Modified stabilisation technique for 2-dimensional recursive digital filter design

Abstract

A new stabilisation procedure is presented, based on the spectral factorisation in the cepstrum domain. Pistor's stability criterion is used for the stability of the resulting filter. It is shown that any unstable causal 2-dimensional filter function can be decomposed into stable filters that recurse in a halfplane. The new procedure has been compared with the planar least-squares inverse of Shanks and the discrete Hilbert transform of Read and Treitel. It has been found that the present stabilisation technique has a better frequency response approximation than the existing ones. The method is therefore an attractive alternative to the least-squares procedure and the discrete Hilbert transform method.