Joint Admission and Power Control for Massive Connections via Graph Neural Network

The sixth-generation mobile communication system (6G) puts forward higher requirement for connection density, which is difficult to meet with the existing resource management schemes in real time. In this paper, we investigate the graph neural network (GNN) based algorithms for supporting the massive connectivity in 6G. Using the power intensity of the received signal or signal-to-interference-plus-noise ratio (SINR) as a measure of communication quality, we aim to maximize the number of links that meet quality of service (QoS) requirements in a given area through joint admission and power control. Specifically, we consider two models. Among them, the blocking interference model presets the transmit power of the link in advance, and only needs admission control. After the original problem is converted to the maximum independent set (MIS) problem, we design a solution based on graph convolution network (GCN) and Q-learning. The accumulative interference model considers all the interference in the scene and controls the power and access jointly. For this model, we propose an algorithm based on graph attention network (GAT). Simulations demonstrate that the proposed GNN based algorithms preserve small computation time and achieve significant performance gain even with large network scale. As such, they are very suitable for the 6G scenario with massive connections.