ABSTRACT: Since the 1980's, a large number of inference methods have been developed for sensor arrays(e.g. estimators of the number of emitting sources, their direction of arrival, their power)and the statistical performances of these tools have been characterized in the asymptotic regime where thenumber of samples N of the observed signal tends to infinity while the number of antennas M remains constant.In practice, these methods are used in the context where N >> M. Nevertheless, it is not always possible to have such an amount of samples,especially when the number of antennas M is large, or when the involved signals are stationnary only for short periods of time. Forthese reasons, it seems more reasonable to consider the context where M and N are both large and of the sameorder of magnitude, and the asymptotic regime to consider is M,N converging to infinity such that the ratio M/N converges to a positiveconstant. In this context, random matrix theory, a branch of probability theory, providesinteresting results which can be exploited to analyze the traditional inference methods and improvetheir performance. In this talk, we will review some of these results and see some straightforwardapplications to the problems of source detection and localization. The talk is self contained and only basic knowledge in probability, linearalgebra and signal processing are required.

 

SPEAKER: P. Vallet received the M.Sc. from ESIEE Paris (France) in 2008 and his Ph.D degree from the University of Paris-Est/Marne-la-Vallée (France) in 2011. He is currently visiting CTTC as a post-doc researcher, and is funded by the French army through the agency Direction Général de l'Armement (DGA). His research interests include array processing for large sensor networks,performance analysis of large MIMO systems and random matrix theory.