Abstract—This letter considers the problem of casual heart rate Many signal processing algorithms proposed so far for MA
tracking during intensive physical exercise using simultaneous 2 reduction from PPG signals consider weak MA scenarios in
channel photoplethysmographic (PPG) and 3 dimensional (3D) which subjects perform small motions such as finger move-
acceleration signals recorded from wrist. This is a challenging
problem because the PPG signals recorded from wrist during ments [6], [7], [8] and walking [8]. Moreover, some prior works
exercise are contaminated by strong Motion Artifacts (MAs). In that consider the scenario of HR monitoring from PPG sig-
this work, a novel algorithm is proposed which consists of two nals in fitness consider low MA PPG signals recorded from
main steps of MA Cancellation and Spectral Analysis. The MA fingertip [8], [9] or ear [10]. Although HR monitoring from
cancellation step cleanses the MA-contaminated PPG signals wrist-type PPG signals during intensive physical exercise is ex-
utilizing the acceleration data and the spectral analysis step
estimates a higher resolution spectrum of the signal and selects
tremely challenging, but it is of great interest to wearable smart
the spectral peaks corresponding to HR. Experimental results on devices such as smart-watches [4].
datasets recorded from 12 subjects during fast running at the peak In order to overcome this challenge, we consider that simul-
speed of 15 km/hour showed that the proposed algorithm achieves taneous acceleration data is also recorded and available for pro-
an average absolute error of 1.25 beat per minute (BPM). These cessing. By using this extra information, we propose an effi-
experimental results also confirm that the proposed algorithm
cient algorithm that accurately tracks HR during high speed
keeps high estimation accuracies even in strong MA conditions.
(12-15 km/hour) running of the subject by simultaneous pro-
Index Terms—Adaptive motion artifact cancellation, causal cessing of the PPG signals taken from two different sensors and
heart rate monitoring, iterative method with adaptive thresh-
the acceleration data along the three axes. The proposed algo-
olding (IMAT), photoplethysmograph (PPG), singular value
decomposition (SVD), sparse spectrum estimation. rithm consists of two main steps of MA cancellation and spectral
analysis.
The MA cancellation step decomposes the acceleration
I. INTRODUCTION signals into periodic MA components using Singular Value
Decomposition (SVD). These MA components are then sup-
1070-9908 © 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
228 IEEE SIGNAL PROCESSING LETTERS, VOL. 23, NO. 2, FEBRUARY 2016
selves, it was observed that utilizing them directly as the MA range. Once the out-of-band MAs are cancelled, we apply adap-
reference (as proposed by [26]) would hamper convergence of tive filters to suppress in-band MAs. The reference MA signal
the applied adaptive filter and degrade the overall performance. components needed for the adaptive filter are extracted from the
Hence, another key innovation of this research is to extract simultaneous acceleration data by Singular Value Decomposi-
the reference MA signals by decomposing the 3D acceleration tion (SVD). These two substeps are further explained in the fol-
signals by SVD prior to adaptive filtering. The significance lowing:
of this decomposition substep is observed by simulations in Reference MA Generation Using SVD: As expected, the si-
Table III, Section III. This substep also provides an additional multaneous acceleration data along the three axes include foot-
design parameter (the number of reference MA signals ex- prints of the MAs. However, the acceleration data are convo-
tracted) to let us trade the estimation error with the simulation luted noisy signals composed of different periodic components
time. The proposed Peak Selection sub step is another key themselves. This will strongly hamper convergence of the ap-
innovative aspect of this research that efficiently tracks HR by plied adaptive filter. Hence, we propose to generate the refer-
simultaneous processing of the two channel PPG signals in ence MA signals by decomposing the three acceleration sig-
comparison with [4], [13] and [15] that consider single channel nals by SVD prior to adaptive filtering as depicted in Fig. 1.
