&&ReWrAp:HEADERFOOTER:0:ReWrAp&&

Date of Award

Fall 2011

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Electrical and Computer Engineering

Supervisor

Sorina Dumitrescu

Co-Supervisor

Jun Chen, Xiaolin Wu

Language

English

Committee Member

Jun Chen, Xiaolin Wu

Abstract

This thesis presents two practical coding schemes based on low-density generatormatrix (LDGM) codes, for two cases of the multiple description (MD) problem. The first one is for the two description problem for finite-alphabet sources with Hammingdistortion measure. The proposed MD code targets the Zhang-Berger region, which is the best inner bound known so far for the corresponding MD rate-distortion region.The coding scheme can be regarded as a practical implementation of a theoreticalsequential coding system for the corner points of a related rate-region, where the random codebooks are replaced by multilevel LDGM codebooks and the encoding at each stage is performed via a message passing algorithm. This coding system is further applied in three notable cases: 1) no excess sum-rate case for binary sources;2) successive refinement for general finite-alphabet sources, 3) no excess marginal ratefor uniform binary sources. Furthermore, in order to assist the code design in the noexcesssum-rate case for binary sources, the exact expression of the distortion regionand of the auxiliary variables needed to achieve its boundary, are derived, which is another important contribution of the thesis.

The second proposed MD code is for the case of L descriptions with individual and central distortion constraints, for the memoryless Gaussian source with squared distortion measure. It is shown first that the coding problem for an arbitrary point on the dominant face of an L-description El Gamal-Cover (EGC) rate region, can be converted to that for a vertex of a K-description EGC rate region for some K<=2L-1,where the latter problem can be solved via successive coding. The practical coding scheme reduces each successive coding step to a Gaussian quantization operation, and implements this operation using multilevel LDGM codes.

The LDGM-based coding schemes are extensively tested in practice for all aforementionedcases. The experimental results show very good performance, verifyingthat the proposed schemes can approach the theoretical rate-distortion bounds.

McMaster University Library