Application of Neural Networks to Simulate a Monopole Microstrip Four-Tooth-Shaped Antenna

Abstract

The application of a multilayer perceptron (MLP) for calculating the electrodynamic characteristics of a monopole microstrip four-tooth-shaped antenna (direct problem) as well as for solving the problem of synthesizing a four-tooth-shaped antenna for given characteristics (inverse problem) is considered. A backpropagation algorithm is used to train MLP. Various MLP architectures are considered. The errors of MLP prediction of various electrodynamic characteristics of the antenna are estimated. The graphs of dependences of a network operation error on the number of neurons in the hidden layers are presented. The proposed MLP architecture with two hidden layers each having 16 neurons is found to give a sufficient accuracy. A conclusion is made regarding a more accurate determination of parameters with a given architecture using MLP in comparison with regression models.