Hybrid equivalent circuit-deep neural network design of optically transparent metasurface with microwave RCS reduction.

Metasurface-based ultra-wideband radar cross-section (RCS) reduction techniques that offer flexibility and optical transparency are strategically important for electromagnetic protection. Although many methods, including machine learning techniques, have been developed to design complex metasurfaces, they typically require prior knowledge and extensive computational resources. Herein, we propose a physics-guided intelligent design approach to develop flexible optically transparent metasurfaces, which integrates the circuit analog optimization method (CAOM) with a deep neural network (DNN). As a proof-of-concept, a flexible, optically transparent metasurface for ultra-wideband RCS reduction was designed, fabricated, and experimentally characterized. The experimental results align closely with the simulations, demonstrating excellent flexibility and wide-angle RCS reduction from 8.9 to 37.2 GHz (123% fractional bandwidth). The metasurfaces exhibit excellent optical transparency with a visible transmittance of approximately 75%. The proposed metasurface shows great potential for integration into the window glass of stealth aircraft and warships, as well as solar-powered vehicles.