On Cause and Cancellation Techniques of Acoustic Echo in Voice Communication

Acoustic echo cancellation is common in voice communication. The interference signal caused by acoustic echo will reduce the quality of voice transmission. The current paper discusses the cause and technique for cancellation of acoustic echo, mostly based on the use of adaptive filter. Adaptive filter iteratively alters its parameters to minimize the difference between its output and the actual signal. Morever, this paper makes a comparison of various adaptive filtering algorithms and elaborates the detailed operational process of NLMS filtering algorithm. Introduction It has many problems in the technology of voice communication, with the most significant being the echo. Echo is from voice speaker which is transferred by microphone again after being reflected many times in voice communication process. It will seriously reduce the quality of the reproduced voice signal. The acoustic echo cancellation is a key technology to eliminate echo so that the quality of voice in the voice signal transmission process could be improved [1]. This paper mainly addresses on introducing the overall generating mechanism and cancellation principle of echo cancellation. Futhermore, this paper draws a conclusion that NLMS filtering algorithm is more effective and uncomplicated than others after making a comparison of various adaptive filtering algorithms. Generating Mechanism of Acoustic Echo The acoustic echo is a phenomenon that the original signal is delayed and anamorphic, and then collected again by microphone. The echo can be divided to two types: the direct echo and the indirect echo. The direct echo is the sound from speaker without any reflection, being directly transmitted into the microphone. The delay time of direct echo is short. The indirect echo is the mixed sounds from the voice speaker which has been reflected many times before transmitted into the microphone together with the original sound. The impulse response of indirect echo will vary quickly and wildly, and the maintained time is longer than direct echo. There is no echo but voice with lower quality when the return time of reflected wave is very short or the amplitude of echo is small. But the echo will be significant when its delay time exceeds tens of milliseconds or amplitude exceeds 30dB [2]. The generating mechanism of acoustic echo is shown in figure 1. Fig. 1 Generating mechanism of acoustic echo 5th International Conference on Education, Management, Information and Medicine (EMIM 2015) © 2015. The authors Published by Atlantis Press 847 Principle of Acoustic Echo Cancellation The principle of acoustic echo cancellation is to build a model which can simulate the actual echo path exactly and the output of this model is the prediction of actual signal. The echo will be eliminated when it subtracts the predicted signal. The most mature method in actual acoustic echo cancellation is to use an adaptive filter to simulate the echo path, with the adaptive filter being the core of echo cancellation. The principle of acoustic echo cancellation is shown in figure 2. Fig. 2 Principle of acoustic echo cancellation In figure 2, the echo path ) (x H can be treated as a function, and it depends on the actual environment. The adaptive filter can also be treated as a function ) (̂x H , and it should approach ) (x H as close as possible. They should produce the same result when their input is identical. The adaptive filer can adjust its parameters with the actual environmental changes, which means that ) (̂x H can keep on simulating ) (x H . The output of filter is ) (̂n y . The ideal error ) (̂ ) ( ) ( n y n d n e   should be zero, with ) (n d as the actual output, which implies that the echo has been eliminated thoroughly. The utility of adaptive filter is illustrated in figure 3. Fig. 3 Function of adaptive filter The principle of adaptive filter is very simple. Finite impulse response filter (FIR), as a frequently-used filter, consists of multipliers, unit delay cells and adders. It usually represents the predicted value of the future output signal by the linear combination of a period of time signals: