International Conference on Communication and Signal Processing, April 6-8, 2017, India

A Frame Work to Estimate Heart Rate and Arterial

Oxygen Saturation (Spo2)
P.Madhan Mohan ,V Nagarajan and A Annie Nisha

Abstract—The pulse oximeter is one of the vital device to

determine the heart rate (HR) and arterial oxygen saturation II. PULSE OXIMETER
(SpO2). The vital parameters are measured based on the signal The pulse oximeter is the device which is used to measure
from the device called photoplethysmogram (PPG). This signal
the heart rate and arterial oxygen saturation. The measurement
gets easily distorted by the movement of the hands which in turn
makes the values unreliable. This paper is a general framework
is based on the photoplethysmogram signal acquired. The
to determine the heart rate and arterial oxygen saturation in pulse oximeter consists of two LEDs and a photodiode [6, 7].
different conditions using motion artifact removing technique. The wavelength of the LED is 660 nm and 940 nm. There are
This paper discusses about the removal of the motion artifact two types of pulse oximeters (as shown in Fig. 1). They are
from the PPG signal. And also the measurement of the HR and described below.
SpO2 using different methods.
A. Reflective pulse oximeter
Index Terms—Heart Rate (HR); Arterial Oxygen Saturation The reflective pulse oximeter has the LED and the
(SpO2); Motion artifact; Pulse oximeter; Photoplethysmogram photodiode adjacent to their sides. The light emitted from the
(PPG) LED is reflected back into the photodiode. Based on the
I. INTRODUCTION intensity level changes the estimation of the parameters is done
The pulse oximeter is a portable device which is used to [8].
determine various vital parameters of the human body. Heart
Rate (HR) and arterial oxygen saturation (SpO2) are one of the B. Transmittance pulse oximeter
parameters to monitor the state of the person [1]-[2]. HR The transmittance pulse oximeter has the LED and the
monitoring from the PPG signal is easier because of the less photodiode on the opposite side. The estimation of the
complexity in the hardware setup. parameters are now based on the light from the LED to the
Heart rate corresponds to the cyclic activity of the heart. It photodiode [9].
is defined as the speed of the heart measured by the number of
contractions as beats per minute (bpm). The heart rate can vary
based on the physical needs. The activities that increases the
heart rate are physical exercise, sleep, anxiety, stress, illness
and injection of drugs [3-5]. The normal range of heart rate is
around 60 to 100 bpm at rest. Any change in the heart rate
beyond the normal range will lead to cardiac arrhythmia.
Oxygen saturation is percentage of oxygen in the blood. It
is defined as fraction of oxygenated hemoglobin to the total
hemoglobin present in the blood. The human body maintains a Fig, 1. a) Transmittance pulse oximeter b) Reflectance pulse oximeter
very precise level of oxygen in the blood stream. The normal
range is between 95 and 100.In section II explains the pulse III. MOTION ARTIFACT REMOVAL METHODS
oximeter and section IV describes the algorithm and section V The motion artifact is the problematic noise which occurs
discuss about the result and conclusion of the paper in section due to the displacement of the sensor probe. The displacement
VI. may be due to many reasons like movement of the finger,
running etc..,. The main effect of this motion artifact is that the
shape of the signal gets affected. This in turn makes the
parameter values unreliable. The frequency range of the PPG
P.Madhan Mohan is with the Satyabama University, Chennai, India (e- signal is 0.5 to 4 Hz and the frequency of the motion artifact
mail: pmadhan.edu@gmail.com) signal is 0.1 Hz. This will cause an overlap in the frequency,
V. Nagarajan was with Adhiparasakthi College of Engineering, which will make the preprocessing stages impractical.
Melmaruvathur, India (e-mail:nagarajanece31@gmail.com) The motion contaminated data is processed in different
A. Annie Nisha is with the Jasmin Infotech, Chennai, India (e-mail: ways to remove the motion artifact noise. The independent
annienisha.arumairaj@jasmin-infotech.com)
component analysis (ICA) is used to remove the noise from the
signal. ICA based algorithm was implemented to determine
arterial oxygen saturation [9, 10].

