A Scalable and Low-Complexity Coordination Framework for Self-Organizing Networks

Self-organizing network (SON) has been introduced as a collection of functions to automatically manage and optimize dynamic networks. For network-wide deployment of SON, coordination among SON functions is indispensable. In addition, since naive coordination among SON functions may lead to high complexity and policy conflict, a carefully designed coordination framework needs to be designed. In this paper, we propose a scalable SON coordination framework, based on deep reinforcement learning (DRL), that manages the operation of SON functions to prevent policy conflicts. To reduce coordination complexity, we formulate a Markov decision process (MDP) problem and introduce an efficient DRL algorithm to solve it. Extensive simulation results show that the proposed framework effectively resolves policy conflicts between SON functions compared to baseline schemes and achieves a faster convergence time.