Compression of ECG using a code excited linear prediction (CELP)

Abstract

A lossy ECG signal compression method using CELP is introduced here. Linear predictive coding (LPC) has been widely used as a compression technique in the field of speech coding. Similar to speech signals, ECG signals can be thought of as being produced by spectrally shaping excitation pulses. The encoder presented uses an analysis by synthesis structure. The short term spectral characteristics of the ECG signal are approximated by an IIR filter using LPC coefficients. Optimal excitations are selected from two codebooks, an adaptive codebook which takes advantage of long term correlation between pulses, and a fixed codebook which models short term changes in the characteristics of individual pulses. Using the MIT/BIH arrhythmia database with a rudimentary fixed codebook the authors were able to achieve a compression ratio of approximately 13:1 with a typical normalized MSE's ranging from 0.017 to 0.033 using a frame size of 200 samples.