結合次頻帶適應性濾波與變步長正規最小均方演算法的迴音消除技巧

Abstract

在本篇論文裡，我們探討在時變環境裡聲學迴音消除的問題。在過去關於適應性濾波技術的研究，都能夠發現次頻帶適應性濾波技術可以享有快速的收斂速度。經由將適應性濾波技術應用至次頻帶，它也可以擁有較低的運算量。另一方面，正規最小均方演算法結合變步長也被發現能在時變環境裡擁有不錯的追縱能力。然而，變步長正規最小均方演算法應用在全頻適應性濾波器上，它的使用通常是耗費運算量的。在本篇論文，我們將變步長正規最小均方演算法應用到使用近似無頻疊失真餘弦調變濾波器組之次頻帶適應性濾波。我們也會分析我們所提出的方法的穩態表現。模擬將會展現我們所提出的方法兼顧追縱能力和收斂速度同時也擁有低量的運算複雜度。

In this thesis, we consider the problem of acoustic echo cancelation for a time varying environment. In the context of adaptive filtering, the subband adaptive filtering (SAF) is known to enjoy a fast convergence rate. By applying adaptive filtering in the subbands, it also has the advantage of a lower complexity. On the other hand, the normalized least mean squares (NLMS) algorithm with a variable step size (VSS)is known to have a good tracking ability in a time varying environment. However the implementation of VSS-NLMS is usually computationally expensive as it is applied to fullband adaptive filter. In this thesis, we apply VSS-NLMS to the subband filters using a approximately alias-free cosine modulated filter bank. We also analyze the steady state behavior of the proposed algorithm. Simulations show that the proposed method achieves a good trade-off between tracking performance and convergence rate while having a low computational complexity.