M2 research internship proposal
Robust joint detection-estimation methodologies for massive radio
Stefano Fortunati, L2S, firstname.lastname@example.org
Lucien Bacharach, SATIE, email@example.com
Laboratoire des signaux et syst`emes (L2S), Bˆat. IBM, Rue Alfred Kastler, 91400 Orsay.
Reception of applications:: until mid-January1
Duration of the internship: between 4 and 6 months
Period of the internship: from February to September 2024.
One of the key features characterizing the new generation of radio telescopes is the large number
of their antenna elements. Built in 2010, the Low-Frequency Array (LOFAR) is currently the
largest radio telescope in operation with 100 000 antenna dipoles distributed across several
European countries . Furthermore, the upcoming Square-Kilometer Array (SKA) will be
made up of more than 130 000 antennas . Such a large number of antennas will make it
possible to acquire increasingly accurate and detailed images of the celestial vault. Such images
will form the basis for promising developments in astrophysics and cosmology in the coming
However, as in any other “remote sensing system”, the signal collected by a radio telescope is
affected by different sources of disturbance that will degrade the quality of the collected image.
Consequently, to take full advantage of the potential of the new radio telescopes, one must
first take the disturbance into account. In general, this disturbance is characterized as a zero-
mean Gaussian random process with possibly unknown correlation structure. Unfortunately,
the Gaussian assumption is often violated by some impulsive noise sources whose presence may
produce a breakdown of the performance of Gaussian-based procedures. In order to take into
1Do not hesitate to enquire with us about other opportunities!
consideration the impulsive behavior of the disturbance, different classes of heavy-tailed distri-
butions (including the Gaussian one as a special case) have been proposed in array processing
literature . However, any choice of a specific heavy-tailed distribution is arbitrary to some
extent, since none of them, taken individually, can fully describe the physical behavior of the
Then, the crucial question is: is it possible to derive robust imaging algorithms, without
any assumption on the specific form of the noise distribution, and that still remain accurate?
If yes, which is the price to pay? In a generic array processing context, a positive answer
has been recently provided in , where a robust test for the detection of sources of interest
in heavy-tailed noise has been proposed. It suggests that it is possible to achieve a desired
detection/estimation performance level regardless the (generally unknown) disturbance model,
provided that the number of antennas is sufficiently large. These results could be leveraged
to derive original imaging algorithms for modern radio telescopes, which naturally fulfill the
requirement of a large amount of sensors.
Description of the expected work
This internship is funded by GDR ISIS 2 and is part of the “SIDEREAL”3 project. The objectives
of the internship are the following:
1. Building upon the existing works such as [5–7], we will adapt the array signal model
used in  to the context of radio telescopes. Particular attention will be devoted to
the disturbance model to be used in astronomical data analysis and on its statistical
description. The main goal here will be to clarify the similarities and the differences
between classical array processing  and radio telescope [5–7] applications (e.g., detection
of a target vs imaging and source reconstruction).
2. After these preliminary investigations, the project will focus on the development of original
image reconstruction algorithms for radio astronomy, based on the detection/estimation
method presented in , and exploiting the massive number of antenna elements available
in modern radio telescopes. Their performance and statistical properties will be assessed
by means of simulated data.
Master 2 or equivalent in machine learning / statistical signal processing or any related field,
programming skills in Matlab or Python.
 van Haarlem, M. P. and al., “Lofar: The low-frequency array,” A&A, vol. 556, p. A2, 2013.
 P. E. Dewdney, P. J. Hall, R. T. Schilizzi, and T. J. L. W. Lazio, “The square kilometre
array,” Proceedings of the IEEE, vol. 97, no. 8, pp. 1482–1496, 2009.
 E. Ollila, D. E. Tyler, V. Koivunen, and H. V. Poor, “Complex elliptically symmetric dis-
tributions: Survey, new results and applications,” IEEE Transactions on Signal Processing,
vol. 60, no. 11, pp. 5597–5625, 2012.
2Groupement de Recherche Information, Signal, Image, viSion: https://www.gdr-isis.fr/.
3“robuSt joInt Detection-estimation mEthodologies foR massivE rAdio teLescopes”
 S. Fortunati, L. Sanguinetti, F. Gini, M. S. Greco, and B. Himed, “Massive MIMO radar for
target detection,” IEEE Transactions on Signal Processing, vol. 68, pp. 859–871, 2020.
 S. J. Wijnholds and A.-J. van der Veen, “Multisource self-calibration for sensor arrays,” IEEE
Transactions on Signal Processing, vol. 57, no. 9, pp. 3512–3522, 2009.
 S. J. Wijnholds and A.-J. van der Veen, “Fundamental imaging limits of radio telescope
arrays,” IEEE Journal of Selected Topics in Signal Processing, vol. 2, no. 5, pp. 613–623,
 A. Leshem and A.-J. van der Veen, “Radio-astronomical imaging in the presence of strong
radio interference,” IEEE Transactions on Information Theory, vol. 46, no. 5, pp. 1730–1747,