Internship Deep learning techniques for radio identification (possibility of PhD afterwards) – Villeurbanne

Internship Deep learning techniques for radio identification (possibility of PhD afterwards)

Inria

Villeurbanne

Postuler

Internship Deep learning techniques for radio identification (possibility of PhD afterwards)

Le descriptif de l’offre ci-dessous est en Anglais

Type de contrat : Convention de stage

Niveau de diplôme exigé : Bac + 4 ou équivalent

Autre diplôme apprécié : engineer degree, master degree

Fonction : Stagiaire de la recherche

A propos du centre ou de la direction fonctionnelle

The Inria research centre in Lyon is the 9th Inria research centre, formally created in January 2022.  It brings together approximately 300 people in 16 research teams and research support services.

Its staff are distributed at this stage on 2 campuses: in Villeurbanne La Doua (Centre / INSA Lyon / UCBL) on the one hand, and Lyon Gerland  (ENS de Lyon) on the other.

The Lyon centre is active in the fields of software, distributed and high-performance computing, embedded systems, quantum computing and privacy in the digital world, but also in digital health and computational biology.

Contexte et atouts du poste

With the popularisation of software defined radios (SDRs), an malevolent actor can deploy radio systems for interfering with legitimate communications, communicating on unlicensed bands and listening to private communications. In this proposed PhD work we target in a first moment spectrum sensing capabilities, that can be used to automatically locate and classify opponent transmissions, characterising it in terms of center frequency, occupied bandwidth, activity pattern, modulation and coding schemes, frame structure and more. Then we will study the identification problem, trying to uniquely single out individual transmitters among all transmitters. The proposed work will (i) create good datasets to train and test systems for spectrum sensing and (ii) develop deep learning (DL) systems for spectrum sensing/classification and identification.

Nowadays, spectrum surveillance is mainly done with relatively simple systems that require intense human intervention. However, as radio communications systems grow more and more complex in nature and can span larger portions of the spectrum, relying on human-based surveillance risks missing out on improper use of the spectrum. Sophisticated means to detect these transmissions, identify them and locate their source is thus necessary, but remains a complicated task to accomplish.

Mission confiée

REMARK: This internship can lead to a PhD offer, that shall start October 2024

Concerning the detection of a radio signal, techniques such as energy detection (ED) [1], cyclostationary detection (CD) [2], matched filter (MF) [3] and random matrix detectors (RMD) [4] have been proposed before, but each carry their own set of problems such as a minimal signal-to-noise ratio requirement fo the ED, and cyclic features for the (CD), just to cite a few. Detecting a signal becomes more challenging when incomplete observation and/or ultra-wideband signals are present [5], which usually spread the transmitted power over a large bandwidth. After a signal has been detected, extracting its characteristics is even a harder task, requiring most of the times, achieving a partial decoding of the target signal and work only for a limited number of kinds of signals at a time. In this work we aim at using deep learning (DL) techniques to jointly address the spectrum sensing and signal classification problems. DL techniques are more adapted to these kinds of problems due to their nature and require no prior knowledge on the signals and their structures. 

The main objective of this internship is to provide DL models to deal with the spectrum sensing/transmission classification problem that are able to perform well in a realistic scenario where challenging channel characteristics exist as well as interference. These DL models can be integrated into a device based on one (or more) SDRs to provide automated detection and characterisation of transmissions, requiring little to none human intervention, and alert law enforcement when stray radio transmissions are detected. 

Using DL for spectrum sensing/transmission classification is not new and has been studied before (i.e. [6], [7] and [8]), however we think we can effectively contribute to the subject due to (i) our ability to produce high quality datasets which will be carefully designed, validated and tailored to training DL models using the CorteXlab testbed, as well (ii) as our prior experience with DL models for radio.
Concerning transmitter identification, aside from using identification fields in packets which are very easily impersonated, we rather base identification through radio fingerprinting, using the transmitter characteristics that are imprinted onto the signal are used to identify the transmitter, much like how we can recognise different people through the tone of their voice. Even though this has been studied before (i.e., [9]), many of the works are plagued with problems like identification through channel characteristics (which are dependent on the position of the radios) or synthetic RF imprinting, and ill-created datasets, like the one in [9], which uses synchronisation to orchestrate the transmissions (we have already shown in [10] that the use of synchronisation biases the dataset). 

