Deep Reinforcement Learning for User Association in Heterogeneous Networks with Dual Connectivity

Abstract

The dual connectivity is emerging as a promising solution to boost capacity in heterogeneous networks. However, it is challenging to obtain an optimal user association in heterogeneous networks with dual connectivity, due to its non-convex and combinatorial nature. In this paper, we propose a user association scheme based on deep reinforcement learning to maximize the overall network utility, which takes both throughput and user fairness into account, in the downlink of a heterogeneous network. Particularly, each user associates with the macro base station (BS) and a micro BS. We apply a deep Q-network (DQN) to obtain the nearly optimal policy to associate the users and micro BSs. Simulation results demonstrate that DQN-based user association performs better compared to the conventional user association schemes in heterogeneous networks with dual connectivity, and it behaves good scalability when the environment changes.