978-1-5090-3800-8/17/$31.00 ©2017 IEEE

1645
 A two stage motion artifact removing method was proposed IV. ALGORITHM
[11-13], in which stage 1 is for the motion detection unit which The flow chart of the algorithm (as shown in Fig. 3) is as
detects the clean frame of data and produces the frequency of follows. The signal from the pulse oximeter is passed through a
the noise signal. The ICA is used to reconstruct the signal [14, 5s window. Then the signal is passed on to a preprocessing
15]. stage where the signal is filtered to remove the external noise
There are many ways to get rid of this motion artifact and dc signal from the PPG.
signal. One of the best method is by using the adaptive noise Then the signal moves to the motion artifact removal stage.
cancellers. In this paper we have applied the adaptive noise Here the noise due to the movement of the patient is removed
cancelling techniques to remove this noise. using the adaptive filters.
Then the heart rate and spo2 estimation stage, where the
A. Adaptive Noise Canceller (ANC) parameter values are estimated using different algorithms
It is the most commonly used method to remove the motion which is explained in the later section.
artifact noise from the signal. It requires an input and a
reference signal. The input signal is from the pulse oximeter
and the reference is from the accelerometer. PPG signals
The reference signal is filtered and subtracted adaptively
from the pulse oximeter signal (as shown in Fig. 2.). The main
advantage of the adaptive filter is that it has the capability to Windowing
change its impulse response adaptively to cancel out the noise
signal. It has the ability to track over the signals under non-
stationary conditions. Band pass filtering

Desired Error Source

signal
signal d(n) e(n) signal Base line removal

Reference Yes
Filter Motion
signal x(n) output
detection
Adaptive filter
No
Motion artifact
removal algorithm

HR estimation

Fig. 2. Adaptive Noise Canceller (ANC)

The main assumption in the adaptive filter is that the signal

and the noise signal are totally uncorrelated. The filter tap HR
Selection SpO2 estimation
weight vector is calculated by the Normalized Least Mean
Square (NLMS) algorithm. During the entire adaptive filtering
process the weight vector is going to be changed.
The advantage of the NLMS algorithm is that they are less HR
SpO2 display
complex and they have the immunity to the fluctuations in the display
signal. In the LMS algorithm the step size of the signal will
vary, but in the in the NLMS algorithm the step size of the Fig. 3. Flow chart of the algorithm
signal is normalized in each step.
The following are the three steps which is going to be A. Preprocesing stage
involved in the adaptive filter In this stage the signal is passed on the initial processing.
The different processing steps are
x The reference signal is filtered
a) Windowing: The signal is made to pass through a 5s
x Reference signal is subtracted from the input signal
window in order to make the processing easier.
x Filter coefficients are adjusted based on the error b) Filtering : External noise frequencies from the signal
signal i.e. the difference between the desired and the reference is removed using the band pass filter.
signal.
c) Base line removal: In this step the DC noise from the
signal is removed.

1646
B. Motion Artifact removal In the frequency domain estimation of the HR. the signal is
As mentioned in the previous section the adaptive filter is down sampled to reduce the number of samples. To the down
applied to remove the motion artifact noise signal from the sampled signal FFT is applied to determine the frequency
pulse oximeter signal. The preprocessed signal is now passed spectrum of the signal. The signal is then normalized to bring
to detect if there is any noise due to motion. If yes then the the amplitude to a target level. And HR is estimated from the
signal is passed to this stage. normalized signal using eq. (3), (4), (5).
In this stage the filter tap weight vector is calculated using D. SpO2 estimation:
the following formula In order to estimate the arterial oxygen saturation the
w(n+1) = w(n) + ((­/|x(n)|^2) x(n)e(n)) (1) normalized ratio of ratios (R) has to be determined. This is
also determined in two ways.
This filter coefficient is varied adaptively in order to match
the match the reference signal. 1) Time domain R estimation
The error signal is calculated using the below formula The ratio of AC and DC of the pulsatile wave component
of the signal can be found. The AC component corresponds to
e(n)=d(n)-wT (n)x(n) (2) the variation in cardiac frequency and DC component
Where d(n) is the desired signal and w(n) is the weight corresponds to the average of the overall light transmitted.
vector. Similar way the ratio of peak and onsets of red infrared
signal can be used to determine R.
Heart rate can be directly determined after the removal of
the noise due to motion artifact. The corresponding ways are 2) Frequency domain R estimation
mentioned below. It is noted that in order to find the arterial In this method the R value is determined by taking the
oxygen saturation it is necessary that the signal has to be a frequency spectrum of the motion artifact free signal. The
clean signal. magnitude values at zero frequency (DC) and cardiac
frequency (AC)
C. Heart rate (HR) estimation
After removing the noise due to motion artifact the heart 3) Spo2 estimation
rate is estimated. Heart estimation can be done in two ways The SpO2 is given by the following formulae
1) Time domain HR estimation: SpO2 = A + B*R + c*R2 (6)
In this method of HR estimation, the signal is preprocessed
to remove the spikes and to smoothen the signal. The Where A, B, C are given by the LED calibration.
corrupted noise signals which cannot be removed by the
analog front end is removed in this step. Then the peaks inside V. RESULTS
the signal is determined by taking differentiation of the signal. The HR and SpO2 obtained from the algorithm is compared
Peaks and onsets is determined by the positive and negative with the reference device output. The output has been shown
zero crossing of the signal respectively. in the following table. The total number of files given as input
Once the peak and onsets in the signal are determined then is 300.
the HR is estimated by the following formulae TABLE I
HRpeak is calculated by RESULTS OF HEART RATE
S.No. Method Files passed Pass percentage
HRpeak(i)=60/((Pint )*10(-3) ) (3) 1 Time domain 297 99%
2 Frequency domain 299 99.7%
HRonset is calculated by
3 Tracking 296 98.7%
(-3)
HRonset(i)=60/((Oint )*10 ) (4)
TABLE II
Where Pint is the peak interval and Oint is the pulse . RESULTS OF OXYGEN SATURATION
interval. S.No. Method Files passed Pass percentage
HRval is given by 1 Time domain 295 98.33%
2 Frequency domain 293 97.7%
ǔHR)_final=1/L (ěHRval (i))) (5) 3 Tracking 290 96.7%

