GNN-Aided User Association and Beam Selection for mmWave-Integrated Heterogeneous Networks

Millimeter-wave (mmWave) and sub-6 GHz hybrid heterogeneous network (HetNet) has gained wide attention for future mobile communications. However, the difference in propagation characteristics of mmWave and sub-6 GHz presents several challenges to achieve efficient utilization of network resources in hybrid HetNets. In this letter, we consider a joint optimization framework that combines user association and beam selection to maximize the network’s sum rate while satisfying access and power constraints at each base station. The problem is highly non-convex and is difficult to solve using standard optimization techniques. To address this issue, we reformulate the problem as an interference graph optimization problem and propose a graph neural network (GNN) aided learning framework to solve the problem. Numerical results demonstrate that the proposed GNN-aided algorithm works well in the HetNets of varying scales and has low complexity while improving the performance of the baseline algorithms by at least 30%.