Two techniques for symmetric multiple description coding with reduced storage space decoder

Abstract

In this work we propose two techniques for symmetric multiple description coding with reduced storage space decoder. The first technique is multiple description scalar quantizer with linear joint de-coders. We propose an optimal design algorithm similar to Vaishampayan's algo-rithm, to which we add an index assignment optimization step. We also solve an additional challenge in the decoder optimization, by proving that the problem is a convex quadratic optimization problem with a closed form solution (under some mild conditions). Our tests show that the new method has very good performance when the probability of description loss is sufficiently low. The other technique is an improvement to the traditional multiple description coding scheme based on uneven erasure protection. We evaluate the asymptotical performance of both schemes for a Gaussian memoryless source. The analysis reveals that the improvement reaches over 1 dB for up to ten descriptions and low probability of description loss. From our experiments we observe that the improved scheme is very competitive comparing to other multiple description techniques as well.