基于自适应神经模糊推理系统的路径损失预测模型的建立(IJWMT-V9-N6-5)

I.J.WirelessandMicrowaveTechnologies,2019,6,40-53PublishedOnlineNovember2019inMECS()DOI:10.5815/ijwmt.2019.06.05Availableonlineat:20August2019;Accepted:15October2019;Published:08November2019AbstractThepredictionofwirelesscommunicationsignalsisofparamountimportanceforpropernetworkplanning.TheexistingpredictionmodelssuchasOkumura-Hata,Co-operativeforScientificandTechnicalResearch(COST-231)andfreespacearelessaccurateforpredictingpath-lossvaluesofwirelesssignalsduetodifferencesinpropagationenvironments.Hence,thispaperdevelopsapath-lossmodelusingAdaptiveNeuro-FuzzyInferenceSystem(ANFIS)foraccuratepredictionofwirelessHighSpeedPacketAccess(HSPA)networksignalinIbadan,Nigeria.ThisisachievedbymeasuringtheReceivedSignalStrength(RSS)fromthreeBaseTransmittingStations(BTS)operatingat2100MHzfrequencyinOjo(longitudeE3’53.1060’,latitudeN7’27.2558’),Dugbe(longitudeE3’50.4361’,latitudeN7’23.0678’)andChallenge(longitudeE3’53.1060’,latitudeN7’21.258’)areasofIbadanusingtheDriveTest.EricsonTestEquipmentforMobileSystem(TEMS)phone,GlobalPositioningSystem(GPS)andComputerSystemareusedtoobtainRSSdataatdifferentdistances.Basestationparameterssuchasthetransmittingantennaheight,receivingantennaheight,carrierfrequencyanddistanceareusedasinputvariablestotrainANFIStodevelopamodel.ThesebasestationparametersarealsousedtoinvestigatethesuitabilityofOkumura-Hata,COST-231andfreespacemodel.AfivelayerANFISstructureisdevelopedandtrainedusingLeastSquareError(LSE)andGradientDescent(GD)methodtoadjusttheconsequentandpremiseparameters.TheperformanceofthedevelopedANFISmodelisevaluatedusingMeanSquareError(MSE)andRootMeanSquareError(RMSE)andcomparedwithOkumura-Hata,COST231andfreespace.TheresultsobtainedforANFISgivelowerRMSEandMSEindicatingthesuitabilityofANFISmodelforpath-lossprediction.ThedevelopedANFISmodelcanbeusedfornetworkplanningandbudgetingintheseenvironments.IndexTerms:HSPA,ANFIS,LeastSquareError,GradientDescentmethod©2019PublishedbyMECSPublisher.Selectionand/orpeerreviewunderresponsibilityoftheResearchAssociationofModernEducationandComputerScience*Correspondingauthor.Tel:[phone]E-mailaddress:[e-mail],[e-mail]-lossPredictionModelUsingAdaptiveNeuro-fuzzyInferenceSystem411.IntroductionWirelesscommunicationisthetransmissionofsignalsoverawirelesschannelthathasnophysicalconnectionbetweenthetransmitterandthereceiver[1].Thereisreductioninsignalstrengthasthedistanceincreasesindicatingthelossinsignalalongthepath.Todeterminetheactualpath-loss,path-lossmodelshavebeendevelopedforpredictingthevalues.Thesemodelsaremathematicalexpressionsusedtopredicttheradiosignalcharacteristicsofaplace[2].Themodelsarecategorizedintothree;random,deterministicandempiricalmodels[1,7,9].Randommodelsgeneratepredictionsrandomlywithoutproperconsiderationsofallthephysicalenvironmentalphenomenawhichmakeitunreliablewhenappliedtoenvironmentthathascomplexstructureandmanyobstacles[9,12].Deterministicmodelsrelyonrayopticallawsforpredictionandarecomputationallycomplexandtimeconsuming[11,12].Althoughthesemodelsseemtobeaccuratethanotherexistingmodelbutaredifficulttoimplementespeciallyinenvironmentswheretherearemanyobstaclesalongthesignalpaths.Ontheotherhand,Empiricalmodelsaredependentontheenvironmentwherethemeasurementistaken.Tomakethesemodelsapplicabletootherenvironments,optimizationapproachmustbecarriedoutonthemodelsforproperadaption[3,8,12,18,19].Themeasureddataisusedasreferenceforcomparisonwithoptimizedmodels.Amajorlimitationofthemodelsislargedeviationwhencomparedtothemeasureddata.Therefore,thispaperinvestigatestheapplicabilityoftheexistingmodelsinIbadan,NigeriaandthedevelopmentofanewmodelusingAdaptiveNeuro-FuzzyInferenceSystem(ANFIS).Theinvestigationiscarriedoutusingthedrivetestmethod.DrivetestequipmentsuchasTestEquipmentforMobileSystem(TEMS),acomputersystemandaGlobalPositioningSystem(GPS)areusedtocarryoutthemeasurementalongthreedifferentlocationsinIbadan,Nigeria.Path-lossvaluesobtainedfrommeasuredRSSareusedtodeterminethedeviationfrompath-lossvaluesoftheexistingmodels.AnANFISstructurewasdevelopedandtrainedwithtransmittingantennaheight,receivingantennaheight,carrierfrequencyanddistanceastheinputvariables.LeastSquareError(LSE)andGradientDescent(GD)methodareusedtoadjusttheconsequentandpremiseparameters.ThenewmodeldevelopedusingANFISiscomparedtothemeasuredpath-lossvalues.NomenclatureAIArtificialIntelligenceANFISAdaptiveNeuro-FuzzyInferenceSystemANNArtificialNeuralNetworkBTSBaseTransmittingStationsCOSTCo-operativeforScientificandTechnicalResearchFLFuzzyLogicGDGradientDescentGPSGlobalPositioningSystemHSPAHighSpeedPacketAccessLSELeastSquareErrorMSEMeanSquareErrorRMSERootMeanSquareErrorRSSReceivedSignalStrengthSMStochasticModelSUIStanford