,nurse,emergencycar,andsoon.)viaagate- way(^g∈G)^.^Typically,êach^pîs^composedôfâ^setôf^body

implantandwearablesensorscalledchildren,andacentral

(4)

Table 2

Characteristics of WBAN cross-layer protocol proposals.

Protocol Energy eﬃciency Traﬃc heterogeneity Channel access scheme Node synchronization Mobility

[13] Low Yes Yes No Yes

TICOSS [16] Good No No No Yes

CICADA [18] Medium No Yes No Yes

WASP [8] Good No Yes No No

DMQoS [9] Medium Yes No No Yes

[19] Medium Yes No No Yes

NME and HME [20] Medium Yes No No Yes

[21] Good No No No No

deviceknownastheCoordinator(C)∗∗whichmaybeasmart phoneora personal digitalassistant (PDA). Inp,children nodessendtheircollecteddatatoC.Infact,Ciscomputa- tionallymorepowerful(intermsofenergy,communication range,memory)thanitssensorsandbehavesasarouterin p.Itisresponsiblefordisseminatingthedatacollectedbyits childrentothemedicalstaff,andpreciselytothesuitableg asaﬁrstdestination.

C isequippedwithmultiple radio interfaces(e.g.,Zig- Beeforcommunicationwithsensors,WiFi, 3G...for Inter- net connection) to allow multiple overlapping transmis- sions[22].Inotherwords,asitsupportsmulti-radiocom- munications,CallowstheWBANtosendandreceivedata simultaneously.

Hence,wemayinferthateachWBANpmaybeconsid- eredasaclusterwhereinCisbydefaultitsclusterheadand itsclustermembersarethesetofitschildrennodes.Con- ventionally,clusteringprotocolsdistinguishthemselves by howtheyelectclusterheads.However,inapersonalWBAN thereisno electionprocessandusuallyC ispre-designed asaclusterhead.Notethatinsomecasesthecommunica- tionbetweenCandgatewaygcannotbeonehopandneeds topassthroughanotherCofaneighbourp.Consequently, eachCmaybebotharelayedand/orarelayingcoordinator.

Wenote by arelayedC, acoordinatorthat communicates withag throughanother C,whilea relayingC isaCoor- dinator throughwhichanother Ccommunicateswithag. Inview ofthis, thecommunicationinournetworkmodel is classiﬁed inintra-body communicationand extra-body communication.Theextra-bodycommunicationinvolvesthe inter-WBANcommunicationaswellasthecommunication betweenthecoordinators andgateways.While, theintra- bodyone referstothecommunicationbetweenp’ssensor nodesandC.Moreover,intra-bodycooperationisrequiredin somecases.Therefore,tocommunicatewiththeirC,some sensors requirethe relaying serviceoftheir neighbouring sensorsbelongingtothesamep.Wenotethatasensornode cannotcommunicatewithanycoordinatororsensorbelong- ingtoanotherp.Torecapitulate,ournetworkmodelismulti- hopclusterbased.Fig.1representsanoverviewofourde- scribednetworkmodel.

3.2. Traﬃccategoriesandpacketclassiﬁcation

UnlikeconventionalWSNs,eachsensornodeinaWBAN hasitsownrequirementsintermsofdelayanddatarate[23]. Table 3 shows the heterogeneous characteristics of some commonlyusedmedicalsensors.

Table 3

Delay and bit rate requirements of healthcare data [23] .

Data source Bit rate (bps) Delay (s) Sampling rate (Hz) Electrocardiogram 10 –100 k < 10 63 –500 Blood pressure 10 –30 > 120 63 Non-invasive cuff 0.05 30 –120 0.025 Cardiac output 1k < 10 63 CO 2concentration 1k 30 –120 63 Temperature (^◦^C) ^0.3 ^>¹²⁰ ^0.02

Furthermore, collected healthcare data are of differ- entimportance.Toguaranteehealthcareservicesefficiency, it is necessary to design a system that can handle such heterogeneitywithdifferentpriorities.Consequently,inour solution we suggest the following traffic differentiation paradigm. Wedefine three classesofdatapackets: EMer- gency (EM),Delay Sensitive (DS) andGeneral Monitoring (GM).Adetaileddescriptionofthesepackettypesaswellas theirpriorityisgiveninTable4.

