比例系数重用自适应滤波算法分析-analysis of adaptive filtering algorithm for proportional coefficient reuse.docxVIP

比例系数重用自适应滤波算法研究中文摘要比例系数重用自适应滤波算法研究中文摘要不同的自适应滤波算法具有不同的收敛速度和稳态失调。收敛速度的快慢意味着自适应滤波器逼近未知系统所需要花费时间的长短，而稳态失调的高低对应自适应滤波器逼近未知系统精度的高低。传统的归一化最小均用方算法（normalizedleastmeansquare，NLMS）和仿射投影算法（affineprojectionalgorithm,APA）在保证一定收敛速度的前提下，具有较大的稳态失调，并且测量噪声越大，稳态失调越大。近年来，HyeonwooCho等提出了基于NLMS和APA算法的滤波器系数重用算法（即NLMS-RC和APA-RC算法）来降低算法的稳态失调。在许多应用中，待估计的未知系统是稀疏的。NLMS-RC和APA-RC算法应用于估计系数未知向系统时，其收敛较慢。为了解决这一问题，本文将比例自适应的概念引入NLMS-RC和APA-RC算法，提出比例系数重用NLMS（IPNLMS-RC）和比例系数重用APA（PAPA-RC）算法。为了进一步提高IPNLMS-RC和PAPA-RC算法的初始收敛速度，本文还将变重用系数阶数的思想应用于IPNLMS-RC和PAPA-RC算法。仿真结果表明，本文提出的算法在估计稀疏的未知系统时，具有很好的收敛性能。关键词:自适应滤波，归一化最小均方算法，仿射投影算法，系数重用，比例自适应，稀疏系统作者：唐学青指导教师：倪锦根AbstractStudyofproportionateadaptivefilteringalgorithmswithreusedcoefficientsStudyofproportionateadaptivefilteringalgorithmswithreusedcoefficientsAbstractTheconvergencerateandsteady-statemisalignmentofdifferentadaptivefilteringalgorithmsdifferfromeachother.Theconvergenceratedeterminesthetimecostbytheadaptivefiltertoapproximatetheunknownsystem,whilethesteady-statemisalignmentdeterminestheaccuracyobtainedbytheadaptivefiltertoapproximatetheunknownsystem.Withcertaindegreeofconvergencerate,theclassicalnormalizedleastmeansquare(NLMS)algorithmandaffineprojectionalgorithm(APA)showrelativelylargesteady-statemisalignment,andthelargerthemeasurementnoise,thelargerthesteady-statemisalignment.Recently,HyeonwooChoetal.haveproposedtheNLMSandAPA-basedreusingcoefficient(RC)algorithms(i.e.,NLMS-RCandAPA-RC)toreducesteady-statemisalignment.Inmanyapplications,theunknownsystemstobeestimatemaybesparse.IftheNLMS-RCorAPA-RCalgorithmisusedtoestimatesuchsystems,itsconvergencerateisslow.Inordertosolvethisproblem,thisthesisincorporatestheconceptofproportionateadaptionintoNLMS-RCandAPA-RCalgorithmsandproposesaproportionateNLMS-RC(PNLMS-RC)algorithmandaproportionateAPA-RC(PAPA-RC)algorithm.Inordertoincreasetheinitialconvergencerate,thisthesisalsoemploystheideaofvariableorderofreusingcoefficientinthePNLMS-RCandPAPA-RCalgorithms.Simulationresultsshowthatthealgorithmsproposedinthisthesishav