In previous work on multiple description lattice vector quantizers (MDLVQ) with L ≥ 3 descriptions, once the central and side lattice codebooks are fixed, the decoding quality is determined for all numbers k of received descriptions. Therefore, it is not possible to achieve tradeoffs between the quality of reconstruction for different values of k, 1 ≤ k ≤ L - 1. This work proposes a flexible MDLVQ capable of overcoming the above drawback. For this, a different reconstruction method is employed and a heuristic index assignment algorithm, which uses L - 2 parameters to control the distortions for 2 ≤ k ≤ L - 1, is developed. The second contribution of this work is a structured index assignment for the case L = 3 and the derivation of asymptotical expressions of the distortions at high resolution. The proposed index assignment has a simple mechanism for controlling the tradeoff between the reconstruction quality when k = 1 and when k = 2, and is able to achieve a wide range of distortion values.

In the previous work on multiple description lattice vector quantizers (MDLVQs) with L ≥ 3 descriptions, once the central and side lattice codebooks are fixed, the decoding quality is determined for all numbers k of received descriptions. Therefore, it is not possible to achieve tradeoffs between the quality of reconstruction for different values of k, 1 k ≤ L - 1. This paper proposes a flexible MDLVQ capable of overcoming the above drawback. For this, a different reconstruction method is employed and a heuristic index assignment (IA) algorithm, which uses L - 2 parameters to control the distortions for 2 k ≤ L - 1, is developed. Experimental results show that the proposed MDLVQ, in addition to achieving the desired tradeoffs, significantly outperforms the classic MD scheme based on unequal erasure protection. The second contribution of this paper is a structured IA for the case of L = 3 and the derivation of the corresponding expressions of the distortions at high resolution. The proposed IA has a simple mechanism for controlling the tradeoff between the reconstruction quality for k = 1, 2. The IA is able to achieve a wide range of distortion values, while keeping the product of the distortions for k = 1, 2 the same as in the prior work.