Proportionate-type Normalized Least Mean Square Algorithms www.it-ebooks.info FOCUS SERIES Series Editor Francis Castanié Proportionate-type Normalized Least Mean Square Algorithms Kevin Wagner Miloš Doroslovački www.it-ebooks.info Firstpublished2013inGreatBritainandtheUnitedStatesbyISTELtdandJohnWiley&Sons,Inc. Apart from any fair dealing for the purposes of research or private study, or criticism or review, as permittedundertheCopyright,DesignsandPatentsAct1988,thispublicationmayonlybereproduced, storedortransmitted,inanyformorbyanymeans,withthepriorpermissioninwritingofthepublishers, or in the case of reprographic reproduction in accordance with the terms and licenses issued by the CLA. Enquiries concerning reproduction outside these terms should be sent to the publishers at the undermentionedaddress: ISTELtd JohnWiley&Sons,Inc. 27-37StGeorge’sRoad 111RiverStreet LondonSW194EU Hoboken,NJ07030 UK USA www.iste.co.uk www.wiley.com ©ISTELtd2013 TherightsofKevinWagnerandMilošDoroslovačkitobeidentifiedastheauthorsofthisworkhave beenassertedbytheminaccordancewiththeCopyright,DesignsandPatentsAct1988. LibraryofCongressControlNumber: 2013937864 BritishLibraryCataloguing-in-PublicationData ACIPrecordforthisbookisavailablefromtheBritishLibrary ISSN:2051-2481(Print) ISSN:2051-249X(Online) ISBN:978-1-84821-470-5 PrintedandboundinGreatBritainbyCPIGroup(UK)Ltd.,Croydon,SurreyCR04YY www.it-ebooks.info Contents PREFACE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix NOTATION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi ACRONYMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii CHAPTER 1.INTRODUCTION TOPTNLMSALGORITHMS . .. . . . .. . 1 1.1.ApplicationsmotivatingPtNLMSalgorithms . . . . . . . . . . . . . . . 1 1.2.HistoricalreviewofexistingPtNLMSalgorithms . . . . . . . . . . . . 4 1.3.UnifiedframeworkforrepresentingPtNLMSalgorithms . . . . . . . . 6 1.4.Proportionate-typeNLMSadaptivefilteringalgorithms . . . . . . . . . 8 1.4.1.Proportionate-typeleastmeansquarealgorithm . . . . . . . . . . . 8 1.4.2.PNLMSalgorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 1.4.3.PNLMS++algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . 8 1.4.4.IPNLMSalgorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 1.4.5.IIPNLMSalgorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 1.4.6.IAF-PNLMSalgorithm . . . . . . . . . . . . . . . . . . . . . . . . . 10 1.4.7.MPNLMSalgorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 1.4.8.EPNLMSalgorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . 11 1.5.Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 CHAPTER 2.LMSANALYSISTECHNIQUES . . . . . . . . . . . . . . . . . . 13 2.1.LMSanalysisbasedonsmalladaptationstep-size . . . . . . . . . . . . 13 2.1.1.StatisticalLMStheory: smallstep-sizeassumptions . . . . . . . . . 13 2.1.2.LMSanalysisusingstochasticdifferenceequations withconstantcoefficients . . . . . . . . . . . . . . . . . . . . . . . . 14 2.2.LMSanalysisbasedonindependentinputsignalassumptions . . . . . 18 2.2.1.StatisticalLMStheory: independentinputsignalassumptions . . . 18 www.it-ebooks.info vi PtNLMSAlgorithms 2.2.2.LMSanalysisusingstochasticdifferenceequationswith stochasticcoefficients . . . . . . . . . . . . . . . . . . . . . . . . . . 19 2.3.PerformanceofstatisticalLMStheory . . . . . . . . . . . . . . . . . . . 24 2.4.Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 CHAPTER 3.PTNLMS ANALYSISTECHNIQUES . . . . . . . . . . . . . . . 29 3.1.TransientanalysisofPtNLMSalgorithmforwhiteinput . . . . . . . . 29 3.1.1.LinkbetweenMSWDandMSE . . . . . . . . . . . . . . . . . . . . 