Electrocardiogram compression by linear prediction and wavelet sub-band coding techniques

Linear Predictive coding (LPC) is extensively used for analysis and compression of speech signal whereas the Discrete Wavelet Transform is widely preferred for electrocardiogram (ECG) compression. In this paper, we present LPC and wavelet based method to encode ECG signals. The compression algorithm has been evaluated with the MIT-BIH Arrhythmia Database, MIT-BIH Compression database and University of Glasgow noisy and normal database. The performance is quantified by computing distortion measures. The percentage root mean square difference (PRD) is found to be below 8% and wavelet-based weighted PRD (WWPRD) below 0.25. The upper quartile value of Wavelet energy based diagnostic distortion (WEDD) is less than 0.3 even for noisy data. It is observed that, a combination of LPC and wavelet subband coding offers fixed compression of 84.09%. Classification before decompression is achieved with accuracy of 97%.