WIP: Multi-connectivity user associations in mmWave networks: a distributed multi-agent deep reinforcement learning method.

Abstract

Multi-connectivity enabled user associations (MCUA) has been believed to be a promising method to enhance the connection between user equipments and base stations in ultra-dense millimeter wave (mmWave) networks. In this paper, the optimal MCUA is investigated from the user-side perspective with the objective of maximizing the overall downlink rate while satisfying the QoS requirements of each user. In view the terribly huge computational cost required by centralized MCUA methods, in this paper, we develop a distributed multi-agent deep reinforcement (MADRL) model to search for the optimal MCUA policy. In the proposed MADRL-MCUA, each UE is regarded as an independent agent and determines the its own association policy according to its own observed benefits and the feedback from the mmWave base stations. Experiment results are presented to demonstrate the effectiveness of the proposed method.