SecureRF-Net: RF Device Classification and Unauthorized Transmitter Detection for Model-Identical Radios

Device and user authentication are fundamental building blocks of secure wireless systems, but attacks, such as spoofing or device impersonation, can bypass higher-layer protections. Radio Frequency (RF) device fingerprinting mitigates this problem by exploiting hardware-generated signal artifacts that are difficult to forge and persist over time. This paper introduces SecureRF-Net, a deep learning framework for robust RF device identification that operates effectively with model-identical radios and under both closed-set and open-set conditions. Unlike existing approaches that maintain separability across varying channels by injecting transmitter-side impairments or relying on receiver equalization feedback, SecureRF-Net achieves robust classification directly from minimally preprocessed raw I/Q data by constructing a channel-independent spectrogram to suppress channelinduced distortions and employing a dual-loss optimization strategy to learn embeddings that are both discriminative and resilient to propagation variability. Experimental validation demonstrates that SecureRF-Net achieves high accuracy in both static and varying channel conditions when classifying six nominally identical transmitters. Furthermore, in open-set scenarios, the framework accurately detects unauthorized transmitters that use the same MAC addresses of authorized devices, with an accuracy exceeding 97%. These results confirm that SecureRF-Net advances the state of the art in physical-layer authentication by providing a practical, hardware-agnostic, and scalable solution for secure RF device identification in real-world wireless environments.