Sub-Nyquist Sampling of ECG Signals Based on the Extension of Variable Pulsewidth Model

To prevent lifestyle diseases, portable heartbeat detection systems by electrocardiograms (ECGs) for daily life monitoring have attracted attentions. For portable systems, power consumption is critical and it is thus beneficial to reduce the sampling frequency at the point of acquisition. Recent developed variable pulsewidth finite rate of innovation (VPW-FRI) framework provides an effective way for low-rate sampling and compression of ECG signals. Unfortunately, noise and model mismatch will induce inaccuracies to such scheme in the reconstruction process. In this article, a sub-Nyquist sampling scheme based on the extension of VPW model is proposed for ECG signals, which reduce the model dismatch error, and improve the reconstruction quality under model mismatch and noise scenario. In the proposed scheme, ECG signal is modeled as the differentiated VPW model to reduce the model mismatch error. Then a FRI sampling framework based on the differentiated VPW model is proposed to sample the ECG signal at a sub-Nyquist rate. After the ECG signal is sampled, some consecutive Fourier series coefficients are calculated according to the samples. Then a modified annihilating filter algorithm is proposed to estimate parameters of the ECG signal and reconstruct the ECG signal. Finally, simulation experiments with real ECG signals from Massachusetts Institute of Technology-Beth Israel Hospital (MIT-BIH) Arrhythmia Database have shown that the proposed scheme outperformed the traditional methods for model mismatch. The signal-to-residual ratio (SRR) of the reconstructed signal of the proposed scheme is higher than that of the VPW-FRI scheme.

I. Introduction

As the World Health Organization (WHO) stated, the cardiovascular diseases are the leading causes of death and disability of people [1], [2]. Heart health is expected to play an increasingly prominent role in health provision and attracts more attention in delay life. Especially, daily life monitoring is important to prevent lifestyle diseases and raise the number of patients requiring nursing care. Since the health conditions of a human’s heartbeat are reflected in the electrocardiogram (ECG) waveforms, it is important to design a portable heartbeat detection system by ECG for daily life monitoring [3]–[7]. For portable systems, power consumption is critical. It is quite intuitive that a higher sampling rate requires more power when converting a continuous-time signal to digital, since a faster and more precise clock is needed [8], [9]. It is thus beneficial to reduce the sampling frequency at the point of acquisition. Furthermore, a low-sampling rate is also beneficial in reducing the transmission power required [10]. Moreover, a 24 h or even longer duration recording for the patient’s ECG is desirable for doctors to detect the human body’s abnormalities [11], [12]. This will result in large amounts of ECG data everyday for storage and transmission. Therefore, there is an urgency to reduce the ECG data stored and transmitted. The current lossless compression methods [13]–[19] for ECG signal compression is based on the traditional Nyquist sampling theory, and the discrete ECG signal is compressed by entropy coding [20], such as Huffman coding [13]–[15], Golomb-Rice coding [16], and Prediction error coding [17]. The lossless compression methods have a high reconstruction accuracy. However, the above methods cannot reduce the sampling rate of ECG signal. Thus, sub-Nyquist sampling of ECG signals are also demanded [21]–[23].