30 3.1.2.RecursivecalculationoftheMWDandMSWD forPtNLMSalgorithms . . . . . . . . . . . . . . . . . . . . . . . . . 30 3.2.Steady-stateanalysisofPtNLMSalgorithm: bias andMSWDcalculation . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 3.3.ConvergenceanalysisofthesimplifiedPNLMSalgorithm . . . . . . . 37 3.3.1.Transienttheoryandresults . . . . . . . . . . . . . . . . . . . . . . . 37 3.3.2.Steady-statetheoryandresults . . . . . . . . . . . . . . . . . . . . . 46 3.4.ConvergenceanalysisofthePNLMSalgorithm. . . . . . . . . . . . . . 47 3.4.1.Transienttheoryandresults . . . . . . . . . . . . . . . . . . . . . . . 48 3.4.2.Steady-statetheoryandresults . . . . . . . . . . . . . . . . . . . . . 53 3.5.Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 CHAPTER 4.ALGORITHMSDESIGNED BASED ON MINIMIZATION OFUSER-DEFINEDCRITERIA . . . . . . . . . . . . . . . . . . . . . . . . . . 57 4.1.PtNLMSalgorithmswithgainallocationmotivated byMSEminimizationforwhiteinput . . . . . . . . . . . . . . . . . . . 57 4.1.1.OptimalgaincalculationresultingfromMMSE . . . . . . . . . . . 58 4.1.2.Water-fillingalgorithmsimplifications . . . . . . . . . . . . . . . . . 62 4.1.3.Implementationofalgorithms. . . . . . . . . . . . . . . . . . . . . . 63 4.1.4.Simulationresults . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65 4.2.PtNLMSalgorithmobtainedbyminimizationofMSE modeledbyexponentialfunctions . . . . . . . . . . . . . . . . . . . . . 68 4.2.1.WDforproportionate-typesteepestdescentalgorithm . . . . . . . . 69 4.2.2.Water-fillinggainallocationforminimizationoftheMSE modeledbyexponentialfunctions . . . . . . . . . . . . . . . . . . . 69 4.2.3.Simulationresults . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 4.3.PtNLMSalgorithmobtainedbyminimizationoftheMSWD forcoloredinput . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 4.3.1.Optimalgainalgorithm . . . . . . . . . . . . . . . . . . . . . . . . . 76 4.3.2.RelationshipbetweenminimizationofMSEandMSWD . . . . . . 81 4.3.3.Simulationresults . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 4.4.Reducedcomputationalcomplexitysuboptimalgainallocation forPtNLMSalgorithmwithcoloredinput . . . . . . . . . . . . . . . . . 83 4.4.1.Suboptimalgainallocationalgorithms . . . . . . . . . . . . . . . . . 84 4.4.2.Simulationresults . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 4.5.Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 www.it-ebooks.info Contents vii CHAPTER 5.PROBABILITYDENSITYOF WDFOR PTLMS ALGORITHMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91 5.1.Proportionate-typeleastmeansquarealgorithms . . . . . . . . . . . . . 91 5.1.1.Weightdeviationrecursion . . . . . . . . . . . . . . . . . . . . . . . 91 5.2.DerivationoftheconditionalPDFforthePtLMSalgorithm. . . . . . . 92 5.2.1.ConditionalPDFderivation . . . . . . . . . . . . . . . . . . . . . . . 92 5.3.ApplicationsusingtheconditionalPDF . . . . . . . . . . . . . . . . . . 100 5.3.1.Methodologyforfindingthesteady-statejointPDF usingtheconditionalPDF . . . . . . . . . . . . . . . . . . . . . . . 101 5.3.2.Algorithmbasedonconstrainedmaximization oftheconditionalPDF. . . . . . . . . . . . . . . . . . . . . . . . . . 104 5.4.Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 CHAPTER 6.ADAPTIVESTEP-SIZE PTNLMS ALGORITHMS . . . . . . . 113 6.1.Adaptationofµ-lawforcompressionofweightestimates usingtheoutputsquareerror . . . . . . . . . . . . . . . . . . . . . . . . 113 6.2.AMPNLMSandAEPNLMSsimplification . . . . . . . . . . . . . . . . 114 6.3.Algorithmperformanceresults . . . . . . . . . . . . . . . . . . . . . . . 116 6.3.1.LearningcurveperformanceoftheASPNLMS,AMPNLMS andAEPNLMSalgorithmsforawhiteinputsignal . . . . . . . . . 