QUANTIZATION FOR MAXIMAL PRESERVATION OF INFORMATION

摘要

In this paper we consider the optimum quantization of uncorrelatedrndiscrete stochastic signals characterized by itrnPDF (Probability Distribution Function). The quantizer isrndesign to maximize the preservation of information fromrnthe original signal to the quantized signal. We present arnsimple characterization of the quantization intervals and therncorresponding quantization values.rnIt is shown that the optimum quantization characteristicrnis in general non-uniform and the optimum quantizationrnintervals depend on the PDF of the stochastic to bernquantized, with smaller quantization intervals (and, consequently,rnsmaller quantization errors) in the regions wherernthe PDF takes higher values (i.e., for the values that arernmore likely to occur. This contrasts with conventional nonuniformrnquantization characteristics where we have smallerrnquantization intervals for signals with smaller amplitude,rnregardless of its PDF.