Graph Neural Networks for Hybrid Beamforming in MIMO Rate Splitting Multiple Access

In this paper, we investigate the use of self-supervised data-driven schemes for digital precoder optimization of hybrid analog-digital beamforming in the downlink of a Multiple-Input Multiple-Output (MIMO) Rate Splitting Multiple Access (RSMA) system. We propose two architectures based on Graph Neural Networks (GNN) and the Multi-Layer Perceptron (MLP) respectively, and analyze their achievable sum-rate performance in the overloaded regime as the system scales up. We use conventional schemes such as Zero Forcing (ZF) and Weighted Minimum Mean Square Error (WMMSE) precoding as benchmarks. Additionally, we compare their complexity in terms of the number of trainable parameters and number of operations, where GNN models have been found out to be scalable solutions.