Coding Optimization of programmable metasurface based on BP neural network

An optimization method based on BP neural network is proposed for the controllable multi-beam generated by a programmable coding metasurface reflector. By adjusting the reflection phase of a reflective array unit through digital coding, a variety of different RCS curves are generated. From which, some key features are extracted as input for the BP neural network. Combining deep learning and optimization algorithm, a beam optimization program is designed to find the best encoding array for the desired beam characteristics. This method provides a possible solution for real-time complex beamforming of a reflective array.