El-Hadi Djermoune Postdoctor (CRAN / IRIS Team)  Research interest  Teaching activities  Curriculum Vitæ  Publications

Research interest

The work I have developed during my thesis from 1999 to 2003 deals with the estimation of the parameters of noisy damped exponentials under strong constraints : high number of components, frequency and dynamic resolution problems, large number of samples, etc. In this context, the direct use of high resolution (HR) methods is generally inefficient. Below, a short abstract of my Ph.D. dissertation is given.

Firstly, after some recalls on HR estimators, we present an approach based on the modified Yule-Walker equations and on the use of a particular estimator of the correlation function, with adjustable summation horizon. It is shown that an ARMA modeling is preferable and that the optimal adjustment of this horizon allows one to improve the signal-to-noise ratio (SNR).
Secondly, we introduce the principles of subband decomposition and, after analyzing the conditions in which this should be carried out (filter bank structure, order selection, polyphase components, etc.), we show the advantages of subband estimation from the point of view of variance, resolution and detection rate. Finally, we underline its limitations concerning the choice of the decomposition depth. The solution adopted to overcome this problem consists in using adaptive schemes.
After having identified some limitations of the adaptive techniques presented in the literature, we propose a new stopping rule based on a spectral flatness measure of the estimation residuals. This approach allows one to obtain the best tradeoff between the variance, the detection rate and the computational complexity. The suggested method is then applied successfully to real-world NMR signals of relatively high complexity. In particular, it is shown that this method allows to deal with signals at lower SNRs than the traditional techniques using a Fourier transform. This allows to expect an appreciable reduction of experimentation times.

• frequency estimation/spectral analysis
• adaptive subband decomposition
• filter banks
• nuclear magnetic resonance spectroscopy

  French version of my Ph.D. dissertation in PDF (3 820 Ko)

Teaching activities

• Analog and digital signal processing
• Java programming
• Industrial informatics
• Mathematics

Curriculum Vitæ

  Download my CV (111 Ko)

Publications

		E.-H. Djermoune Estimation of the parameters of damped exponentials via an adaptive subband decomposition. Application to NMR spectroscopy. Estimation des paramètres de sinusoïdes amorties par décomposition en sous-bandes adaptative. Application à la spectroscopie RMN.