Study of an adaptive quantizer

An adjustable uniform quantizer dependent on observation of blocks of quantized samples is investigated. In this quantizer, the range the samples are expected to occupy within a block is predicted from observation of the previous block. Then the upper and lower saturation levels of the quantizer are adjusted independently to correspond to this predicted range. This procedure is repeated for each new observed block of samples. The adaptive quantizer is evaluated by means of a computer simulation, comparing it to a uniform quantizer with fixed saturation levels. The system is evaluated for television signals, spacecraft engineering sensor signals, and a multiple Gaussian Markov process. For the television signals, the adaptive quantizer acquires a "variable-range" mode of operation, making use of coherence between successive lines in a frame to achieve a reduction in error. When processing the other signals, the adaptive quantizer acquires a "fixed-range variable-mean" mode of operation achieving reductions in mean squared quantization error from 30 to 90 percent. A comparison with an ideal quantizer illustrates the ability of the adaptive quantizer to make effective use of coherence between samples to achieve a reduction in quantization error.