The Study Of Subband And Block Sparse Adaptive Filtering Algorithms For Acoustic Echo Cancellation

With over fifty years development,acoustic echo cancellation(AEC),which is based on adaptive filtering filter technique,has been widely used in various kinds of technological applications,such as vehicular voice system,telephone conference system,hands-free phone,hearing aids and artificial cochlea.The present demand for high quality AEC technique brings two tasks to adaptive filtering algorithm designers: 1.How to achieve the tradeoff between algorithm properties of adaptive filtering algorithm.Because lower complexity or higher steady-state is achieved at the cost of convergence rate.2.How to design and develop the advanced efficient algorithm which is optimized and unified on the basis of convergence rate and computational complexity.This thesis employs subband structure and block sparse system to analyze the steady-state,robustness and complexity property.Its main works are summarized as follows:Firstly,to achieve the optimization of convergence rate and steady-state error,an improved switch adaptive filtering algorithm is proposed for AEC system.The proposed algorithm utilizes voice activity detection technique based on fast and slow envelope tracking method to distinguish the presence of speech signal.Then,the difference of fast and slow tracking envelopes leads to a power threshold which can be used to switch the algorithm according to the status of speech.The multi-subband structure not only overcomes aliasing components,but also obtains lower steady-state error.Computational complexity is also analyzed theoretically.In addition,this thesis realizes the implementation of the improved algorithm on AEC module embedded Web Real-Time Communication(WebRTC)platform.Secondly,concerning the problem that the existing block-sparse system identification algorithm based on mean square error criterion shows poor performance under impulsive interference,an block-sparse normalization least mean square algorithm based on the inverse hyperbolic sine function was proposed.A new cost function was first constructed and the additive value was obtained by steepest-descent method which is robust and effective in suppressing the adverse effects of the impulsive interference.Meanwhile,mean convergence behavior was analyzed theoretically.The simulations demonstrate that in comparison with the block sparsity normalized least mean square algorithm.The proposed algorithm has faster convergence rate and less steady-state error under non-Gaussion impulsive interference and abrupt change environment.Therefore,it is suitable for AEC problem in impulsive noise scenario.