Improved convergence of an adaptive identification algorithm for a large sparse system

Estimation of small tap-coefficients of large sparse system using μ-law based proportionate normalized least square (MPNLMS) algorithm yields slow converges, since the proportionality of these coefficients is ignored in the updated process. The individual activation factor-MPNLMS (IAF-MPNLMS) algorithm solves this problem by assigning new gain distribution factor while updating the tap-coefficients. However, this algorithm also suffers from slow convergence if an input signal is highly correlated. In this paper, a modified activation factor is suggested to improve the convergence speed of the MPNLMS and IAF-MPNLMS algorithms for sparse system identification while the input signal is highly correlated. For correlated input signals, the simulation results show that the proposed algorithm acquires better convergence speed than the other two algorithms.