optimal datalink selection for future aeronautical ... fileoptimal datalink selection for future...
TRANSCRIPT
1
Optimal Datalink Selection for Future AeronauticalTelecommunication Networks
Atm S. Alam, Member, IEEE, Y.Fun Hu, Sr. Member, IEEE, Prashant Pillai, Sr. Member, IEEE, Kai Xu, andAleister Smith
Abstract—Modern aeronautical telecommunication networks(ATN) make use of different simultaneous datalinks todeliver robust, secure and efficient ATN services. This paperproposes a Multiple Attribute Decision Making based optimaldatalink selection algorithm which considers different attributesincluding safety, QoS, costs and user/operator preferences.An intelligent TRigger-based aUtomatic Subjective weighTing(i-TRUST) method is also proposed for computing subjectiveweights necessary to provide user flexibility. Simulation resultsdemonstrate that the proposed algorithm significantly improvesthe performance of the ATN system.
Index Terms—Aeronautical communications, datalinkselection, intelligent algorithm, Multiple Attribute DecisionMaking.
I. INTRODUCTION
THE capacity demand in aeronautical telecommunicationnetworks (ATN) for air traffic control (ATC) and air
traffic management (ATM) systems has grown dramaticallyover the past years. The rise in the capacity demand isreflected by the growing number of aircraft and passengers,which is expected to double by 2035 [1], as wellas the introduction of new high data rate aeronauticalcommunications services, in particular Internet applications.As a result, the need to improve the capacity, safety andefficiency of the global airspace has prompted several globalinitiatives to modernize ATM systems, most notably theEU Single European Sky ATM Research (SESAR) and theUS Next-Generation Air Transportation System (NextGen)programmes, with a major focus on Air Traffic Services(ATS) and Aeronautical Operational Control (AOC) operatingconcepts and requirements [2].
The ATM modernization requires a paradigm shift fromthe current analogue voice to digital data communicationsto handle the increasing amount of and more complexinformation exchange between controllers and pilots forfuture ATM operational procedures [3]. ATM applicationsinclude both safety-critical, such as AOC and ATS, andnon-safety critical services, such as Airline AdministrationControl (AAC) and Aeronautical Passenger Communications(APC) services. The AOC and APC services are expectedto be the major motivation for broadband communicationsdue to the higher transmission rate requirement [4]. Inaddition, future integration of unmanned air vehicles (UAVs)
A. S. Alam, is with the Institute of Communication Systems, University ofSurrey, Guildford, GU2 7XH, UK. (e-mail: [email protected]).
Y.F. Hu, K. Xu, and A. Smith are with the School of Electrical Engineering& Computer Science, University of Bradford, West Yorkshire, BD7 1DP, UK.
P. Pillai is with the Department of Computing & CommunicationsTechnologies, Oxford Brookes University, Oxford, OX33 1HX, UK.
is expected to add further complexity to the alreadycongested civil air space. Eurocontrol and the AmericanFederal Aviation Administration (FAA) have identified twoprimary drivers for anticipating the increasing demand ofthe future aeronautical broadband services: i) an appropriatecommunication infrastructure to support emerging and futureradio technologies, and to cater for future air traffic growth,and ii) a consistent global solution to support a seamlessATM system [3]. Therefore, an integrated ATN solution isbecoming imperative for aircraft operators to improve thecapacity, efficiency and cost-effectiveness of the ATM systemwhile ensuring a high degree of its flexibility, scalability,modularity and reconfigurability.
The EU project NEWSKY (Networking the Sky) [2]specified a single system based on IP technology with thecapability of transmitting data through multiple datalinks,directly to the ground via satellites. Another EU projectSANDRA (Seamless Aeronautical Networking throughIntegration of Data Links, Radios and Antennas) [5],[6] defined an Integrated Modular Radio (IMR) buildingon sophisticated software-defined radio (SDR) technology.The SANDRA project integrated multiple datalinks directlyto the ground networks such as including VHF DigitalLink mode 2 (VDL 2) and Aeronautical Mobile AirportCommunications System (AeroMACS) and/or via satellitesto provide aeronautical communication services. The IMRconcept of SANDRA project was further validated in the UKproject SINCBAC (Secure Integrated Network Communic-ations for Broadband and ATM Connectivity) to providesecure voice and data connectivity for both ATM andpassenger services through a heterogeneous set of radioaccess technologies (VDL 2, Iridium and Inmarsat BGAN)to connect with the ground ATN infrastructure. Eachdatalink has its own transmission characteristics providingcommunication services and is configured to deliver eithersafety critical or non-safety critical communication servicesor a combination of both. However, in a heterogeneous radioenvironment in an aircraft, different applications have differentQoS requirements, security requirements and user/operatorpreferences. The on-board terminal when forwarding the datais required to select the right datalink due to the safety andhigh demand of the expensive datalinks. Therefore, it is animperative decision to select the best datalink to transfer datawhen multiple datalinks are available in order to provide anefficient, reliable, secure and cost-effective communicationsolution, and none of the previous projects have consideredan optimized datalink selection process.
Authors in [7] have discussed the importance of the
Ground SegmentAircraft Segment
Safety Services
Cockpit Applications
Cabin Data
Radio Access Network N
Radio Access Network 1
Transport Segment
Radio Access Network 2
SWIM AGDLGMS
Ground ATN Network
Cloud Computing
AOC Service Provider ATS Service
Provider
Passenger Services
Security Service
Mobility Information Service
...
RaS
RoS
RaS – Radio Subsystem
IMC – Integrated Modular Communication RoS – Routing Subsystem
IMC
2
Fig.1. AnintegratedmodularcommunicationATMarchitecture.
optimiseddatalinkselectioninthiscontext,buttheyneitheranalyzednorsuggestedresultsforit.Currently,linkselectioninheterogeneous mobilecommunicationsenvironmentisusuallymadebasedonthereferencesignalstrength[2].Thechallengingtaskistoconsidermultiplecriteriaintermsofthedatalinkcharacteristics,typeofservicesorapplicationstogetherwithuserandaircraftoperatorpreferences[8].Sinceitisnecessarytoconsider manyattributessimultaneously,a multipleattributedecision making(MADM)approach[9]–[11]isrequired. MADMbasedtechniquesaregainingpopularityduetoitsrelativelyhighprecisionandcapabilityofadoptinguserandoperatorpreferences[10],[12]. Whenmultipleattributesareconsidered,decision makers mustcomputethe weightofeachattributerepresentingtheirrelativeimportanceinordertorankavailablealternatives.Therearetwobroadcategoriesofdetermining weights:objectiveandsubjective methods[9]. Whilethelatterisrelatedtouserexperience,theobjectivemethoddoesnottakeanyuserpreferenceintoaccount.Twoobjectiveweightingmethodssuchasentropyandvariance[8],[10]andthreewell-knownsubjectiveweightingmethodssuchaseigenvector[10], weightedleastsquare[8],[12]and Trigger-basedaUtomaticSubjectiveweighting(TRUST)[8],[11]methodsarecommonlyused. Thefirsttwosubjective weightingmethodsdonotworkwellforthedatalinkselectionproblemsincetheirpair-wisecomparisonprocessisslowandnotautomatic[8],[9],[11].TheTRUSTmethod[11]hastheabilitytoautomaticallycomputethesubjectiveweightsanditiscomparativelyfastandefficient.
