Postsmoothing for the LMS algorithm and a fixed point roundoff error analysis

The effect of post-algorithm smoothing on digital implementations of the least mean square (LMS) algorithm is studied. An expression is derived for the mean square error (MSE) with post-algorithm (PA) smoothing but without finite wordlength effects. It is shown that the MSE can be reduced from that of LMS without PA smoothing. PA smoothing results in reduced convergence speed. The effect of fixed-point finite precision on PA smoothing is studied. An expression for the MSE is derived. Monte Carlo simulation results that support the result that digital LMS with PA smoothing performs better than digital LMS without PA smoothing are presented.<
>