Dynamical optimal training for behavioral modeling of nonlinear circuit elements based on radial basis function neural network

The thorough electromagnetic compatibility (EMC) and signal integrity (SI) modeling of a complete system is a very challenging task due to the extreme complexity of high-speed applications. This paper describes a new methodology for the behavioral modeling based on radial basis function neural network (RBFNN). The stable and optimal learning rate, in the sense of maximum error reduction, for each iteration during the back propagation process can be found. The performance of the dynamic optimal learning rates (DOLR) for RBFNN is discussed by minimizing the mean square error (MSE) between actual and measured outputs for the set of neural network. From the experimental results, we can find that the proposed modeling approach enables to generate accurate behavioral model from measurement carried out on devices operating in their application environments.