The Research And Implementation Of Adaptive Echo Cancellation Algorithm

Adaptive echo cancellation algorithm is an effective method to solve echo problem in the fields of multimedia communication and etc. In the video conference, hand-free communication and other multimedia communication systems, the existence of echo badly influences communication quality. Especially for telecommunications system, because transmission delay increases, the impact of echo is much more distinct. So how to cancel echo in the multimedia communication systems already becomes a very important problem. The echo in communication systems commonly denotes"acoustic echo". In the local end, the voice of remote end caller is played in the form of sound wave by loudhailer, sound wave forms repetitious voice through the transmission and reflection in the house. The echo signal is received by microphone in the local end, a feedback loop is formed. Through this loop, echo is transmitted back, so the caller hears his own voice, this is called echo. The paper introduces three methods for echo cancellation. Acoustic echo canceller (AEC) is the main approach for echo cancellation. AEC is based on the relativity of loudhailer signal and multi-path echo produced by it, establishes voice model of feedback loop, utilizes the model to estimate echo, modifies the coefficient of filter continually and makes the estimation value much more approach actual echo. Then, subtract echo evaluation value from input signals of microphone, consequently achieve the goal of cancelling echo. AEC compares the input of microphone and the past values of loudhailer, consequently cancels acoustic echo which is prolonged delay and multi-reflected. Based on the past output values of loudhailer which are stored in memory, AEC can cancel various delayed echo.Adaptive filter researched by the paper is adaptive FIR (finite-duration impulse response) filter which is the most suited to cancel echo signal, it is also called tapped delay line filter or finite impulse response filter. Adaptive FIR filter is a special filter which can automatically adjust its parameters. In the design phase, FIR filter doesn't need to know the knowledge about input signal and noise statistic characteristics in advance, it can gradually"realize"or estimate the needed statistic characteristics in its work process, based on these it automatically adjusts its parameters in order to achieve optimal filter effect. Once the statistic characteristics of input signal are changed, it is capable of tracking the change, automatically adjusts parameters and makes the performance of filter optimal over again. The optimal denotes to satisfy certain optimal rule (namely goal function).ParallelλVSS-LMS algorithm proposed by the paper is a kind of adaptive FIR filter, the goal function adopted by it is mean squared error (MSE). Through the discussion of this paper it can be known that the optimal solution of adaptive FIR filter's goal function is wiener solution. The most effective method to gain wiener solution is to utilize steepest descent algorithm to search. Actually in the process of echo cancellation, the obtaining of exact estimation of correlation matrix R and cross correlation vector P between tapped input and expected vector is impossible, so the exact measure of gradient vector cannot be obtained. Therefore, the use of steepest descent algorithm in echo canceller is localized. In the end of fifties, Windrow and Hoff, et al, first proposed least mean square (LMS) adaptive algorithm. The algorithm is a search algorithm based on steepest descent algorithm. It obtains gradient estimation through instantaneous estimation of gradient vector, consequently gets wiener solution. LMS algorithm adoptsinstantaneous estimation, implements simply, doesn't need to calculate relational correlation function and doesn't need matrix inversion algorithm, so we adopt the kind of LMS algorithms in echo canceller.The paper compares LMS algorithm and its improved algorithms using experimentation. Through the experimentations, it is known that LMS algorithm is simple and easy to implement. But in LMS algorithm, step factorμis invariable, the requirement of step factorμon convergence rate and tracking ability is inconsistent with convergence precision in adaptive algorithm, the inconsistency is difficult to solve. The step adjustment principle of variable step adaptive filter algorithm is that in the phase of initial convergence of algorithm and when changes occur in unknown system, adopt biggish step factor. In this way, quick convergence rate and strong tracking ability can be obtained in initial phase. After the convergence of the algorithm, adopt lesser step factor, obtain lesser homeostasis imbalance. In recent years, there are many improved researches on variable step adaptive filter algorithm.λVSS-LMS algorithm is a new variable step adaptive filter algorithm, it improves NLMS algorithm. Calculation quantity increases little, but through experimentation it is known that it can evidently quicken convergence rate, after convergence it can achieve smaller and steady least mean square error (MSE).ParallelλVSS–LMS algorithm proposed by the paper is based onλVSS-LMS algorithm. LMS algorithm and its improved algorithms are simple, robust, easy to implement, so they are widely applied in the design of adaptive filter. But the convergence rate of them is slow, moreover they are difficult to satisfy the fast process requirement on great data quantity in real project. The paper proposes a new adaptive filter algorithm—parallelλVSS-LMS algorithm. The algorithm is on the basis of parallel LMS algorithm, introduces modifier coefficientρand oblivion factorλi ofλVSS-LMS algorithm, utilizes current and past error information in the total number of M to decide the iteration step of the next step and then adjusts tapped coefficient of filter in order to quickly converge to optimal value wopt. The algorithm possesses steady performance, fast convergence rate, is easy to implement and can satisfy fast process requirement.The paper presents the whole definitions of parallelλVSS–LMS algorithm, includes expressions of basic elements and algorithm description. Expressions of basic elements are as follows: input signal of parallelλVSS-LMS algorithm, output function, error function, update formula of step factor and weight recurrence formula. Through the detailed algorithm description of parallelλVSS-LMS algorithm, it is easy to see implementation process and iteration condition of parallelλVSS-LMS algorithm. In the implementation process of adaptive echo canceller, the paper presents the flow chart of adaptive filter based on parallelλVSS-LMS algorithm.The paper implements an adaptive echo canceller based on the thought and interrelated algorithms of adaptive FIR filter. The echo cancellation system is composed of two main modules: adaptive filter module and voice activity detector module. Voice activity detector module is divided into three parts: local-end voice detector, remote-end voice detector and dual-end voice detector. It makes echo canceller have different work states in different voice modes and makes the algorithm much more integrated and robust. The paper presents the run process of the echo cancellation system and detailed descriptions of main modules, introduces functions and usages of main methods, and shows the flow charts of echo canceller and adaptive filter.