A robust adaptive filtering algorithm with variable step size

A varible-step adaptive normalized least mean square (VS-NLMS) filtering algorithm which is robust in the presence of interferences is presented. An estimated cross-correlation function between the input reference signal and the error signal is used as the variable step size. An exponential weighting recursive estimation algorithm is proposed to update the cross-correlation coefficients. Compared with the conventional fixed-step adaptive filtering algorithm (NLMS), the VS-NLMS algorithm can achieve much lower misadjustment noise in the steady state with the same convergence speed and tracking ability. The properties of the VS-NLMS algorithm are confirmed by theoretical analysis and computer simulation