Multiple description trellis-coded quantization

Hamid Jafarkhani, Vahid Tarokh

Abstract: We present a construction of multiple description trellis-coded quantizers. We use the tensor product of trellises to build a trellis which is applicable to multiple description coding. The problems of index assignment and set partitioning for the resulting trellis are considered. The Viterbi algorithm provides the best path for encoding and the design procedure utilizes a generalized Lloyd algorithm. The encoding process simultaneously generates all the transmitted sequences. Furthermore, the complexity of the scheme is almost independent of the rate. The quantizer provides remarkable performance with little encoding complexity.