ACOUSTIC ECHO CANCELLATION : A COMPARISON OF SEVERAL ALGORITHMS

Abstract

Acoustic echo cancellation (AEC) plays a key role in successfully implementing a teleconferencing system. Since echo cancellers are basically adaptive filters, the effectiveness of the adaptive filtering algorithms often determines the performance of the echo cancellers. In this thesis, we present a novel adaptive filtering scheme which assumes that a coarse estimate of the power spectral density of the filter input is known a priori. The proposed algorithm uses this information to whiten the input prior to its application in the adaptation process. We emphasize on the specific application of acoustic echo cancellation and experimentally show that the proposed scheme can readily be adopted in the application of AEC. The time-domain and frequency-domain algorithm s or the proposed scheme are presented and compared with the normalized least-mean.-square (NLMS), generalized NLMS (GNLMS), and frequency-domain block LMS (FBLMS) algorithms in terms of convergence and tracking behavior, through computer simulations and actual real-time implementations on a digital signal processing (DSP) processor in a teleconferencing setup The frequency-domain algorithm has similar convergence behavior, better or similar tracking behavior, and close to minimum computational complexity, in comparison with the GNLMS and PFBLMS algorithms. For the time-domain realization, the proposed scheme significantly improves the convergence of the AEC which employs the NLMS algorithm and has a superior convergence as compared to the frequency-domain realizations. However, the time-domain realization is more expensive to implement, as it requires more processing power.