Advances in subband adaptive filtering using a low-resource oversampled filterbank implementation

Advanced adaptive filtering algorithms for low-resource real-time subband implementations are discussed. First, normalized LMS (NMLS) and an associated whitening by decimation (WBD) convergence improvement technique are discussed. Next, complexity reduction techniques termed sequential partial update normalized LMS (S-NLMS) algorithms are introduced. Treatment of these algorithms from a multirate filterbank viewpoint provides a novel and valuable insight into its performance. A hybrid algorithm is then presented combining the advantages of WBD and S-NLMS. Both S-NLMS and hybrid algorithms are implemented on a low-resource subband adaptive filter utilizing oversampled filterbanks. Performance evaluations show that considerable computation cost reduction is achieved without materially reducing the convergence properties or the steady-state performance of the subband adaptive system.