在使用免手持通訊系統時,一個主要的目的是要清楚的將聲音訊號傳送到對方,也能很清楚的聽到對方的聲音,聲學迴音是這個系統所必須克服的重要問題。本論文先介紹適應演算法NLMS與FGMDF,接下來利用電腦模擬在不同系統架構下,比較適應演算法所表現性能之優缺點,進而利用DSP的技術,實現迴音消除器,因此提高系統的聲音傳送品質。 在模擬及實作過程中,我們發現採用雙濾波器架構,能有效的判斷雙邊談話,並把迴音消除。利用FGMDF演算法可以自由的選擇更新區塊的大小,同時由於採用頻率域能量的正規化,我們的系統具有快速收斂的特性,而它的運算量不會隨濾波器長度倍數成長,並改善了MDF時間延遲及NLMS運算量的問題。
In hands-free communications systems, it is essential to transmit voice signal to the other side and hear the other side''s voice clearly. Therefore we have to overcome the acoustic echo problem of the system. In this thesis, we briefly review the algorithms of NLMS and FGMDF. We use computer simulation to compare the performance of several adaptive filtering algorithms. Finally, we use DSP to implement acoustic echo canceller. We employ the dual filter structure to determine double talk state and to cancel acoustic echo. The FGMDF algorithm provides the flexibility of choices of update buffer size. The employment of frequency power normalization makes the filter converge fast. In summary, our DSP Implementation of acoustic echo cancellation with frequency power normalized generalized multidelay adaptive filters enjoys fast convergent performance and affordable time delay and computational complexity.