Spatial filtering using linear analog parallel architectures

Cellular Neural Network (CNN) type analog parallel architectures have been studied both for their potential in high-speed image processing applications and for exhibiting rich and exciting nonlinear spatio-temporal dynamics among which waves and patterns. Such architectures have been also used as linear spatial filters for piecewise linear cells working in the linear central part of their characteristics. The spatio-temporal dynamics of homogeneous CNN's working in the above region can be studied using the decoupling technique. Since, for nohomogeneous arrays, this technique is no more valid, numerical simulations were used. The paper presents results regarding the filtering capabilities of analog parallel architectures of CNN simulated at system and transistor level.