A method for identifying non-Gaussian parametric model with time-varying coefficients

A method for identifying a non-Gaussian AR model with time-varying parameters is addressed. The proposed approach is based on the application of higher-order spectra (HOS) and wavelet analysis. To solve the problem and identify the characteristics of the time-varying linear system, a time-varying parametric model is proposed as a non-Gaussian AR model. The model coefficients that characterize the time-varying system are the functions of time and can be represented by a family of wavelet basis functions, having the invariant basis coefficients. This method can well track the changes of the model coefficients. The experimental results show the effectiveness of the proposed approach.