Blind identification and deconvolution algorithms using higher-order cumulants

Abstract

The use of second-order statistics in identification or deconvolution incurs error due to additive Gaussian noise. In contrast, higher-order statistics (HOS) are insensitive to Gaussian perturbation and can be used to characterize nonminimum phase (NMP) systems. In this paper we propose batch, recursive and adaptive form algorithms based on third- and fourth-order cumulants to solve the identification and deconvolution problems. A new order determination procedure is also presented. Simulation results demonstrate that our algorithms do have superior performance when compared with existing algorithms.