Design of Asymmetric Multiple Description Lattice Vector Quantizers Suhas N. Diggavi, N. J. A. Sloane, and Vinay A. Vaishampayan AT&T Shannon Laboratories, 180 Park Avenue, Bldg 103, Florham Park NJ 07932, USA. Tel: (973)-360-8492, FAX: (973)-360-8178. Email: {suhas,njas,vinay}@research.att.com Abstract We consider the design of asymmetric multiple description lattice quantizers that cover the entire spectrum of the distortion profile, ranging from symmetric or balanced to successively refinable. We present a solution to a labeling problem, which is an important part of the construction, along with a general design procedure. This procedure is illustrated using aZZ 2 lattice. We also evaluate its rate-distortion performance and compare it to known information theoretic bounds. I Introduction A multiple description source coder generates a set of binary streams or descriptions of a source sequence, each with its own rate constraint. The transmission medium may deliver some or all of the descriptions to the decoder. The objective is to minimize the distortion between the source sequence and the decoded sequence when all the descriptions are available, while ensuring that the distortion obtained by decoding only a subset of the descriptions remains below a pre-specified value that depends on the subset. With D descriptions, we associate a vector of length 2 D , each element of which is a distortion constraint for a given description subset. This distortion vector is referred to as the distortion profile. In recent years, multiple description coders have received considerable attention, driven by the interest in packet voice and video communications. Most of the work has centered around the successively refinable and balanced cases (with the exception of [1]), which are in a sense two extreme cases of the distortion profile. Successive refinement coders find application in networks with a priority structure whereas bal- anced codes are useful in networks that do not have such a structure, the best example being the Internet. In this paper we propose a structured scheme that bridges the two schemes, in the sense that it allows for an arbitrary distortion profile (within reason). By making 1