Deep Learning Optimized Sparse Antenna Activation for Reconfigurable Intelligent Surface Assisted Communication

Reconfigurable intelligent surface (RIS) is a revolutionary technology for achieving high rate and large coverage in future wireless networks by smartly reflecting the signals with adjustable phase shifts. To design reflection beamforming, accurate individual channel state information required at RIS, which challenge task due to lack of signal processing ability passive mode. In this paper, we add units few antennas RIS partially acquire channels extrapolate them full channels, active antenna selection key point but has not been addressed yet. We construct an network that utilizes probabilistic sampling theory select optimal locations these antennas. With network, further two deep learning-based schemes, i.e., extrapolation scheme beam searching scheme. The former convolutional neural from partial while latter adopts fully-connected achieve direct mapping beamforming vector maximal transmission rate. Simulation results show proposed outperforms trivial uniform selection, performance more stable than fewer