Reaserch Of Adaptive Signal Processing Algorithms For Echo Cancellation

Adaptive filtering is an important part in the filed of signal processing. Adaptivealgorithms are responsible for estimating and tacking the unknown external system byusing the filter whose filter coefficients can be adapted flexibly, and optimize someperformance parameters. The main purpose of adaptive algorithms chiefly includesthree parts: improving the convergence speed，decreasing the steady-state error andreducing the computational complexity. Therefore, the design of an adaptive algorithmwhich holds fast convergenece speed，small steady-state error and low computationalcomplexity has always been the main task of adaptive filtering.The system identification and interference cancellation are the primary applicationsof the adaptive filtering. Both echo cancellation in wireless repeaters and acoustic echocancellation firstly need to identify the unkown external echo channel with the help ofthe adaptive filter. And then, based on the echo channel identification, the incomingsignal and eatimation of the echo channel are applied to acquire the estimantion of echosignal. Finally, the echo estimation is substracted from the received signal to produceerror signal and the echo cancellation is accomplished, meanwhile the error signal is fedback to the adaptive filter, which adjusts its filter coefficients in order to minimize theerror signal. The adaptive algorithm is the key role for the echo cancellation, therapidity and accuracy for the echo cancellation depends on the convergenceperformance and steady-state performance of the adaptive algorithm.On the backround of the system identification and interference cancellation, thestudy works for several adaptive algorithms based on the three purposes mentionedabove are as follows,1. The mean performance, mean square deviation, mean square error andconvergence condition of normalized subband adaptive filtering algorithm are analyzedbased on the energy conservation principle. A dynamic selection variable step-sizesubband affine projection algorithm is proposed and its computational complexity isanalyzed.2. An optimazed method for the selection of freqnecy bins is proposed in the frequency domain adaptive algorithm, which updates only the most significantfrequency bins for every filter coefficient updating. Compared with the conventionalfrequency domain adaptive algorithm, it decreases the steady-state mean squaredeviation and reduces the computational complexity.3. A varaible step-size fractional tap-length algorithm is proposed. It can controlthe increase or decrease of the tap-length on the process of the filter coefficient updatingflexibly. It improves the convergence rate and robustness compared with the previousvariable tap-length algorithms.4. A convex combination of normalized subband adaptive filter with selectivepartial updates algorithm is designed based on the existing convex combinationalgorithms. Its main advantage is reducing the computational complexity when thememory effect of the unkown system is long and the large number of taps is needed. Ithas an ability of making a compromise between the convergence speed andcomputational complexity.5. An existing control updating logic for the two-path updating algorithm isimproved in acoustic echo cancellation. It not only can distinguish between thedoubletalk and the sudden change of the echo channel, but also holds the similarconvergence performance compared with the existing updating logic. Furthermore, itgains a low computational complexity.