Codage de sources avec information adjacente et connaissance incertaine des corrélations

Abstract : In this thesis, we considered the problem of source coding with side information available at the decoder only. More in details, we considered the case where the joint distribution between the source and the side information is not perfectly known. In this context, we performed a performance analysis of the lossless source coding scheme. This performance analysis was realized from information theory tools. Then, we proposed a practical coding scheme able to deal with the uncertainty on the joint probability distribution. This coding scheme is based on non-binary LDPC codes and on an Expectation-Maximization algorithm. For this problem, a key issue is to design efficient LDPC codes. In particular, good code degree distributions have to be selected. Consequently, we proposed an optimization method for the selection of good degree distributions. To finish, we considered a lossy coding scheme. In this case, we assumed that the correlation channel between the source and the side information is described by a Hidden Markov Model with Gaussian emissions. For this model, we performed again some performance analysis and proposed a practical coding scheme. The proposed scheme is based on non-binary LDPC codes and on MMSE reconstruction using an MCMC method. In our solution, these two components are able to exploit the memory induced by the Hidden Markov model.
Complete list of metadatas

Cited literature [97 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-01136663
Contributor : Abes Star <>
Submitted on : Monday, March 3, 2014 - 6:37:24 PM
Last modification on : Friday, September 14, 2018 - 1:02:32 AM
Long-term archiving on : Tuesday, June 3, 2014 - 11:00:50 AM

File

VD2_DUPRAZ_ELSA_03122013.pdf
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-01136663, version 2

Collections

Citation

Elsa Dupraz. Codage de sources avec information adjacente et connaissance incertaine des corrélations. Autre [cond-mat.other]. Université Paris Sud - Paris XI, 2013. Français. ⟨NNT : 2013PA112301⟩. ⟨tel-01136663v2⟩

Share

Metrics

Record views

550

Files downloads

747