Channel Optimized Distributed Multiple Description Coding

TL;DR

This paper introduces channel optimized distributed multiple description vector quantization schemes for efficient source coding over noisy channels with packet loss, employing iterative decoding and source selection methods.

Contribution

It presents novel CDMD encoding and decoding algorithms, including deterministic annealing design, asymmetric MMSE decoding, and source selection strategies for improved distributed source coding.

Findings

Enhanced reconstruction accuracy over noisy channels.

Effective source selection methods improve decoding performance.

Comparative analysis demonstrates advantages of proposed schemes.

Abstract

In this paper, channel optimized distributed multiple description vector quantization (CDMD) schemes are presented for distributed source coding in symmetric and asymmetric settings. The CDMD encoder is designed using a deterministic annealing approach over noisy channels with packet loss. A minimum mean squared error asymmetric CDMD decoder is proposed for effective reconstruction of a source, utilizing the side information (SI) and its corresponding received descriptions. The proposed iterative symmetric CDMD decoder jointly reconstructs the symbols of multiple correlated sources. Two types of symmetric CDMD decoders, namely the estimated-SI and the soft-SI decoders, are presented which respectively exploit the reconstructed symbols and a posteriori probabilities of other sources as SI in iterations. In a multiple source CDMD setting, for reconstruction of a source, three methods are proposed to select another source as its SI during the decoding. The methods operate based on minimum physical distance (in a wireless sensor network setting), maximum mutual information and minimum end-to-end distortion. The performance of the proposed systems and algorithms are evaluated and compared in detail.