| Electromagnetic Fingerprinting uses unique emissions from wireless devices to identify them via
machine learning and deep learning algorithms. Two applications of UHF RFID EM
fingerprinting pertaining to IoT security are explored, one of which involves detecting SQL
injection virus malware in UHF RFID user memory in a hypothetical supermarket supply chain.
The frequency domain RSS data is analyzed using supervised classification techniques via
Python. Feature reduction is achieved through observing power threshold crossing and number
of maxima within smaller feature bands. Random Forest is the best model, being able to
successfully predict malicious and normal tags 82% of the time, where the low and high
frequency ranges contribute the most within the observed spectrum. A similar system was used
as a conceptual digital twin web resolver that could differentiate tags based on the Electronic
Product Code (EPC) with 99% accuracy.
Keywords: UHF RFID, Malware, Maxima Detection, Threshold Crossing, Supervised Learning,
Digital Twin, Electromagnetic Fingerprint, Resolver. |