Enhancement of residual echo for improved acoustic echo cancellation

This paper investigates the use of a signal enhancement technique, namely a noise suppressing nonlinearity, on the adaptive filter error in order to increase the stability and the performance of acoustic echo cancellation (AEC) when there is a continuous distortion to the acoustic echo signal. The algorithm presented here differs from others in that the enhancement of signal is done in the adaptation loop, rather than as a post-processing technique for further reduction of residual echo in the signal, and that the resulting nonlinearity for the cancellation error is formulated as a solution to the signal enhancement problem. Combining the nonlinear error suppression method with NLMS and other adaptive step-size algorithms based on NLMS shows an improvement of between 5 to 15 dB in the average ERLE for additive white noise and around 2 dB for speech coding distortion when a simulated acoustic echo is used. The reduction of the misalignment of 5 dB or more for both noise cases can be expected. The technique is shown to be beneficial also with a real acoustic echo. The new method is seen as a viable technique for improving the existing AEC algorithms when the acoustic echo is corrupted by linear distortion in the form of additive noise or by nonlinear distortion in the form of speech coding.