【作者】 高鹰；

【导师】 谢胜利；

【作者基本信息】 华南理工大学 ， 通信与信息系统， 2002， 博士

【摘要】 自适应信号处理理论和技术已发展成信号处理领域里的一个新的分支,并且在系统辨识、回波消除、自适应谱线增强、自适应信道均衡、语音线性预测、预测解卷、信号检测、自适应噪声消除、自适应天线阵等方面获得了越来越广泛的应用。对自适应滤波算法的研究是当今自适应信号处理中最为活跃的研究课题之一。信号的统计量在自适应信号处中起着极其重要的作用,累积量是一种重要的统计量之一。本论文主要研究基于累积量的自适应滤波算法及其在系统辨识、回波消除等方面的应用。首先,我们对建立在二阶累积量(自相关,互相关)基础之上的自适应滤波算法进行了研究和讨论,针对固定步长的LMS算法存在的收敛速度、时变系统跟踪速度与收敛精度之间的矛盾,建立了步长因子μ与误差信号e(n)之间的一种非线性函数关系,该函数关系简单,且在误差e(n)接近零处具有缓慢变化的特性。由此函数我们得出另一种新的变步长自适应滤波算法,同时分析了参数α,β取值原则及对自适应滤波算法收敛性能的影响。该算法有较好的收敛性能且比其它的变步长自适应滤波算法如SSVLMS算法、SV-LMS算法和L.E-LMS算法的计算量少。其次,我们对累积量域自适应滤波原理,误差准则和相应自适应滤波算法进行详细的讨论研究,通过线性寻优改进了基于误差准则J1 (n)的CDLMS算法和基于误差准则J 2 (n)的HOS4-MSEA算法。对导出CDRLS算法的误差准则J3 (n)采用梯度法而得到了CDEFWLMS算法。通过定义一个新的累积量域误差准则J(n)而获得了基于此误差准则的CDSWLMS算法。我们还详细研究和讨论了对高斯噪声不敏感的基于累积量的均方误差准则及其自适应滤波算法。根据各阶统一形式的基于高阶累积量的误差准则和基于高阶累积量的Wiener-Hopf等式,我们从新推导了CLMS算法和CRLS算法。在应用方面,我们除了通过计算机模拟仿真把以上算法应用于系统辨识中之外,还对基于二阶累积量的多路声回波消除自适应滤波算法进行了分析研究,讨论了多路声回波消除问题及原因,对现有的多路声回波消除自适应滤波算法进行了介绍。改进了扩展LMS算法和带旋转因子的多路声回波消除自适应滤波NLMS-RF算法和APA-RF算法。把更多还原

【Abstract】 The theory and technology of adaptive signal processing have become a popular topic insignal processing field, and have been widely used in signal processing practices, such assystem identification、echo cancellation、adaptive line enhancement、adaptive channelequalization、speech signal linear prediction、predictive deconvolution、signal detection、adaptive noise cancelling and adaptive beamforming etc. Research for adaptive filteringalgorithms is one of the most active research topics in adaptive signal processing. Signalstatistics are of major importance in adaptive signl processing. Cumulant is one of theimportant statistics. The dissertation is to focus on the research of cumulant-based adaptivefiltering algorithms and their application in system identification and echo cancellation. Firstly, the author of the dissertation discusses about some variable step size adaptivefiltering algorithms based on second-order cumulant (self-relation、cross-relation), andestablishes another non-linear functional relationship between μ and e(n);which is notonly simple, but also has the property of slight change e(n) near to zero. On the basis of thefunctional relationship, the author presents a new variable step size LMS adaptive filteringalgorithm, and analyzes the algorithm with various α and β . Besides good convergenceproperties, the algorithm has less computational complexity than other variable step sizeadaptive filtering algorithm, such as SSVLMS algorithm、SV-LMS algorithm、L.E-LMSalgorithm,. Secondly, the author improves CDLMS algorithm based on error criteria J1 (n) andHOS4-MSEA algorithm based on error criteria J2 (n) by linear searching optimum afterdiscussing about cumulant domain adaptive filtering principle 、error criteria and adaptivefiltering algorithms. He also presents CDEFWLMS algorithm by using descent method onerror criteria J3 (n) from which the CDRLS algorithm was derived. Then he defines a newcumulant domain error criteria J (n)and presents CDSWLMS algorithm based on J(n).Thirdly, the author discusses about Gauss noise insensitive MSE criteria and adaptivefiltering algorithm, and re-derives CLMS algorithm and CRLS algorithm on the basis of allhigh-order cumulant-based error criteria and Wiener-Hopf equation.Fourthly, After analyzing the problem in second-order cumulant-based multichannelacoustic echo cancellation and introducing multichannel acoustic echo cancellationalgorithms, the author improves two channel Extend LMS algorithm and adaptive filteringalgorithm with rotating factor for miltichannel echo cancellation. He also presents athird-order cumulant-based two channel echo cancellation algorithm by applying cumulantdomain adaptive filtering principle to multichannel acoustic echo cancellation.Lastly,The author discusses correlation function adaptive filtering algorithms(thesecond-order cumulant domain adaptive filtering algorithm) and their application in echocancellation, and then presents a correlation function recursive least squares algorithm andanalyzes convergence of the algorithm. The correlation function adaptive filtering algorithmsare based on the processing of correlation function of the input signal, instead of theprocessing of input signal itself. In echo cancellation. The correlation function adaptivefiltering algorithms can continue the tap adaptations under double-talk conditions.更多还原