Research On Stereo Acoustic Echo Cancellation Algorithm

In multimedia communication systems such as conference calls,adaptive echo cancellation technology can be used to eliminate the coupling echo originating from the speaker and the microphone,thereby ensuring the quality of real-time full-duplex communication.In the case of multiple participants,the multi-channel acoustic echo cancellation system can provide users with a very good spatial positioning effect and bring participants an immersive experience.Stereo echo cancellation technology is not a simple extension of the mono model,and its theory and practice need more in-depth research and exploration;In addition,with the improvement of users' requirements for sound quality and the development of hardware technology,stereo echo cancellation algorithm(SAEC)will be one of the important research directions in the field of multimedia communication.Compared with the case of mono echo cancellation,in addition to requiring more adaptive filters,stereo algorithms also have problems such as slow convergence of adaptive filter coefficients and large misalignment.The latter is mainly because the two far-end reference signals are generated by the same sound source,so they have a high degree of correlation,which makes the condition number of the autocorrelation matrix very large,which in turn affects the convergence of the adaptive filter coefficients.If an adaptive filtering algorithm such as the least mean square(LMS)is highly dependent on the condition number of the covariance matrix,then the correlation algorithm can be used to reduce the correlation between the two far-end signals as much as possible to improve the performance of the algorithm.However,the use of decorrelation algorithms will inevitably affect the stereo sound quality.Therefore,when choosing a decorrelation algorithm,both the effect of decorrelation and the impact on stereo sound quality should be considered.If recursive least squares algorithm(RLS)or affine projection algorithm(AP)is used,although the two are not sensitive to the condition number of the covariance matrix,the large amount of calculation affects their application in real-time acoustic systems.Even if they have fast algorithms,there will be problems such as numerical instability.This article mainly focuses on the in-depth research and analysis of stereo acoustic echo cancellation algorithms,and proposes several algorithms for the problems of stereo echo cancellation technology.The main content and innovation of this article are explained from the following aspects:(1)Using LMS-like adaptive algorithms as a model,the research status of decorrelation algorithms is deeply investigated,and the effects of different decorrelation algorithms in stereo echo cancellation applications are compared and analyzed.The principles of other adaptive algorithms whose performance is less affected by the condition number of the covariance matrix are studied,such as recursive least squares(RLS)and affine projection(AP).(2)The problems of slow convergence and large misalignment in the adaptive solution of the stereo echo cancellation algorithm are studied.The relationship between the correlation between two remote reference signals and the condition number of the corresponding covariance matrix is established,and how this relationship affects the misalignment of the adaptive solution of the frequency domain adaptive algorithm is explained.(3)Based on the segmented block frequency domain adaptive algorithm(PBFDAF),a decorrelation algorithm with little effect on stereo sound quality is proposed.The decorrelation algorithm uses a time-varying all-pass filter in the high-frequency part,and uses the theory based on the lack of fundamental frequency in the low-frequency part.First,using a timevarying all-pass filter to change the phase of a high-frequency signal will not have a great impact on the sense of hearing.Secondly,the human brain has the ability to process the information present in the overtones to calculate the "missing" fundamental wave.Therefore,it is possible to adaptively track and cancel the pitch,that is,the fundamental frequency,so as to weaken the correlation of low-frequency signals without affecting the stereo perception.(4)A frequency-domain block stereo echo cancellation algorithm using the coherence between channels is proposed,which is an approximation of the RLS algorithm.Compared with the RLS algorithm,its calculation has been greatly simplified.And it uses the correlation of two far-end reference signals to update the filter coefficients,so compared with the LMS algorithm,it can better process highly correlated speech signals,especially stereo signals.(5)A stereo echo cancellation algorithm based on simplified fast transversal filter is proposed.It evolved on the basis of the Fast Transverse Filtering(FTF)algorithm,using only the forward prediction error of the input signal,and completely discarding the forward and backward predictors.The calculation amount is equivalent to that of the NLMS algorithm,which is greatly simplified compared to the calculation amount of the RLS algorithm.