免手持通訊系統可以應用在免手持聽筒電話、視訊會議、遠距教學及遠端醫療等系統。而此種通訊系統必需解決聲學迴音之問題。可適性濾波器提供這個問題的有效解決方法,這些方法可概分為LMS家族及RLS家族演算法,本論文就是針對這些可適性濾波器進行比較分析。 首先我們詳列LMS(NLMS、BLMS、FBLMS、MDF、GMDF演算法)及RLS(RLS、快速RLS、快速牛頓演算法)家族演算法的步驟及運算量,進而建議各演算法在各階數可採用之DSP處理器之選擇。最後我們將這些演算法建構在白色雜訊、AR模型及實際語音之輸入並分析其收斂特性及結果。 我們發現MDF與GMDF演算法對於白色訊號及AR訊號有不錯的表現,然而在真實語音下,NLMS表現的比MDF、GMDF以及快速牛頓演算法都要好,因此在實際之迴音消除應用上我們建議使用NLMS演算法。
Hand-free communications system finds several applications such as speaker-phone, tele-video conferencing systems, distance learning and medicine systems. A major problem in hands-free system is that of acoustic echo. Adaptive filtering has been considered as a promising way in solving this problem. It is the purpose of this thesis to study several adaptive algorithms including LMS family (NLMS, BLMS, FBLMS, MDF, and GMDF) and RLS family (RLS, Fast RLS, and FNTF). We tabulate each algorithm with operation counts. Intensive simulation has been conducted with different type of input signals including white noise, AR process, and speech. A real acoustic echo path is employed to evaluate the performance of each algorithm. Based on our experiments, the NLMS algorithm seems to be best solution in acoustic echo cancellation.