Impulsive noise elimination using polynomial iteratively reweighted least squares

Abstract

A new nonlinear filtering technique is introduced for the elimination of impulsive noise modelled with a symmetric /spl alpha/-stable (S/spl alpha/S) distribution. The new algorithm, called polynomial iteratively reweighted least squares (PIRLS), employs a Volterra filter the coefficients of which are estimated by minimizing the l/sub p/-norm of the estimation error. The filter, hence constructed, is used to estimate the clean data from the corrupted data. Simulation results obtained for audio data corrupted by synthetic S/spl alpha/S noise indicate that PIRLS is very successful in removing impulsive noise.