Notethattrafficprioritiesmayvarydependinguponthe valuesgeneratedbythesensors.Forinstance,bodytemper- aturereadingsmayproduceEMtrafficflowsiftheirvalues exceedthenormalthreshold.Togivemeaningtotheclassi- ficationmade,anefficientchannelaccessaswellasaccurate routeselectionareneeded.Accordingly,inthenextsection weproposePCLRPthatoperatesincompliancewiththede- finedtrafficcategoriesaswellasourmulti-hopclusterbased networkmodel.

4. PrioritybasedCrossLayerRoutingProtocol(PCLRP)

PCLRPincorporatesalooselycoupledinteractionbetween theapplication, MAC,network andphysicallayers. It is a mechanismwhereintheMACslotallocationiscustomized fortheunderlyingroutingprotocolstakingintoaccountthe QoSrequirementsoftherunninghealthcareapplicationand physicalparameters.Infact,PCLRPexploitstheexchanged MACframestobuildtheroutingschemeforfreeandwith- outaroutediscoverymessageexchange.Moreover,therout- ingalgorithmsexploitthenode’sbatterylevelandtheMAC framestructure(numberofTDMAslots)intherelayselection process.Ontheotherside,theMACsuperframedurationde- pendsonthenetworktopology(numberofnodes)andthe QoSrequirementsoftherunninghealthcareapplication(EM, DS,GM).

Section 4 details thePCLRP operationsalong withthe MACandroutingprotocolsthatitencompasses.

(5)

Fig. 1. Network model architecture.

Table 4

Data packets type classiﬁcation and ﬁeld encoding.

User priority Packet subtype Description

1 EM They are the most critical data packets. They should be forwarded in a short time and reliable way. They are used to report alerts and warnings (e.g., packets reporting data values that exceed normal thresholds).

2 DS The video traffic type defined by the standard is designed for nonmedical applications, e.g., video gaming. For this reason, we defined the DS type as the medical video type, e.g., video streaming of elderly monitoring and motion control. DS packets must be delivered in stringent deadline, while a reasonable packet loss is tolerated.

3 GM The lowest priority is given to GM packets. They correspond to regular measurements of patient physiological parameters that typically indicate normal values.

4.1. PriorityCrossLayeredMediumAccessChannelprotocol

4.1.1. Generaldescription

ThePCLMACprotocoloperatesinabeaconenabledmode, wherebeaconsaretransmittedatthebeginningofeachsu- perframefollowedbyanactiveandanoptionalinactivepe- riod. Allcommunications take place in the active period;

inthe inactive period,nodesare allowed to power down andconserveenergy. As describedabove, inour network model,inter-WBAN cooperationisrequiredinsomecases.

Consequently,ourPCLMACsuperframestructuremaycon- tainportionsoftimededicatedforinter-WBANcommuni- cation.Moreover, we have mentioned that C is the most

powerfuldeviceinWBANpandactsasarouter.Byanalogy, Cistheaccesschannelschedulerandbeaconsgeneratorin PCLMAC.Hence,asdescribedinFig.2,thePCLMACsuper- frameconsistsof:

• Mandatoryelements:aBeacon(B),aChildrenContention AccessPeriod(CCAP),aChildrenContentionFreePeriod (CCFP)andaDownLinkperiod(DL).

• Optionalelements:aNeighbourContentionAccessperiod (NCAP),aNeighbourContention-FreePeriod(NCFP)and aninactiveperiod.

Theactivepartofasuperframeisdividedintosixmain portions:

(6)

Fig. 2. Superframe structure.

• Beacon(B)isusedforresourceallocationinformationand synchronization.Italsocarriessecuritycodesinorderto allocateresourcestolegitimateusers.

• TheDownlink(DL)isusedbytheCtotransmititsqueries andcontrolframestoitsassociatedclustermembers.

