Master 2013 2014
Stages de la spécialité SAR
Correlation machine

Site :Equipe Audition, LSP
Lieu :Ecole normale supĂ©rieure, 29 rue d'Ulm
Encadrant : Alain de CheveignĂ©
Dates :Avril-Aout 2014
Rémunération :statutaire
Mots-clés : Parcours ATIAM : Acoustique, Parcours ATIAM : Traitement du signal


Covariance and correlation are building blocks of signal processing. Calculating them is a computational bottleneck, and integrating this constraint may be at the expense of performance. This project aims to encapsulate the calculation so as to free the algorithm designer from computational constraints. Given an array of signals, the "correlation machine" calculates on demand arrays of covariance and cross-correlation coefficients, with arbitrary window size and shape, at arbitrary temporal offsets, and with arbitrary ranges of lag parameters. The "machine" is responsible for dealing with computational and memory constraints and tradeoffs. The project involves the following tasks (a) define the programming interface, (b) implement the machine in matlab, (c) write test routines, (d) write up the results. Depending on the skill and tastes of the intern, the following additional tasks may be addressed : reimplement the machine in C, callable from matlab and python, (e) explore solutions for optimization (GPU, cluster, etc.), (f) explore applications to multichannel audio (source localization, source separation, indexing), (h) explore the feasability of real-time calculation. This project requires excellent programming and writing skills.