A selective normalized subband adaptive filter exploiting an efficient subset of subbands

In this paper, we present a subband adaptive filter which selects a subset of subbands and utilizes them in updating the adaptive filter weight. The NSAF algorithm has a tradeoff between the number of subbands and convergence speed. The proposed algorithm, thus, increases the number of subbands to acquire improved convergence speed. However, employing an increased number of subband filters raises computational complexity. We use only a subset of extended subbands so as not to have redundant computational complexity, while we maintain performance. To minimize performance degradation from the extended subbands, we show that the larger ratio of the corresponding squared error to an input power should be selected through a geometric interpretation. Throughout the experiments, we show that the proposed NSAF algorithm has good convergence performance compared with the conventional NSAF algorithm.