Les commentaires sont clos.

Stage M2 / PhD : Machine learning for wireless telecommunication signal classification in LEO Satellites.

8 December 2021

Catégorie : Stagiaire

This internship is to be followed by a PhD Thesis application, partly funded by CNES. An advantage will be given to candidates who have a strong motivation to go for applying for the PhD afterward.


Low-earth orbit (LEO) satellites, Internet of Things (IoT), interference management, classification, machine learning, neural network, selective channel


K. Amis, F. Guilloud (IMT Atlantique, Brest campus)


CNRS UMR 6285 - Lab-STICC / T2I3 / Cosyde




March – August 2022

Application procedure

Send CV / grades / motivations by email to karine.amis and frederic.guilloud at



Low Earth Orbit satellite constellations can be made of small satellites (CubeSat) to reduce deployment and operating costs. This kind of constellations is well suited to operate IoT services (Sat-IoT) [JDN21, VIPRESS21, CCG+21]. The advantage of Sat-IoT is its wide area coverage without the need to deploy any ground base stations to mesh the IoT network. In return, the satellite receives a superposition of multiple signals he both from the IoT system itself (self-interference) s and from other interfering systems operating in neighbouring frequency bands . Moreover, due the low orbit, the speed of the satellite induces strong signal distortions among which non-negligible Doppler shifts.

Internship objective and contents

Since the onboard processing is a scarce resource, one has to use it wisely. The superposition of many different signals and interferences makes the detection and decoding work highly complex and reduced-complexity solutions have to be found. We propose during this internship to set up a machine learning-based signal classifier [JPJ+19, DZG19] to help separate first the signals corresponding to different standards that operate in the same frequency band, and then the different users within a single standard. This preprocessing could be a precious help to the decoders onboard the satellite.

The first part of the internship will be to decide the IoT standards (like e.g. AIS, LoRa, Sigfox, etc.) that will be considered to define the first-level classes and simulated the associated waveforms.

Then a state-of-the art will enable to select some machine learning algorithms, which will be implemented and compared according to judicious criteria (classification error rate regarding the standards and the users).

Simulations will be run on synthetized signals. Depending on the progression and on the results, simulations might also be run on real signals provided by CNES.


[JDN21], accessed the 12/11/2021.

[VIPRESS21] accessed the 21/11/2021.

[CCG+21] M. Centenaro, C. E. Costa, F. Granelli, C. Sacchi and L. Vangelista, "A Survey on Technologies, Standards and Open Challenges in Satellite IoT," in IEEE Communications Surveys & Tutorials, vol. 23, no. 3, pp. 1693-1720, thirdquarter 2021, doi: 10.1109/COM

[JPJ+19] Jithin Jagannath, Nicholas Polosky, Anu Jagannath, Francesco Restuccia, Tommaso Melodia, Machine learning for wireless communications in the Internet of Things: A comprehensive survey, Ad Hoc Networks, Volume 93, 2019, 101913, ISSN 1570-8705, https://doi.

[DZG19] X. Li, F. Dong, S. Zhang, and W. Guo, “A survey on deep learning techniques in wireless signal recognition,” Wireless Comms. and Mobile Computing, vol. 2019, pp. 1–12, 02 2019.