Design of tree-structured multiple description vector quantizers

We present a new multiple description source coding scheme based on tree-structured vector quantization (TSVQ). In this scheme, the codebook of each decoder is organized in a binary tree. The encoding is greedy and based on a sequence of binary decisions as in traditional TSVQ. Each binary decision of the encoder corresponds to adding information on one of the available channels and the encoding complexity can be shown to be proportional to the total bitrate. We describe the encoder structure for the two-channel case, and propose an entropy-constrained design algorithm based on marginal return analysis. Experimental results on a Gaussian source are presented for various design parameters and the generalization of the scheme to more than two channels is outlined