Prediction of a signal from earlier samples is widely used to reduce the information that must be transmitted for a suitably accurate reconstruction at a receiver. Linear prediction is a common means of effecting the prediction, but it does not accommodate well signals that include dominant innovations from time to time, as in the case of speech, or signals that are not well suited to modeling as the output of a linear dynamical system. We have investigated a scheme in which a set of recent samples is matched to an earlier set of samples and then the sample following the earlier set is used as the prediction. Error behavior is compared with linear prediction, theoretically and experimentally, and found to be reasonably similar on a mean square basis.
展开▼
