兩段式非線性回音消除之收斂性分析與最佳步伐調適

Abstract

免持聽筒或是視訊會議系統常驅動功率放大器和喇叭操作在飽和非線性區，導致一般使用的線性回音消除器性能降低。在本篇論文中，我們將對非線性回音消除建構出一個無記憶式片段線性(PWL)處理器串聯一個線性濾波器的模型，它能有效降低運算量。另外，為了克服在串聯式系統的發散問題，我們採取兩段式調適的方式，首先只有線性濾波器開始更新，接著PWL和線性濾波器的係數同時進行調適。更進一步，推導出它的收斂分析及穩定條件，並在電腦模擬得到驗證。最後，我們發展出LMS演算法在不同時間、不同時間及不同閥(tap)的最佳步伐調適來加強收斂速度，同時提供他們實用的實現方式和電腦模擬。

Hand-free phone or teleconferencing system drives the power-amplification and loudspeaker commonly into saturated nonlinear region, leading to that the performance of conventional acoustic echo cancellation (AEC) reduced. In this thesis, we will build a cascade model which consists of a memoryless piece-wise linear (PWL) processor and a linear filter for AEC. It is beneficial to reduce the computation. Besides, in order to overcome the divergence problem in a cascade system, we will adopt the two-stage adaptation that starts with a linear filter, and then joint adaptation of PWL and linear coefficients follows. Further, the convergence analysis and stability criterion will be derived and the computer simulation will justify our analysis. Finally, LMS algorithms with optimum time-variant and time-&tap- variant step-sizes are developed to improve convergence rate. Their practical implementations and computer simulations are also provided.