A Neuro-Computational Approach to Linear Array Synthesis

Abstract

The field of electromagnetics benefitted a lot due to learning and estimating capability of the
artificial neural network as for determination of most of the parameters, direct relationships between
various variables are missing or involved lengthy computations. In this paper, we try to study and
model the relationship of gain and beamwidth with respect to the number of antenna elements in a
linear antenna array system by proposing a neuro-computational model. The neuro-computational
model is based on a multi-layer perceptron network trained by the Levenberg-Marquardt algorithm.
We found that the multi-layer perceptron network modeled the relationship to a very high degree of
accuracy with MSE as low as of the order of 10-9.