On order statistic least mean square algorithms

The order statistic least mean square (OSLMS) algorithm is a modification of the least mean square (LMS) algorithm, that uses an order statistic (OS) filtering operation to the gradient estimates. There are many OS filters that can be applied to the gradient estimates and each of them is optimal for a certain gradient distribution. In this paper a new OSLMS algorithm is introduced, permitting an automatic selection of the “optimal” OS filter. In this algorithm, an adaptive L filter is employed for filtering the gradient. Simulations conducted in a system identification framework show the improvements of the new algorithm comparing with known OSLMS algorithms.