Multiple Description Coding of Discrete Ergodic Sources

TL;DR

This paper introduces a new stationary coding approach for multiple description coding of discrete ergodic sources, establishing its relation to traditional block coding and proposing a universal algorithm based on simulated annealing.

Contribution

It defines MD stationary coding, relates it to block coding, and develops a universal simulated annealing-based algorithm for ergodic source coding.

Findings

Established relationship between stationary and block MD coding.

Proposed a universal coding algorithm using simulated annealing.

Enhanced understanding of rate constraints in MD coding.

Abstract

We investigate the problem of Multiple Description (MD) coding of discrete ergodic processes. We introduce the notion of MD stationary coding, and characterize its relationship to the conventional block MD coding. In stationary coding, in addition to the two rate constraints normally considered in the MD problem, we consider another rate constraint which reflects the conditional entropy of the process generated by the third decoder given the reconstructions of the two other decoders. The relationship that we establish between stationary and block MD coding enables us to devise a universal algorithm for MD coding of discrete ergodic sources, based on simulated annealing ideas that were recently proven useful for the standard rate distortion problem.