Performance comparisons of the forward-backward LMS adaptive filters and the split-path LMS adaptive filters for adaptive line enhancement

Abstract

Performance comparisons of the forward-backward LMS (FBLMS) adaptive algorithm and the split-path LMS (SPLMS) adaptive algorithm are presented. Typical simulation results are shown for the adaptive line enhancement application. It is shown that the FBLMS algorithm uses both forward and backward linear prediction errors in two different filters. The FBLMS algorithm converges at the same speed as the LMS, but gives about half the misadjustment and about twice the number of computations. The SPLMS algorithm also uses both forward and backward linear prediction errors, but splits the filter in two to maintain about the same number of computations. The SPLMS algorithm converges faster than the LMS, but gives some misadjustment and requires a few more additions. >