Distributed source coding (DSC) refers to the compression of the outputs of two or more physically separated sources that do not communicate with each other. Driven by a host of emerging applications (e.g., distributed sensor networks and wireless video), DSC has assumed renewed interest in recent years. Although the theoretical result given by the Slepian-Wolf theorem has been known for more than 30 years, practical approaches to DSC did not appear until 1999.
Aiming to advance the field, this research covers the theory, algorithms, and applications of DSC, with focus on the hard problem of practical code design. The basic element of a distributed source code is the binning scheme. For Wyner-Ziv code design, the PI proposes a paradigm based on Slepian-Wolf coded quantization (SWCQ). Preliminary analytical results show that the performance limits of SWCQ mirror those in classic source coding. Initial code design based on TCQ and LDPC codes gives results that are significantly better than any previously reported. Inspired by these results, the PI proposes to study successive refinement for the Wyner-Ziv problem and apply SWCQ to layered video coding.
This research will also tackle the more general problem of multiterminal source coding under the same powerful paradigm of SWCQ. While asymmetric Slepian-Wolf coding is used in SWCQ for Wyner-Ziv coding, for multiterminal source coding with SWCQ, the PI will explore symmetric Slepian-Wolf coding after quantization of each source. Both direct multiterminal source coding and indirect multiterminal source coding (e.g., the CEO problem that arises in distributed sensor networks) will be treated.
The theme of the research is to view lossy DSC as a source-channel coding problem and perform algebraic binning via the combination of quantization and Slepian-Wolf coding based on channel codes. The main novelty lies in using a single SWCQ framework as the embodiment of algebraic binning to solve different code design problems and address potential applications.
