In adaptive system identi.cation, exploitation of sparsity that may be inherent in the system leads to improved performance of the identi.cation algorithms. The recently proposed ZA-LMS algorithm achieves this by introducing a 搝ero attractor?term in the update equation that tries to pull the coef.cients towards zero, thus accelerating the convergence. For systems whose sparsity level, however, varies over a wide range, from highly sparse to non-sparse, the ZA-LMS algorithm, however, performs poorly, as it can not distinguish between the zero and the non-zero taps of the system. In this paper, we propose a modi.ed ZA-LMS algorithm for tackling the case of variable sparseness, which selectively chooses the zero attractors only for the 搃nactive?taps. The proposed method is very simple, easy to implement and well supported by simulation studies.
展开▼
