Semester Award Granted

Spring 2025

Submission Date

May 2025

Document Type

Thesis

Degree Name

Master of Science (MS)

Thesis/Dissertation Advisor [Chair]

Georgios Sklivanitis

Abstract

As 5G networks expand, ensuring secure identification and authentication of user devices is critical. This paper explores machine learning-based Radio Frequency Fingerprinting (RFF) to identify and distinguish trusted and rogue devices in 5G networks. We evaluate ResNet, Transformer, and LSTM architectures using channel-isolated (CI) spectrogram and raw IQ slice inputs across varying packet sizes. Results show that ResNet with CI spectrogram inputs achieves the highest device classification accuracy and scalability while mitigating the Next-Day Effect, while the same architecture with IQ Slices is best for rogue device detection. Unlike related works, we emphasize the role of spectrograms in accurately capturing discerning features in 5G signals for scalable RFF applications. These fingerprints strengthen authentication processes against device impersonation at the physical layer of 5G networks. Using real-world 5G datasets from an outdoor wireless network testbed, this study demonstrates the feasibility of AI-driven RFF for secure device authentication.

Share

COinS