Deep Learning for Arbitrary Backscattering Programmable Metasurface

We propose a programmable metasurface based on deep learning to realize arbitrary backscattering radar cross section. This method encodes the metacell into binary values of “1”, “2”, “3,” and “4” respectively, using a 2-bit programmable digital matesurface unit that relies on a switch diode. Enable amplitude prediction across a broad angle range from -60 to 60 degrees. We successfully validate the effectiveness of this method through various experiments