ENHANCED STOCHASTIC TAPS NLMS FILTER WITH EFFICIENT SPARSE TAPS LOCALIZATION

An efficient adaptive filtering algorithm for network echo cancellers is proposed in this paper. Employing the primary and auxiliary filter that operate parallel, an algorithm recently proposed named stochastic taps NLMS (ST-NLMS) estimates and locates the active region of the unknown system impulse response, therefore results a rapid transient behavior while estimating long, sparse, echo path like system. However, ST-NLMS still fails to give satisfactory estimation of effective system response region, hence leads to prohibitive computational complexity and further enhance potential. In this paper, based on the analysis to ST-NLMS, a new algorithm named ST-NLMS++ is proposed, which originally adopts hybrid active taps re-initialization and fake active tap elimination to enhance the estimation of active tap region. In stationary environment and small step-size assumption, theoretical examination reveals that ST-NLMS++ leads a superior estimation of effective impulse location and achieves faster tap-weight convergence as well as the same level of misadjustment compared to ST-NLMS and traditional NLMS. Experiments further verify the analysis and demonstrate that the tracking performance of proposed algorithm is also satisfying in both stationary and non-stationary scenarios