Graph Convolutional Network-Based Channel Extrapolation for Hybrid RIS-Aided Communication

Reconfigurable intelligent surface (RIS) is a promising technology to improve network coverage and transmission rate by intelligently modifying the scattering environment for future communication networks. However, acquiring the channel state information (CSI) in passive RIS scenarios may lead to significant overhead. Therefore, we focus on the channel acquisition of hybrid RIS-aided communication systems based on deep learning (DL). Specifically, we utilize graph convolutional network (GCN) to extract the spatial correlation features for selecting the optimal activation RIS elements and channel extrapolation along the antenna domain. Additionally, weakly connected edges are pruned to reduce computational cost. Simulation results validate the effectiveness of the proposed channel extrapolation scheme.