Robust Nonnegative Least Mean Square Algorithm Based on Sigmoid Framework

Impulsive noise causes nonnegative algorithms to yield excessive error during iterations, which will damage the stability of the algorithm and causes performance degradation. In the paper, a NonNegative Least Mean Square algorithm based on the Sigmoid framework (SNNLMS) is proposed. The algorithm embeds the conventional nonnegative cost function into the Sigmoid framework to receive a new cost function. The new cost function has the characteristics of suppressing the impact of impulse noise. In addition, in order to enhance the robustness of the SNNLMS algorithm under sparse system identification, the Inversely-Proportional Sigmoid NonNegative Least Mean Square (IP-SNNLMS) is proposed based on the inversely-proportional function. Simulation results demonstrate that the SNNLMS algorithm effectively solves the problem of misadjustment caused by impulsive noise. IP-SNNLMS enhances the robustness of the algorithm and improves the defect of the convergence rate of the SNNLMS algorithm under the sparse system identification.