Radio Frequency Fingerprinting mit Software Defined Radios

  • Manuel Lang

Student thesis: Master's Thesis

Abstract

Radio-based communication is an integral part of advancing digitalization. Wireless
transmission channels are widely used in Internet of Things (IoT) devices, smartphones,
drones, self-driving cars or industrial applications. The inherent characteristic of wireless
transmissions, which is to transmit information invisibly through modulated electromagnetic waves, introduces new risks and attack vectors. Unauthorized foreign devices,
devices masquerading as others, or compromised devices can communicate unnoticed in
the background. Therefore, detecting and identifying radio communications using passive means is particularly relevant.
In this work a prototypic Radio Frequency Fingerprinting (RFF) identification system was developed. The identification system, based on a Convolutional Neural Network
(CNN), identifies transmitters based on the characteristics of the emitted radio signal.
It operates on raw signal data recorded with a Software Defined Radio (SDR). The
CNN uses the raw signal data of the recorded radio transmissions without preprocessing the signal data, meaning no demodulation of the received signal takes place. For
the evaluation, devices transmitting IEEE 802.11 frames were recorded using the GNU
Radio framework and an Ettus N200 USRP. Additionally, reference datasets were used
to evaluate the identification system.
The evaluation of the prototypic implementation shows good results in identifying
transmitters with an accuracy of over 91% in environments with limited training data.
With a larger amount of training data, a higher accuracy of over 95% was achieved.
Date of Award2024
Original languageGerman (Austria)
SupervisorRobert Kolmhofer (Supervisor)

Cite this

'