AI-driven blockchain for secure and scalable federated learning in 6G-NTN-enabled UAV networks

The convergence of 6G Non-Terrestrial Networks (NTNs) and Unmanned Aerial Vehicles (UAVs) holds transformative potential for mission-critical applications such as real-time disaster response and autonomous urban mobility. Yet, the deployment of UAV-assisted 6G NTNs faces a persistent trilemma: ensuring security in decentralized settings, preserving data privacy without sacrificing model accuracy, and maintaining energy efficiency in dynamic environments. To address this, we propose a unified framework that integrates (1) a proof of Adaptive Trust consensus mechanism using long short-term memory-based behavioral modeling to detect Sybil and poisoning attacks with 99.2% accuracy, (2) a hybrid Cheon–Kim–Kim–Song secure multi-party computation encryption scheme that enforces ɛ = 1.0 differential privacy while preserving the utility of 98%, and (3) a quantized deep reinforcement learning beamforming strategy using an 8-bit policy network that aligns the mmWave energy by 25%. Simulations on a 1000-node UAV testbed demonstrate 480 TPS throughput, 110 ms latency, and 350 Wh total energy use, surpassing FedBeam and De-Trust-FL by 15%–30% in privacy, security, and efficiency metrics. The framework is validated through high-fidelity simulations calibrated with Da-Jiang Innovations Matrice 300 Real-Time Kinematic UAV specifications, and all core modules are released as open-source for reproducibility. This work establishes a scalable, privacy-preserving, and resilient architectural foundation for secure UAV operations in next-generation NTN environments.