Comparison between two non-minimum phase system estimators

The most commonly used moving average (MA) parameter estimators are approximations to the maximum likelihood estimator (MLE). The use of the MLE for parameter estimation can be justified for large data records because of its asymptotic unbiased and minimum variance properties. Moreover, it is well known that the MLE has a zero error asymptotically. A new estimator of the MA parameters was also developed. The main property of this estimator is to give, in absence of noise, a zero error for a finite signal point number with a high known probability. The performance of this estimator in the presence of an additive Gaussian or bounded noise have been studied. It is not restricted to minimum phase systems. The aim of this paper is to compare its performance with those obtained with a higher order estimator (GM estimator) developed by Giannakis and Mendel (1989) for identification of non-minimum phase systems driven by binary inputs