Exploring Capacity-Approaching Channel Codes In Distributed Source Coding

点击翻译按钮获取中文摘要

ABSTRACTDistributed source coding (DSC), also known as distributed compression or the Slepian-Wolf problem, refers to the compression of two or more physically separated but statistically correlated information sources, where the sources send the (compressed) data to a common destination without communicating to each other. Having a close relation to a wealth of network information theory problems, distributed source coding is particularly appealing to sensor networks due to its ability to compress out the inter-source redundancy without explicit inter-sensor communication. Of specific interest to this research is the algorithmic design of practical and efficient DSC solutions using powerful channel codes. In view of the fairly mature status of the advanced channel coding technologies in a theoretical context and the very pervasive scope of their well-proven practical applications, exploring what cutting-edge channel codes can offer in this new field is particularly exciting. This research focuses on the theoretic and algorithmic study of lossless and lossy DSC for symmetric and asymmetric memoryless sources. The goal is to exploit the code binning idea in capacity-approaching linear channel codes, including turbo codes and low-density parity-check (LDPC) codes, to achieve compression rates that are close to the theoretical limit. Specific attention is paid to sources that have non-uniform distributions and/or source-dependent correlations, which are common in practical applications like sensor networks and multimedia streaming. Techniques including source decomposition, code splitting, rate combining and post-processing are used to design efficient coding strategies to match to the sources. In addition to analysis and simulations, practical issues are also addressed and a hardware sensor testbed is built to verify and demonstrate the true performance of these DSC prototypes.