HAL Id: hal-01516795
https://hal.sorbonne-universite.fr/hal-01516795
Submitted on 2 May 2017
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Distributed under a Creative CommonsAttribution - NonCommercial - ShareAlike| 4.0 International License
Non-uniform sampling & denoising applied to nuclear magnetic resonance
Guillaume Laurent, Christian Bonhomme
To cite this version:
Guillaume Laurent, Christian Bonhomme. Non-uniform sampling & denoising applied to nuclear magnetic resonance. Spring School on Sparse Representations and Compressed Sensing, Apr 2016, Ilmenau, Germany. �hal-01516795�
Non-Uniform Sampling & Denoising
applied to Nuclear Magnetic Resonance
G. LAURENT, C. BONHOMME
Laboratoire de Chimie de la Matière Condensée de Paris
Spring School on Sparse Representations and Compressed Sensing, 4th-8th April 2016, Ilmenau, Germany
●
NUS and SVD are useful to increase sensitivity
●
Efficient algorithms are critical
●
Graphic card is a low cost option (Nvidia GTX 750 = 120 €)
Highlights
●
Automatic SVD thresholding
●
Sparse matrix SVD
●
Combining NUS and SVD
Future work
●
V. BARRET-VIVIN, F. PORTIER and M. ROBIN for samples
●
P. P. MAN for Java application
●
W. WOELFFEL for python application
Acknowledgements Non-Uniform
Sampling (NUS) Spectra denoising
[2]
Singular Value Decomposition (SVD) [3]
Low rank approximation 1D Signal
Hankel Matrix
Tools Context Nuclear Magnetic
Resonance Increasing sensitivity References
SANOFI
Physico-chemical spectroscopic analysis
Precise but poor sensitivity
↓
100 mg of sample needed Broad noisy peaks
for distributed environments
Sensitivity gain
= time gain
or sample size gain Microcoils (MACS) [1]
[1] D. Sakellariou, G. L. Goff, and J.-F. Jacquinot, ‘High-
resolution, high-sensitivity NMR of nanolitre anisotropic samples by coil spinning’, Nature, vol. 447, no. 7145, pp. 694–697, Jul.
2007.
[2] M. Mobli and J. C. Hoch, ‘Nonuniform sampling and non- Fourier signal processing methods in multidimensional NMR’, Prog. Nucl. Mag. Res. Sp., vol. 83, pp. 21–41, Nov. 2014.
[3] J. A. Cadzow, ‘Signal enhancement-a composite property mapping algorithm’, IEEE T. Acoust. Speech, vol. 36, no. 1, pp.
49–62, Jan. 1988.
[4] P. P. Man, C. Bonhomme, and F. Babonneau, ‘Denoising NMR time-domain signal by singular-value decomposition
accelerated by graphics processing units’, Solid State Nucl. Mag., vol. 61–62, pp. 28–34, Jul. 2014.
In st ru m en ta tio n
Proc essin g Ac qui siti on
Sensitivity
2E+5 2E+6 2E+7
0,1 1 10 100 1000
P4 530 PM 745 C2D E6400 C2Q Q8200 Ci3 4005U C2D T9600 Ci5 4670K NVS 160M 8400 GS FX 570 FX 770M 820M GTX 260 GTX 660 Matrix size
Time (s)
CP U G PU
* * ¤ * * ¤ ¤ ¤ Matrix shape
Results
CPU: Central Processing Unit GPU: Graphics Processing Unit
Massively parallel (Nvidia CUDA) [4]
GPU limited by memory transfers CPU time / 100:
●
Divide and conquer → / 8
●
SSE3 instructions → / 3
●
Intel MKL library → / 2
●
Single Precision → / 2
Java
0,1 1 10 100 1000
CPU
Time (s) GPU
Windows Ci5 4670K + GTX 660 (2013)
Optimised CPU GPU
Unoptimised CPU
29
