An Ultrathin Filtering Transmitarray Antenna Based on Machine-Learning-Enabled Phase Synthesis

A machine-learning (ML)-enabled phase synthesis method is proposed in this letter to optimize filtering transmitarray antennas (TAs). First, a single-layer filtering TA element is proposed, which not only offers a sufficient phase-shift range and low transmission loss, but also exhibits high selectivity, with transmission zeros below −19 dB. To enhance the gain filtering characteristics, an ML-enabled phase synthesis method is proposed, utilizing convolutional neural network for forward gain prediction and genetic algorithm for inverse optimization. To verify the effectiveness of the proposed method, filtering TAs with an aperture size of 252 mm × 252 mm and a profile height of only 0.073λ are constructed and simulated. Compared to the conventional method, the 1 dB and 3 dB gain bandwidths of the proposed filtering TA increased by 112% and 36%, respectively, with frequency selectivity increased from 48% to 61%. Fabricated filtering TA measurements confirmed these results.