Neural network simulation of a dielectric ring resonator antenna

This paper discusses the application of neural network (NN) techniques in the modelling of an electromagnetic (EM) field problem consisting of a dielectric-ring resonator antenna. The NN takes into consideration the geometrical and material parameters of the antenna, and the input reflection coefficient values for different frequencies in the bandwidth of interest. The NN is trained off-line using results obtained with the finite element method (FEM). The NN is considered as a nonlinear function approximator and the Levenberg-Marquardt algorithm is used to solve the corresponding least-squares optimization problem. The NN approach is found to provide a tremendous computational speed improvement as well as computer memory reduction compared to the traditional numerical methods of analysis.