DespitetheautomaticsubjectiveweightcomputationbytheTRUSTmethod,itcompletelyignorescertainattributesthatarenotimportanttotheusers,buttheymaybeimportantfromotherperspectivessuchastheoperator,forexample,toimprovethesystemefficiency.Consequently,thelinkselectiondecisionmaynotbeoptimal.AnotherlimitationoftheTRUSTmethodisthatthereisnoprovisionforuserstoprioritizeoneeventoverothersasintheeigenvectormethod,which
leadstotheinaccuracyofobtainingthesubjectiveweights.ToalleviatethelimitationsoftheTRUST methodandtoaddressthenecessityoftheautomaticweightcomputation,anintelligentTRUST(i-TRUST)methodisproposedinthispaper.Theproposedi-TRUSTalgorithmisintelligentenoughtoprioritizeuserrequirementsaccordingtorelativeimportanceandbyintroducingaparameter,( )addedtoeachattributetoreflecttheuserpreferenceinapreciseway.Itisworthnotingthatthecontributioninthispaperistwo-fold:firstly,thederivationandintegrationofbothobjectiveandsubjectiveweightswhileformulatingthedatalinkselectionprobleminthefutureaeronauticaltelecommunicationnetworksasanMADMbasedproblemformakingoptimaldatalinkselectiondecisions;andsecondly,anintelligentalgorithmisproposedtoautomaticallycomputethesubjectiveweightsthatreflectstheimportanceofuserpreferencesinordertoimprovetheoverallsystemperformance.Thepaperisorganizedasfollows:SectionIIpresentsanoverviewoftheATMarchitecture.SectionIIIexplainsthedatalinkselectiontechniqueswhiletheMADM-baseddatalinkselectionprocedurewiththeproposedi-TRUSTmethodisdescribedinSectionIV.SectionVdescribesthecomparativeanalysisandperformanceoftheproposedoptimizedalgorithmfollowedbyconclusioninSectionVI.
II.ATMARCHITECTURE
ThedesignofATMarchitectureadoptsanopenstandardapproachutilizingIP, ETSIandIEEEstandardsforanintegratedcommunicationsprotocolarchitecture.
A.ArchitectureOverview
The ATMsystemarchitectureisresponsibleforthecontrolandmanagementofcommunicationsandinformationexchangesrequiredfortheoperationprimarilyconsistsofthreedistinctactivities[13]: ATC, Air Traffic flowManagement(ATFM)andAeronauticalInformationServices(AIS).Inordertoperformtheseactivities,thereisa
3
TABLEITRAFFICDOMAINMAPPINGOFCOMMUNICATIONSERVICESAND
APPLICATIONS.
TrafficDomains
ATNCommunicationServices
Applications
Safetyservicetraffic
ATS(critical)
Radio(voice)ADS-B/ADS-CTrajectory-negotiationMeteoCPDLCAir-Air(Freer)
Cabin datatraffic
APC(non-critical)AAC(non-critical)
EmailInternetTel-fax-dataMultimediaetc.
Cockpitapplicationstraffic
AOC(critical)
FleetmanagementEngineeringactivitiesMaintenance
requirementtohaveareliableandefficientcommunicationenvironment.Therearethreemainelementsofcommunicationenvironmentforactivities:thegroundATNnetwork,radiolinktechnologiesandaterminalnode(TN)on-boardaircraft.Inthispaper,theconceptoftheATMarchitecturedistinguishesthesethreeelementsasground,transport,andaircraftsegments,respectively,asshowninFig.1.
The TNon-boardtheaircraftisreferredtoastheintegratedmodularcommunication(IMC)terminalconsistingofaroutingsubsystem(RoS)andaradiosubsystem(RaS)equippedwith multipledatalinks.TheRoSrepresentsthenetworklayeroftheprotocolstackanditconsistsofasecurerouter.Inthisarchitecturaldesign,thesegregationbetweendifferenttrafficdomainshastobetakenintoaccountduetothehighsecurityrequirementsandthesecurerouterintheRoSisresponsibleforthesecuresegregationandsecurityawarerouting.Differenttypesofgroundapplicationstothecorrespondingairapplicationsoverdifferentair-groundsub-networkscanbeprovided.TableIprovidesthetrafficdomainmappingofcommunicationservicesandapplications.Ontheotherhand,theRaS,whichrepresentsthedatalinkandthephysicallayersoftheprotocolstack,supportsradioresourceallocation,radiolinkestablishment,QoSmappingandprotocoladaptation. The TNhasthepossibilitytochooseadatalinkamongavailabledatalinkssubjecttospecificrequirements.Asthecommunicationsinvolvebothoperational(ATC,AOC,AAC)andnon-operational(APC)servicesandtheysharenetworkresources,theoptimaldatalinkselectionforaspecificservicecanbechallengingduetomanychallengesincludingsafetyandsecurityconcerns.
B.DatalinkParametersandPreferences
TheTNon-boardtheaircraftcollectsinformationaboutavailabledatalinksandmaintainsadatabaseofinformationthatisnecessaryfordecision-makinginregardstotheselectionoftheoptimaldatalinkselection.Boththeuserandoperatorpreferencesaswellasthedatalinkconditionsareconsideredinmakingtheoptimaldecision.Thefollowingare
identifiedasthemostimportantdecisionparametersinthedatalinkselectioncontext:
Linkquality:Thelinkqualityis measuredintermsofreceivedsignalstrength(RSS),whichevaluatestheavailabilityofdatalinks[14].QoSrequirements: In ATM,thesatisfactionoftherequiredQoSintermsofbandwidth,delaysandpacketlossrateisanimportantmeasureforsafetyandsecurity.Linkcosts: Differentdatalinks mayhavedifferentchargingpolicies.Therefore,itisimportanttoconsiderthedatausagecostinthedatalinkselectiondecision.Thisisespeciallythecaseforaeronauticalpassengerservices.Resource utilization: Sincethe number of userssupported by each datalink for aeronauticalcommunicationsisrestricted,another key decisionmakingparameterisresourceutilizationtoensurethatthelinkselectionalgorithmnotonlyimprovesthedatarateandcost,butalsotheresourceutilization.Security: Safety-related communicationsrequire asecurityconcepthandlingthreatsandattackstothesystem[15],[16].TheIMCsecurityconceptsupportsthesegregationoftrafficintodifferentsecuritydomainsintheaircraftsegment,whichcandictatethedatalinkselectiondecisiontoagivenservice.
Additionally,itisalsoimportanttoprovideflexibilitytobothusersandaircraftoperatorsin makingthedatalinkselectionprocess.Therationaleforprovidingtheflexibilityistoallow,forexample,theoperatortopreferdifferentsettingsdependingupontheirflighttypessuchaseconomicand/orbusinessclassflights,thustheuserandoperatorpreferencesarealsorequiredtoconsiderinthedatalinkselectionprocess.Inthefollowingsections,thenecessaryproceduresfortheoptimaldatalinkselectionprocesswillbedescribed.
