A Machine-learning-based Compact Metalens Antenna with Improved Directivity and Pattern Symmetry

In this work, a machine-learning based compact metalens antenna is proposed with enhanced directivity and improved pattern symmetry in E - and H-plane. An artificial neural network (ANN) is trained as a surrogate model to relate the incident angle, polarization, and the required phase shift of a unit cell to its geometric dimension. After that, a metalens antenna with a focal-to-diameter ratio of 0.29 is designed and validated through simulation. Compared to a traditional design, the proposed ANN enabled design results in a directivity enhancement of 0.5 dB and a more symmetry E- and H-plane pattern with the 3-dB beam width difference in both planes reduced by 60%. The proposed method is promising for automatic metalens antenna design.