ICNet: GNN-Enabled Beamforming for MISO Interference Channels With Statistical CSI

Abstract

This paper proposes a graph neural network (GNN) based model, termed interference channel net (ICNet), for multiple-input single-output (MISO) interference channels with statistical channel state information (CSI). After a graphic representation of MISO interference channels, the ICNet directly maps the statistical CSI to the beamforming vectors, and the attention mechanism is adopted to jointly utilize the node and edge features. Via unsupervised learning, the ICNet is able to solve the outage-constrained energy efficiency maximization problem. Numerical results show that the ICNet is with millisecond-level inference time compared to the second-level inference time of the convex optimization based algorithm and achieves less than 5% average optimality loss to the convex optimization based algorithm.