Cumulant‐based approach to FIR system identification
Adnan M Al-Smadi · International Journal of Circuit Theory and Applications · 2003
Abstract In this paper, a non‐recursive approach is developed for estimating the coefficients of a moving average (MA) model from only third‐order cumulant statistics of a finite realization of the observations of the output data. The signal observations may be noisy. The excitation signal is assumed to be zero mean, non‐Gaussian stationary sequence that is not observed. The noise is additive and may be coloured Gaussian or non‐Gaussian. This novel technique is based on forming a third‐order cumulant composite data matrix. The method presented here requires the solution of a system of linear equations, which can be achieved using least‐squares methods. The proposed approach is illustrated via computer simulations and is shown to be consistent. Copyright © 2003 John Wiley & Sons, Ltd.