Multiterminal Source Code Design Based on Slepian-Wolf coded quantization∗

Multiterminal (MT) source coding refers to separate lossy encoding and joint decoding of multiple correlated sources. This paper presents two practical MT coding schemes under the same general framework of Slepian-Wolf coded quantization (SWCQ) for both direct and indirect quadratic Gaussian MT source coding problems with two encoders. The first asymmetric SWCQ scheme relies on quantization and Wyner-Ziv coding, where a quantized source is compressed using asymmetric Slepian-Wolf coding (with side information at the decoder). It is implemented via source splitting with one classical source coding component and two Wyner-Ziv coding components. In the second symmetric SWCQ scheme, the two quantized sources are compressed using symmetric Slepian-Wolf coding. We show that each scheme can potentially achieve any point on the inner bound of the rate region for both direct and indirect MT coding problems. Practical designs based on entropy-coded TCQ for classic source coding and LDPC code based asymmetric Slepian-Wolf coding of TCQ indices for Wyner-Ziv coding in the first scheme, and arithmetic code and turbo code based symmetric SlepianWolf coding of TCQ indices in the second scheme perform only 0.29 bit per sample away from the inner bound of the rate region. This work thus represents a major step towards limit-approaching MT code designs.