Design of entropy-constrained quantizers based on coset codes

Abstract

Asymptotic vector quantization (VQ) theory indicates that at high rates, the optimal VQ uniformly places its reproduction vectors in the high dimensional space in a manner that each of the quantization cells is made as spherical as possible. As compared with scalar quantizers whose quantization cells are cubic, VQ exploits the space-filling efficiency of its constituent convex polytope and hence achieves the so-called granular gain.<>