III.DATALINKSELECTIONTECHNIQUES
Futureaeronauticalcommunicationsystemsareexpectedtosupportmultipledatalinksforair-groundcommunications.Inordertoimprovetheoverallnetworkperformance,adatalinkselectiontechniqueisrequiredtoselectthebestlinktobeusedforanyconnectionrequest.TheselectionofthemostsuitablelinkissubjectedtodifferentflightphasesandotherATMoperationalrequirementssuchassecurityandsafetyrequirements.Therefore,itisimportantforthesystemtofulfilthoserequirementswhenconsideringthepossibleoptimizationaspects.Assuch,abaselineandanoptimizeddatalinkselectiontechniquesareconsideredinthissectionforcomparison.Inbothcases,apre-linkselectionprocessillustratedinFig.2takesplacetoidentifycandidatedatalinksoutoftheavailabledatalinksthat wouldsatisfyasetofpre-definedcriteriaincludingthecurrentflightphase,thelinkcharacteristicsandtheQoSrequirementsbeforethefinalselectionofatargetdatalink.Thecurrentflightphaseinformationisusedtoconstraintheeligiblelinkalternatives whilethe QoSrequirementsgivetherequest’s QoSspecificationastheofferedQoSshouldmatchtherequestQoSrequirements.Thesecurityrequirementslimittherequestthatcanbeusedfora
Get attributes of available data link i
i++ <= iMax
Select flight phase attribute
Eligible for current flight phase?
Select QoS attribute
Satisfy QoS requirements?
Yes
Select security attribute
Satisfy security requirements?
Yes
Yes
Number of candidate links
Discard link alternative i from list
No
No
No
Yes
No
4
Fig.2.Pre-linkselectionflowchart.
particulartrafficdomain.Wheneveranattributeofalinkfailstomeetthoserequirements,itisremovedfromthecandidatelinklist.Therationaleofthepre-linkselectionprocessisthatdatalinksthatdonotfulfiltherequirementscannotbeselectedinordertoreducethecomplexity.
A.BaselineDatalinkSelection
Thebaselinedatalinkselectionalgorithmisastaticprocessthatselectsthetargetdatalinkoutofthecandidatedatalinksbyorderingthembasedonapredefineddatalinkprioritylist,L= {l1,l2,...,ln}.Inthispaper,fourtypesofradiolinksareconsidered:InmarsatBGAN(BroadbandGlobalAreaNetwork)Class4andClass6,VDL2andIridium.ThetwoBGANClassescanbeusedforbothcockpitandpassengervoiceanddatacommunications,butClass6offershigherdatarate(432kbps)thanthatofClass4(200kbps).VDL2isdedicatedforvoiceandshortmessagesbetweenthepilotandthecontrollerontheground,whileIridiumcanbeusedforbothcockpitandpassengervoicecommunicationstransport.Ithastobenotedthatsecuritymeasurestosegregatecockpitandpassengers’communicationsareoutsidethescopeofthispaper,butreaderscanreferto[17]formoreinformationonrisksanalysisinrelationtothistopic.Inthebaselinealgorithm,ifacockpitvoiceapplication,forexample,isrequested,alistofcandidatedatalinks whichsatisfythe minimum
TABLEIIATTRIBUTEVALUESFORAVAILABLEDATALINKS.
❳❳❳❳❳❳❳❳AttributesDatalinks
VDL2 Iridium BGAN
BR(kbps) 31.5 2.4Class4:Upto200Class6:Upto432
PD(ms) 2000 748 950(800to1100)BER 10 5 10 6 10 5
RMF - - -RSS - - -
CST($/MB1) 1 7 3
requirementsoftherequestedapplicationsareselectedfromthepre-selectionprocess.Thebaselinealgorithmwillthenorderthecandidatelistaccordingtoapre-definedpriorityfactor.Ifthelinkusagecostisusedastheprioritycriteria,thenanorderlistcanbeformedasfollows:
1)VDL22)BGANClass4(BGAN1)3)BGANClass6(BGAN2)4)Iridium
ItimpliesthatVDL2isthetopchoicefortherequestedapplicationsandIridiumhasthelowestpriorityasthelinkusagecostisthepriorityparameterinthiscase.TherelativeusagecostofeachdatalinkisshowninTableII,whichhasbeendrawnfromavailablepackagesofferedbythecorrespondingoperator.However,theprioritycanvarydependingonthepreference.Ingeneral,thestepsinvolvedinthebaselinealgorithmaresummarizedasfollows:Step1:Ifthereisonlyonedatalinkonthecandidatelinklistafterthepre-linkscreening,itwillbechosenasthetargetlinktoestablishconnection.Step2:Iftherearemorethanonecandidatedatalinks,thecandidatedatalinklistissortedtoformadesignatedprioritylinklist,L.Thealgorithmfirstchoosesthedatalinkwiththehighestpriorityfromthesortedcandidatelistasthetargetlink.AconnectionestablishmentrequestiscreatedbasedontheQoSparametermappingforthetargetdatalink.Inthecaseoffailurefortheconnectionestablishment,thesecondhighestprioritycandidatelinkischosenasthetargetlinktoestablishconnection.Thisprocessisrepeateduntilaconnectionisestablishedsuccessfullyoralllinksinthecandidateprioritylisthavebeentried.Step3:Ifthereisnocandidatedatalinkinthelist,noconnectionwillbeestablishedandtherequestwillbedropped.ThecorrespondingbaselinedatalinkselectionflowchartisillustratedinFig.3,wherethecandidatedatalinksarefoundfromthepre-linkselectionprocess.
B.OptimisedDatalinkSelection
Thebaselinelinkselectionalgorithmmainlyreliesonthepre-definedlinkprioritylistbasedonasinglespecificattributeandtheweaknessofthebaselineapproachisthatmultipleattributescannotbeusedforthelinkselectionprocess.Thus,thebaselinealgorithmisnotanoptimallinkselection
1Thesearerelativevaluescalculatedbasedavailablepackagesforeachlink.
Candidate data links
Sort candidate links according to the priority list
Select link with highest priority
Number of candidate data links?
= 1
Select the data link
= 0
Request dropped
MADM selection process
Baseline Optimal
5
Fig.3. Baselineandoptimallinkselectionalgorithms.
algorithmasitonlyreliesonstaticallyconfiguredpreferencesbutnotonawidersetofattributes,whichareimportanttoconsiderinmakingtheselectiondecision.Inthenextsection,anoptimaldatalinkselectionalgorithm
applyingthe MADM methodintegrated withaproposedi-TRUSTalgorithmforsubjectiveweightevaluationtoselectthemostoptimumlinksubjecttomultipleattributesisderived.Thisreferstomakingpreferencesdecisionoverthecandidatedatalinksthatarecharacterizedbymultipleattributeswitheachdatalinkhavingdifferentcharacteristicsandattributevalues.Theimportanceofoneattributeoveranotherisconsideredbychoosingtheirweightvalues.ThedatalinkselectionalgorithmispresentedinFig.3whileFig.4illustratesthe MADMprocess,whichincludestwomainsteps:Initial step:Thisisaninformation gatheringstep,
whereallrequirementsandinformationaboutattributesofeverycandidatelinkarecollectedincludinguser/operatorpreferencesandapplicationrequirements.MADMstep:Thisprovidesallthenecessarystepsfor
the MADM-baseddatalinkselectiondecisionincludingtheadjustmentofattributevaluesthroughnormalizationmethod,computingbothobjectiveandsubjectiveweights.Thefinalweightforeachattributeisobtainedbycombiningboththeobjectiveandsubjectiveweights.Inordertomakethelinkselectiondecision,normalizedvaluesofmultipleattributesforeachcandidatelinkarecombinedtoobtaintheutilityscoreforeachdatalink.Finally,anoptimaldatalinkisdeterminedbasedontherankoftheutilityscoresfromallcandidatedatalinks.ThedetailanalysisoftheMADMtheoryunderpinningthose
stepsisexplainedinthefollowingsection.
