Resource Allocation with Graph Neural Networks-Multi Agent Reinforcement Learning for 6G HetNets

Abstract

The sixth generation (6G) mobile communication system is expected to meet the different service needs of modern communication scenarios. In this paper, the spectrum sharing problem in the 6G HetNets is addressed by combining multi-agent reinforcement learning (MARL) with Graph Neuronal Networks (GNN) for graph-structured problems. A problem is formulated to maximize the sum of the data rate of all the links in the network. In particular, GNN extracts the characteristics of the different types of communication links that coexist in the network taking into account the network topology and the observations of the neighbors. Then, we model the spectrum access of the multiple links as a multi-agent problem and exploit recent progress of multi-agent Reinforcement Learning to develop a distributed spectrum that guarantees good quality of service QoS, with low levels of interference and good data rate. Extensive simulation demonstrates that the proposed method is near-optimal compared with other relevant previous research work.