Analysis of Training Data Sets in Artificial Neural Networks Applied to a Radio Frequency Problem

In this paper, a novel study has been carried out to understand the functionality of artificial neural network (ANN) algorithms on a radio frequency (RF) problem. Determining the dominant TM 0 resonant frequency of a rectangular microstrip patch located above dielectric substrate backed by a ground plane has been chosen as our RF problem. Extensive training has been carried out using two widely used ANN algorithms: radial basis function (RBF) and feed-forward backpropagation network (FFBNET). Later, a different input testing data set was generated and fed into the ANN algorithms to determine the resonant frequency. Comparative analysis indicated that the RBF is suitable for small data sets whereas the FFBNET is applicable to large data sets.