116 6.3.2.LearningcurveperformanceoftheASPNLMS,AMPNLMS andAEPNLMSalgorithmsforacolorinputsignal. . . . . . . . . . 117 6.3.3.LearningcurveperformanceoftheASPNLMS,AMPNLMS andAEPNLMSalgorithmsforavoiceinputsignal . . . . . . . . . 117 6.3.4.Parametereffectsonalgorithms . . . . . . . . . . . . . . . . . . . . 119 6.4.Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124 CHAPTER 7.COMPLEXPTNLMS ALGORITHMS . . . . . . . . . . . . . . 125 7.1.Complexadaptivefilterframework . . . . . . . . . . . . . . . . . . . . . 126 7.2.cPtNLMSandcPtAPalgorithmderivation . . . . . . . . . . . . . . . . 126 7.2.1.Algorithmsimplifications . . . . . . . . . . . . . . . . . . . . . . . . 129 7.2.2.Alternativerepresentations . . . . . . . . . . . . . . . . . . . . . . . 131 7.2.3.StabilityconsiderationsofthecPtNLMSalgorithm . . . . . . . . . 131 7.2.4.Calculationofstepsizecontrolmatrix . . . . . . . . . . . . . . . . . 132 7.3.Complexwater-fillinggainallocationalgorithm forwhiteinputsignals: onegainpercoefficientcase . . . . . . . . . . . 133 7.3.1.Derivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133 7.3.2.Implementation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136 7.4.Complexcoloredwater-fillinggainallocationalgorithm: onegainpercoefficientcase. . . . . . . . . . . . . . . . . . . . . . . . . 136 7.4.1.Problemstatementandassumptions . . . . . . . . . . . . . . . . . . 136 7.4.2.OptimalgainallocationresultingfromminimizationofMSWD . . 137 www.it-ebooks.info viii PtNLMSAlgorithms 7.4.3.Implementation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138 7.5.Simulationresults . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139 7.5.1.cPtNLMSalgorithmsimulationresults . . . . . . . . . . . . . . . . 139 7.5.2.cPtAPalgorithmsimulationresults. . . . . . . . . . . . . . . . . . . 141 7.6.TransformdomainPtNLMSalgorithms . . . . . . . . . . . . . . . . . . 144 7.6.1.Derivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145 7.6.2.Implementation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146 7.6.3.Simulationresults . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147 7.7.Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151 CHAPTER 8.COMPUTATIONALCOMPLEXITYFOR PTNLMS ALGORITHMS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153 8.1.LMScomputationalcomplexity . . . . . . . . . . . . . . . . . . . . . . 153 8.2.NLMScomputationalcomplexity . . . . . . . . . . . . . . . . . . . . . 154 8.3.PtNLMScomputationalcomplexity . . . . . . . . . . . . . . . . . . . . 154 8.4.ComputationalcomplexityforspecificPtNLMSalgorithms . . . . . . 155 8.5.Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157 CONCLUSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159 APPENDIX1.CALCULATIONOFβ(0),β(1) AND β(2) . . . . . . . . . . . . 161 i i,j i APPENDIX2.IMPULSERESPONSELEGEND . . . . . . . . . . . . . . . . . . 167 BIBLIOGRAPHY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169 INDEX . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173 www.it-ebooks.info Preface Aimsofthisbook Theprimarygoalofthisbookistoimpartadditionalcapabilitiesandtoolstothe fieldofadaptivefiltering.Alargepartofthisbookdealswiththeoperationofadaptive filters when the unknown impulse response is sparse. A sparse impulse response is oneinwhichonlyafewcoefficientscontainthemajorityofenergy.Inthiscase, the algorithmdesignerattemptstousetheaprioriknowledgeofsparsity.Proportionate- type normalized least mean square (PtNLMS) algorithms attempt to leverage this knowledgeofsparsity.However,anidealalgorithmwouldberobustandcouldprovide superior channel estimation in