IV. MADMFOROPTIMALDATALINKSELECTION
A.IdentifyingRelevantAttributes
Upward attributes
Downward attributes
Attributes Requirements
Operator
preferences
User
preferences
Application
QoS levels
Objective weighting
Subjective weighting
Combined weightsAdjustment
Combined multiple attributes
Utility score for each link
Optimal link (highest score)
Adoptedattributesarethekeytothe MADM-basedproblemandtheattributesshouldbethe mostimportant
Fig.4. MADM-baseddatalinkselectionprocess.
onesdeemedrelevanttothe finaldecision. Alotofstudieshavebeendoneinnetworkselectionissuesusingutilityfunctionsinterrestrialheterogeneousnetworkscontext[9]–[12],[14]andthey maynotbedirectlyapplicabletotheaeronauticalcommunicationscontext.Thisisduetotheverydifferentcriteriainlinkselectioninaeronauticalcommunications,whichmustconsiderthetypeofaeronauticalcommunicationsservicesfortheQoSpurpose,thepolicyandrulesthat mayoverruletheuseofaspecificdatalinkforsafety-criticalservicesforsafetyandsecuritypurposes,aswellastheairlinecontractwiththeaeronauticalcommunicationserviceproviders.Inaddition,avarietyofdifferentlinktechnologies maybeavailableforaircraftcommunicationssimultaneouslyandaeronauticalcommunicationnetworksmayconsiderattributesthataredifferentfromthoseconsideredbyterrestrialnetworks[9]–[12],[14]. Mostimportantly,thechosenattributesshouldbemeasurableinameaningfulandpracticalwayforeachoftheproposeddatalinks.However,thenumberofadoptedattributes, mustberestrictedfortheMADMmethodtoreachatrade-offbetweencomputationcomplexity,requirementsandsafetyservices[9]. Whenconsideringtheattributesforaeronauticalcommunications,oneshouldfirstthinkofguaranteeingtherequired QoS,safetyandresourceutilisation,whichhavebeendiscussedinSectionII.B.Sincetheaeronauticalcommunicationsinvolvebothterrestrialandexpensivesatellitelinks,itisthereforeimportanttoconsidersuchattributesthathavetheabilitytoreflectthelinkusagecost,linkquality,resourceutilization,delaysaswellastoprovideflexibilitytoallowbothoperatorsanduserstosetpreferences. Moreover,theperformanceofeachdatalinkisaffectedbythereceivedsignalstrength,whichisneededtoconsiderinthedecisionmaking.
Todetermineanoptimallinkamongadiversesetoflinks,agivensetofattributesareconsidered,whichhavebeencarefullyidentifiedforthescenariounderconsiderationinthispaper,namely:bitrate(BR),packetdelay(PD),biterrorrate(BER),relativelinkcosts(CST),resourcematchingfactor
6
(RMF)andRSS.Itisassumedthataplayoutdelaybufferisusedtode-jitterthepacketvoicestreamand,hence,jitterisnotconsidered.Securityrelatedattributesuchassafetyornon-safetyserviceshasalreadybeentakencareofinthepre-linkselection.Theattributesareclassifiedintoupwardanddownward.Theattributes whichhavethehigherpreferencerelationinfavourofthehighervaluesarecalledupwardattributes,i.e.,thehighertheattributevaluethebetter.Ontheotherhand,downwardattributesareinfavourofthelowervalues,i.e.,thelowerthevaluethebetteritis[10].Forexample,BRandRSSareupwardattributeswhilePD,BER,CSTandRMFaredownwardattributes.Frombothoperator’sanduser’spointofviews,highervaluesofupwardattributesandlowervaluesofdownwardattributesarepreferred.Inthisregard,anadjustmentofthedownwardattributesintoupwardattributesisrequiredduringnormalizationprocessasexplainedinthefollowingsub-section.Thedefinitionandimportanceofeachattributearedescribedbelow:
Maximumbitrate(BR): Thisupwardattributemeasuresthemaximumtransferofbitsdeliveredperunitoftimeanditistheupperlimitanapplicationbeingprovidedfromadatalink.
Packetdelay(PD):Thisdownwardattributeindicatesthemaximumdelayfordeliveringpackets withinthedatalink,whichismeasuredinmilliseconds.
Biterrorrate(BER):Thisdownwardattributeistheratioofthenumberoferrorbitstothetotalnumberoftransmittedbits.Itisanimportantparameterinmeasuringtheperformanceofdatalinksandisusedtoconfigureradiointerfaceprotocols,algorithmsanderrordetectioncoding.
Resource matchingfactor(RMF): Itisdefinedasthedifferencebetweentheaverageoftheofferedbitrateperuserandtherequiredbitrate.Thisindicateshowcloselyfitstherequiredbitrateintothelink,i.e.,thelowertheRMFvaluethebetter match.Itwillprovideanimprovedresourceutilizationofthesystem.Therefore,thisattributeisconsideredasadownwardattributeandcanbedefinedas:
(1)
where, =averageofferedbitrateperuserineachdatalink,
=maximumbitrateineachdatalink,=occupiedbitrateineachdatalink,
=maximumnumberofsupportedusersbyeachdatalink,=currentnumberofuseroccupiedbyeachdatalink.
and, =requiredbitrateofarequest.Receivedsignalstrength(RSS): Thisupwardattribute
indicatesthelinkqualityandisa measureofpowerinareceivedsignalin .
Linkcost(CST): Thisdownwardattributeindicatestheaveragecostofusingadatalink,anditcanplayanimportantroleinthelinkselectionprocess.Theaveragecostvalueforeachdatalinkisderivedbasedontheavailableusagepackagecostsofferedinthemarket.
TableIIrepresentsthe measuresofeveryattributesfortheconsidereddatalinksandthevaluesarechosenbasedonthespecifications[15].Sincethe RMFand RSSattributesaredynamicattributesandtheirvalueschangeovertimedependingupon(1)andthechannelconditionoftherespective
datalink,respectively,theircorrespondingvaluesintheTableIIareleftblank.
B. Normalization
Differentattributeshavedifferent measurementunits,sonormalizationisanecessarystepinthe MADM-basedmethodtoavoid anomalyinthe decision making process. Thenormalizationmethodisusedtoscaledifferentcharacteristicsofdifferentunitstoacomparablenumericalrepresentation.Foragivenattribute , representsthevalueofthisattributeinthethdatalink,and representsitsnormalizedvalue,where isthenormalizedfunction.Thereareseveralnormalization methodssuchas Max-Min,Sum,Square-RootandEnhancedMax-Min[10].However,thefirstthreemethodsdonotconsiderthedifferencebetweenupwardanddownwardattributes whiletheenhanced Max-Minadjustsdownwardattributesintoupwardattributes. Therefore,theoutputoftheenhanced Max-Min methodareallconsideredasupwardattributesandtheenhanced Max-Min methodisadoptedinthispaperasfollows:
forupwardattributes
fordownwardattributes(2)
C. Modellingthe Weights
The weighting of an attributerepresentsitsrelativeimportanceinregardstootherattributesanditdetermineshowdifferentcandidatelinksareranked.Asindicatedinprevioussections,there aretwo broad categories of determiningweights:objectiveandsubjective methods. Thesubjectiveweighting methodisrelatedtouserexperiencetoobtainsubjectiveweights,whiletheobjective methodforobtainingobjectiveweightsdonotconsideruserpreference.
