Identification of non-Gaussian parametric model with time-varying coefficients using wavelet basis

Many time series in practice turn to be the time-varying (TV) non-Gaussian processes. In this paper, we address the problem of how to describe these non-stationary non-Gaussian time series. A non-Gaussian AR model with TV parameters is proposed to track the non-stationary non-Gaussian characteristics of the signal. Since wavelet has flexibility in capturing the signal´s transient characteristics at different scales, a set of wavelet basis is employed so that the model parameters can effectively track the variations of TV signals and be used to estimate the corresponding TV bispectrum. The experiments results confirm the superior performance of the presented model over the previous method.