Parallel Nlms Filters with Stochastic Active Taps and Step-Sizes for Sparse System Identification

Within the framework that two filters are working in parallel, stochastic taps NLMS (ST-NLMS) effectively chooses only active taps for adaptation, resulting in a good transient behavior when identifying long, sparse, echo path like systems. However, ST-NLMS still suffers from the inherent limitation of LMS. This necessitates a compromise between the opposing fundamental requirements of fast convergence rate and small misadjustment. Following the same block diagram as ST-NLMS, the stochastic step-size NLMS (SS-NLMS) scheme is proposed and integrated into the ST-NLMS framework. The combination leads to a novel algorithm called STS-NLMS, which adjusts step-size and active taps simultaneously. Extensive experiments demonstrate that substantial improvements in the speed of convergence are achieved by using the proposed algorithm in stationary environment outperforming both NLMS and ST-NLMS with the same small level of misadjustment. In addition, the proposed algorithm shows superior tracking capability when the system is subjected to an abrupt disturbance. Furthermore, if nonstationary environment is considered, the performance of the proposed algorithm is still satisfying