Learning-Based Radio Fingerprinting for RFID Secure Authentication Scheme

Jiaqi Xu, Xingya Zhao, Arjun Bakshi, Kannan Srinivasan

Abstract: In recent years, many low-cost identification systems have been proposed using RFID tags. Due to their limited compu-tational capability, passive RFID tags cannot provide secure links with cryptographic algorithms, and are vulnerable to cloning and replaying attacks. RF fingerprinting techniques based on distinctive hardware imperfections of devices have been explored for RFID systems. However, since RFID tags send their payloads by reflecting the carrier wave signal from readers, the received tag response signal will contain reader-specific fingerprints which can hinder the authentication accuracy using RF fingerprints. The rich dynamics of the wireless channel also present significant challenges for RF fingerprinting. To address these issues, we present FILES, a learning-based RF fingerprinting scheme for RFID systems. It is built upon the top of the EPC C1G2 protocol and works with any COTS tags. FILES analyzes the channel and reader information from the received carrier wave signal and decouples it from the tag response signal. It then extracts features from the preprocessed signal and classifies them with a neural network. We build a prototype of our system and conduct extensive experiments. The results show that our system can achieve an accuracy of up to 99% across readers, across tag locations, and across environments with 50 COTS RFID tags.