A Fast Converging Algorithm for System with Highly Sparse Impulse Response

The proportionate normalized least-mean-square (PNLMS) algorithm with individual activation factors (IAFPNLMS) converges fast when the echo path is highly sparse, and has been used in system identification. Unfortunately, it suffers from slow convergence speed after the fast initial process. To solve the problem, in this paper, the idea of mu-law PNLMS (MPNLMS) algorithm is introduced into the IAFPNLMS algorithm, which results in IAF-MPNLMS algorithm. Simulations show that the proposed algorithm performs better than MPNLMS, improved PNLMS (IPNLMS) and IAFPNLMS algorithms for system identification with highly sparse impulse response.