Research On Adaptive Filtering Method For Acoustic Echo Cancellation

The acoustic echoes ubiquitously exist in communication systems as internet telephony and video conferencing etc,which seriously affect the quality of voice communication and degenerate intelligibility.Therefore,in the situation of the high requirement for the voice quality,the acoustic echoes must be suppressed or canceled.At present,the echo cancellation technology is mainly based on the software processing,whose core is the adaptive filter method.This thesis focuses on study on the adaptive filtering methods in echo cancellation.Considering the factors that affect the performance of the filtering algorithm in different occasions,this thesis aims to find new solutions for the algorithm in echo cancellation under different environmental backgrounds.The main research work includes:(1)For the basic proportional minimum mean square(IPNLMS)adaptive filtering algorithm,when the filter converges,the filter weights may be allocated to larger increments by the fixed parameters,steady-state imbalance is brought out.And the performance of the following algorithm is affected by the presence of near-end speech signals.In view of this situation,an improved time-varying parameter IPNLMS algorithm is proposed.In this method,the near-end signal energy estimation is used to construct intermediate variables,so that the IPNLMS algorithm can accurately change and adjust the parameter value of the weight increment with the state of the filter.(2)Aiming at the nonlinear echo that is difficult to be eliminated by the basic linear echo canceller,two combinational schemes of filters are introduced.A collaborative spline adaptive filter combination structure is also proposed.In which,a linear and nonlinear spline adaptive filter are collaboratively combined in a manner similar to the convex combination.The different error signals are used to update the corresponding weight coefficients for the two filters.In the nonlinear echo cancellation application,the linear adaptive filter is used to simulate the entire echo path,and the nonlinear one is used to simulate the nonlinearities introduced by the echo path and make compensations,so as to give full play to the advantages of the two filters in different scenarios.(3)In non-Gaussian noise scenarios,the performance of the basic adaptive filtering algorithm using MMSE as the error criterion is limited.In combination with the method of nonlinear echo cancellation mentioned above,an echo cancellation method in ?-stable distributed noise environment is proposed.The collaborative functional link adaptive filters are taken as the basic model and the idea of the normalized Least Mean p-norm(NLMP)algorithm is used to update the weight coefficients of the two filters,so as to achieve a better echo cancellation effect.(4)For the problems of non-unique solution and misalignment in the multi-channel echo cancellation different from the single-channel case.Three representative multi-channel adaptive filter algorithms for echo cancellation are studied.The performance differences of three algorithms are analyzed through simulation experiments.