Inverse Design of Dielectric Superstrates Based on PINNs for High-Gain Antenna

This work presents an inverse-design method for high-gain dielectric superstrates, integrating the reciprocity principle with physics-informed neural networks (PINNs). The method optimizes the dielectric distribution within a finite design region under the constraint of the wave equation, enabling the incident wave to form a localized field enhancement at a prescribed focal point. A straight-through discriminator layer is incorporated into the structure network, providing a binary representation of dielectric units and significantly reducing the extensive parameter sweeping and full-wave simulations required in conventional phase-compensated superstrate design. Upon completion of the inverse design, placing a radiating antenna at the focal point with polarization aligned to the designed wavefront yields a significant increase in main-beam gain. Simulation results show a maximum main-beam gain enhancement of up to 5.97 dB while maintaining the original operating bandwidth. Experimental fabrication and measurements further confirm improved antenna gain across the operating band after introducing the dielectric superstrate, with a maximum measured gain of 12.75 dBi.