Digital Twin-Assisted Adaptive Multi-Agent DRL for Intelligent Spectrum and Resource Management in Open-RAN UAV-Enabled 6G Networks
Abstract Overview
This paper studies spectrum sharing and resource management in Open-RAN UAV-enabled 6G networks, where UAVs act as aerial radio units to support ground radio units and distributed ground users. The authors formulate a joint optimization problem over UAV positions, RU-GU associations, transmission power, and bandwidth allocation under energy, latency, SINR, mobility, and collision-avoidance constraints. Their solution decomposes the problem into particle swarm optimization for UAV trajectory planning and a digital twin-assisted multi-agent DRL scheme for adaptive association, power, and bandwidth control. The digital twin is integrated with the Open-RAN non-RT and near-RT RIC hierarchy to support centralized training and decentralized low-latency inference, and the framework is evaluated through simulation against several multi-agent baselines.
Novelty
The main novelty is the combination of a digital twin-assisted Open-RAN control architecture with a hybrid PSO plus multi-agent DRL optimization framework for UAV-assisted 6G resource management. The work is distinctive in jointly addressing UAV trajectory design, spectrum sharing, association, power allocation, and bandwidth allocation while explicitly incorporating latency and UAV energy constraints in the same framework.
Results
In simulation, the proposed method converges faster and reaches higher average rewards than MADDPG, MAPPO, multi-agent actor-critic, and a greedy baseline. It also achieves higher average data rates across training and varying cluster counts, while reducing average total latency to around 60 ms and maintaining stable performance.
Key Points
- The paper formulates joint UAV trajectory and radio resource management as a constrained nonconvex optimization problem in an Open-RAN UAV-assisted 6G setting.
- The proposed solution combines PSO for feasible UAV positioning with a digital twin-assisted multi-agent DDPG-style framework for RU-GU association, bandwidth allocation, and power control.
- Simulation results indicate faster convergence, higher data-rate performance, and lower latency than the reported benchmark methods.