Theobjective weightsarecalculateddirectlybasedontherelativedifferencebetweenattributes,givenby forattribute . Therearetwocommon methodsforobjectiveweightingsuchasentropy-basedandvariance-based[10].Theentropy-based methodisnotsuitablefordatalinkselectionproblembecauseitgiveshigherweightstotheattributesthathavesimilarvaluesamongalllinksandlowerweightstothoseattributeswithvaluesvaryingacrossdifferentlinks[9].Theoptimaldatalinkselectionneedstogivehighweightstotheattributesthatcandistinguishonelinkfromothers.Therefore,thevariance-based methodisadopted,whereifoneattributehasexactlythesamevalueineverylink,itsweightissetto.Thevariance-basedobjectiveweightingisgivenby[10]:
(3)
where:
(4)
isthenumberofattributes,and isthetotalnumberofdatalinks.
7
Onthecontrary,subjectiveweightsareusuallycalculatedbasedonthesubjectivefeelingsofthedecisionmakertakingintoaccountsomesubjectiveinformationsuchascustomerpreferences,operatorpoliciesandsoonasdirectinputs[11].Assumingthesubjectiveweightvector, canbeobtainedbasedonthatinformationusingasubjective weightingmethod,andthevector, canberepresentedas:
(5)
Thesubjectiveweight ,where canbeobtainedbyusingthewidelyusedpair-wisecomparisonmatrixBcontainingallthecomparisonvalues betweenthethandthethattributes.Sinceitspair-wisecomparisonisaslowprocessandnotautomatic,theTRUSTmethod[11]isproposedtocalculatethesubjectiveweights.ThemotivatingpartoftheTRUSTmethodisitsautomaticcomputationofsubjectiveweightsinacomparativelyfastandefficientway.However,ithassomelimitationsexplainedinSectionI.TocompensatethelimitationoftheTRUST method,anewi-TRUSTmethodisproposed,whichtakesadvantageoftheautomaticsubjectiveweightingfromTRUSTandcapturesthesubjectivepreferencesbyusers/operators.Inthei-TRUST,theparticularrequirementsfromauserwiththeirrelativeimportancearegivenbythefollowingvector:
(6)
where isthenumberofrequirements.Thevalues isintherangebetween and ,where
.If ,thecorresponding threquirementisdemandedwithitsimportancelevelbetweenand , whileif ,thereisnodemandforthe
correspondingrequirement. Moreover, indicatesthehighestimportantwhereas indicatestheleastimportantrequirement whileintermediatevaluesindicate mid-levelimportancebetweenthehighestandthelowest.Abinaryvector isnowdefinedfrom where
fornon-zeroelementsin .Adiagonalmatrixisthengeneratedfrom as:
......
... (7)
where and .Arequirement-to-attributecorrespondencematrix, isdefinedbasedonthepre-definedrelationshipbetweenrequirementsandattributesdefinedinTableIII(inSectionV)as:
......
... (8)
where indicatestheeffectofthethrequirementonthethattribute,i.e.,either ,or .Beforeconsideringthesubjectivepreference,abaseorlocalweightvectorisusedas:
(9)
where isthebaseweightofthethrequirementandthesearemanuallysetinadvancebytheoperatororthealgorithmdesignerwithpossiblyusingeigenvectormethodplusanalytichierarchyprocess(AHP).TableIIIrepresentstherelationshipbetweentriggereventswithsuchbasesubjectiveweightvectorandthisisstoredintheaircraftterminal(i.e.,TN).Inordertoreflecttherelativeimportanceoftherequirementsbytheuser,anewbasesubjectiveweightingvector isobtainedbyusing(6),(7),(8)and(9)as:
(10)
where isthe element-wise multiplicative operator.Moreover,inordertoavoidtheundesirablesituationofunderminingcertainattributesintheTRUSTmethod,ascalarvalue isaddedto turningallzeroelementsintonon-zerosdenotedby .Thefinalsubjectiveweightingvectorcanbeobtainedas:
(11)
where isthenormalizationfunctionandtheweightofthethattributeisobtainedas:
(12)
Theobjectiveandsubjectiveweightsarethencombinedintoasingleweight, ofthe thattributeusing(4)and(12)as:
(13)
Thesecombinedweightsarethenusedtorankthecandidatelinkstoobtainthetargetdatalinkasdescribedbelow.
D.Ranking-ObtainingtheOptimalDatalink
MADMalgorithmscanbeoftwocategories:compensatoryandnon-compensatoryalgorithms.Compensatoryalgorithmcombines multipleattributestofindthebestalternativewhereas the non-compensatory algorithm is used tofindacceptablealternatives, whichsatisfythe minimumrequirements, but may not bethe optimal one[18].Compensatory MADMalgorithmsare more widelyusedandtheseincludealgorithmslikesimpleadditiveweighting(SAW), multiplicativeexponentialweighting(MEW),grayrationalanalysis(GRA)andTechniqueforOrderPreferencebySimilaritytoanIdealSolution(TOPSIS)[18].AmongstthemtheSAWmethodiscommonlyusedforthenetworkselectionproblemduetoitssimplicity[19]andhenceisadoptedinthispaper.TheutilityscoreofthethdatalinkbyemployingtheSAWmethodisgivenby:
(14)
where isthe weightvectorofattributes, ,and .Clearly,animportantstepoftheSAWoperatoristocomputetheweightsofallthegivenattributevalues,whichcanbe
8
obtainedby(13),andthenaggregatetheseweighedattributesbyaddition.Finally,thebestlinkconfigurationisobtainedbyrankingthemandselectingtheonethathasthehighestutilityscore, andthisisgenerallyobtainedby:
(15)
where istheselectedandthebestdatalinkforagivenrequestatanyinstant.
V.PERFORMANCEEVALUATIONANDDISCUSSION
A.SimulationScenariosandParameters
Toevaluatethe performance ofthe proposed optimaldatalinkselectionalgorithmforaeronauticalcommunications,ascenarioisdefinedwhereanaircraftisequippedwiththeIMCterminal with multipledatalinksandusers, who makerequests,areallowedtoprovidepreferences.The BR,PD,BER,andRSSattributesareaffectedbytheuserchoiceofthequalitypreferencewhiletheusercostpreferenceaffectstheCSTattributeaccordingtoTableIII.Bothreal-time(RT)andnon-RT(NRT)applicationsaswellassafetyandnon-safetyservicesareconsidered.Thesafetyandnon-safetyservicesarecategorizeddependinguponthedomainoftherequestedsessionfromdifferenttrafficdomainssuchascockpit,cabinandpassengers.Inthissimulation,arandomnumberofapplicationrequestswithdifferentrequirementsisgeneratedandthedataraterequirementsareuniformlydistributedwiththeminimumandmaximumvaluesof1.5kbpsand256kbps,respectively,whilethedatalinkselectionisperformedforeachrequest.Thesessionarrivalratereferstothenumberofnewapplicationsrequestedperunittime.Foreacharrivalrate,thesimulationisrunfor100timesandtheaverageoftheresultsgeneratedfromeachsimulationistakenasthefinalresultforthesimulation.Itisworthmentioningthatthealgorithmalsoprovidestheflexibilitytotheoperatortotunethebaseweight,
inTableIIItosetprioritydependingupontheflightclassinselectingthedatalink.
