Digital Twin-Driven Cross-Layer Orchestration Using Multi-Agent Reinforcement Learning for 6G Drone-NTN Networks

As Sixth Generation (6G) networks evolve toward a unified terrestrial and non-terrestrial ecosystem, maintaining optimal Quality of Service (QoS) across highly dynamic environments remains a formidable challenge. This paper proposes a high-fidelity Digital Twin (DT)-driven framework for Automatic Quality, Coverage, and Capacity Optimization (AQCCO), specifically tailored for integrated drone and Non-Terrestrial Network (NTN) systems. By leveraging a multi-layer DT architecture, we construct a real-time virtual mirror of the physical network, enabling predictive fault diagnosis and proactive resource orchestration. At the core of our solution is a Multi-Agent Reinforcement Learning (MARL) approach that jointly optimizes drone trajectories, satellite beamforming, and Reconfigurable Intelligent Surface (RIS) configurations. Extensive simulations conducted in diverse urban and rural scenarios validate the framework, demonstrating an average improvement of 15.3 dB in SINR, a 23.6% boost in coverage, and an 89.4% increase in network throughput. Furthermore, the system achieves an energy efficiency gain of 31.2% while maintaining a sub-50 ms decision-making latency. The framework ensures rapid service recovery under emergency conditions with sub-second response times, providing network operators with an intelligent, demand-responsive tool for optimizing key performance indicators (KPIs) in the 6G era.