Evaluating the Performance of RF Fingerprinting Feature Extractor for Proximity Integrated Circuit Cards

Physical Layer Security (PLS) has been a longstanding area of research, with Radio Frequency Fingerprinting (RFF) as one of its core components. Recently, the development of Zero Trust Architecture (ZTA) has incorporated RFF to enhance security further. RFF is also used in the Radio Frequency Identification (RFID) field to add another layer of device authentication and security. With advancements in the field of machine learning, RFF technology has benefited from the integration of machine learning techniques. However, several challenges remain unresolved, such as the variation in Channel State Information (CSI), which can significantly impact identification accuracy, particularly in Near Field Communication (NFC) systems. This paper analyzes a channelrobust feature extractor using a ResNet model trained on channel-independent spectrograms. This analysis is conducted within the framework of the ISO/IEC 14443 Type A protocol, a widely used NFC air-interface protocol with Proximity Integrated Circuit Cards (PICC) and Proximity Coupling Device (PCD).