Graph Neural Network for Multi-User MISO Secure Wireless Communications

This paper propose a graph neural network (GNN) framework to achieve physical layer security. We consider the secure communication between a multi-antenna base station and multiple users, in the presence of multiple eavesdroppers, where the GNN-based beamforming is conducted for secure transmissions. Particularly, we reinterpret the networks roles as graph elements and track the inter-user interference through the graph structure, and thus the secrecy rate maximization is obtained through neural network training. Numerical results indicates that the proposed GNN approach approximate the secrecy performance as compared with the conventional optimization techniques, while obtaining the solution in a more efficient manner, with the ability to adapt and scale in dynamic wireless networks.