A New Square-Error-Regularized NLMS

The purpose of a variable step-size normalized LMS (VSS-NLMS) filter is to solve the problem of fast convergence or low mis-adjustment. While most VSS-NLMS algorithms need to tune several parameters for better performance, we introduced a tuning-free square-error-based VSS-NLMS algorithm recently . In this paper, we propose a more general structure of that of the generalized square-error-regularized LMS (GSER-LMS) algorithm. Extensive simulation results demonstrate that our GSER-LMS algorithm outperforms several existing schemes in speed of convergence, tracking ability, and low mis-adjustment.