Study On Algorithm Of Echo Cancellation For Tele-Communication Systems

Echo cancellation is a bottleneck in telecommunication systems. This text introduced first two kinds of Echo Cancellations for the telecommunication systems-line echo cancellation (LEC) and acoustic echo cancellation (AEC), describing their history of the Creation and development, basic principles and problems that need to be resolved. The next in order, it introduced in detail the most basic and important algorithm of Adaptive Echo Cancellation is The LMS algorithm. A major disadvantage of the LMS algorithm is that its convergence behavior is highly dependent on the power spectral density of filter input. This leads to slow convergence rate when the input signal exhibits large power fluctuation. To solve this problem and lead to a fast convergence, a new algorithm that updates the LMS algorithm by introducing a forgetting factor is presented in this paper. The new algorithm improves the gradient estimation of the LMS algorithm by introducing a forgetting factor. So the convergence direction of filter is formed with the gradient of the time and the departed convergence direction. This updates the gradient with the departed convergence direction, and prevents the convergence direction from greatly leaving the fastest convergence direction, and overcomes the disadvantage that convergence behavior of the LMS algorithm depend the spread of the eigenvalues of the correlation matrix in some degree. So the convergence rate is faster, and the convergence behavior is more stabile. The Comparison between the new algorithm and NLMS is made under the same simulations in LEC system. The Comparison results prove the advantages of the new algorithm in fastening the convergence rate.At last the algorithm protections and implementations of line echo cancellation (LEC) and acoustic echo cancellation (AEC) are discussed based on the applications such as hands-free telecommunications and teleconference systems. An improved spectral subtraction method is introduced for speech enhancement as the post-processing of AEC. Some new methods are proposed to protect the AEC/LEC adaptive filter from divergence in double talk situation and over-adjusting near convergence, such as the weighted-sum normalization and the step-size control by cross-correlations. Both methods are trying to keep the system in steady through "soft-decision".