A variable step-size-based ICA method for a fast and robust acoustic echo cancellation system without requiring double-talk detector

摘要

In this paper, we propose a novel method for improving performance of the acoustic echo canceller (AEC) employed in the hands-free communication. The main objective is to realize an improved performance without requiring a double talk detector (DTD). The basic idea is to employ a gradient-based independent component analysis (ICA) method with a generalized Cauchy distribution-based flexible score function. In order to achieve an improved convergence and steady-state error performance of the ICA algorithm, a variable step-size (VSS) is proposed for the adaptation rule. Experimental results demonstrate that the proposed method outperforms the existing methods, such as the natural gradient ICA-based AEC method and the VSS normalized least-mean-square (VSS-NLMS) algorithm.