Learning Resilient Radio Resource Management Policies With Graph Neural Networks

We consider the problems of user selection and power control in wireless interference networks, comprising multiple access points (APs) communicating with a group equipment devices (UEs) over shared medium. To achieve high aggregate rate, while ensuring fairness across all users, we formulate resilient radio resource management (RRM) policy optimization problem per-user minimum-capacity constraints that adapt to underlying network conditions via learnable slack variables. reformulate Lagrangian dual domain, show can parameterize RRM policies using finite set parameters, which be trained alongside variables an unsupervised primal-dual approach thanks provably small duality gap. use scalable permutation-equivariant graph neural (GNN) architecture based on topology derived from instantaneous channel conditions. Through experimental results, verify configurations further demonstrate that, such adaptation, our proposed method achieves superior tradeoff between average rate 5th percentile -- metric quantifies level allocation decisions as compared baseline algorithms.