Study On The Method Of Nonlinear Acoustic Echo Cancellation

The Acoustic Echo Canceller?AEC?based on the adaptive filtering techniques is the main approach to remove echo in the communication systems,however,some difficulties still wait for breakthroughs in AEC technology,the identification of nonlinear acoustic echo path is one of them.As acoustic devices employ smaller and cheaper components,the nonlinear characteristics in the communication systems become more and more significant,resulting in the degradation of the traditional AEC method based on the linear adaptive filters,and affecting the communication quality.In addition,noise is an unavoidable issue in all communication systems.Besides the ubiquitous Gaussian noises,non-Gaussian noises are widely existed in real world.The impact characteristic of these noises destroys the Normalized Least Mean Square?NLMS?algorithm and other adaptive filtering algorithms based on the l₂ norm optimization criterion.At the same time,considering that the echo path often has obvious sparse characteristics in hands-free calls,video conferences and other communication occasions,the performance of the adaptive filtering algorithms is further improved by combining the system sparsity with the weight of proportional terms.Therefore,aiming at the above issues,a nonlinear acoustic echo cancellation scheme that cascades a memoryless nonlinear filter and a linear Finite Impulse Response?FIR?filter is proposed in this thesis.First,in the nonlinear filter part,the nonlinear system is modeled by a memoryless polynomial filter based on Taylor series;Secondly,in the linear filter part,considering the non-Gaussian noise widely existing in real life and the sparse characteristics of the echo path,this thesis proposes a Modified Improved Proportionate Normalized Sign Algorithm?MIPNSA?by combining the proportional matrix with Sign Algorithm?SA?.The algorithm can not only adapt to different background noises,but also enhance the ability to adapt to sparse systems.Thus,we effectively remove the nonlinear acoustic echo under different noise backgrounds,and a more detailed theoretical analysis of the convergence is made.A large number of simulation results show that compared with the existing Volterra method,the nonlinear acoustic echo cancellation method used has the characteristics of smaller calculations and more flexible design;At the same time,compared with the traditional algorithms such as Improved Proportionate NSA?IPNSA?and Normalized SA?NSA?,the proposed algorithm has better convergence and robustness.