On the digital simulation of linear cellular neural networks

Cellular nonlinear/neural networks (CNN's) are one of the analog systems that is hard to emulate or simulate on digital systems. It is known that CNN systems are linear for Gabor-type spatial filters. Although it is possible to represent the state equations of the discrete CNN in matrix notation, it is almost impossible to implement the huge state matrix on a digital system without optimization. In this paper some well known linear equation solving methods are optimized for CNN and required computational powers and memories are compared.