The cardiovascular health status of the human body can be reflected through pulse waves. Important physiological parameters such as heart rate, blood pressure, and the degree of vascular sclerosis can be obtained through the analysis of these waves. The sensor predominantly used for pulse measurement is the photoelectric sensor, which is capable of detecting pulses at various measurement positions, thus making it extensively used in wearable sports equipment for heart rate detection. However, during the process of measuring pulse waves with photoelectric sensors, there are often various noise interferences such as motion artifacts, power interference, and respiratory effects. Moreover, this measurement method is primarily invasive, which can make people uncomfortable. Therefore, it is necessary to select appropriate sensors to avoid discomfort to the human body during the measurement process and denoise the collected signals.
We designed a pulse wave signal acquisition platform based on fiber Bragg grating (FBG) sensors. The platform was composed of FBG sensors embedded in nylon wristbands. Initially, the FBG wristband was secured at the radial artery of the left hand to gather pulse wave signals for demodulation. The collected pulse wave signals were subject to baseline drift. Hence, integrated empirical mode decomposition (EMD) and cubic spline interpolation were used for detrending prior to denoising. Subsequently, the amplitude of Gaussian white noise added to the complementary ensemble EMD (CEMMD) was optimized using particle swarm optimization (PSO) algorithm. The CEEMD algorithm decomposed the pulse wave signal into a series of intrinsic mode function (IMF) components. An improved wavelet threshold function was then applied to process these IMF components. The correlation coefficient between each IMF component and the original pulse wave signal was calculated, and this coefficient was used to determine the effectiveness of each component. Finally, all effective signals were reconstructed to obtain a smooth pulse wave signal.
To validate the performance of the method proposed in this study, simulation experiments are conducted using three comparative algorithms. The denoising performance is evaluated using signal noise ratio (SNR) and root-mean-square error (RMSE). Gaussian white noise with an SNR ranging from 5 to 25 dB is added to the simulation signal. The denoising performance is also verified on actual collected pulse wave signals. The simulation results (Table 1 and Table 2) show that even when 5 dB noise is added, the SNR after denoising can still reach 15.785 dB, and RMSE can be reduced to 1.251. When 25 dB noise is added, the SNR after denoising is 31.959 dB, and RMSE is 0.215. Even if the SNR is low, compared with other methods, the algorithm proposed in this study performs better on these two evaluation indicators and has better denoising performance. The results of determining the amplitude of Gaussian white noise (Fig. 4) intuitively display that when the amplitude of Gaussian white noise added in CEEMD is 0.35, the average mutual information of IMF components is the lowest. This indicates that the denoising effect is the best at this time. The actual experimental results are shown in Fig. 9. The signal obtained after denoising by the proposed algorithm is smoother; the amplitude is not distorted, and it effectively removes spikes and high-frequency noise in the signal. This is because the PSO algorithm optimizes the amplitude of white noise added to CEEMD, overcoming problems such as modal aliasing, endpoint effects, and new harmonic components introduced by inappropriate Gaussian white noise in the decomposition process of CEEMD. Using correlation coefficients to select valid and invalid signals successfully removes most invalid signals (Table 5). In general, the proposed algorithm can better remove noise in signals than other algorithms.
We propose a method for collecting pulse wave signals using FBG sensors. By considering the various noise interferences in the pulse wave signal, a joint denoising algorithm of PSO-CEEMD-IWT is proposed. Different amplitudes of white noise are added to both the simulation signal and the actual signal. We determine 0.35 as the optimal amplitude of white noise added to CEEMD, which further suppresses the modal aliasing phenomenon, compared with the amplitude selected based on experience. The average mutual information obtained by the method in this paper is lower than that obtained by selecting the white noise amplitude according to experience. The results show that the SNR, RMSE, and other indicators obtained by the proposed algorithm are the best; there is no waveform and amplitude distortion, and the denoised signal is smoother, which proves that the performance of the pulse wave denoising proposed in this paper is more outstanding. The signal has a higher degree of restoration to the pulse wave signal, which is of great significance for later combination with feature extraction and objectification of pulse diagnosis. We also propose a feasible way to obtain high-quality pulse waves.
