Automated Design of Millimeter-wave Dielectric Transmitarray Antenna Using Tandem Network

This letter introduces a novel approach that utilizes deep learning for automatic design of millimeter-wave transmitarray (TA) antennas. Instead of focusing on meta-atoms' design and optimization, the proposed method enables the direct design of entire TA antenna and eliminates the need for the time-consuming optimization and fine-tuning process in TA antenna design. Furthermore, an innovative training strategy was employed and yielding a high prediction accuracy exceeding 90%. The well-trained model can generate TA antennas with specified criteria in just 1 ms. As an example, a TA antenna was designed and validated through simulation and experimentation. A measured antenna gain of 24.9 dBi, a side lobe level of -17 dB, and an aperture efficiency of 64% at 80 GHz was achieved, which demonstrate excellent agreement with the input parameters of TA antenna. Our proposed method exhibits remarkable intelligence and speed compared to other TA antenna design methods. IEEE