非線性與立體聲回音消除之收斂性分析

Abstract

由於現代通訊系統的多樣化應用，愈來愈多的聲訊回音消除(AEC)設計必須考量非線性特性(喇叭)及立體聲環境(高相關輸入)。已經有許多針對非線性適應濾波器所提出的演算法，但卻缺少相對應的演算法收斂分析。根據無記憶多項式適濾波器與NLMS演算法，我們提出了非線性聲訊回音消除的收斂分析式；另一方面，為了尋找非線性聲訊回音消除較佳的收斂方式，我們也提出並比較幾種不同的係數更新方法。在立體聲回音消除方面，我們利用Doğancay在分析單聲道選擇式係數更新收斂性時所採用的方法，提出應用於立體聲回音消除部分係數更新之收斂分析式。另外，結合非線性喇叭與立體聲回音消除的實驗驗證了相關性削減與非線性適應濾波器的功效。最後，電腦模擬用來驗證支持之前的分析。

Because of diversity applications of modern communication system, more and more acoustic echo cancellation (AEC) designs need to consider the nonlinearity characteristic (loudspeaker) and stereophonic environment (high correlated inputs). There are many proposed algorithms for nonlinear adaptive filter but lack of convergence analysis for these algorithms. We propose the convergence analyses for nonlinear AEC based on memoryless polynomial nonlinear filter and NLMS algorithm. Several adaptation strategies for nonlinear AEC are also provided and compared for finding better convergence behavior. In stereophonic AEC, we analyze the convergence of partial update stereophonic AEC by taking use of the Doğancay’s method used in selective partial update monophonic AEC study. The combinations of nonlinear loudspeaker and stereophonic AEC are demonstrated to verify the coherence reduction and nonlinear adaptive filter effect. Finally, computer simulations are presented to support the analysis.