Experimental evaluation of ECG signal denoising methods based on HRV indices and their application in indoor thermal comfort study under different temperatures

Abstract

The heart rate variability (HRV) features extracted from an electrocardiogram (ECG) signal have been widely used to help improve the performance of thermal comfort models, with the ECG signal denoising method being very important. In this study, 20 volunteers were recruited for experiments in a well-controlled chamber. Their ECG signals (a total of 6,000,000 sets of data) were collected under stable and variable temperature environments. With the use of the setup database, four denoising methods were compared: the fast Fourier transform (FFT), wavelet transform, Chebyshev filter, and Butterworth filter. When the performance of these methods was evaluated from ECG images, the Butterworth filter exhibited a more comprehensive and smoother representation of the original signal. Judging from denoising indices, the Butterworth filter exhibited the best performance, with its denoising indices surpassing those of fast Fourier transform and wavelet transform by differences ranging from 26% to 122%. Judging from HRV index and R-peak positioning, the data processed by Butterworth exhibited the highest reliability (93.5%), outperforming FFT (30.0%), the wavelet filter (72.5%), and the Chebyshev filter (78.5%). Furthermore, the 5th-order Butterworth filter exhibited the largest signal-to-noise ratio (SNR) with 1.1–34.4% larger and the smallest percent root-mean-square difference (PRD) with 15.2–66.1% smaller compared to other orders. Next, the characteristics of denoised HRV features under different temperatures were studied. During a temperature rise stage, the fitting R2 values of rMSSD, SDNN, and HR with temperature were 0.85, 0.64, and 0.78, respectively; meanwhile, during the subsequent temperature drop stage, the fitting R2 values were 0.78, 0.84, and 0.78, respectively. This suggests that changes in temperature can be effectively reflected by HRV indices. When gender was taken into consideration, the HR of females showed a 55.8% higher fitting level compared to males.