Model-Agnostic Uncertainty Quantification for Fast NFC Tag Identification using RF Fingerprinting

Near Field Communication (NFC) is widely used in security applications such as door access systems and ID cards. However, clone attacks can replicate digital information, enabling unauthorized access. RF fingerprinting offers a robust defense by extracting unique physical-layer features from NFC cards that cannot be cloned. While RF fingerprinting has been extensively applied to Internet of Things (IoT) device authentication, NFC tags present distinct characteristics that require specialized approaches. This paper focuses on RF fingerprinting for the ISO15693 NFC tag, which is a widely used international standard, by leveraging multi-channel, multi-rate data sampling to enhance accuracy. Deep learning and Random Forest models are employed to identify NFC tags, while uncertainty quantification, particularly Conformal Prediction, accelerates the identification process with high confidence and precision. A software-defined radio (SDR) testbed is developed to transmit customized commands and collect multi-channel multi-rate NFC signals. The multi-channel multi-rate NFC signals are progressively collected to ensure fast and accurate identification. Experimental results demonstrate that the proposed system achieves high accuracy by adaptively utilizing the optimal combination of NFC signals. The developed solution is model-agnostic which can be utilized for any machine learning-based NFC tag identification.