Fixed block rate quantizer using entropy coding

When applied to signal quantization, it is well known that entropy coding allows to obtain results closest from the limit of the rate distorsion function, than optimum quantizers. However, the entropy coding results in a variable number of bits per signal sample. This drawback can be partially avoided by using an algorithmic procedure which allows to keep the number of bits constant per block of signal samples. In the first part of the paper, three conventional feed-forwardly adapted quantizers: namely the Block Companded PCM (BCPCM) quantizer, the Max optimum quantizer, and the optimum uniform quantizer, are briefly reminded and the algorithmic procedure of the proposed quantizer is described. Then, the performances of the four methods are compared in terms of signal to quantizing noise ratio, when operating on signals having a Laplacian probability density function, assuming various block lengths of samples, and bit rates. The quantizer using entropy coding is shown to provide significant improvements over the other quantizers. Finally, the four quantizers are incorporated in an adaptive predictive coder (APC), and used to quantize residual signals produced by inverse filtering. Again, the quantizer using entropy coding is shown to provide the best performances.