A predictive updating scheme to improve the NLMS algorithm for acoustic echo cancellation

This paper presents the normalized least mean square (NLMS) algorithm with an predictive updating scheme to improve the performance of an acoustic echo canceler. According to the grey system theory, input speech and echo signals are preprocessed by an accumulated generating operation, and then a low order polynomial fitting is utilized to generate the predicted data. By using these predicted data with some consistent properties, the scheme proposed herein adjusts coefficients of the adaptive FIR filter based on the NLMS updating process for a good convergence performance. The computer simulation results demonstrate that an AEC using the proposed scheme can obtain a 6 dB improvement over the conventional repetitive updating one