Learning Robust Radio Frequency Fingerprints Using Deep Convolutional Neural Networks
Abstract
Radio Frequency Fingerprinting (RFF) techniques, which attribute uniquely identifiable signal distortions to emitters via Machine Learning (ML) classifiers, are limited by fingerprint variability under different operational conditions. First, this work studied the effect of frequency channel for typical RFF techniques. Performance characterization using the multi-class Matthews Correlation Coefficient (MCC) revealed that using frequency channels other than those used to train the models leads to deterioration in MCC down to 0.05 (random guess), indicating that single-channel models are inadequate for realistic operation. Second, this work introduced, developed, and demonstrated Fingerprint Extraction through Distortion Reconstruction (FEDR), a neural network-based approach for quantifying signal distortions in a relative distortion latent space. Coupled with a Dense network, FEDR fingerprints were evaluated against common RFF techniques for up to 100 unseen classes, where FEDR achieved best performance with MCC ranging from 0.945(5 classes) to 0.746 (100 classes), using 73 percent fewer training parameters than the next-best technique.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 2022
- Accession Number
- AD1181181
Entities
People
- Jose A. Gutierrez Del Arroyo Perez
Organizations
- Air Force Institute of Technology