A Theory For Vector Quantization
S.C. Greer, James George Dunham · 2005
A DPCM-based vector quantizer is derived as an approximation to the recently developed step-by-step optimal quantizer algorithm. The new quantizer requires only a fraction of the computation needed by the optimal quantizer. While the approximation is derived under the assumption of a large bit rate, simulation studies have shown that for certain sources the DPCM-based vector quantizer provides mse results equivalent to that of the optimal quantizer at a bit rate as low as .5 bit per sample for a vector length of two. This derivation provides strong motivation for a vector DPCM structure with a quantizer designed for a specific pdf and a predictor designed using the Generalized Wiener-Hopf equations.