Successive refinement of vector sources under individual distortion criteria

Jayanth Nayak, Ertem Tuncel, Deniz Gündüz, Elza ErkipIEEE Trans. Information Theory 2010 · DOI · OpenReview

Abstract

The successive refinement problem is extended to vector sources where individual distortion constraints are posed on each vector component. For vector Gaussian sources with squared-error distortion, a single-letter rate-distortion characterization is inherited from the previously studied Gaussian multiple descriptions problem with covariance distortion constraints. Though this characterization is amenable to well-known numerical convex optimization techniques, an analytical solution is difficult to obtain in full generality even for 2-D sources. In this work, the special case of successive refinability is addressed analytically. Specifically, vector Gaussian sources are shown to be <i xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">not</i> successively refinable everywhere unlike scalar Gaussian sources. It is also shown that, for 2-D Gaussian sources, the rate loss at the second stage can be as high as 0.5 b/sample in a ¿degenerate¿ scenario corresponding to what is known as sequential coding of correlated sources. Finally, analysis of 2-D binary symmetric sources with Hamming distortion reveals that the behavior of these sources with respect to successive refinability exhibits remarkable similarity to their 2-D Gaussian counterparts.
BibTeX
@article{nayak2010successive,
  author = {Jayanth Nayak and Ertem Tuncel and Deniz Gündüz and Elza Erkip},
  title = {{Successive refinement of vector sources under individual distortion criteria}},
  journal = {IEEE Trans. Information Theory},
  year = {2010},
  doi = {10.1109/tit.2010.2040858},
}
Figure