Machine Learning-Enabled Wide-Beam Metasurface Antenna for Wide-Angle Scanning Antenna Array

A machine learning-enabled wide-beam metasurface (MTS) antenna is proposed for the wide-angle scanning array using a fully-connected neural network. The wide-angle scanning based on MTS antenna elements feature the challenges of not only rich degrees of freedom but also the requirements of port matching and specific radiation patterns so that the multi-objective-multi-parameter optimization suffers from high complexity and high computing costs. To alleviate these challenges, an inverse model mapping the performance into the structural parameters is constructed, trained, and verified. With the aid of the inverse surrogate model, the proposed MTS antenna achieves a 3-dB beamwidth from 100° to 125° with a 20% operating bandwidth and is used to construct an eight-element linear array with equal spacing. The designed ML-enabled MTS antenna array achieves beamscanning from −62° to 62° with stable peak gains from 14.2 to 14.5 dBi and scan loss from 2.4 to 2.9 dB over the band ranging from 6.6 to 7.4 GHz.