Aunifiedscenarioconsideringacombinationofpossibleapplicationrequirements,userandoperatorpreferencesshowninTableIVisconsideredtohelpexplaintheproposedlinkselectionalgorithmforitsperformanceevaluation. Onthedatalinkside,threetypesofdatalinks,i.e.,VDL2,IridiumandBGANandsixattributeshavebeenconsideredasillustratedinTABLEII.OneVDL2,oneIridiumandtwoBGANdatalinks(explainedin SectionII.Aanddenotedby BGAN1andBGAN2)areused.Twodifferentapplications(RTandNRT)withdifferentdataraterequirementsareassumed.Tensessionrequests withacombinationofdifferentuserpreferences,suchasmoneyfirstandqualityfirst,andsafetyornon-safetyservicesareconsidered,asshowninTABLEIV.BasedontheabovestudiesinSectionIV.Cforselectingtheoptimaldatalink,theenhanced Max-Minmethodisusedfornormalization.Thevariancebased methodisusedforobjectiveweightingwhilethei-TRUSTmethodisusedforsubjectiveweighting.Inthebaselinealgorithm,thepredefinedlinklistdefinedinSectionIII.Ahasbeenusedtoselectthetargetdatalinkforagiven
request withoutconsideringanypreferences.ThevaluesofattributesforeachdatalinkdefinedinTABLEIIareusedfornumericalanalysis.Thereceiversensitivityof-98dBm(VDL2),-125dBm(BGAN),-115dBm(Iridium-voice)and-112dBm(Iridium-data)areconsideredaccordingtothecorrespondingspecifications[20]–[22].
B.ResultsandDiscussion
Firstly,the proposed i-TRUST methodfor computingsubjectiveweightshasbeenvalidatedandcomparedwiththeexistingTRUST methodusing Matlabsimulation.Sincetheeigenvector method[23]iscommonlyusedforcomputingsubjectiveweightsbecauseofitsaccuracy,butaslowprocessduetoitsuseofpairwisecomparisons,theideahereistofindwhichmethodproducestheweightvaluesclosesttothosecomputedbytheeigenvector method.Thesubjectiveweightvector, employingtheeigenvector methodisobtainedbyusingpair-wisecomparisonplusAHP.Dependingontheindividualrequest,bothTRUSTandi-TRUST methodsuseTableIIIfortriggeringthecorrespondingattributesaccordingtothepreferredeventsmarkedbycrosses(x)thatallowtoformthe matrixandevaluatesallproceduresin SectionIV.C.Thesubjectiveweights canthenbecomputedby(11).Inordertoseethecomparisonbetweenthesetwomethods,Fig.5showsthecorrelationofthecomputedsubjectiveweighsbybothTRUSTandi-TRUST methodsanditisclearlyevidentthati-TRUSTprovides moreaccurate weightsthanTRUST.Thisisduetothefactthatthei-TRUSThastakentheuserpriorityinconsiderationwhilecomputingtheweights.Fortherestoftheresultsinthispaper,thei-TRUSTwillbeusedforcomputingthesubjectiveweights.
Thedetailedproceduretoobtaintheoptimaldatalinkusingtheproposed methodforalltheconsideredunifiedscenarios(TableIV)isdescribedandcompared withthebaselinealgorithm.ThesummaryofalltensessionrequestswiththeselectedlinkforeachsessionbyemployingboththeoptimalandbaselinealgorithmsisshowninTable V.Theselecteddatalinkforeachrequestbytheproposedoptimalalgorithmisdifferentfromthatbythebaselinealgorithmandthesearethebestselection.
Thedetailedoptimallinkselectionprocessforfiveselectedsessionrequests(RID#1,4,5,6and10)willbediscussedandisillustratedinTABLE VI.Thelinkcharacteristicsshowninthetablearebasedon TableII,equation(1)andtheinstantaneousreceivedsignalstrengthoftheassociatedlink.For RID#1,only BGANlinks(BGAN1and BGAN2)aretreatedascandidatelinksduringthepre-linkselectionprocess.BothVDL2andIridiumwerenotconsideredascandidatelinksbecauseIridiumdoesnotsatisfythedataraterequirementandVDL2onlyprovidesnon-safetyservices.Sinceallattributesexceptthe RSShavethesamevaluesforbothlinks,theoptimalalgorithmgavehigher weightontheRSSattribute,i.e.,0.50793.Finally,thealgorithmcomputestheutilityscoresforbothlinksandselectstheBGAN2,whichhasthehighestutilityscoreandthisisduetothebetterRSSvalueoftheBGAN2link, i.e.,-82dBmhigherthanthatof BGAN1(-116dBm).ForRID#4,theuserpreferredthecheapestlink
9
TABLEIIIRELATIONSHIPBETWEENTRIGGEREVENTSANDSUBJECTIVEWEIGHTSOFATTRIBUTES.
Eventsand Weights AttributesLevel1 Level2 Baseweights, BR PD BER RMF RSS CST
ApplicationsReal-time(RT) 0.25 x xNon-RT 0.20 x
Customerpreferences Lowprice 0.30 xOperatorpreferences Loadcondition 0.15 xDynamicattributes Signalstrength 0.10 x
TABLEIVUNIFIEDSCENARIOSFOR10SESSIONREQUESTS.
RequestID(RID)
Dataraterequire-ments(kbps)
Applications Users OperatorsReal-time
(RT)Non-RT
Qualityfirst
Moneyfirst
Safetyfirst
1 20
2 15
3 1.5
4 12
5 2
6 64
7 32
8 256
9 128
10 256
Fig.5. Comparisonresults(correlationofweights)betweeneigenvector,TRUSTandi-TRUSTmethods.
TABLEVOPTIMALLINKSELECTIONBYEMPLOYINGTHEPROPOSEDMADM-BASEDALGORITHM WITHUSINGTHEI-TRUSTSUBJECTIVEWEIGHTING
METHOD.
RequestID
Applicationtype
Safety/Non-safety
Dataratereq.
(kbps)
Qualityimpor-tance
Priceimpor-tance
Optimal BaselineRSS
(dBm)Targetlink
RSS(dBm)
Targetlink
1 RT Safety 20 9 1 -82 BGAN2 -100 BGAN12 RT Non-safety 15 9 1 -94 BGAN1 -98 VDL23 RT Safety 1.5 1 9 -80 BGAN2 -80 BGAN24 RT Non-safety 12 1 9 -84 VDL2 -123 BGAN15 NRT Safety 2 9 1 -89 IRIDI -113 BGAN26 NRT Non-safety 64 9 1 -83 BGAN1 -83 BGAN17 NRT Safety 32 1 9 -82 BGAN2 -82 BGAN28 NRT Non-safety 256 1 9 -93 BGAN1 -93 BGAN19 RT Safety 128 9 1 -76 BGAN2 -76 BGAN210 NRT Non-safety 256 1 9 - Dropped! - Dropped!
10
TABLEVIDETAILEDOPTIMALLINKSELECTIONPROCESSFORFIVESELECTEDREQUESTS.
