Passive User Profiling Using Array of Sustainable Backscatter Tags

• Published in: IEEE communications letters Vol. 29; no. 8; pp. 1824 - 1828
• Main Authors: Wang, Haoming, Lai, Hao, Goay, Amus Chee Yuen, Mishra, Deepak, Seneviratne, Aruna, Ambikairajah, Eliathamby
• Format: Journal Article
• Language: English
• Published: IEEE 01.08.2025
• Subjects: Accuracy; Antennas; Backscatter; Backscatter communication; experimental demonstration; machine learning; Privacy; Radio frequency; Sensors; Support vector machines; Training; Wireless communication; wireless sensing; Wireless sensor networks
• ISSN: 1089-7798; 1558-2558
• DOI: 10.1109/LCOMM.2025.3576750

Wireless technology is increasingly used for real-time identity verification through active sensors, but traditional biometrics like fingerprint scanning raise privacy concerns due to their invasive nature. To address this, we propose a first-ever Backscatter Communication (BackCom)-based user profiling and commodity RFID height or weight profiling demonstration, using backscatter signals from RFID tags placed in the environment rather than on individuals. These battery-free, energy-harvesting tags offer a sustainable, privacy-preserving method for passive identity recognition. By applying a pre-trained linear machine learning algorithm to the Received Signal Strength Indicator (RSSI) data from RFID tags, we can identify individuals based on the modulation of the backscatter signal caused by their unique physical characteristics. Our system achieves up to 90.2% accuracy in identifying individuals from a set of seven. Additionally, we employ an unsupervised anomaly detection method that combines ResNet-18 feature extraction with Principal Component Analysis (PCA), yielding over 90% overall accuracy in distinguishing between known and unknown subjects.