both sparse and non-sparse (dispersive) channels. In addition,itwouldbepreferableforthealgorithmtoworkinbothstationaryandnon- stationaryenvironments.Takingallthesefactorsintoconsideration,thisbookattempts toaddtothe stateofthe artinPtNLMS algorithmfunctionalityforall thesediverse conditions. Organizationofthisbook Chapter1introducestheframeworkofthePtNLMSalgorithm.Areviewofprior workperformedinthefieldofadaptivefilteringispresented. Chapter 2 describes classic techniques used to analyze the steady-state and transientregimesoftheleastmeansquare(LMS)algorithm. In Chapter 3, a general methodology is presented for analyzing steady-state and transient analysis of an arbitrary PtNLMS algorithm for white input signals. This chapterbuildsonthepreviouschapterandexaminesthattheusabilityandlimitations ofassumingtheweightdeviationsareGaussian. InChapter4,severalnewalgorithmsarediscussedwhichattempttochooseagain atanytimeinstantthatwillminimizeuser-definedcriteria,suchasmeansquareoutput error and mean square weight deviation. The solution to this optimization problem www.it-ebooks.info x PtNLMSAlgorithms resultsinawater-fillingalgorithm.Thealgorithmsdescribedarethentestedinawide varietyofinputaswellasimpulsescenarios. In Chapter 5, an analytic expression for the conditional probability density function of the weight deviations, given the preceding weight deviations, is derived. This joint conditional probability density function is then used to derive the steady-state joint probability density function for weight deviations under different gainallocationlaws. In Chapter 6, a modification of the µ-law PNLMS algorithm is introduced. Motivated by minimizing the mean square error (MSE) at all times, the adaptive step-size algorithms described in this chapter are shown to exhibit robust convergenceproperties. In Chapter 7, the PtNLMS algorithm is extended from real-valued signals to complex-valuedsignals.Inaddition,severalsimplificationsofthecomplexPtNLMS algorithm are proposed and so are their implementations. Finally, complex water-fillingalgorithmsarederived. InChapter8,thecomputationalcomplexitiesofalgorithmsintroducedinthisbook arecomparedtoclassicalgorithmssuchasthenormalizedleastmeansquare(NLMS) andproportionatenormalizedleastmeansquare(PNLMS)algorithms. www.it-ebooks.info Notation The following notation is used throughout this book. Vectors are denoted by boldfacelowercaseletters,suchasx.Allvectorsarecolumnvectorsunlessexplicitly statedotherwise.ScalarsaredenotedbyRomanorGreekletters,suchasxorν.The ith component of vector x is given by x . Matrices are denoted by boldface capital i letters, suchasA.The(i,j)thentryofanymatrixAisdenotedas[A] ≡ a .We ij ij frequentlyencountertime-varyingvectorsinthisbook.Avectorattimekisgivenby x(k). For notational convenience, this time indexing is often suppressed so that the notation x implies x(k). Additionally, we use the definitions x+ ≡ x(k + 1) and x− ≡x(k−1)torepresentthevectorxattimesk+1andk−1,respectively. ForvectorawithlengthL,wedefinethefunctionDiag{a}asanL×Lmatrix whosediagonalentriesaretheLelementsofaandallotherentriesarezero.Formatrix A, we define the function diag{A} as a column vector containing the L diagonal entries from A. For matrices, Re{A} and Im{A} represent the real and imaginary partsofthecomplexmatrixA. Thelistofnotationisgivenbelow. x avector x ascalar A amatrix x theithentryofvectorx i [A] ≡a the(i,j)thentryofanymatrixA ij ij Diag{a} adiagonalmatrixwhosediagonalentriesarethe elementsofvectora diag{A} acolumnvectorwhoseentriesarethediagonal elementsofmatrixA I identitymatrix E{x} expectedvalueofrandomvectorx www.it-ebooks.info