Structured spherical codes for Gaussian quantization
Jon Hamkins, K. Zeger · 2002
We construct a vector quantizer for the Gaussian source which outperforms known existing quantizers for rates above 2 bits per sample, and of equivalent computational complexity. In particular it outperforms "trellis-based scalar-vector quantization" and "trellis coded quantization". The quantizer is a gain-shape quantizer, whose shape codebook consists of a "wrapped spherical code" in 25 dimensions, constructed from the Leech lattice in 24 dimensions.