• CCAP is used by children nodesfor transmitting both controlframesandpendingdata.Infact,newchildren sendtheir joinrequests intheCCAP.Also,failedpack- etsareretransmittedinthisportionoftime.Sincetradi- tionalCSMA/CAschemesuseafixedcontentionwindow forallnodetypes,andhencetheyareunsuitableforhet- erogeneousnetworkslikeWBANs,inCCAPweconsider a priority-guaranteed CSMA/CAprocedure. In fact, the CCAPisdividedintothreeparts:UplinkContentionfor EMergencytraffic(UCEM), UplinkContentionforDelay Sensitive(UCDS)andUplinkContentionforGeneralMon- itoring (UCGM) traffic. Duringthe CCAP period,nodes withemergencytrafficcontendfortransmissionthrough- outalltheCCAPperiod,whilemultimedianodescontend throughoutboththeUCDSandUCGM.Whereas,nodes withnormaltrafficcontendfortransmissiononlyinthe UCGM.Moreover,thepriority-guaranteedCSMA/CApri- oritizesallthenodesbyusingthreedifferentInter-Fame Spaces(IFS):EmergencyIFS(EMIFS),DelaySensitiveIFS (DSIFS) and the General Monitoring IFS (GMIFS) with EMIFS < DSIFS < GMIFS.Moreover, prioritized random back-off isapplied duringCCAP. Indeed,theback-off time dependsonthetrafficclassandisrandomlychosen as follows:

backoff time∈[0,2^BE⁺^T^{ra f f ic}^_^pr^ior^ity−1] (1)

whereBEdenotestheback-off exponent.Usingthepro- posedEq.(1),weguaranteethatahighprioritytraﬃchas higherprobabilitytransmissionopportunity.

• CCFPconsistsofmultipleTDMAslotsreservedbytheco- ordinator’schildrennodes.Itiscomposedofthreesub contentionfreeperiods: CFP1: slotsreserved byemer- gencynodes,CFP2:slotsreservedbymultimedianodes, andCFP3:slotsreservedbygeneralmonitoringnodes.

• NCFPisanoptionalperiodcomposedofTDMAslotsre- servedby relayed neighbourcoordinators.This timeis suﬃcientenoughforsendingdatafromrelayedcoordi- natorsandreceivingitbytherelayingC.

• NCAP is usedby the set ofC’s relayed neighbours for transferringbothcontrolframesandpendingdata.Infact, neighbouringcoordinators,aimingtorelaydataviathe currentcoordinator,sendtheirjoinrequestsintheNCAP.

Failedpackets arealsoretransmittedinthisportionof time.

4.1.2. Priorityguaranteedresourceallocationand synchronization

Asmentionedbefore,forresourceallocationandsynchro- nization,thecoordinatorcontinuouslybroadcastsbeaconsto allchildrenandneighboursatthebeginningofeachsuper- frame.Onlyactivenodesareabletoreceivethebeacons.

4.1.2.1.Synchronization. Inpractice,clocksindifferentsen- sor nodes suffer from random drifts, causing slot mis- alignmentovertime.Theycontinuouslykeepdriftingaway from each other even if they initially start at the same time[24,25,27].Consequently,ifthereisnosynchronization amongthenodes,thewholeideaoftimedivisionmultiplex- ingmaynotbeproductive.Thismayresultincollisionsand hugedatalossinWBANs.Onemorechallengeencountered isthedecisionaboutthestartofthesuperframeandhow toensuresleepandwakeupscheduleaccuracytoprevent messagesbeingmissedout.Ontheotherside,whenWSNs areusedforcriticalapplications,likehealthcare,timestamping andavery accurate timeinformation arean absolute

requirement.

All that shows the need for clock synchronization in WBANs.Inother words,forPCLMACTDMA slotallocation tomaterialize,fairlygoodtimesynchronizationneedstobe present.

Thus,toensurenodesynchronization,thePCLMACbea- conmessageistimestampedrightbeforeitissentwithT0. Uponreceiptofthebeacon,concernednodestimestampthe ACKmessagebothwithTr(beaconreceptiontime)andT_ack (Acksendingtime).Thethreetimestamps(^T⁰,Tr,T_ack)^form^a

datapointtocomputetheGuardtimeGT [27].Hence,toen- sureslotalignment,ﬁrstthecoordinatorcomputesthemax- imumclockskewinitsﬁeld[28].LetS={^S1,...,Sn}^be^both

thesetofallitschildrennodesandthelistofitsrelayedco- ordinatorsC.TheclockskewbetweenS_iandC,denotedbyδi

isdeﬁnedtobethedifferencebetweentheclocksofCandS_i. Usingtheprevioustimestamps,anodeS_icomputestheδias follows:

δ

i=Tr−T₀−T_b (2)

WhileCcomputestheδiasbelow:

δ

i=T_ack−T0−T_b (3)

with

T_b=B_data+O

v

D ⁽⁴⁾

(7)

Table 5

Notations and deﬁnitions.

