Modeling Broadband Microwave Structures by Artificial Neural Networks

Abstract

The paper describes the exploitation of feed-forward neural networks and recurrent neural networks for replacing full-wave numerical models of microwave structures in complex microwave design tools. Building a neural model, attention is turned to the modeling accuracy and to the efficiency of building a model. Dealing with the accuracy, we describe a method of increasing it by successive completing a training set. Neural models are mutually compared in order to highlight their advantages and disadvantages. As a reference model for comparisons, approximations based on standard cubic splines are used. Neural models are used to replace both the time-domain numeric models and the frequency-domain ones.