Theoretical And Applied Research, Based On Wavelet Transform And Adaptive Algorithm For Acoustic Echo Cancellation

With rapid advances in communication technology, hands-free systems are widely used in VOIP, handheld mobile terminals and remote conferencing systems. However, along with the multifunction and multiplicative of communications emerge in endlessly, people also have been in trouble own to acoustics echo caused by the coupling between the speaker and microphone while enjoy in the communication quality which has been more convenient, morn abundance and more comfortable. The acoustic echo seriously affects the clarity of the voice.In order to improve the quality of voice communication, a commonly adopted approach to deal with the problem brought by acoustic echoes in speech communication is to apply adaptive echo canceller. To identify and simulate echo route by adaptive filter is the basic principle of acoustic echo cancellation. Because of the insatiability of communication system and phonic itself, together with the special environment requirement of adaptive filter, the high performance of adaptive arithmetic has to be pursued, such as fast convergence, better robust, and computation simply.This thesis conducts a detailed study on a variety of widely applied adaptive filter algorithms---NLMS algorithm, APA algorithm and RLS algorithm. These different are theoretically and empirically compared. In terms of efficiency, NLMS is better than APA; While, in terms of convergence rate, APA is the faster than NLMS, while input signals are relative. How to deal with the strong relevance of the input signal in order to improve the convergence performance of NLMS algorithm? How to get a new algorithm owns fast convergence rate and good efficiency? How to consider comprehensively, and play the advantages of various algorithms to achieve more effective in acoustic echo cancellation? This is the priorities the article will consider.In this paper, input signals of the adaptive filter are firstly orthogonal transformed by wavelet transforming. The input signals are orthogonal decompounded to a multi-scale space using time-frequency localization properties of wavelet in time-frequency, so that, we can reduce the decentralization of the vector auto-correlation matrix character value. That can greatly increase the convergence of the algorithm step size, and improve the convergence speed and stability of NLMS algorithm. In addition, we also find that signal of different frequencies have different sampling points, and the computational complexity is different. The computational complexity of low frequency signal is large, and the computational complexity of high frequency signal is small. If we use different adaptive filter algorithms in different frequency bands, that will be more rational allocation,that will economize the limited resources and use less computation to improve the convergence rate results. So this article proposes a new algorithm (WL-APNLMS algorithm) which based on wavelet transform and adaptive theory for acoustic echo cancellation. From the computer simulation, we known the new algorithm owns a good performance. The new algorithm is superior to NLMS algorithm and APA algorithm which are used frequently in acoustic echo cancellation.Most hands-free systems are multi-channel systems, which are consisted with speakers and microphones. In principle, multi-channel echo cancellation and single-channel echo cancellation are essentially different, but the principle of two-way echo cancellation and multi-channel echo cancellation is the same, only complexity is different. In this article, we explore the performance of two-way NLMS algorithm, two-way APA algorithm, as well as two-way WL-APNLMS algorithm. At present, two-way echo cancellation (SAEC) has been the hot spots in international research.