简介:Baum-Welchalgorithmmostlikelyresultsinunderflowinpractice.Insomeliteratures,suchas"Scaling"algorithmwasintroducedtosolvetheproblem.Inapplications,however,somemistakeswerefoundintheequationspresentedintheseliteratures.Thepracticalcalculationsshowthattheoriginalalgorithmoftenresultsinpoororevennoneconvergenceandratherhighererrorrateinspeechrecognition.Themistakesintheseliteraturesandbringsforwardthecorrectequationsareanalysed.Thespeechrecognitionsystemusingtherevisedalgorithmcanconvergewellandhaslowererrorrate.
简介:Inthispaperamodeloftransvcrsalfilterispresentedtostudytheadaptivematchofthetimevariantchannel.Theleastmeansquareerrorfil-teringmethodisusedtoobtaintheweightingcoeffcientsofthefilter.Withthepurposeofspeedinguptheconvergenceoftheiterationequationofadaptivefiltering,anadaptivefactoroftheiterationstepsizeμ_aisderivedinthispaper.Theresultofcomputersimulationshowsthatinthecaseofusingadaptiveμ_a,theconvergencespeedoftheiterationequationisincreased2timesapproximatelyincomparisonwithconstantμ_f.Thestudysuggeststhattheadaptivefilterwithadaptiveμ_a.havetheperformancetofollowthechangeoftime-variantcharacteristicsofthechannel.
简介:Thequadratictransformationmethodisproposedtoestimatethetrabecularspacing(Tb.Sp),animportantindexforosteoporosisdiagnosis.Theperformanceofthisalgorithmwasinvestigatedbyscattermodel,two-dimensionfinitedifferencetimedomain(2D-FDTD)simulationandinvitroexperimentsofbovinecancellousbonespecimens.Comparingwiththeotherfourmethods—autoregressivecepstrum(AR),adaptivefilter-autoregressivecepstral(AFAR),inversefilter-autoregressivecepstrum(InvAR),andsimplifiedinversefiltertracking(SIFT),quadratictransformationismuchmorestableandaccurate.TheresultsdemonstratedthatquadratictransformationisagreatalgorithmforTb.Spestimation.
简介:Themodeloflinearfrequencymodulationcontinuouswave(LFMCW)appliedinunderwaterdetectionandthemethodforthedetectionofechosignalandtheestimationoftargetparameterswerestudied.Byanalyzingtheheterodynesignal,analgorithmwiththestructureofheterodyne-FractionalFourierTransform(FRFT)wasproposed.Toreducethecomputationofsearchingtargetsinatwo-dimensionalFRFTresult.theheterodynesignalwouldbeprocessedbyFRFTataspecificorder,afterRadon-AmbiguityTransform(RAT)wasappliedtoccstimatethesweeprateofthesignal.SimulationsprovedthatthealgorithmcaneliminatethecouplingphenomenonofdistanceandvelocityofLFMCW,andestimatetargets’parametersaccurately.ThelaketrialresultsshowedthattheprocessinggainofLFMCWprocessedbythealgorithminthispaperwas13dBbetterthanthatoftheLFMprocessedbymatchedfilter.TheresearchresultsindicatedthatthealgorithmappliedinLFMCWunderwaterdetectionwasfeasibleandeffective,anditcouldestimatetargets’parametersaccuratelyandobtainagooddetectionperformance.
简介:
简介:常规MVDR适应beamformer是用真实的声学的地的narrowbandCSDM估计最佳的beamforming重量的高分辨率的narrowbandbeamformer。在实际应用,MVDR算法需要长观察时间估计协变性矩阵。这个固有的性质使本地化移动迅速的目标困难。为宽带信号,MVDR算法需要转换增加计算要求的每CSDM。为相关来源,因为信号将取消对方,MVDR的性能将戏剧性地降级。一个快会聚的MVDR算法基于subbandsubarray处理被建议。完整的频率乐队被划分成subbands的集合,线数组被划分成subarrays的集合。为每subband,减少的尺寸的STCM是计算的。快会聚的MVDR算法和空间光谱评价的然后适应的beamforming重量被获得。同时,空间光谱评价能用两方面的空间变光滑方法为相关来源被做。模拟和试用数据的结果证明建议方法有高分辨率、在即时附近的集中性质,两方面的空间变光滑有decorrelation的令人满意的有效性。
简介:
简介:Amorerelaxedsufficientconditionfortheconvergenceoffiltered-XLMS(FXLMS)algorithmispresented.Itispointedoutthatifsomepositiverealconditionforsecondarypathtransferfunctionanditsestimatesissatisfiedwithinallthefrequencybands,FXLMSalgorithmconvergeswhateverthereferencesignalislike.Butiftheabovepositiverealconditionissatisfiedonlywithinsomefrequencybands,theconvergenceofFXLMSalgorithmisdependentonthedistributionofpowerspectraldensityofthereferencesignal,andtheconvergencestepsizeisdeterminedbythedistributionofsomespecificcorrelationmatrixeigenvalues.ApplyingtheconclusionabovetotheDelayedLMS(DLMS)algorithm,itisshownthatDLMSalgorithmwithsomeerroroftimedelayestimationconvergesincertaindiscretefrequencybands,andthewidthofwhicharedeterminedonlybythe"time-delayestimationerrorfrequency"whichisequaltoonefourthoftheinverseofestimatederrorofthetimedelay.
简介:Withtheincreasingofdetectionabilityofpassivesonar,theweaksignaldetectionprobleminmultipleinterferencesbecomesmoreandmoreimportant.Inthetime/bearingrecord(TBR)displayofsonardetection,whenthereexisttracesofmultipleinterferences,theidentificationofweaksignalisdifficultorimpossible.Theadaptivenoisecancellationtechniqueprovidesthetheoreticalbasisforsuppressingstronginterferences.Butthesolutionforfindingthesteady-stateoptimumfiltermatrixisquitedifficultduetotherealtimecalculationofinversematrixofinputdatacorrelationmatrix.Theiterativeinversebeamforming(IBF)algorithmforsolvingtheoptimumfiltervector,whichisexpressedbyinversematrixoftheoceanenvironmentdata,isderivedinthispaper,bywhich,theoptimumfiltercanbeeventuallyexpressedasasumofseriessimplematricesofconstructedfromsensordata.Basedonthealgorithmproposedinthispaper,someexamplesofatseaexperimentareprovided.Thestronginterferencesarecancelledandtheweaksignalisemerged,evenitdidn'tappearintheconventionalbeamforming(CBF)processing.
简介:
简介:Toaimattheproblemthatthehorizontaldirectivityindexofthevectorhydrophoneverticalarrayisnothigherthanthatofavectorhydrophone,thehigh-resolutionazimuthestimationalgorithmbasedonthedatafusionmethodwaspresented.TheproposedalgorithmfirstemploysMUSICalgorithmtoestimatetheazimuthofeachdividedsub-bandsignal,andthentheestimatedazimuthsofmultiplehydrophonesareprocessedbyusingthedatafusiontechnique.Thehigh-resolutionestimatedresultisachievedfinallybyadoptingtheweightedhistogramstatisticsmethod.TheresultsofthesimulationandseatrialsindicatedthattheproposedalgorithmhasbetterazimuthestimationperformancethanMUSICalgorithmofasinglevectorhydrophoneandthedatafusiontechniquebasedontheacousticenergyfluxmethod.Thebetterperformanceisreflectedintheaspectsoftheestimationprecision,theprobabilityofcorrectestimation,thecapabilitytodistinguishmulti-objectsandtheinhibitionofthenoisesub-bands.