Adaptive filters based on the high order error statistics

This paper presents convergence analyses of the stochastic gradient adaptive algorithms based on high order error power criteria. In particular, our attention has focused on investigating the statistical behaviour of the least mean absolute third (LMAT) and the least mean fourth (LMF) adaptive algorithms. For each algorithm, under a set of mild assumptions, we have derived nonlinear evolution equations that characterize the mean and mean-squared behaviour of the algorithm. Computer simulation examples show fairly good agreement between the theoretical and actual behaviour of the two algorithms