A FAMILY OF NORMALIZED LEAST MEAN FOURTH ALGORITHMS

Abstract

In this work, a family of normalized least mean fourth algorithms is presented. Unlike the LMF algorithm, the convergence behavior of these algorithms is independent of the input data correlation statistics. The first proposed algorithm uses a simple normalization of the regressor and is called simply the NLMF. The second algorithm consists of a mixed normalized LMF (XE-NLMF) algorithm which is normalized by the mixed signal and error powers. Finally, the third algorithm, called the variable XE-NLMF, is a modified version of the XE-NLMF where the mixed-power parameter is time-varying. An enhancement in performance is obtained through the use of these techniques over the LMF algorithm. Moreover, the simulation results obtained confirm the theoretical predictions on the performance of these normalized LMF algorithms.