Experimental results of subband acoustic echo cancelers under spherically invariant random processes

The subband adaptive filter system has been applied to the acoustic echo cancellation problem in order to overcome the problems of slow convergence due to spectrally dynamic input and high computational costs associated with a single, long adaptive filter. Research into the convergence characteristics of subband acoustic echo cancelers (AECs) have either used Gaussian white noise, USASI noise (as an approximation for speech signals), or short term speech signals. Research in the statistical modeling of speech signals using spherically invariant random processes (SIRPs) and analyses of LMS and NLMS algorithms under SIRPs has led to a better understanding of adaptive filter performance under more realistic input in the context of AEC. We present experimental results using a subband AEC under SIRP input. These results yield a better understanding of the performance in actual acoustic echo cancellation applications and highlight the benefits of subband techniques