Contact: Robin Gerzaguet, Matthieu Gautier ( firstname.lastname@example.org email@example.com )
Laboratory: IRISA -- GRANIT
Keywords: Software Defined Radio (SDR), Radio Frequency, Artificial Intelligence
The internship could be followed by a PhD position.
Security of information systems has become a critical challenge at multiple scales with both strategic interests for protection (defense) and interception (attack). One of the most common way to protect data is to encrypt them: Encrypted data is also qualified as "black data" in opposition to non-encrypted data ("red data"). To prevent any compromising, "red data" should not be emitted or accessed (in any way) from outside. The reasons of this non-encryption may be diverse: either encrypting is not possible, occurs later or seems unnecessary.
Among the large list of potential threats, TEMPEST scenarios are one of the most problematic: it occurs when "red data" is emitted un-intentionally due to the presence of a non-legitimate channel. This channel may have different nature (light, sound, or electromagnetic (EM)) and due to different causes (EM leakage, coupling, air gap bridging...). Different studies in the literature have focused on TEMPEST scenarios and it has been recently shown that regular wireless transmission (e.g WiFi or Bluetooth) can hide an EM TEMPEST channel . Recently in Granit, we have shown that Bluetooth signal can be tracked with the use of a Software Defined Radio (SDR) and that un-intentionally emitted signal can be extracted from it .
Anyway, before any attack, one has to first be able to recognize a device inside a network. The purpose of the bibliography phase is to study the main approaches used for RF Fingerprint. RF Fingerprint consists in being able to recognize a device based on received signals. Each device has some unique properties (associated to analog components) that can be used as “fingerprint” for identification. This fingerprint can then be used to be able to recognize that a device is present and active within a network.
Most of the state-of-the-art RF fingerprint methods are parametric, offline and based on narrowband descriptors . Recently, primarily works [3,4] use Machine Learning (ML) to better leverage complex models (memory-based power amplifier models) for identification. In the proposed project (bibliography and internship), we will pave the way for efficient ML uses for RF fingerprint dedicated to SDR. The proposed methods will be based wideband non-linear feature extraction (Intermodulation products, coefficient of Volterra kernels…) based on the acquisition of several hundreds of MHz based on new SDR architectures.
 G.Camurati,S.Poeplau,M.Muench,T.Hayes,andA.Francillon,“ScreamingChannels:WhenElectromagneticSideChannelsMeetRadioTransceivers,”inProc. ACM SIGSAC Conference on Computer and Communications Security,CCS’18,(NewYork,NY,USA),pp.163–177,Association for Computing Machinery, 2018.
 O. Gungor and C. E. Koksal, “On the Basic Limits of RF-Fingerprint-Based Authentication,” IEEE Transactions on Information Theory, vol. 62, no. 8,pp. 4523–4543, 2016.
 S. Riyaz, K. Sankhe, S. Ioannidis, and K. Chowdhury, “Deep learning convolutional neural networks for radio identification,”IEEE CommunicationsMagazine, vol. 56, no. 9, pp. 146–152, 2018.
 X. Zhou, A. Hu, G. Li, L. Peng, Y. Xing, and J. Yu, “Design of a Robust RF Fingerprint Generation and Classification Scheme for Practical DeviceIdentification,” in Proc. IEEE Conference on Communications and Network Security (CNS), pp. 196–204, 201
 C.Lavaud,R.Gerzaguet,M.Gautier,andO.Berder,“TowardRealtimeinterceptionofFrequencyHoppingSignals,”in Proc. IEEE International Workshop on Signal Processing Systems, 2020.
(c) GdR 720 ISIS - CNRS - 2011-2020.