BI Beacon interval

T b Beacon transmission time

T 0 Beacon message sending time T r Beacon message reception time S C’s cluster members (children and

relayed C) B data Beacon sampled bits

O v Overhead bits

D Communication data rate

δ ^Maximum^clock^skew

δi Clock skew between node i and coordinator C

slot d Slot duration

GT Guard time

DLD Down Link Duration

P data Sampled data bits

CFP Contention Free Period intended for reservation

Ack Acknowledgement bits

ASD Active Superframe Duration CCAP Children Contention Access Period CCAP _ min Minimal CAP period (used by neighbour

nodes)

NCAP Neighbour Contention Access Period A _ NCF P Number of slots intended for neighbour

reservation

NCAP _ min Minimal Neighbour Contention Access Period

NAS Number of Slots that a node aims to book

TCF Turned around Calibration Factor Sec i Conﬁdence level of gateway i ( Sec i∈ {1,

0.75, 0.5, 0})

RTT i RTT evaluation value of gateway i Dir Evaluation of the patient movement

vs the gateway position α ^The^security^weighting^coeﬃcient

β The RTT weighting coeﬃcient

γ The direction weighting coeﬃcient

λ The MF coeﬃcient

σ The RH coeﬃcient

NB _ needed _ Slots The number of slots needed by a relayed C

NB _ f ree Number of free slots that a relaying C may provide for its neighbour C AVG _ Walking _ speed The average walking speed of a patient

(m/s)

WT The walking period(s)

g _ prob Probability to encounter a g each walked meter

whereB_data andOvarethebeaconsampledandoverhead bits,andDisthedatarate.

Thebasicnotationusedinthispaperissummarizedin Table5.

Wenoteherethateach nodemeasuresonlyitsownδi

whiletheclusterheadCcomputestheδiofeachassociated nodetoobtainδ^,^whichîsêxpressedâs^below:

∀

^Si∈S;

δ

=Max

| δ

i

|

⁽⁵⁾

InordertopreventpossibleTDMAslotoverlap,thecoordi- natorCusestheδâs^the^guard^time^to^beînsertedîntoêach

slotduration.Incontrast,allS_i,∀ⁱ^,^makeûseôfδitosettle theirwakeupandsleeptimers.TheGT canbedefinedas thetimedurationinaslot,inwhichthepackettransmission isnotcarriedouttoaccountforanycollisionduetoclock

Fig. 3. PCLMAC slot structure.

errors [25]. Tonegate the possible interference effects of slowestandfastestclockednodeshavingadjacentCFPslots, theGT isfairlytwodividedandinsertedintheeithersides ofthetransmissiondurationineachTDMAslot.ThePCLMAC slotstructureisaspresentedinFig.3.

The ideabehindusingδ ^asGT isto handlethemaxi- mum possibleclock driftsbetweenanytwo nodes.More- over, evenGT ischosen astheworst case,nodesare set- tlingtheirclocksovertimeandmaybeconvergingtowards amoreuniformclock.Infact,nodescontinuouslysettletheir clocksaccordingtothecoordinator’sclockusingthebeacon timestamping.Further,usingadynamicguardtimetolerates nodefailures,adoptsnewlyjointnodestothenetworkand accommodatesclockdriftscausedbyenvironmentalfactors, whicharetime-varying[26].Theproposedsynchronization schemeachievessynchronizationandresynchronizationfor free,withoutanyexplicitsynchronizationmessageexchange.

The PCLMAC protocol allows low duty cycle children nodestoremaininsleepmodeandsavetheirenergy(con- sumedbybeaconoverhearing).

4.1.2.2. Activesuperframecalculation. TheActiveSuperframe Duration(ASD)iscomposedoffivemainportions:DL,CCFP, CCAP,NCFP,andNCAP.Thedurationofthesedifferentpor- tions is flexible and may vary from one superframe to another. It dependson nodestraffic andeventsvariation.

Hereafter,wedetailhowtheclusterheadCperformsthetime assignmentofeachcitedportion.

