Abstract
In this paper a procedure is developed for the spatial-domain design of two-dimensional recursive digital filters having separable denominator. The designed digital filter gives the best approximation, in least squares sense, to desired two-dimensional Markov parameters over a finite interval. In the design the digital filter is expressed by a canonical-form statespace model and the conjugate gradient method proposed by Davidon, Fletcher, and Powell is used to optimize a performance index. Since the resulting filter has a separable denominator, it is easy to check the stability. Finally two examples are solved to illustrate the proposed technique.
