GCN-Based Multi-Station Collaborative Predictive Beamforming for Integrated Sensing and Communication in Vehicular Networks

To mitigate the junction effect of road side units (RSUs) in integrated sensing and communication (ISAC) predictive beamforming for vehicular networks, we propose a graph convolutional network (GCN)-based multi-station collaborative predictive beamforming. An optimization problem is formulated to maximize the communication sum rate and a penalty-based method is employed to convert this non-convex problem into an unconstrained form. Furthermore, we introduce an attention-based GCN that integrates both graph convolution and attention mechanisms to effectively capture local and global dependencies of graph nodes, thereby enhancing predictive beamforming performance. Extensive simulations demonstrate that the proposed method achieves excellent results, especially with an average communication rate improvement of 5.3 dB at the junction.