PPG signals available. Finally, avoiding the high complexity In this technique, each acceleration signal goes through em-
Multiple Measurement Vector (MMV) model utilized by [14] bedding, SVD and grouping steps [4]. During embedding, a
and utilizing the Iterative Method with Adaptive Thresholding so called L-trajectory matrix is formed from each acceleration
(IMAT) that proved to maintain acceptable spectrum estimation signal. Subsequently, these L-trajectory matrices are decom-
performance while reducing the computational complexity in posed to linearly independent rank-one matrices. These ma-
comparison with FOCUSS [27] used by [4] is a key improve- trices are classified into groups with the same or harmoni-
ment achieved by this research.” cally related oscillatory components. Finally, groups with fre-
The rest of this letter is organized as follows. Section II ex- quency components inside [0.4-5] Hz band are utilized as the
plains the proposed method. Section III presents and discusses MA reference signal components for adaptive cancellation.
the simulation results achieved by the proposed method on the Adaptive MA Cancellation: During this procedure, all the
12 benchmark datasets. And finally Section IV concludes this reference MA components derived by SVD are removed from
letter. both PPG signals by successive application of adaptive filters as
For further reproduction of the reported results, MATLAB depicted in Fig. 2. Each adaptive filter stage receives the residual
codes of the proposed algorithm have been made available on- signal resulting from its prior stage and a reference signal com-
line at: ee.sharif.edu/~imat/ ponent as input. Finally, the resulting cleansed signal is input to
the spectral analysis step for HR tracking.
II. THE PROPOSED ALGORITHM
As mentioned earlier, the proposed algorithm consists of
two main steps of MA cancellation and spectral analysis. B. Spectral Analysis
Fig. 1 shows a block diagram of the proposed HR tracking
algorithm. The blocks used in the proposed algorithm are Spectral Analysis is a key step in the proposed HR tracking
explained in the following subsections. algorithm. As observed in Fig. 1, this step consists of two sub-
steps of High Resolution Spectrum Estimation using the Itera-
A. MA Cancellation tive Method with Adaptive Thresholding (IMAT) and Peak Se-
In this step, prior to further processing, we filter both PPG sig- lection. IMAT is a fast and efficient algorithm for sparse signal
nals in [0.4-5] Hz band to reject the MAs outside the natural HR reconstruction [11], [12] that proved to outperform some other
BOLOURSAZ MASHHADI et al.: HEART RATE TRACKING USING WRIST-TYPE PHOTOPLETHYSMOGRAPHIC (PPG) SIGNALS 229
(1) (4)
TABLE I
THE RESULTS ACHIEVED BY THE PROPOSED ALGORITHM ON 12 SUBJECTS’ RECORDINGS
TABLE IV
COMPARISON BETWEEN THE AAE AND ASTPF VALUES ACHIEVED FOR
SUBJECT 1 WHILE USING IMAT AND FOCUSS SPECTRUM ESTIMATION
ALGORITHMS
defined as by [4]. This results yield an average absolute error Finally, Table IV compares the AAE and ASTPF values
of 1.25 BPM. achieved for Subject 1 while using IMAT [11] and FOCUSS
Table II compares the performance of the proposed technique [27] for high resolution spectrum estimation in our proposed
with other state-of-the-art techniques that utilize the same algorithm. It can be concluded from this table that IMAT
datasets and performance measures. significantly reduces the simulation time while maintaining
Table III reports the AAE values achieved for different acceptable estimation accuracy.
number of MA reference signals ( ) utilized for adaptive MA
cancelation. Note that represents the case where 3D IV. CONCLUSION
acceleration signals are directly fed to the adaptive filters. It
is observed that direct utilization of the acceleration signals In this letter we proposed a novel algorithm for real-time heart
as MA reference will hamper the convergence of the adaptive rate estimation using wrist-type PPG signals when subjects are
filter and hence degrade the overall performance. This table performing intensive physical exercise. In order to deal with the
shows the significance of the proposed reference MA genera- strong MAs caused by subjects’ fast running, we recorded and
tion algorithm. Overally, we can conclude that is an utilized simultaneous acceleration data as the MA reference sig-
optimum choice regarding both performance and complexity. nals. The proposed algorithm consists of two key steps of MA
We also observe that the algorithm’s simulation time can be Cancellation and Spectral Analysis and shows high robustness
decreased by decreasing the design parameter and tolerating against strong motion artifacts caused by intensive physical ex-
higher AAE values. ercise.
