Air-Ground Coordination Communication by Multi-Agent Deep Reinforcement Learning

In this paper, we investigate an air-ground coordination communication system where ground users (GUs) access suitable UAV base stations (UAV-BSs) to maximize their own throughput and UAV-BSs design their trajectories to maximize the total throughput and keep GU fairness. Note that the action space...

Full description

Saved in:
Bibliographic Details
Published inICC 2021 - IEEE International Conference on Communications pp. 1 - 6
Main Authors Ding, Ruijin, Gao, Feifei, Yang, Guanghua, Shen, Xuemin Sherman
Format Conference Proceeding
LanguageEnglish
Published IEEE 01.06.2021
Subjects
Benchmark testing
Privacy
Reinforcement learning
Simulation
System performance
Throughput
Transforms
Online AccessGet full text

Cover

More Information
Summary:In this paper, we investigate an air-ground coordination communication system where ground users (GUs) access suitable UAV base stations (UAV-BSs) to maximize their own throughput and UAV-BSs design their trajectories to maximize the total throughput and keep GU fairness. Note that the action space of GUs is discrete, and UAV-BSs' action space is continuous. To deal with the hybrid action space, we propose a multi-agent deep reinforcement learning (MADRL) approach, named AG-PMADDPG (air-ground probabilistic multi-agent deep deterministic policy gradient), where GUs transform the discrete actions to continuous action probabilities, and then sample actions according to the probabilities. The proposed method enable the users make decisions based on their local information, which is beneficial for user privacy. Simulation results demonstrate that AG-PMADDPG can outperform the benchmark algorithms in terms of fairness and throughput.
ISSN:1938-1883
DOI:10.1109/ICC42927.2021.9500477