A statistical method for selecting block structures based on estimated Volterra kernels

摘要

A method for classifying systems as appropriate block structured models is developed based on estimates of Volterra kernels and their variances. While the traditional methods for classifying systems as block structured models are qualitative, quantitative criteria are proposed. Exact least squares regression is used to estimate Volterra kernels so that the statistics associated with these regressions can be applied to the kernel values and to any linear function of the estimates. Methods for LNL, Hammerstein and Wiener cascades are developed and the results for Wiener system classification are presented in detail.