BOLOURSAZ MASHHADI et al.: HEART RATE TRACKING USING WRIST-TYPE PHOTOPLETHYSMOGRAPHIC (PPG) SIGNALS 231
REFERENCES [16] S. Zhu, K. Tan, X. Zhang, Z. Liu, and B. Liu, “MICROST: A mixed
approach for heart rate monitoring during intensive physical exercise
[1] E. N. Bruce, Biomedical Signal Processing and Signal Modeling. using wrist-type PPG signals,” in 37th Annu. Int. Conf. IEEE Engi-
New York, NY, USA: Wiley, 2001. neering in Medicine and Biology Society (EMBS 2015), Aug. 2015.
[2] G. Hertzman, “The blood supply of various skin areas as estimated by [17] A. Temko, “Estimation of heart rate from photoplethysmography
the photoelectric plethysmograph,” Amer. J. Physiol., vol. 124, no. 2, during physical exercise using Wiener filtering and the phase vocoder,”
pp. 328–340, Oct. 1938. in 37th Annu. Int. Conf. IEEE Engineering in Medicine and Biology
[3] K. A. Reddy, B. George, and V. J. Kumar, “Use of fourier series Society (EMBC), Aug. 25–29, 2015, vol. , no. , pp. 1500–1503.
analysis for motion artifact reduction and data compression of photo- [18] A. K. Ahmadi, P. Moradi, M. Malihi, S. Karimi, and M. B. Shamsol-
plethysmographic signals,” IEEE Trans. Instrum. Meas., vol. 58, no. lahi, “Heart Rate monitoring during physical exercise using wrist-type
5, pp. 1706–1711, 2009. photoplethysmographic (PPG) signals,” in 37th Annu. Int. Conf. IEEE
[4] Z. Zhang, Z. Pi, and B. Liu, “TROIKA: A general framework for heart Engineering in Medicine and Biology Society (EMBC), Aug. 25–29,
rate monitoring using wrist-type photoplethysmographic signals during 2015, vol. , no. , pp. 6166–6169.
intensive physical exercise,” IEEE Trans. Biomed. Eng., vol. 62, no. 2, [19] N. K. Lakshminarasimha Murthy, P. C. Madhusudana, P. Suresha, V.
pp. 522–531, Feb. 2015. Periyasamy, and P. K. Ghosh, “Multiple spectral peak tracking for
[5] J. Yao and S. Warren, “A short study to assess the potential of heart rate monitoring from photoplethysmography signal during inten-
independent component analysis for motion artifact separation in sive physical exercise,” IEEE Signal Process. Lett., vol. 22, no. 12, pp.
wearable pulse oximeter signals,” in 27th Annu. Int. Conf. IEEE 2391–2395, Dec. 2015.
Engineering in Medicine and Biology Society (EMBS 2005), Sep. [20] E. Khan, F. Al Hossain, S. Uddin, S. Alam, and M. hasan, “A ro-
2005, pp. 3585–3588. bust heart rate monitoring scheme using photoplethysmographic sig-
[6] R. Krishnan, B. Natarajan, and S. Warren, “Two-stage approach for nals corrupted by intense motion artifacts,” IEEE Trans. Biomed. Eng.,
detection and reduction of motion artifacts in photoplethysmographic vol. , no. 99, Aug. 2015.
data,” IEEE Trans. Biomed. Eng., vol. 57, no. 8, pp. 1867–1876, 2010. [21] T. Schack, C. Sledz, M. Muma, and A. M. Zoubir, “A new method
[7] M. Ram, K. V. Madhav, E. H. Krishna, N. R. Komalla, and K. A. for heart rate monitoring during physical exercise using photoplethys-
Reddy, “A novel approach for motion artifact reduction in PPG sig- mographic signals,” in 23rd Eur. Signal Processing Conf. (EUSIPCO
nals based on AS-LMS adaptive filter,” IEEE Trans. Instrum. Meas., 2015), Sep. 2015, pp. 2716–2720.
vol. 61, no. 5, pp. 1445–1457, 2012. [22] P. Mullan, C. M. Kanzler, B. Lorch, L. Schroeder, L. Winkler, L. Laich,
[8] R. Yousefi, M. Nourani, S. Ostadabbas, and I. Panahi, “A motion-tol- F. Riedel, R. Richer, C. Luckner, H. Leutheuser, B. M. Eskofier, and C.
erant adaptive algorithm for wearable photoplethysmographic biosen- Pasluosta, “Unobtrusive heart rate estimation during physical exercise
sors,” IEEE J. Biomed. Health Informatics, vol. 18, no. 2, pp. 670–681, using photoplethysmographic and acceleration data,” in 37th Annu. Int.
