A Wideband Double-Layer Rectangular Dielectric Transmitarray Designed Using Deep Learning

A simple, wideband double-layer dielectric transmitarray (DTA) is proposed and systematically investigated. The unit cell comprises two stacked rectangular dielectric blocks with variable dimensions. Both transmission and resonance behaviors of the unit cell are analyzed. It is found that the phase-variation range increases monotonically with frequency. Weak transmissions can be observed at the lower and upper gain edge of the DTA, which is attributed to small phase-variation range of non-resonance mode, as well as the energy confinement of the resonance modes, respectively. It is favorable that the lower and upper gain edge can be adjusted independently, which greatly facilitates the wideband DTA design. A deep learning framework of both forward and inverse neural networks is used for fast unit-cell dimension prediction, and genetic algorithm is subsequently applied to optimize the multi-frequency reference phases. A double-layer 20× 20 DTA is designed and fabricated using ceramic-based 3-D printing technology. The measured peak gain is 24 dBi at 33GHz with aperture efficiency of 46%. As compared to the reported wideband DTAs, wider 1-dB and 3-dB gain bandwidths of 34.8% and 42.6% are obtained, respectively.