IoT Device Security via Deep Learning-Based RF Fingerprinting: A Survey of Channel Robustness, Adversarial Attacks, and Edge Deployment

The rapid growth of the Internet of Things (IoT) demands authentication mechanisms that remain effective even when cryptographic credentials are compromised or difficult to manage at scale. Radio frequency (RF) finger printing addresses this challenge by exploiting device-specific hardware imperfections as unclonable physical-layer identi fiers. Deep learning has emerged as the dominant paradigm for RF fingerprinting because it can automatically learn dis criminative hardware signatures directly from raw in-phase/ quadrature (I/Q) samples, consistently outperforming tradi tional feature-engineering approaches. Despite significant ad vances, practical deployment remains limited by three major challenges: sensitivity to channel variations and low signal to-noise ratio (SNR) conditions, vulnerability to adversarial attacks, and insufficient evaluation on resource-constrained edge hardware. This survey reviews more than 120 studies published between 2015 and 2026 and presents a compre hensive taxonomy of deep learning architectures, including convolutional, recurrent, transformer-based, meta-learning, and generative approaches. The survey further provides a comparative analysis of public datasets, systematically re views channel-robustness and domain adaptation techniques, and offers a comprehensive overview of adversarial attacks and defences in RF fingerprinting. In addition, hardware-in the-loop and edge-deployment studies are analysed to char acterise accuracy–efficiency trade-offs for practical IoT de ployment. Key findings indicate that hybrid CNN-RNN ar chitectures provide superior robustness under low-SNR con ditions, channel-aware augmentation significantly improves generalisation, and black-box adversarial attacks pose the most realistic threat to deployed systems. Finally, open chal lenges and future research directions are identified to guide the development of robust, secure, and deployable RF fin gerprinting systems.