2014. Conf. IEEE Engineering in Medicine and Biology Society, Aug. 2015,
[9] S. M. Lopez, R. Giannetti, M. L. Dotor, J. P. Silveira, D. Golmayo, pp. 6114–6117.
F. Miguel-Tobal, A. Bilbao, M. Galindo Canales, and P. Mart´in Es- [23] Z. Chengzhi and R. Jafari, “Robust heart rate estimation using wrist-
cudero et al., “Heuristic algorithm for photoplethysmographic heart based PPG signals in the presence of intense physical activities,” in
rate tracking during maximal exercise test,” J. Med. Biol. Eng., vol. 37th Annu. Int. Conf. IEEE Engineering in Medicine and Biology So-
32, no. 3, pp. 181–188, 2012. ciety (EMBC), Aug. 2015, pp. 8078–8082.
[10] M.-Z. Poh, N. C. Swenson, and R. W. Picard, “Motion-tolerant [24] K. Chan and Y. Zhang, “Adaptive reduction of motion artifact from
magnetic earring sensor and wireless earpiece for wearable photo- photoplethysmographic recordings using a variable step-size lms
plethysmography,” IEEE Trans. Inf. Technol. Biomed., vol. 14, no. 3, filter,” in Proc. IEEE Sensors, 2002, vol. 2, pp. 1343–1346.
pp. 786–794, 2010. [25] H. Han and J. Kim, “Artifacts in wearable photoplethysmographs
[11] F. Marvasti and A. Amini et al., “A unified approach to sparse signal during daily life motions and their reduction with least mean square
processing,” EURASIP J. Adv. Signal Process., 2012. based active noise cancellation method,” Comput. Biol. Med., vol. 42,
[12] [Online]. Available: http://ee.sharif.edu/~imat/ no. 4, pp. 387–393, 2012.
[13] Z. Zhang, “Heart rate monitoring from wrist-type photoplethysmo- [26] S. H. Kim, D. W. Ryoo, and C. Bae, “Adaptive noise cancellation
graphic (PPG) signals during intensive physical exercise,” in IEEE using accelerometers for the PPG signal from forehead,” in 29th Annu.
Global Conf. Signal and Information Processing (GlobalSIP), Dec. Int. Conf. IEEE Engineering in Medicine and Biology Society (EMBS
2014, pp. 698–702. 2007), Aug. 2007, pp. 2564–2567.
[14] Z. Zhang, “Photoplethysmography-based heart rate monitoring in [27] S. F. Cotter, B. D. Rao, K. Engan, and K. Kreutz-Delgado, “Sparse so-
physical activities via joint sparse spectrum reconstruction,” IEEE lutions to linear inverse problems with multiple measurement vectors,”
Trans. Biomed. Eng., vol. 62, no. 8, pp. 1902–1910, Aug. 2015. IEEE Trans. Signal Process., vol. 53, no. 7, pp. 2477–2488, 2005.
[15] B. Sun and Z. Zhang, “Photoplethysmography-based heart rate [28] F. Marvasti, M. Azghani, P. Imani, P. Pakrouh, S. J. Heydari, A. Gol-
monitoring using asymmetric least squares spectrum subtraction and mohammadi, A. Kazerouni, and M. M. Khalili, “Sparse signal pro-
bayesian decision theory,” IEEE Sensors J., no. 99, pp. 1–1, Aug. cessing using iterative method with adaptive thresholding (IMAT),” in
2015. 19th Int. Conf. Telecommunications (ICT’12), Apr. 2012, pp. 1–6.