TheASDcontainsafixedminimalCCAPdurationdenoted byCCAP_min.CCAP_minisusedtoguaranteethatnewand disconnectedchildrennodesmaycontendtojointheirclus- terheads.Moreover,ifthecoordinatorisactingasarelay,it specifiesalsoafixedminimalNCAPforneighbourcontention (NCAP_min).TheexistenceofDLdependsonthefactthatthe coordinatorhasdatatotransmittoitsmembersornot.That is,theDLDuration(DLD)maybeequaltozero.

ForTDMAslotsreservation:

• ﬁrst,CcomputesCFPasfollows:

CFP=ASD−CCAP_min−NCAP_min−DLD (6)

• Second,itcomputesthemaximumnumberofTDMAslots (NB_S)thatmaybereserved.

NB_S= CFP

slot_d ⁽⁷⁾

with

slot_d=

δ

⁺^P^_^data⁺^O

v

+Ack

D ⁽⁸⁾

HavingdeterminedtheNB_S,Cmaymanagethereserva- tionrequestsofitschildrenaswellasofitsrelayedcoordina- tors.Infact,aimingtolimitnetworkoverheadandcollisions,

(8)

unlikethestateoftheartexistingprotocols,CFPslotreser- vationisdonewithoutsendingslotallocationqueriesinthe CAP.Specifically,anodemakesarequestforCFPslotsfrom theclusterheadbysettingaNumberofAskedSlots(NAS) flaginitstransmittedframe.Obviously,thisflagissettozero ifthenodeisnotintendedtoperformareservation.Theclus- terheadreplieswithanacknowledgementthatcontainsthe numberofthegrantedslots.Consequently,thereservation fornextdatatransmissionsisperformedwithincurrentdata transmissionandnoadditionalsignallingisrequired.

Thecoordinatorassignsslotsaccordingtotheiravailabil- ityandtheorderofrequestsarrival.Whatisevidentisthat,to sustainthehealthofpatient,eachcoordinatorservesitschil- drenﬁrst.Thismaybeexplainedbythefactthat,asmallloss ofcollectedinformationmaybedevastatingforthepatient’s life.Moreover,mostofhealthcareapplicationsarecostlyand itismandatorythatwemanageourresourceseffectively.An- otherimportantpoint,thereisariskthattheneighbourCisa maliciousnodethatintendstodeterioratetherelayingC’sre- sources.Accordingly,Cprovidesitschildrenalltheslotsthey ask,thenifthereremainslots,itservesitsneighbours(re- layedC).

Let A_NCFP denotes the number of slots intended for neighbourreservation.A_NCFPisobtainedasfollows:

A_NCFP=NB_S−CCFP (9)

Wementionthatthenumberoftotalslotsthatareeffectively reservedmaybelessthanNB_S.Hence,theremaineddura- tionwillbetwodividedandaddedtotheCAPportions.In facttheCCAPandNCAPperiodsarecomputed,respectively, asfollows:

CCAP=

(

^A^_^NC^F^P⁻^NC^F^P

)

^×^slotd

2 +CCAPmin (10)

NCAP=

(

^A^_^N^CF^P−NCFP

)

×slot_d

2 +NCAP_min (11)

Formoredetails,wepresent,respectively,inAlgorithm1and Algorithm2thepseudocodesofPLCMACprotocolattheco- ordinatorandatthechildrensides(ateachnode_iside).

4.2. Routingestablishmentphase

Ournetworkmodelisadistributedmobilepublic-BAN whereintraandinter WBANcooperationsarerequired.In nextsubsectionswedetailtheintraandextraroutingpaths establishmentprocedures.

4.2.1. Intra-bodyrouting

Sinceinhealthcareapplicationsbiosensorsaredeployed onthehumanbody,whichisaverylossymediumforpropa- gatedwaves,thepropagationlossaroundthehumanbodyis high.Consequently,usingrelaynodesbecomesadvantageous andsometimesevenanabsoluterequirementtoensurecon- nectivityofthenetwork[29].Ontheotherhand,WBANnet- worksaresmallinsize.Indeed,intheworstcaseintra-body communicationwillbetwohops.Hence,weassumethatthe topologyisatwo-hopextendedstarBANtopology.Asourap- proachiscross-layer,inourtopologysetupandroutingpaths establishmentwemakeuseofthedescribedPLCMACBeacon exchange.Therefore,whenacoordinatorCisswitchedonfor

Algorithm1AlgorithmexecutedbycoordinatorC.

