Novel Neural Network Models of Q-Type Integrals and Their Use for Circular-loop Antenna Analysis

In this paper, the neural network methodology was applied to develop analytical models, deflned by means of weighted sums of basis functions, to approximate/interpolate the class of Q-type power-radiation integrals arising in antenna theory. The associated problems in this proposed approach, i.e., the choice of the type and number of basis functions, and the model's parameters optimization, were replaced by the problem of training weights in neural networks. The better neural model resultant of this research (obtained through of the damped-sinusoid basis functions set) was used for e-cient evaluations of the circular-loop antenna radiation resistance and directivity. The neural model accuracy and computational e-ciency were compared with recent approaches published in literature. In the recent years, the analysis of circular-loop antennas has been revisited. The analytical ap- proximations for radiation resistance and directivity have been presented by means of series of Bessel functions (1{2) and approximate expressions (3). A class of power-radiation integrals named Q-type integrals that arise in antenna theory were considered. Such integrals often appear when cylindrical coordinates are used in the analysis of the radiated power (particularly, in the case of a circular-loop, a circular microstrip antenna, and a circular aperture). The Q-type integrals representation was given in (1), using the usual Bessel function notation, as follows: