3L-BC: a three-layer blockchain architecture for collaborative machine learning in secure drone communications

This paper presents 3L-BC, a novel three-layer blockchain architecture designed to enhance security and privacy in Unmanned Aerial Vehicle (UAV) communications through collaborative machine learning. UAVs face significant challenges, including data confidentiality breaches, single points of failure, and computational resource constraints that traditional frameworks struggle to address. The proposed architecture strategically partitions functionality across three layers: the Drone Layer handles data acquisition with minimal preprocessing, the Fog Layer manages local training with privacy-preserving noise injection, and the blockchain layer ensures tamper-resistant storage and consensus-based validation. A trust-based fusor node selection mechanism, utilizing pilot accuracy and role indices, governs global model fusion, ensuring that only reliable nodes contribute to the collaborative intelligence. Experimental validation on the UAV123 dataset demonstrates that 3L-BC outperforms existing systems, achieving higher throughput (5700 transactions per second (TPS)), reduced latency (0.2567 s), shorter processing time (0.475 s), and superior accuracy (99.98%). Comprehensive threat modeling confirms resilience against eavesdropping, tampering, model poisoning, and consensus attacks. The architecture strikes a balance between security requirements and computational efficiency, making it particularly suitable for mission-critical applications in resource-constrained environments. The system's real-world applicability spans disaster response, precision agriculture, smart city traffic management, and infrastructure inspection. 3L-BC's strategic integration of blockchain and collaborative learning advances the frontier of secure, resource-aware, and privacy-preserving frameworks for dynamic drone networks.