Resource Allocation in UAV-Assisted Networks: A Clustering-Aided Reinforcement Learning Approach
Shiyang Zhou,Yufan Cheng,Xia Lei,Qihang Peng,Jun Wang,Shaoqian Li
DOI: https://doi.org/10.1109/tvt.2022.3189552
IF: 6.8
2022-11-15
IEEE Transactions on Vehicular Technology
Abstract:As an aerial base station, unmanned aerial vehicle (UAV) has been considered as a promising technology to assist future wireless communications due to its flexible, swift and low cost features, where resource allocation is the basis for ensuring energy-efficient UAV-assisted networks. This paper formulates a joint optimization problem of user association, UAV trajectory design and power control to maximize the channel capacity among all ground users at a limited power level in a downlink transmission. To tackle the mixed-integer non-linear programming problem, this paper proposes a clustering-aided reinforcement learning approach consisting of three consecutive stages. Firstly, modified expectation-maximization unsupervised learning algorithm is investigated to cluster the ground users, which reduces the dimensions and hence, the association complexity is reduced as well. Then, Kuhn-Munkres algorithm is incorporated for user association, which associates a UAV with the ground users via matching to the cluster, and assigns the UAVs to the centroid of the matching cluster for pre-placement, with the aim of speeding up the convergence of the following deep reinforcement learning algorithm. Finally, a multi-agent twin delayed deep deterministic (MATD3) policy gradient is proposed to solve the non-convex sub-problem, which determines the transmit power and designs the fine-tuned trajectory of UAVs. By incorporating low-bias value estimation technique, the reward of the proposed MATD3 algorithm is improved. Simulation results have demonstrated that our proposed approach achieves higher reward as well as converging faster than existing reinforcement algorithms. Besides, the clustering-aided reinforcement learning has lower computational complexity than the benchmark schemes.
telecommunications,engineering, electrical & electronic,transportation science & technology