Selective Normalized Subband Adaptive Filter With Subband Extension

Abstract

We present a novel subband adaptive filtering (SAF) algorithm that selects a subset of subbands and uses them to update the adaptive filter weight. The normalized SAF (NSAF) algorithm has a tradeoff between the number of subbands and the convergence speed. As the number of subbands increases, the convergence speed gets faster. However, employing an increased number of subbands raises the computational complexity. To improve the convergence speed, we first extend the number of subbands and then develop a selective scheme exploiting an efficient subset of the extended subbands so as to remove redundancy in the computational complexity. We show that subbands with a larger ratio of the corresponding squared error to an input power should be selected to achieve a similar performance to that of the extended subband adaptive filter. Experimental results show that the proposed NSAF algorithm has better convergence performance compared with the conventional NSAF algorithm.