Low-overhead Beam Training Scheme for Extremely Large-Scale RIS in Near Field

Extremely large-scale reconfigurable intelligent surface (XL-RIS) has recently been proposed and is recognized as a promising technology that can further enhance the capacity of communication systems compensate for severe path loss. However, pilot overhead beam training in XL-RIS-assisted wireless enormous because near-field channel model needs to be taken into account, number candidate codewords codebook increases dramatically. To tackle this problem, we propose two deep learning-based schemes systems, where residual networks are employed determine optimal RIS codeword. Specifically, first far-field beam-based (FBT) scheme which received signals all fed neural network estimate In order reduce overhead, partial (PNBT) proposed, only corresponding XL-RIS input network. Moreover, an improved PNBT performance by fully exploring network’s output. Finally, simulation results show outperform existing sweeping approximately 95%.