Date of Award
6-2026
Document Type
Thesis
Publisher
Santa Clara : Santa Clara University, 2026
Department
Electrical and Computer Engineering
First Advisor
Kurt Schab
Abstract
Conventional cybersecurity protocols authenticate devices using digital credentials that can be stolen, copied, or extracted from compromised hardware. Radio frequency (RF) fingerprinting offers a complementary physical-layer authentication mechanism that binds device identity to the unforgeable manufacturing variations present in every transmitter’s analog hardware. This thesis explores the application of convolutional neural networks (CNNs) to RF fingerprinting, focusing on the identification of nominally identical IoT transmitters from raw I/Q samples of the LoRa preamble’s turn-on transient.
We developed an end-to-end system consisting of a modular data collection testbench using a USRP B210 software-defined radio, a 1D CNN trained directly on raw I/Q data, and a real-time inference pipeline deployed on an NVIDIA Jetson Orin Nano for live edge classification. The final system achieved 93% average recall and 94.2% average precision across eight nominally identical LoRa transmitters during live inference, successfully meeting the project’s 90% accuracy target. Subsequent analyses using t-SNE feature clustering and saliency mapping further showed that the CNN’s learned representations correspond to physically meaningful regions of the LoRa preamble, providing qualitative evidence that the model is learning genuine hardware-intrinsic features rather than dataset artifacts.
Recommended Citation
Chainani, Pranav; Gertner, Maxwell; and Patmore, Genevieve, "RF Fingerprinting: Neural Networks for Device Identification" (2026). Electrical and Computer Engineering Senior Theses. 122.
https://scholarcommons.scu.edu/elec_senior/122
