Parameterization of unit-pulse response model

A parameterization of the unit-pulse response model using the eigenvectors of the autocorrelation matrix of the input is proposed. It is shown that the parameters of the parameterized model are directly related to the frequency characteristics. This model makes it possible to develop effective identification algorithms in the time domain, considering the estimation accuracy in the frequency domain. In practice, on the basis of this model, the authors have proposed a modified least-square estimation which is more robust than the ordinary least-square estimation against observation noise