Heart Rate Monitoring From Photoplethysmographic Signals During Motion State

The real-time heart rate can reflect a person’s heart ability. From the other hand, heart rate could also represent the general level of health. Hospital often measures the heart rate apply by the electrocardiogram(ECG), which is inconvenient to measure during the daily activities and exercise. Photoplethysmographic(PPG) is a method of measuring the pulse wave. It bases on the LED diodes and the light detectors, measures the lights attenuated and reflected by human’s tissues and arteries. At the same time, the detector traces the arteries’ fluctuation and records the pulse waves. Because the signals is easy to obtain、 the detector is wearable, PPG has became the major method of measuring the pulse and blood oxygen during the non-hospital environment.Since the PPG signal influenced by motion artifact(MA) and deformation of arteries, the accurate heart rate is hard to be extracted from PPG signal during motion states. Through a depth study and implementation, this paper compares several de-noising methods in motion, and propose a denoising process based on singular value decomposed(SVD) which can be used for periodic and aperiodic motion. Wavelet transform according to the PPG signal can be decomposed into different frequency wavelet coefficients. Some wavelet coefficients contain pulse signal component and they can be chosed to do empirical mode decomposition(EMD). Adaptive filter(AF) is a common method for processing non-stationary signals. AF utilizes continuous iteration to make the error between desired signal and input signal minimize, obtains the target signal which closed to the desired signal. AF’s performance influenced by the convergence rate and steady-state error. Variable step size least mean square algorithm can effectively improve the computational efficiency. AF usually uses the acceleration signal as the reference signal. Independent component correlation(ICA) is an algorithm for bilnd source separation, which can isolate independent components from multi-channel observation signals. The PPG signals recorded from dual-channel can be separated to pulse signal and motion artifact signal. Singular value decomposition(SVD) expands the PPG signal to a matrix, and decomposes it to a singular value matrix. Accroding to the characteristic of sigular value matrix, the singular values in matrix sorts in descending order. We can select the frequency of component close to the pulse frequency, reconstructing them into pulse wave signal.After de-noising the real PPG signal sampled in motion by the above method, this paper compares their abilities of de-noising in frequency domain and proves the SVD method has the highest accuracy.In calculating the heart rate, we use the method calculating heart rate in frequency domain called spectral peak tracking(SPT). SPT adopt the spectral peak’s characteristic of heart rate in spectrum to calculate heart rate, adjusting through last heart rate continuously. Conbined with dual-channel PPG, we can use a new SPT method called DSPT. This approach inproves the algorithm’s fault tolerance and accuracy. This paper propose a new method called dual spectral peak tracking based on bayesian decision theory(DSPT-B) to improve DSPT method, and combining the asymmetric smoothing method to predict some inaccurate heart rate point. The results prove that DSPT-B has a higher accuracy and lower computational complexity.In order to verify the accuracy of all algorithm used before, we designed a experimental procedure which contains periodic and aperiodic motions. Then, we measured 12 subjects’PPG signals under these motions. After analysising the average absolute error, average absolute error percentage, correlation coefficient and Bland-Altman plot we can draw the conclusion that the SVD reconstruction and DSPT-B method has the best ability of calculating heart rate in motion.