Recap antenna synthesis and optimization using backpropagation and radial-basis function artificial neural networks.

摘要

A 4x2 microstrip square patch antenna array, designed to operate in the 5.3 GHz range, was characterized and simulated using finite-element method (FEM) based models in COMSOL Multiphysics as a reconfigurable aperture (RECAP) antenna by controlling the excitation of each element individually. Based on the FEM models, backpropagation (BP) and radial-basis function (RBF) artificial neural networks (ANNs) were developed to: a) synthesize the response parameters, based on changes in the operating parameters (reconfigurable state and frequency), and b) optimize the reconfigurable state based on desired response parameter levels and frequency. The ANNs were tested using the training data (6630 patterns), and with test-only data (78 patterns).;The results show that the RBF ANN architectures generate more favorable results in terms of reproducing the outputs used for training. However, the BP ANN architectures generated better results in terms of generalizing the outputs used only for testing. In terms of synthesis, the ideal balance of efficiency and accuracy was found by using multiple networks in tandem to synthesize the corresponding response parameters, with almost no loss in generality.