NMP-LTI System Identification Using Evolutionary Spectral Representation

摘要

A common and well known procedure in identification and modeling of stochastic processes is based on the application of the power spectrum. Although this procedure has a wide range of applications, it has certain limitations. The fact that the power spectrum carries only the magnitude information and no phase information implies that min/max phase characteristics had to be assumed a priori to identify the LTI system parameters. Since non-minimum phase systems are systems having zeros inside and outside the unit circle, the identification of non-minimum phase LTI systems cannot be done using the power spectral analysis. Furthermore, for non-stationary processes, processes having at least one time-varying parameter, the conventional power spectrum does not reflect the time variation of the process characteristics. Higher-order bispectrum has been applied in identification of LTI systems under stationarity assumptions and restricting the signal to have non-symmetric distribution. System identification when the input/output are non-stationary requires new techniques and a great deal of work is needed in this area. In this paper, a new method based on the evolutionary spectral representation is proposed for the identification of an LTI system, using a non-stationary input. Examples will be depicted to show the powerful of the proposed method.