Research themes

GPU computing applications in image processing
 Implementation of a DCTbased Inpainting algorithm for Nvidia GPU using CUDA. Joint work with Yassir Moudden and Antoine Pedron.
 Implementation of a CDF (CohenDaubechiesFeauveau) lifting scheme discrete wavelet transform algorithm
See also the following slides : JTE GPU 04/12/2008  slides

Waveletbased multifractal analysis
 solar magnetograms and active regions (collaboration with the Trinity College of Dublin Astrophysics Group). See the slides of the talk given at conference ADA6 in May 2010: ADA6 2010
 study of the multifractale properties of simulations of the interstellar medium (collaboration with E. Audit from CEA/IRFU/Sap).

data acquisition system developpment for highenergy physics:
design of data acquisition systems and digital signal processing systems for highenergy physics and astrophysics experiments.
 software and hardware developments for the UTS subsystem of Eclairs With S. Schanne from CEA/IRFU/SaP, we develop software code under RTEMS for the scientific trigger of Eclairs.
 FPGA digital signal processing study for BAOradio project (dark energy survey through Baryon Acoustic Oscillations power spectrum measuments):
 development of a VHDL 4096points streaming FFT operator dealing with a continuous 8bit, 500 MSPS data stream.
 (in progress) development of a FFTbased correlator system (a 256points FFT every 2 ns).
 The BAO Radio acquisition system
GPU Implementation of numerical simulation code for astrophysics : RAMSESGPU
In 20092010, with my colleagues F. Château et R. Teyssier, we have implemented several numerical schemes to solve Euler equations (gas dynamics) on GPU : 1. a finite volume code using a Godunovtype Riemann solver 2. a Riemannfree solver, the TurganovTadmor centered scheme. First results are reported in a proceeding presented at HPCTA 2010.
This first attemp was further developped to give the highly efficient multiGPU magnetohydrodynamics simulation code named RAMSESGPU.
PhD thesis manuscript (advisor: Alain Arnéodo):
Abstract :
Since the end 80’s, wavelet transform has been recognized as a privileged tool to study fractal objects, providing a unified multifractal formalism for both functions and measures. In the first part, we use the 2D WTMM (Wavelet Transform Modulus Maxima) methodology to study mammography. We illustrate the usefulness of the methodology in the study of texture segmentation of rough surfaces and in the geometric characterization of clusters of microcalcifications, which are early signs of breast cancer. In a second methodologic part, we generalize the WTMM method to provide a multifractal description of both 3D scalar and vectorial data fields, introducing the tensorial wavelet transform. We show that a recursive filter technique allows to save between 25% and 60% of computing time, as compared with FFT based filtering techniques. Then we apply the 3D WTMM method to Direct Numerical Simulations (DNS) of the NavierStokes equations in turbulent regime with moderate Reynolds numbers. By mesuring a significantly nonzero cancellation exponent, we bring evidence that multifractal properties of both 3D dissipation and enstrophy fields are well accounted for nonconservative multiplicative cascading processes. Moreover, we observe that the cancellation exponent decreases as the Reynolds number increases. Finally, we present the first results of a fully vectorial multifractal analysis of both velocity and vorticity fields on the same numerical simulations showing that the value of the intermittence parameter , as measured by the tensorial 3D WTMM method, is significantly larger than the one obtained by studying 1D longitudinal velocity increments.
Keywords
Multidimensionnal wavelet transform, rough surface, singularity, H”older exponent, multifractal, scale invariance, autosimilarity, WTMM methodology, singularity spectrum, fractional Brownian motion, stochastic process, random cascade, mammography, microcalcifications, fully developped turbulence.