1: // Start of superframe 2: // synch _ t: synchronization time 3: CFP = ASD −DLD −CCAP _ min −NCAP _ min 4: NB _ S = _slot^{CF P}_{_}_d

5: Next _ f rame _ star t = cur rent _ time + synch _ t + T b

6: ∀^received _ Ack i;δ^_ⁱ⁼^T^_^ack⁻^T^_⁰⁻^T^_^b 7: broadcast Beacon new

8: for i = 1 to N 9: δ= Max |δi| 10: end 11: if ( DL _ d > 0 ) 12: send data to nodes 13: end if

14: if (receive ( d ata nodei, and d ata.NAS > 0 ) 15: if (NB_S)

16: if ( N B _ S > = data.N AS)

17: NB _ S− = data.NAS 18: else

19: NB _ S = 0 20: end if

21: switch node_type:

22: case EM: add CFF1 slots 23: case DS: add CFF2 slots 24: case GM: add CFF3 slots 25: case Relayed C: add NCFP slots 26: end

27: send (ACK, node i) 28: end if 29: else

30: discard request 31: end if

32: if ( CAP and recei ve(sleep _ request nodei, NB _ slots)) 33: if ( node iis allowed to sleep)

34: send (ACK, node i) 35: else

36: discard request 37: end if

38: end if

39: if (CFP1 or CFP2 or CFP3 or NCFP) 40: keep listening to receive data 41: if (inactive period starts)

42: go to sleep and wake up at Next _ f rame _ start 43: C CAP = C CAP _ min + ^NB₂^_^S∗slot _ d

44: N CAP = N CAP _ min + ^NB₂^_^S∗slot _ d

thefirsttime,itgeneratesaPLCMACbeaconmessageanddif- fusesittothewholenetwork.Aspreviouslymentionedinthe PLCMACdescription,uponbeaconreceipt,childrennodesre- spondbyatimestampedACK.Knowingthatwemeanbya relaycapablenodeachildnodethathasagoodbatterylevel andisdirectlyconnectedtoC.Morethantimestamps,the ACKcontainsabooleanflagcalledRthatindicatesifthenode isarelaycapableornot.Thisflagisusedinthetopologyex- tensionprocedure.Infact,upontheirdeployment,children nodeswaitforbeaconreceptiontojoinC.Theykeeplisten- ingforabeacontimeoutperiod.Uponthisperiod,children thathavenotreceivedthebeaconsendjoinrequeststotheir closerrelaycapablenodes.Relaynodesarechosenaccording totheReceivedSignalStrengthIndicator(RSSI)andδ^values.

Theuseofδ^helps^a^node^to^choose^the^relay^with^which^is

moresynchronized.Algorithm3describesthepseudocodeof ourintra-bodyroutingalgorithm.

4.2.2. Extra-bodyrouting

Asdescribedinournetworkmodel,theextra-bodycom- municationinvolvesinter-WBANcommunicationaswellas

(9)

Algorithm2Algorithmexecutedbyachildnode.

1: // Start of superframe 2: wake up to receive beacon 3: if (receive(Beacon)) 4: δ_ i = T _ r −T _ 0 −T b

5: Next _ f rame _ star t = cur rent _ time + δ_ i

6: if ( DLD > 0 and there is data to me)

7: // node will receive data from the coordinator 8: keep listening

9: else 10: go to sleep 11: end if

12: if ( its reserved _ slot > 0 )

13: wake up when its reserved_slots starts 14: set data.NAS = NAS

15: send data

16: go to sleep when its reserved_slots ends 17: end if

18: end if

19: if (( CCAP starts || NCAP starts)^&(∃^data^or^sleep^_^request^to^send

||Not connected)

20: // nodes wake up to contend for slot reservation request, sleep request and/or data transmission

21: switch node_type:

22: case EM: wake up

23: case DS: keep sleeping and wake up when UCDS starts 24: case GM: keep sleeping and wake up when UCGM starts 25: end

26: end if

27: if (inactive period starts)

28: go to sleep and wake up at Next _ f rame _ start 29: end if

30: else if (Beacon timeout) 31: NB_lost_beacons++

32: if ( NB _ lost _ beacons > max )

33: // max is maximum authorized number of beacons that a node can loose 34: Keep listening to ﬁnd a relay node

35: else

36: go to sleep and wake up at Next _ f rame _ start 37: end if

38: end if

Algorithm3Intra-bodyroutingalgorithm.

