Effects of quantizer mismatch

If the probability density function of a signal to be quantized is known exactly, it is possible to match the quantizer characteristic to the source statistics for a given criterion, e.g. minimum mean-squared quantizing noise. However, if sources with unknown or nonstationary statistics have to be coded, we have to take into account a possible mismatch between quantizer and probability density function of the source signal. The effects of such a quantizer mismatch are investigated for various density functions, which are often used for modeling the quantizer input signal of speech or picture encoding systems. The results have partly been gained analytically and are partly based on numerical calculations. A comparison of signal-to-noise ratios is presented as well as quantizer output entropies for different mismatch situations.