Bilinear time series in non-Gaussian signal modeling

Non-Gaussian processes are taken to be the output of a bilinear system driven by a Gaussian white noise. The authors develop a 2D quarter-plane bilinear model as a nontrivial generalization of a 1D bilinear time series model. A maximum-likelihood-based parameter estimation method is then developed. Finally, the validity of the model is illustrated by simulation examples.<<ETX>>