Machine-Learning-Based Optimization for Wideband Metasurface Mosaic Antenna

This paper proposes a method using machine learning algorithm to realize wideband design of metasurface Mosaic antennas. An effective artificial neural network (ANN) is set up to optimize geometry parameters and realize desired impedance matching performance. The input data of the proposed ANN model is the target| S 11 | and gain performance of the antenna using the proposed non-uniform Mosaic structures. The output results of the model are geometry parameters satisfying the required performances. The proposed antenna is based on a metasurface Mosaic antenna. By optimizing the geometry parameters of the cutting slots, the proposed antenna realizes wideband property. The optimized metasurface Mosaic antenna was fabricated and tested. Measurement results demonstrate that the impedance bandwidth is from 4.91 GHz to 7.18 GHz (37.6%). The measured gain results are 8.90 dBi, 8.24 dBi, and 9.16 dBi at 5.5 GHz, 6.0 GHz, and 6.5 GHz, respectively.