—references—

[1]    Shen, J., Liu, S., Wang, Y., Xie, G., Rashvand, H. F., & Liu, Y. (2009). Robust energy detection in cognitive radio. IET communications, 3(6), 1016-1023.

[2]    Kim, K., Akbar, I. A., Bae, K. K., Um, J. S., Spooner, C. M., & Reed, J. H. (2007, April). Cyclostationary approaches to signal detection and classification in cognitive radio. In 2007 2nd ieee international symposium on new frontiers in dynamic spectrum access networks.

[3]    Giannakis, G. B., & Tsatsanis, M. K. (1990). Signal detection and classification using matched filtering and higher order statistics. IEEE Transactions on Acoustics, Speech, and Signal Processing, 38(7), 1284-1296.

[4]    Cardoso, L. S., Debbah, M., Bianchi, P., and Najim, J. (2008). Cooperative spectrum sensing using random matrix theory. 3rd International Symposium on Wireless Pervasive Computing, Santorini, Greece, 2008.

[5]    Sahin, M. E., Guvenc, I., & Arslan, H. (2005, May). Optimization of energy detector receivers for UWB systems. In 2005 IEEE 61st Vehicular Technology Conference.

[6]    Gao, J., Yi, X., Zhong, C., Chen, X., & Zhang, Z. (2019). Deep learning for spectrum sensing. IEEE Wireless Communications Letters, 8(6), 1727-1730.

[7]    O'Shea, T., Roy, T., & Clancy, T. C. (2017, October). Learning robust general radio signal detection using computer vision methods. In 2017 51st asilomar conference on signals, systems, and computers (pp. 829-832).

[8]    Riyaz, S., Sankhe, K., Ioannidis, S., & Chowdhury, K. (2018). Deep learning convolutional neural networks for radio identification. IEEE Communications Magazine, 56(9), 146-152.

[9]    Al-Shawabka, A., Restuccia, F., D’Oro, S., Jian, T., Rendon, B. C., Soltani, N., … & Melodia, T. (2020, July). Exposing the fingerprint: Dissecting the impact of the wireless channel on radio fingerprinting. In IEEE INFOCOM 2020-IEEE Conference on Computer Communications.

[10]    Morin, C. (2021). Approches d'apprentissage profond pour la détection en couche physique de télécommunications multi-accès (Doctoral dissertation, Université de Lyon).

Principales activités

Main activities may include:

  • Exploit the current mathematical models and theory concerning spectrum sensing/transmitter classification and transmitter identification to provide a design for both the dataset creation and DL models to be used.
  • Create high quality datasets, different from the ones currently available in the
    literature, for spectrum sensing and transmitter classification.
  • Create DL models starting from semantic segmentation ones and evolving toward complex radio situations including partially observed signals (in time and frequency) as well as interfering signals.
  • Implementation and experimentation on CorteXlab.

Additional activities:

  • High quality papers and report writing
  • Seminars and internal collaboration within the team MARACAS

Compétences

Technical skills and level required : machine learning (at least basic), probabilities and statistics (basic), signal processing (intermediate), scientific programming in Python – scipy, matplotlib, numpy, panda, xarray (intermediate).

Additional values : digital communications, expertise in programming with tensor flow or PYtorch

Languages : English, French (optional)

Relational skills : hability to collaborate, to communicate with pairs. Strong motivation, enthousiast, curious. 

Avantages

  • Subsidized meals
  • Partial reimbursement of public transport costs
  • Possibility of teleworking (after 6 months of employment) and flexible organization of working hours
  • Professional equipment available (videoconferencing, loan of computer equipment, etc.)
  • Social, cultural and sports events and activities

Rémunération

Minimum legal gratification

Postuler

Voir tous les emplois