Research On Echo Cancellation Technology Based On Adaptive Algorithm

With the rapid development of the communication technology, more and moreequipments provide hands-free function, such as video conference based on VoIPcommunication technology, telephone conference, and hands-free telephone. However,echo is usually included in these communication systems, and it needs to be canceled.The research and the design based on FPGA for the algorithms of the echocanceller are implemented, which is based on the causes of the acoustic echo, thehistory and current status of the research for echo cancellation technology and theresearch for the principles of the echo canceller. The theory of the echo cancellation incommunication system is researched first, then the algorithms of each modules in echocanceller are researched, especially the core adaptive filter module, including Wienerfilter algorithm, the steepest descent algorithm, LMS, NLMS, DLMS and theBLMS adaptive algorithms. The LMS and NLMS algorithms are simulated andanalyzed by using Matlab. According to the theoretical research and the simulation andanalysis, the length and the update step-size of the echo canceller based on FPGA areselected. After the algorithm is simulated successfully by using Matlab, the design andimplementation of the echo canceller based on FPGA are researched. And the adaptivefilter module and the remote voice detection module are designed by using Quartus IIand Verilog HDL. The function of the adaptive filter module and the remote voicedetection module which are designed by using Verilog HDL is simulated in Quartus II.Then according to ITU-T G.167, the adaptive filter module is simulated by usingQuartus II and Matlab co-simulation approach and its performance is tested.The algorithm used in this paper is DLMS which is based on delay. The value ofthe D is not1, which is different form the common DLMS algorithm, instead, its valueis adjusted to using parallel processing and pipeline design approach in the design, andthe efficiency of the system is improved.