A method for identification of time-varying non-Gaussian model using wavelet basis functions

A novel method was proposed to addresses the issue of identification of time-varying linear system with non-Gaussian input. A non-Gaussian AR model with time-varying coefficients was developed to track the non-stationary non-Gaussian characteristics of the signal. For system identification and coefficients estimation, each transient model coefficients was expanded onto a finite set of basis sequences. Wavelet basis function was employed so that the model parameters can be effectively tracked and used to estimate the corresponding local parametric bispectrum. Finally, the performance of the proposed approach was assessed with some simulations. The experimental results show the flexibility and the effectiveness of the presented method.