A Robust Sparse System Identification via Integrating an Individualized Time-Varying Stepsize Adjustments

In this paper, we present a new adaptive filter solution that mitigates the impact of highly correlated input signals to a system with sparse and long impulse response, reduces negative effect of sparsity measures on the performance of sparse related algorithms, and improves the trading-off between achieving a fast convergence rate and attaining a small steady-state MSE. We based this new solution on a recently proposed technique that innovatively employs a tap-individualized time adjusting stepsize. Based on the total input power and input signal associated with the underlying tap, each tap of the adaptive filter has its own variable stepsize adjustment. This adjustment is based on a logarithmic formulation designed to deliver a consistent reduction in stepsize in each iteration by leveraging the low input excitation and, at the same time, compressing the high input excitation into a relatively moderate input range. Simulation results show that the proposed algorithm delivers a superior performance, at various system sparsity levels and input types, compared with known sparse-aware algorithms.