Convex Combined Adaptive Filtering Algorithm for Acoustic Echo Cancellation in Hostile Environments

Abstract

Research into combined adaptive filters has been attracting considerable attention; the performance of which is generally superior to that of the individual component filters. Formulating an adaptive filtering algorithm to handle a range of cases can be very difficult, particularly in complicated scenarios. This paper applies a robust convex combination method to the problem of acoustic echo cancellation. According to the minimum mean square error criterion, we first derive the optimal combination factor for a combination adaptive filter with two component filters. Next, we provide an approximation method to the inaccessible optimal combination factor. Finally, we extend this approximated factor to a robust one so that it is applicable even in situations with strong a ambient impulsive noise (IN) and abrupt changes in the echo channel. To do this, we non-linearly transfer the approximated combination factor so that this combination factor exhibits less transient time during the adaptation process of the combination factor and a fewer fluctuations in its steady state. Moreover, we detect the presence of the strong ambient IN and force the combined system output mainly contributed by one of the two component adaptive filters which is less sensitive to the the strong IN. We adopted computer simulations combining two affine projection sign algorithms with different step sizes as an example. Simulation results show that the proposed method is superior to comparable algorithms in terms of the normalized mean squared deviation and echo return loss enhancement.