Entropy-constrained index assignments for multiple description quantizers

Multiple description coding has revealed itself highly useful for the transmission of digital signals over packet-switched networks or diversity systems. The index assignment problem is central in the theory and practice of multiple description quantizers. We propose a novel local optimization algorithm that produces index assignments for such quantizers. It locally minimizes the distortion due to description losses with constraints on the entropy of each description. It works in any vector dimension and is easily generalizable to various cases such as asymmetric descriptions or m-description systems. We present experimental comparisons with previously known index assignment methods in both scalar and vector cases and show that the proposed algorithm performs significantly better.