Vous êtes ici : Accueil » Kiosque » Annonce


Mot de passe : 

Mot de passe oublié ?
Détails d'identification oubliés ?


12 mars 2018

PhD DGA/ONERA: Robust statistical learning for adaptive processing in MIMO RADAR

Catégorie : Doctorant

This thesis is funded by DGA and ONERA. The subject is about the development of new robust statiscal learning algorithms with applications to different RADAR systems.


Keywords : learning, robust statistics, non convex optimization, MIMO RADAR, real data

1. Scientific background

The increase of sensors number and the development of new technologies have led to large size of RADAR data. Since decades, statistical approaches have shown their good capacity to deal with RADAR data. In particular, the statistical learning methods are the most likely used in many applications. But they are based on a learning support which depends on the data size. Therefore, since this size is strongly increased with the new RADAR systems, most of classical algorithms become inefficient. Development of new approaches adapted to large data size is a current topic both in research and in application fields. More precisely for statistical methods, new algorithms have to correctly run with a limited learning support. For this purpose, the best solution is to take into account the mathematical and physical properties of the data in the new algorithms (integration of prior knowledge). From an application point of view, it is needed to develop new efficient and robust algorithms for data of large size from new RADAR systems.

The MIMO STAP (Multiple-Input Multiple Output Space-Time Adaptive Processing) RADAR for the detection of moving target (GMTI for Ground Moving Target Indication) deals with the large data dimension problematic. Indeed, the multi-sensor, the multi-Doppler and the multi-arrays characteristics of the MIMO STAP lead to large size of data. Moreover, most of classical approaches in MIMO-STAP assume that the different transmitted waveforms are orthogonal but this is no longer verified in practice. Therefore, we have to develop new algorithms for data of large size and without the assumption of orthogonality of the transmitted waveforms.

2. Objectives

The general objective of the thesis is the development of a GMTI algorithm for MIMO-STAP RADAR. The thesis is organized around three main focus areas. In the first one, new statistical learning strategies for adaptive processing [1] will be proposed for large dimension data. The low rank and the Kronecker structure properties of the covariance matrix of the disturbance will be exploited to strongly reduce the learning support size [2,3].

The second topic deals with the problematic of transmitted waveforms which are not orthogonal. The goal is to develop new algorithms whose performances are not degraded by considering realistic MIMO transmitted waveforms (not orthogonal). The sidelobes reduction has already been proposed with mismatched filters [3]. We propose to include such approaches in the adaptive processes developed in the previous step. If possible, we will derive the theoretical performances of the new developed algorithms by using asymptotic methods [4,5].

Real MIMO RADAR data have been recently acquired by ONERA for moving target detection. The final objective of the thesis will then be to apply the previous developed algorithms on these real data which will allow to evaluate their performance for the detection and the localization of moving targets in several conditions (small SNR, slow moving target, ...). New scenarii and transmitted waveforms could be considered if necessary.

3. Informations

Pr. Guillaume Ginolhac
(Ph.D advisor)

LISTIC (EA3703),
Université Savoie Mont Blanc


Dr. Frédéric Brigui



Dr. Arnaud Breloy

LEME (EA4416),
Université Paris Nanterre


Jean François Nouvel

Salon de Provence


The thesis will take place at ONERA in Palaiseau (Paris area) and will start the october 1st 2018. The funding is assured by the DGA and the ONERA.

4. Candidate profile

The candidate must have the equivalent of a master degree in the field of signal and image processing, applied mathematics or data science. A solid background in statistical signal processing and/or numerical optimization is required as well as programming skills in Matlab and/or Python. Only applicants with E.U. nationality will be considered, due to funding constraints.

5. Contact

Applicants must send via e-mail to guillaume.ginolhac@univ-smb.fr, frederic.brigui@onera.fr a CV as well as a transcript of the last year study. Recommendation letters will be a plus. Do not hesitate to send an e-mail for more information.


[1] J. Ward, “Space-time adaptive processing for airborne radar,” Tech. Rep., Lincoln Lab., MIT, Lexington, Mass., USA, December 1994.

[2] A. Breloy, Y. Sun, G. Ginolhac, D. Palomar, "Robust Rank Constrained Kronecker Covariance Matrix Estimation", ASILOMAR 2016, Pacific Grove, USA.

[3] Y. Sun, A. Breloy, P. Babu, D. Palomar, F. Pascal, and G. Ginolhac, “Low-complexity algorithms for low rank clutter parameters estimation in radar systems,” Signal Processing, IEEE Transactions on, to appear.

[3] W. Roberts, P. Stoica, J. Li, T. Yardibi and F. A. Sadjadi, "Iterative Adaptive Approaches to MIMO Radar Imaging," in IEEE Journal of Selected Topics in Signal Processing, vol. 4, no. 1, pp. 5-20, Feb. 2010.

[4] G. Ginolhac, P. Forster, F. Pascal et J.-P.Ovarlez, "Performances and Robustness of a Robust Low-Rank STAP Filter", IEEE Transactions on Sig. Proc., Vol. 61, no1, pp 57 - 61, Janvier 2013

[5] A. Combernoux, F. Pascal, G. Ginolhac et M. Lesturgie, "Asymptotic performance of the Low Rank Adaptive Normalized Matched Filter in a large dimensional regime". ICASSP 2015, Brisbane, Australia.


Dans cette rubrique

(c) GdR 720 ISIS - CNRS - 2011-2018.