Transmitarray Design of Dielectric Resonator Antennas Using Deep Learning

This study explores a transmitarray design of dielectric resonator antennas operated in X-band (8-12 GHz). In our design, a 3x3 unit that includes the mutual coupling effect is employed to account for the coupling effect, with each sharing the same Floquet port. A two-head deep learning model based on the Res-Net architecture is developed to efficiently determine the phase value and transmission coefficient. With the height and diameter distribution being the inputs, the model successfully achieves a minimal mean squared error (MSE) loss of 0.2855 and a high Pearson correlation coefficient of 0.9954 on the test dataset. The effectiveness of the trained model is demonstrated through the optimization of a transmitarray using a sliding window approach, resulting in a synthesized window-transmitarray with superior gain performance compared to the conventional calculation methodologies.