An Adaptive Predictor for Date Compression

摘要

The problem of predicting the Nth sample of a periodically stationary random sequence (a digitized ECG) using a set of L prior samples is considered. The entropy of the source is calculated, using a Markov source model, to find that entropy decreases rapidly with source order. Only a very short predictor should be needed. The linear, least-mean-square estimator is derived and computer simulated. It is shown to be short (L=1), relatively robust, moderately accurate (usually within 10%), and adaptive in that the estimator improves from period to period. Data compression ratios of about 4:1 can reasonably be expected from direct application of the predictor; however, by judicious deletion and later regeneration of samples, it is felt that an additional 4:1 compression is achievable. (Author)