Polarization Insensitive Programmable Metasurface RCS Controlling Based on Deep Learning

Radar Cross Section (RCS) control based on programmable metasurfaces has significant research prospects in radar stealth and deception. However, traditional optimization algorithms cannot accomplish the metasurface encoding design for given RCS control indicators. Therefore, this paper proposes an RCS control inverse design encoding scheme based on a tandem neural network. The training set is based on a programmable metasurface composed of 1-bit polarization-insensitive metasurface units with good phase and amplitude characteristics in the wide frequency band of 7.87-12.44GHz. By combining forward prediction and inverse design in a tandem network architecture, real-time encoding design can be achieved. Simulation results show that by performing spatial encoding throughout the X-band, dynamic RCS control of the normally incident plane wave can be achieved, and the maximum RCS attenuation value of a single frequency point exceeds 20dB.