Fast Beamforming Adaptation for Distributed NOMA in D2D Communication Using MetaGNN

Device-to-device (D2D) communication, when combined with non-orthogonal multiple access (NOMA), holds promise for enhanced spectral efficiency. However, it demands careful radio resource allocation, particularly under topology changes. In this letter, we propose a fast one-shot beamforming adaptation scheme based on a graph neural network (GNN) integrated with meta-learning. The approach computes meta-gradients to derive general meta-parameters optimized for rapid adaptation, rather than direct deployment, thereby enabling effective adaptation to target network conditions. Evaluation results show that the proposed solution achieves sum-rate gains of up to 27% over GNN-based schemes and up to 86% over the multi-agent reinforcement learning (MARL) baseline.