Graph Neural Networks to Enable Scalable MAC for Massive MIMO Wireless Infrastructure.

Next generation wireless networks are expected to run efficiently on general purpose hardware to enable virtual radio access network (RAN). As the demand for data rate increases and the number of wireless devices explodes over time, conventional methods fall short as they suffer from high complexity/latency or significant performance loss. Therefore, scalable, intelligent, and low complexity algorithms that extract complex features of large-scale networks are needed. An emerging problem for next generation wireless networks is user equipment (UE) scheduling for multi-UE massive multiple input multiple output (MIMO) systems. In this paper, we propose an artificial intelligence (AI) based solution for massive MIMO base stations to reduce complexity of UE scheduling in the medium access control (MAC) layer. We propose a scalable and low complexity one-shot (single inference without loops) UE scheduling using graph neural networks (GNN). Our results show that our GNN based solution reduces complexity by about 90% compared to greedy search techniques while achieving up to 96% weighted sum rate of optimal UE grouping.