Some Properties of a Predictive Quantizing System

This paper presents some theoretical properties of a predictive quantizing system for the transmission of random signals. Expressions for the mean-squared error are derived for some input processes. It is shown that for the Gaussian case the predictive system gives no improvement over the conventional system. The class of linear Markov processes is also considered, and an example is presented which clearly illustrates the limited region of improvement for the predictive system. The relation of the improvement to the correlation of the input samples is also illustrated.