Requestrequirements(RID:1): Requestrequirements(RID:5):Datarate : 20 Datarate : 2.0Application : Real-Time(RT) Application : Non-Real-Time(NRT)Safety/Non-safety : Safety Safety/Non-safety : SafetyQualityimportance : 9 Qualityimportance : 9Priceimportance : 1 Priceimportance : 1
Linkcharacteristics: Linkcharacteristics:BR PD BER RMF RSS CST BR PD BER RMF RSS CST
BGAN1 32.0 950.0 10 5 24.73 -100 3.0 IRIDI 2.10 750.0 10 6 0.40 -89 7.0BGAN2 32.0 95.0 10 5 24.73 -82 3.0 BGAN2 32.0 950.0 10 5 24.73 -113 3.0
Weights: Weights:BR PD BER RMF RSS CST BR PD BER RMF RSS CST
W= 0.12577 0.12580 0.02510 0.10040 0.50793 0.11500 W= 0.01257 0.01260 0.50730 0.05030 0.35969 0.05760
Utilityscores: 0.246038(BGAN1) 0.753962(BGAN2) Utilityscores: 0.929833(IRIDI) 0.070167(BGAN2)Optimallink: BGAN2 Optimallink: IRIDI
Requestrequirements(RID:4): Requestrequirements(RID:6):Datarate : 12 Datarate : 64Application : Real-Time(RT) Application : Non-Real-Time(NRT)Safety/Non-safety : Non-safety Safety/Non-safety : Non-safetyQualityimportance : 1 Qualityimportance : 9Priceimportance : 9 Priceimportance : 1
Linkcharacteristics: Linkcharacteristics:BR PD BER RMF RSS CST BR PD BER RMF RSS CST
VDL2 31.5 1200 10 5 19.5 -84 1.0 BGAN1 64 950 10 5 2102 -83 3.0BGAN1 32.0 950 10 5 30.8 -123 3.0
Weights: Weights:BR PD BER RMF RSS CST BR PD BER RMF RSS CST
W= 0.086 0.0861 0.0239 0.0675 0.05594 0.6808 W= 0.01257 0.0126 0.5073 0.0503 0.35969 0.0576
Utilityscores: 0.81607(VDL2) 0.18393(BGAN1) Utilityscores: 1.0(BGAN1)Optimallinks: VDL2 Optimallink: BGAN1
Requestrequirements(RID:10):Datarate : 256Application : Non-Real-Time(NRT)Safety/Non-safety : Non-safetyQualityimportance : 1Priceimportance : 9
REQUESTDROPPED!
withoutcaringaboutthelinkquality.Inthiscase,theoptimalalgorithmgavemoreweighttothecost(CST)attributethanothersandfinally,itselectstheVDL2linkwhichhasthehighestutilityscore.Twodatalinks(IridiumandBGAN2)werepre-screenedascandidatelinksforthefifthrequest(RID#5).TheBERandRMFattributesweregiventhehigherweightvaluesthanothersduetotherequestrequirements,i.e.,theNRTapplication,whichmainlyaffecttheBERattributeandtheRMFattributeisfortheoperator’schoiceofinteresttoenhancetheoverallsystemutilization.ForRID#6,thealgorithmfoundonlyonecandidatelinkduringthepre-linkselectionprocess,sothishasbeenselectedastheoptimallink.Therequestedsession,RID#10hasbeendroppedduetotheunavailablebandwidthofthelinksinbothcases.
Thecalldroppingprobabilityisdefinedastheprobabilitythatcertainsessionrequests(orcallrequests)areblockedduetolackofresources(i.e.,iftherearenotavailabledatalinksforservingthosesessions)andthisisusuallycalculatedusingErlang-Bformula.Fig.6showsthecalldroppingprobabilitywithrespecttodifferentarrivalrate,λinordertovalidatethesimulationandtheoreticalresults.Itisseenthatthecalldroppingprobabilityincreasedwithλ
0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Arrival rate, 6
10-8
10-7
10-6
10-5
10-4
10-3
10-2
10-1
100
Call dropping probability
TheorySimulated
asexpected.However,
Fig.6. Calldroppingprobabilityfordifferentarrivalrates.
thereisa minorvariationforlowerarrivalratesbetweenthesimulatedandtheoreticalresults.ThisisattributedtotheintroductionoftheRMFattributeintheproposedmethod,whichoffersbetterdatalinkutilizationefficiency,resultingin
0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.50
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1Avearge relative cost per unit data
Arrival rate, λ
Lowest cost priority (a)Highest cost priority (b)
11
0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.50
100
200
300
400
500
600
700
800
900
Avearge throughput [kbps]
Arrival rate, λ
Lowest quality priority (a)Highest quality priority (b)
(a)
(b)
Fig.7.Illustratingtheuserpreferenceforimproving(a)theservicecostand(b)thequalityperformance.
lesscallsbeingdropped.Athigherλ,theimpactoftheRMFattributebecomeslessandhence,thesimulationresultisclosertothetheory.1)Costandthroughputanalysis:Fig.7(a)showsthecost
performanceoftheproposedalgorithmfromtheuser’spointofview,i.e.,howtheuserpreferenceaffectstheperformance.Theresultsshowtheaveragerelativecostsperunitdatausagefordifferentarrivalrateswhilesettingthehighestandthelowestcostprioritybytheuser.Itisseenthatwhentheusergavethehighestprioritytothecostparameter(CST),theuserwasabletosavenoteworthydatausagecosts.Forexample,over36%ofsavingcanbeachievedbysettingthehighestcostprioritycomparedtothatbysettingthelowestcostpriorityatlowerλ.However,therelativesavingdecreasedwiththeincreaseofλduetotheloweravailabilityofthecheapestdatalinkatthehigherλ.Similarly,theuserpreferenceforthequalityattributesshowsthesimilarresultsintermsofanimprovedaveragethroughput(upto12%)asshowninFig.7(b),wheretheusersprefereitherthehighestorlowestpriorityforthequalityattributeswhilethecostprioritywaschosenrandomly.2)Resourceutilizationanalysis:
0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.50
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Arrival rate, λ
Resource utilization
Baseline algorithmOptimised algorithm
Fig.8illustratestheperformanceofresourceutilizationofbothbaselineandproposedoptimalalgorithms.Intheoptimalalgorithm,theRMFattributehasbeenconsideredinordertoimprovethe
−200 −150 −100 −50 00
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
RSS [dBm]
CDF
Baseline algorithmOptimised algorithm
Fig.8. Resourceutilizationfordifferentarrivalrates.
Fig.9. Cumulativedistributionfunctionofthereceivedsignalstrength.
overallsystemresourceutilization.Theresultsdemonstratethattheresourceutilizationincreasedwithλinbothcasesandtheoptimisedalgorithmalwaysperformedbetter. Whenthearrivalrateislow,animprovementofupto81%intheutilizationcanbeachievedwiththeproposedalgorithmscomparedtothebaseline.ThisisduetothefactthattheRMFattributeplaysaroletoimprovetheresourceutilizationbyfittingtherequestedbitraterequirementswiththelinkbandwidth.Therewerefastincreaseintheresourceutilizationofthebaselinecaseasλincreases.Itisimportanttomentionthattheresourceutilizationofthebaselineapproachedtothatoftheoptimisedonewhenthearrivalrateincreased.
