A numerical algorithm for stable 2D autoregressive filter design

Based on previous theoretical results we present in this paper a global estimation scheme for solving the stable 2D autoregressive filter problem. The different algorithms are based on the traditional Newton method and on the log barrier method that is employed in semi-definite programming. The Newton method is the faster one but the barrier method ensures that the iterates stay in the cone of positive semidefinites. In addition, a numerical test for the existence of a stable factorization of a two-variable squared magnitude response function is presented. Corresponding author: Hugo J. Woerdeman [email protected] Mailing address until July 1, 2003: ESAT-SISTA Katholieke Universiteit Leuven Kasteelpark Arenberg 1