1: if (Recv(Beacon new, node j))

2: // child node jreceives new beacon from C 3: if (next_hop == NULL)

4: next_hop = coordinator_id 5: end if

6: else if (beacon_timeout and next_hop == NULL)

7: // node does not receive a beacon from the coordinator: not in its communication range

8: ∀list ened AC K node_i−>coordinator

9: if ((new ACKi.R == true) and RSSI (new ACK i) > RSSI (last ACKi)) 10: Next_hop = sender of (new ACK i)

11: else if ((new ACKi.R == true) and RSSI (new ACK i) == RSSI (last ACKi))

12: Node jcomputes the clock skew δVS node i

13: Chooses the node having the smallest δ 14: end if

15: end if

thecommunicationbetweenthecoordinatorsandWiFigate- ways.Ourextra-bodyroutingalgorithmconsistsoftwodis- tinctandcooperatingparts:

• WiFigatewaysselection

• WBANrelayselection(forinterWBANcommunication) 4.2.2.1.WiFigatewaysselectionalgorithm. Theselectionpol- icy often used by mobile terminals for automatically

Table 6

RTT evaluation values.

Obtained RTT Corresponding value RT T ≥Max _ RT T 0

AVG _ RT T ≤RT T < Max _ RT T 0.5 Min _ RT T ≤RT T < AVG _ RT T 0.75 RT T < Min _ RT T 1

selectinganaccesspoint,simplyconsistsofscanningAPsand thenchooses theunencryptedonewiththehighestsignal strength.Thispolicyignoresimportantfactorsthatimpact ontheobtainedQoS,sinceitdoesnottakeintoaccountthe networkloadthatnearbyAPshave,orsecurityproblemsthat mayexist.ThiscanresultinunbalancedloadonsomeAPs, leadingtolowdatathroughputandanunsatisfactorynet- workperformance.Consequently,tocopewiththeseshort- comings,we consideranddescribeindetailthefollowing keyparameters(RoundTripTime,Patientdirection,Gateway conﬁdencelevel):

• RoundTripTime(RTT):Itisthetimebetweenstartingthe transmissionofapacketandreceivingthecorresponding immediateacknowledgement[30].TheuseofRTTisad- vantageousandmaybejustiﬁedasfollows:

– ComputingtheRTTis animplicitestimationofthe gatewaydistanceandnetworkdensity.Obviously,a distantgatewayrequiresmuchtimetorespondtoits associatedclientsthananearone.Moreover,accord- ingtoGüntherandHoene[30],thedistanceseparat- ingaWBANpfromagatewaygmaybeobtainedas follows:

Distance=

(

^RTT⁻^TCF

)

^×^speed^of^light

2 , (12)

whereTCFistheTurnedaroundCalibrationFactor.Itis anadjustmentdelayforerrorsthatmayoccurduring datatransmissions[30].

– Inthiscase, we do notrequirestrong assumptions toestimate the gatewaydistance,like each patient knowsormaydeterminethepositionofeachgateway.

– UseofGPSisenergyconsuming.

We deﬁne three thresholds (Max_RTT, AVG_RTT and Min_RTT)toevaluatetheRTTparameteraccordingtothe proceduredescribedinTable6.

• Patientdirection(Dir):thisparameterindicatesifthegate- wayisinthesamedirectionofthemobilepatientornot.

Ithelpsustoavoidthesocalledping-pongeffect[31].In fact,whenadeviceiswithintherangeofmultipleWiFi AccessPoints(APs),thepingpongeffectcanbedeﬁned astheseriesofconsecutivetransitionsfrom/todifferent accesspoints.Itistheresultoftheaggressivenatureof 802.11interfacesthattrytoconnect toanaccesspoint withabettersignaloncethesignalfromthecurrentac- cesspointdropsbelowagiventhreshold[31,32].Thisis- suearisesmainlywhenthedeviceismobile,andcanbe thesourceofextraenergyconsumptionanddataloss.Dir takestwopossiblevalues:1,ifgatewaygisinthesame directionofthepatient,and0,otherwise.

Wemayestimate ifagatewayisinthemobilitydirec- tionofthepatientornot,basedontwosuccessivecap- turedRSSIvaluesfromsuchgateway.Infact,iftheRSSIis