3)Userexperience:Todemonstratethecomparativeuserexperiencebyusingtheproposedoptimalalgorithm,thecumulativedistributionfunction(CDF)ofthereceivedsignalstrengthandtheaveragepacketdelayexperiencedbyeachuserareshowninFig.9andFig.10,respectively.TherequestedsessionshavebeengeneratedrandomlyandthesamebaseweightasinTABLEIIIareusedforcomputingsubjectiveweights.ItcanbeseenfromFig.9(theCDFofRSS)thatthehighernumberofusersexperienceshigherreceivedsignalstrengthwiththeoptimisedalgorithmthanthatwiththebaselinealgorithm.Forexample,the median,i.e.,0.5onthey-axisforbothalgorithmsshowsthat50%ofthe
0 200 400 600 800 1000 12000
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Average Packet Delay [ms]
CDF
Baseline algorithmOptimised algorithm
760770780790
0.18
0.2
0.22
0.24
0.26
12
Fig.10. Cumulativedistributionfunctionoftheaveragepacketdelays.
requestedsessionsexperiencesthesignalstrengthof-95dBmorhigherand-75dBmorhigherwhenemployingthebaselineandoptimalalgorithms,respectively.Theimprovementisderivedfromtheintelligentdecision makingprocessoftheoptimisedalgorithmbasedonmultipleattributevalues,wheretheproposedi-TRUSTsubjectiveweightingmethodinthe MADM-basedschemefurtherensuresthattheselecteddatalinkisabletoofferthegoodqualitysignalevenifallotherlinkcharacteristicsarethesame.Ontheotherhand,thebaselinealgorithmonlyconsidersapre-definedlist.Similarly,theCDFoftheaveragepacketdelayforthesamesettings(Fig.10)alsoconfirmsthattheoptimisedalgorithmperformsbetterthanthebaselinealgorithm(animprovementof 2%).
VI.CONCLUSION
Inthispaper,anintegratedmodularcommunicationsystemforanaircraftterminalequippedwithanoptimaldatalinkselectionalgorithmisproposed.Inordertoselectanoptimaldatalinkfromamulti-radioterminal,anMADM-baseddatalinkselectionalgorithm withanintelligentsubjectiveweighting method(i-TRUST)isproposed.Thealgorithmprovidesgreaterflexibilitytobothusersandoperatorintermsofsettingtheirpreferenceswiththecapacityofautomaticdecision making. Resultsdemonstratethattheproposedi-TRUSTmethodprovidessubjectiveweightvaluesascloseastheeigenvectormethod,whichvalidatedtheaccuracyofcomputingsubjectiveweightsascomparedtotheexistingTRUSTmethod.Finally,thedetailedanalysisoftheproposedMADM-basedmethodhasbeendiscussednumericallythroughregoroussimulationanalysis,whichshowstheimprovementofthesystemperformanceintermsofcost(over36%),throughput(upto12%)andresourceutilization(upto81%).Theuserexperienceofusinghighqualitydatalinkisalsoimproved withoutcompromising withthelinkcost.ItisexpectedthattheproposedMADM-basedalgorithmtogethertheproposedi-TRUSTsubjectiveweightingmethodcanbeemployedinotherubiquitousnetworks.
ACKNOWLEDGMENT
ThisworkispartiallyfundedbytheInnovateUKprojectSINCBAC-SecureIntegrated NetworkCommunicationsfor
Broadbandand ATM Connectivity: Application number18650-134196.
REFERENCES
[1]EUROCONTROL,“Fourthstudyon’Challengesofgrowth2013’,”pp.14–30,2013.[Online].Available:https://www.eurocontrol.int/articles/challenges-growth
[2]F.Schreckenbach, M.Schnell,S.Scalise,andP.Piatt,“NEWSKY-Networkingtheskyforaeronauticalcommunications,”inProceedingsoftheIntegratedCommunications,NavigationandSurveillance(ICNS)Conference,2007.
[3]Eurocontrol/FAA, “Communications operating concept andrequirementsforthefutureradiosystem,”inCOCRv2.0,2002.
[4]C. Zhang, Y. Zhang,J. Xiao,andJ. Yu,“Aeronauticalcentralcognitivebroadbandair-to-groundcommunications,”SelectedAreasinCommunications,IEEEJournalon,vol.33,no.5,pp.946–957,2015.
[5]The European Commission, SANDRA: Seamless aeronauticalnetworkingthroughintegrationofdatalinks,radiosandantennas.GrantAgreementNo.233679,ACP8-GA2009-233679,2013.
[6]InnovateUK,SINCBAC:SecureintegratednetworkcommunicationsforbroadbandandATMconnectivity. ProjectNo.18650-134196,2015.
[7]C. Kissling, “Data link selection in mobile aeronauticaltelecommunication networks,” in 27th AIAA InternationalCommunicationSatelliteSystemsConference,2009,pp.1–8.
[8]L. Mohamed,L.Cherkaoui,andA.Adib,“Anintelligentnetworkselection strategy based on MADM methods in heterogeneousnetworks,”InternationalJournalofWireless&MobileNetworks,vol.4,no.1,pp.83–96,2012.
[9]L. Wang and D. Binet, “MADM-based network selection inheterogeneous wireless networks: a simulation study,” in 1stInternational Conference on Wireless Communication, VehicularTechnology,InformationTheoryandAerospace&ElectronicSystemsTechnology,2009,pp.1–6.
[10]L.WangandG.-S.Kuo,“Mathematicalmodelingfornetworkselectioninheterogeneous wirelessnetworks-atutorial,”CommunicationsSurveys&Tutorials,IEEE,vol.15,no.1,pp.271–292,2013.
[11]L. WangandD.Binet,“TRUST:atrigger-basedautomaticsubjectiveweightingmethodfornetworkselection,”inProceedingsofthe5thAdvancedInternationalConferenceonTelecommunications(AICT),2009,pp.362–368.
[12]L.Li,X.Chu,andJ.Zhang,“Ahierarchical MADM-basednetworkselectionschemeforsystemperformanceenhancement,”inIEEEInternational Conference on Communication Systems, 2014, pp.122–126.
[13]“Airtrafficmanagement.”[Online].Available:https://www.eurocontrol.int/articles/what-air-traffic-management
[14]F.BariandV.Leung,“Automatednetworkselectioninaheterogeneouswirelessnetworkenvironment,”IEEENetwork,vol.21,no.1,pp.34–40,2007.
[15]ETSI, “VHF air-ground digital link (VDL) mode 2, Technicalcharacteristics and methods of measurement for ground-basedequipment,Part1:PhysicallayerandMACsub-layer,”pp.1–53,2010.
[16] Aiaa,“AFrameworkforaviationcybersecurity,”AnAIAADecisionPaper,no.August,pp.1–16,2013.
[17] AeronauticalRadioInc.,CommercialAircraftInformationSecurityConceptsofOperationandProcessFramework. ARINCSpecification811,Dec.2005.
[18] X.Gelabert,J.Pérez-Romero,O.Sallent,andR.Agustí,“AMarkovianapproach to radio access technology selection in heterogeneousmultiaccess/multiservice wirelessnetworks,” IEEETransactionsonMobileComputing,vol.7,no.10,pp.1257–1270,2008.
[19] O.Ormond,J.Murphy,andG.M.Muntean,“Utility-basedintelligentnetworkselectioninbeyond3Gsystems,”inIEEEInternationalConferenceonCommunications,2006,pp.1831–1836.
[20] Nick Witt and I. Herail, “Generic requirements for anATN/VDL Mode2air/groundcommunicationsserviceprovider,”EUROCONTROL/VFC/Datalink/ACSPGEN,EditionNo.1.009,2013.
[21]ThraneandT.Vehicular,“BGANTerminal,TT-3712AExplorer527DD1494Information,”Tech.Rep.,2007.
[22]“Iridium - technical details.” [Online]. Available: http://www.decodesystems.com/iridium.html
[23] H.Pervaiz,Q.Ni,andC.C.Zarakovitis,“UseradaptiveQoSawareselection methodforcooperativeheterogeneouswirelesssystems:adynamiccontextualapproach,”FutureGenerationComputerSystems,vol.39,pp.75–87,2014.