2026-05-20T16:57:33Zhttps://keep.lib.asu.edu/oai/request

oai:keep.lib.asu.edu:node-2018862025-07-17T19:39:31Zoai_pmh:alloai_pmh:repo_items

201886 https://hdl.handle.net/2286/R.2.N.201886 http://rightsstatements.org/vocab/InC/1.0/ All Rights Reserved 2025 142 pages Doctoral Dissertation Academic theses en Li, Jiawei Zhang, Yanchao Reisslein, Martin Xue, Guoliang Zou, Shaofeng Arizona State University Partial requirement for: Ph.D., Arizona State University, 2025 Field of study: Computer Engineering Radio-frequency identification (RFID) tags are now common in stores, factories, and everyday devices. These small, low-cost tags make tracking and access easy, yet they also open new doors for security and privacy problems. This dissertation shows how modern machine-learning tools can both protect RFID systems and power new attacks that use RFID tags in unexpected ways. This work contains three main chapters and a final chapter. Chapter 2 introduces RF-Rhythm (Radio Frequency Rhythm) for two-factor authentication. Chapter 3 presents RCID (Reflection Coefficient-based RFID Fingerprint). Chapter 4 describes TagStroke, a keystroke inference attack using an RFID tag array under a keyboard. Chapter 5 presents the conclusion and future directions. RF-Rhythm turns a standard RFID card into a two-factor authenticator. Users tap a personal rhythm, producing unique phase patterns in the backscatter. A machine learning model learns the pattern, while phase hopping prevents eavesdropping. User tests show near-zero false accepts or rejects. RCID secures the RFID system with each tag’s wideband RF fingerprints. A special Orthogonal Frequency Division Multiplexing (OFDM) powered RFID reader captures a 20 MHz profile in milliseconds, and a convolutional neural network tells tags apart. In tests with 600 off-the-shelf tags, RCID reached 97% accuracy and offered about 202 bits of fingerprint entropy. TagStroke shows the other side of RFID use. By hiding a small tag array under a keyboard, an attacker can read tiny signal changes caused by typing, even from a few meters away. A temporal-convolution and transformer model spots keystrokes, and a large language model improves text recovery. Experiments with 11 volunteers achieved 87% key accuracy and a 21% word-error rate. Computer Engineering Electrical Engineering Computer Science RFID Security AI-Based RFID System Security and Privacy: Challenges and Solutions