Blind identification of FIR systems using third order cumulants

Abstract

A variety of real systems can be represented (either completely or in part) by the finite impulse response (FIR) systems. The parameters of such a system as well as the phase are often required in many practical situations. Non-minimum phase (NMP) systems may be identified correctly from the higher order cumulants which contain the non-minimum phase information. Five estimators based on third-order cumulants are discussed for the identification of (minimum or nonminimum phase) FIR systems from the observed noisy output which has been contaminated by zero-mean symmetric white noise of unknown variance. Two of these algorithms employ both non-linear optimisation and third-order cumulant matching techniques, and the other three are based on the method of least squares but using a specific relation between the third-order cumulants and the parameters of FIR systems. Results from extensive simulations are presented to highlight the salient features and to make comparison with some published results.