Low-complexity lossless multichannel ECG compression based on selective linear prediction

Abstract

In this paper we present low-complexity method for multichannel lossless compression, dedicated for portable ECG acquisition systems. Each channel is selectively decorrelated using two linear predictors, separately for the heart beats and the background signal. Next, cross-channel correlations are used to determine pairs of strongly dependent channels, which allows for further decorrelation of signals using cross-channel predictors. Prediction is therefore obtained at small computational cost, using a simple QRS detector and three short linear predictors per channel, updated periodically using fast Levinson-Durbin recursion. Finally, the prediction error and the estimate of its variance is used in entropy encoding with highly effective method of asymmetric numeral systems. The average compression ratio depends on the similarity between channels and varies from 2.92 (MIT-BIH ADB, low similarity) to 3.43 (INCART, high similarity).