Experiments with linear prediction in television

The correlation present in a signal makes possible the prediction of the future of the signal in terms of the past and present. If the method used for prediction makes full use of the entire pertinent past, then the error signal — the difference between the actual and the predicted signal — will be a completely random wave of lower power than the original signal but containing all the information of the original. One method of prediction, which docs not make full use of the past, but which is nevertheless remarkably effective with certain signals and also appealing because of its relative simplicity, is linear prediction. Here the prediction for the next signal sample is simply the sum of previous signal samples each multiplied by an appropriate weighting factor. The best values for the weighting coefficients depend upon the statistics of the signal, but once they have been determined the prediction may be done with relatively simple apparatus. This paper describes the apparatus used for some experiments on linear prediction of television signals, and describes the results obtained to date.