Radio frequency fingerprinting: a survey of AI generalization challenges and solutions for wireless security

Radio frequency (RF) fingerprinting has emerged as a promising physical-layer security technique for device authentication in wireless networks, offering protocol-independent identification by exploiting hardware-induced signal characteristics and imperfections. This paper explores its challenges and potential within the heterogeneous wireless networks, including UAV communications and IoT deployments which are characterized by dynamic operating conditions, heterogeneous device profiles, and elevated security demands. Although deep learning-based methods have substantially improved controlled recognition accuracy from 70–90% to as high as 95–99%, the generalization gap, identified as the severe performance degradation when deployment conditions differ from training environments, remains the fundamental barrier to reliable deployment in practical environments. Systems frequently demonstrate performance degradation, ranging from 30–70%, when confronted with unseen datasets and real-world operating conditions that differ from those encountered during training. This survey systematically traces the evolution of RF fingerprinting, from traditional manual feature engineering to cutting-edge learning architectures, with a particular emphasis on generalization challenges. Key research gaps are identified in multi-factor generalization modeling, real-world validation under diverse conditions, scalable architecture design for broad device deployments, adaptive learning mechanisms for dynamic contexts, and computational efficiency for resource-constrained platforms. The analysis provides foundational insight for advancing next-generation RF fingerprinting techniques capable of sustaining robust and consistent performance across highly variable and complex wireless environments. By clarifying current limitations and future directions, this work contributes to the practical realization of secure RF-based authentication for heterogeneous wireless networks.

Mots clés

• RF fingerprinting
• physical-layer security
• generalization challenges
• deep learning
• wireless authentication
• IoT security
• UAV communications