UAV RF Fingerprinting with Power Spectra Estimates

Abstract

There is a vital need to detect and classify different radio-controlled Unmanned Aerial Vehicles (UAVs), and that need will only grow as they gain wider use. Using Radio Frequency (RF) fingerprinting is a stealthy way to classify drone types and their operating modes without making any RF transmissions. In this work, we perform several tests to advance the current literature in detecting and classifying UAVs. We find that image classifiers can perform similarly to and even outperform 1D coefficient-based classifiers. We also demonstrate the effect of different Discrete Fourier Transform (DFT) averaging, a common technique used to reduce the noise variance of power spectra estimates. We found that increasing the averaging before creating our images provides increased accuracy and reduced training time.