Research On The Algorithms Of Adaptive Echo Cancellation

With the fast development of VoIP (Voice over IP) and WLAN (Wireless Local Area Network) technology, the integration of WLAN with VoIP has become the mainstream for the future of communication. It is anticipated that VoIP over WLAN will be one of the most critical technologies employed in next generation communication system. Compared with traditional telephone, the fatal disadvantage of VoIP is its worse speech quality. The influence factors are various and one of the most important is the echo. So, in order to enhance the speech quality of VoIP, echo cancellation plays a critical role in the VoIP systems.At present, echo cancellation is mostly implemented by adaptive echo cancellation techniques. However, double talk, in which the near-end and the far-end speakers talk simultaneously, causes a serious problem in the adaptive estimation of the adaptive echo cancellation filter coefficients. To tackle this problem, almost all current adaptive echo cancellation techniques first detect the presence of double talk and then take appropriate actions such as disabling the adaptive algorithm during the double-talk period. But these techniques result in two problems: 1) Since the coefficients of the filter are not updated during the double-talk, once echo path changes during this period, it is impossible for the adaptive filter to trace the change of echo path and reduces the echo cancellation effect; 2) Most of these detectors are too complex in order to achieve satisfactory performance. Therefore, it is possible to make the coefficients of the filter diverge from the optimum due to the time required by the detector to detect double-talk.A doubletalk detector (DTD) is used with an echo canceler to sense when far-end speech is corrupted by near-end speech. Its role is to freeze the adaptation of the model filter when near-end speech is present in order to avoid divergence of the adaptive algorithm. Several authors have proposed to describe many different ways of doing so ,including the Giegel Algorithm, the cross-correlation algorithm and the coherence algorithm. We show in this paper that all the algorithms and the calculated data of the algorithm.