2-D system identification from bispectrum samples

The problem of modeling 2-D system transfer functions given only bispectrum samples of the noisy output is considered. This identification procedure is a three-step process. First, the output bispectrum samples are calculated along a 2-D slice in the 4-D bispectrum frequency region. Next, these samples are used to obtain a rational model of the bispectrum slice via a 2-D divided difference approach. Finally, two methods for finding the underlying system transfer function parameters from this rational model are developed. This system identification procedure is general in that it does not impose causality, recursibility, or minimum phase restrictions on the system to be identified.<
>