Comparing performance of entropy constrained scalar quantizer and other scalar quantizers

Abstract

The optimal entropy constrained scalar quantizer (ECSQ) gives the best possible rate-distortion performance of any scalar quantizer. While at high bit rates the optimal ECSQ converges to the uniform quantizer, optimization at low-bit regions demands usually numerical solutions with high complexity. The performance of optimal ECSQ is compared with three simpler scalar quantization configurations: uniform threshold quantizer, uniform quantizer with Wiener filter and the zero input correlation quantizer with Wiener filter. Gaussian distributed inputs are used in the simulations. Results show that the simpler quantizers, are able to match the optimal ECSQ performance, even at low bit rates while the uniform quantizer with Wiener filter provides the simplest configuration.