A model-based analysis of scalable Multiple Description Coding

Abstract

This paper proposes a novel Rate-Distortion (R-D) model for a generic Embedded Multiple Description Scalar Quantizer (EMDSQ). The proposed model is experimentally validated for several quantizer instantiations, showing notable accuracy. Driven by the proposed model, a detailed theoretical analysis of a general EMDSQ-based Multiple Description Coding (MDC) system is carried out. In this context, an optimized packet transmission strategy for a scalable MDC system is theoretically investigated. The analysis reveals important insights into the optimal overall redundancy and average Signal-to-Noise-Ratio (SNR) variations of the system. Our analysis, being independent of the quantizer realization, allows for drawing valuable conclusions on the theoretically achievable performance of EMDSQ-based MDC systems.