Efficient And Robust Affine Projection Algorithms For Echo Cancellation

With the development of signal processing applications and the increasing complex of modern electromagnetic interference environment, the noise usually shows a strong impulsive property due to the impact of nature and human in many practical environments. The solution of adaptive algorithm will no longer be the optimal solution when the noise obeys non-Gaussian impulsive distribution. Therefore, the study of efficient adaptive algorithm against impulsive noise is special significant under non-Gaussian environment. This paper will focus on the study of efficient and robust APA under non-Gaussian impulsive environment.In order to improve the poor performance of traditional APA under impulsive interferences, on the basis of the research on non-Gaussian impulsive noise signal and APA algorithm, a new anti-outliers error kernel function which is based on the M-estimate APA algorithms is proposed. The performance of the proposed algorithms is analyzed. The proposed algorithm improves the steady-state misalignment and reduces computational complexity by introducing the μ guidelines and proportionate memory principle.To solve the high update rate and steady-state misalignment problems of traditional robust set-membership APA, this paper designs a new robust set-membership error bound to ensure the convergence performance, the algorithm also reduces the update rate as well as steady-state misalignment effectively. The new error bound solve the unstable shortcomings at initial point of the original algorithm by adopting smooth transition method.Considering the slow convergence rate and high misalignment of affine projection sign algorithm, an adaptive regularization parameter is applied to proportionate affine projection sign algorithm by using the normalized gradient descent method, which ensures the stability of the regularization parameter against error changes. To further ensure the stability of the algorithm, the paper deduces the lower bound of the smallest regularization parameter.In echo cancellation applications, the background noise or disturbance of echo cancellation system not always obeys Gaussian distribution. On the contrary, non-Gaussian noise is very common. This paper proposes several efficient and robust algorithms for echo cancellation to overcome the above problems, which can effectively eliminate the echo under network and acoustic echo channel.