Minimum-norm realization of 2D recursive filters: a quasiconvex programming approach
W.-S. Lu · 2003
The minimum-norm realization (MNR) of a 2D recursive digital filter is considered and it is shown that an MNR problem can be formulated as a quasiconvex optimization problem in which the largest generalized eigenvalue of a certain matrix pencil is minimized. It is demonstrated that efficient interior-point convex programming techniques such as Nesterov and Nemirovskii's projective method can be used to perform the optimization with considerably reduced computational complexity compared to the existing minimization techniques.