Quantização vetorial utilizando códigos esféricos em camadas de toros

Fabiano Boaventura de Miranda, Cristiano Torezzan

DOI: 10.14209/sbrt.2017.89

Evento: XXXV Simpósio Brasileiro de Telecomunicações e Processamento de Sinais (SBrT2017)

Keywords: Vector quantization Gaussian sources Spherical codes Data compression

Abstract

In this paper we present a novel approach to the problem of vector quantization for gaussian sources, combining a spherical code in layers of flat tori and the shape/gain technique. The basic concepts of spherical codes in tori layers are reviewed and exemplified by a 48-dimensional vector quantizer for the shape, which uses the 24-dimensional Leech lattice as its pre- image. A scalar quantizer is optimized for the gain using Lloyd- Max algorithm for a given rate. The main complexity cost refers to the lattice decoding, which is done in the half of the code dimension. In the proposed example, we apply a polynomial algorithm for the Leech lattice decoding. The proposed quantizer is described in details and some numerical results are presented showing that our approach may outperform some competitive state-of-art technique for vector quantization.

Download