Successive Wyner-Ziv Coding for the Binary CEO Problem under Log-Loss

TL;DR

This paper introduces a practical successive coding scheme for the binary CEO problem under log-loss, transforming it into multiple Wyner-Ziv problems and achieving near-theoretical performance with graph-based codes.

Contribution

It proposes a novel successive coding approach converting the CEO problem into multiple Wyner-Ziv problems, utilizing compound LDGM-LDPC codes for asymptotic optimality.

Findings

Achieves sum-rate distortion performance close to theoretical bounds.

Employs a successive decoding scheme with the Sum-Product algorithm.

Demonstrates practical encoding and decoding for the binary CEO problem.

Abstract

An $l$-link binary CEO problem is considered in this paper. We present a practical encoding and decoding scheme for this problem employing the graph-based codes. A successive coding scheme is proposed for converting an $l$-link binary CEO problem to the $(2l-1)$ single binary Wyner-Ziv (WZ) problems. By using the compound LDGM-LDPC codes, the theoretical bound of each binary WZ is asymptotically achievable. Our proposed decoder successively decodes the received data by employing the well-known Sum-Product (SP) algorithm and leverages them to reconstruct the source. The sum-rate distortion performance of our proposed coding scheme is compared with the theoretical bounds under the logarithmic loss (log-loss) criterion.