Neural Network–Assisted Metasurface Retroreflector With Wide‐Angle Adaptivity

ABSTRACT Retroreflectors that scatter the incident electromagnetic (EM) wave along the incoming direction are widely used for backscattering enhancement and intensifying echo signals for many applications, including collision avoidance systems, lifesaving, and satellite communications. Recently, metasurface has been applied for achieving retroreflectors with extremely low profile, but they always require a predetermined incident angle information and have limitations for wide‐angle and polarization‐independent incidences. Here, we propose an adaptive retroreflector composed of an active anisotropic metasurface for EM waves incidences of wide‐angle and multiple polarizations. The reflective phases of the constituent meta‐atoms under orthogonal polarized incidences can be controlled independently and tuned continuously by individually altering the anisotropic characteristic of the meta‐atoms. The proposed metasurface first estimates the direction of arrival (DOA) based on a pre‐trained artificial neural network with space–time coding–modulated metasurfaces and then modulates the phase gradient accordingly to achieve the adaptive retroreflection. As a proof of concept, we construct a tunable metasurface and experimentally demonstrated at microwave frequency that high‐efficient on‐demand adaptive retroreflection for multiple polarizations can be achieved over a continuous incident angle range from −50° to +50°. The proposed retroreflector can be extended to other frequencies and may find promising applications in communication systems, remote sensing, and radar scattering enhancement techniques.