Index Assignment for N Balanced Multiple Description Scalar Quantization

In this paper, we address the design of any number of balanced multiple descriptions using the multiple description scalar quantization(MDSQ) technique. The proposed scheme has the advantages of low complexity, the possibility of being extended easily to any number of descriptions and the possibility to trade off between the side, partial and central distortions. Unlike existing schemes, it can produce balanced descriptions at low rates, at the price however of a slightly higher distortion. The behavior of the proposed index assignment at high rate is in the same time similar to state-of-the-art schemes. The proposed scheme offers the possibility to adapt to loss probability, and rate constraints, in playing with both the number of descriptions, and the rate of each of them, to minimize the average distortion. The comparison with the systematic FEC (N, k) scheme shows that the FEC scheme in general gives smaller average distortion, but that our scheme seems to be more robust to sudden changes in network conditions and that receiving all the descriptions in general gives smaller distortions.