Electrocardiogram signal modeling based on regular-pulse excitation

Abstract

The growing demand for modern wearable sensors and internet of things (IoT) telemedicine systems requires lightweight data compression methods aimed at reducing transmission bandwidth and wireless power consumption while maintaining high clinical diagnostic quality. Traditional methods usually demand complex processing or distort electrocardiogram (ECG) waveform characteristics, especially when high compression ratios (CRs) are targeted. This paper presents an efficient, lightweight ECG compression framework combining predictive coding with N-residual sparse sampling, which retains only one residual sample for every N data points. By integrating linear predictive coding (LPC), sparse residual sampling, and streamlined quantization and encoding, the proposed approach eliminates the need for computationally heavy transformations, prior model training, or predefined codebooks. Experimental evaluations demonstrate perfect reconstruction of the ECG signal at CRs up to 4. Optimal subsampling achieves CRs between 14.6 and 28.4, with signal-to-noise ratio (SNR) values ranging from 15.73 dB to 42.5 dB and percent root-mean-square difference (PRD) between 1.9% and 5.6%, while consistently keeping PRD below 9%. With its minimal computational overhead, low memory footprint, and preserved diagnostic integrity, this method is well-suited for energy-efficient, real-time implementation on resource-constrained wearable healthcare devices.