A Novel Variable Step-Size NLMS Algorithm and Its Analysis

Abstract By building the generalized Sigmoid function relationship between normalized step-size and the power of error signal, a novel variable step-size NLMS algorithm is proposed. It is proved that the step-size of NPVSS-NLMS changes as the new algorithm does when A=σv -m and B=2. The physical meanings of the parameters in this algorithm are explored. The theoretical analysis illustrate that this algorithm combine the virtues of NPVSS-NLMS and Sigmoid function, and it leads to faster convergence rate and lower final misalignment. The computer simulation results support the theoretical analysis.