Robust equalization based on a network of Kalman filters in impulsive noise environments

We investigate the problem of channel equalization in "ε-contaminated" impulsive noise environments. We show that the equalizer can be structured into a network of Kalman filters (NKF) operating in parallel. It is based on a state space description of the communication system, the approximation of the a posteriori pdf of the plant noise (related to the transmitted symbols in our case) by a weighted sum of Gaussian (WSG) density functions, and the knowledge of the pdf of the "ε-contaminated" noise which can be written as a sum of two Gaussians weighted by the probability of the appearance of impulsive and Gaussian noise in the observations. The useful information can be extracted from the received sample at any time, even when impulses occur, by exploiting the knowledge of the probability of appearance of the impulsive noise and its variance without using any clipping or localisation mechanism for impulses in the observations. Simulation results show that the performance of the proposed algorithm is less vulnerable to impulsive noise and is more robust than the conventional NKF algorithm based on impulse clipping.