Final HR value can be arrived by taking the average of the

HRval.. From the table it can be seen that the algorithm is giving a
reliable results. Pass percentage of the algorithm is given
2) Frequency domain HR estimation: based on the comparing of the results with a reliable reference
device results.

1647
 VI. CONCLUSION Artifact Reduction in PPG Signals Based on AS-LMS Adaptive Filter”,
IEEE Transactions On Instrumentation And Measurement, Vol. 61, No.
Since the heart rate and oxygen saturation are vital 5, May 2012.
parameters to monitor critical patients. Also its application in [8] Rasoul Yousefi, and Mehrdad Nourani, “Separating Arterial and
diagnosis is growing day by day. Therefore a reliable Venous-Related Componentsof Photoplethysmographic Signals for
Accurate Extraction of Oxygen Saturation and Respiratory Rate”, IEEE
algorithm has been developed to determine the heart rate and Journal Of Biomedical And Health Informatics, Vol. 19, No. 3, May
oxygen saturation. The performance is compared with the 2015.
other methods and results are more reliable. Hence this can be [9] M. Raghu Ram, K. Venu Madhav, Ette Hari Krishna, Nagarjuna Reddy
implemented in real-time applications. Komalla, Kosaraju Sivani, and K. Ashoka Reddy, “ICA-Based
Improved DTCWT Technique for MA Reduction in PPG Signals With
Restored Respiratory Information”, IEEE Transactions On
Instrumentation And Measurement, Vol. 62, No. 10, October 2013.
REFERENCES
[10] Thomas Jensen, Sune Duun, Jan Larsen, Rasmus G. Haahr, Mette H.
[1] Mahdi Boloursaz, Ehsan Asadi, Mohsen Eskandari, Shahrzad Kiani, and Toft, Bo Belhage, and Erik V. Thomsen, “Independent Component
Farokh Marvasti, “Heart Rate Tracking using Wrist-Type Analysis Applied to Pulse Oximetry in the Estimation of the Arterial
Photoplethysmographic (PPG) Signals during Physical Exercise with Oxygen Saturation (SpO2)- a Comparative Study”, 31st Annual
Simultaneous Accelerometry”, Signal Processing Letters, IEEE 23.2 International Conference of the IEEE EMBS Minneapolis, Minnesota,
(2016): 227-231. USA, September 2-6, 2009.
[2] Zhilin Zhang , “Photoplethysmography-Based Heart Rate Monitoring in [11] R. Krishnan, B. Natarajan, and S. Warren, “Two-Stage Approach for
Physical Activities via Joint Sparse Spectrum Reconstruction”, IEEE Detection and Reduction of Motion Artifacts in Photoplethysmographic
Transactions On Biomedical Engineering, Vol. 62, No. 8, Pp. 1902- Data”, IEEE Transactions On Biomedical Engineering, Vol. 57, No. 8,
1910, August 2015. August 2010.
[3] Andrei V. Gribok,1,2 Xiaoxiao Chen,1 and Jaques Reifman, “A Robust [12] R. Krishnan, B. Natarajan, and S. Warren, “Analysis and detection of
Method to Estimate Instantaneous Heart Rate from Noisy motion artifacts in photoplethysmographic data using higher order
Electrocardiogram Waveforms”, Annals of Biomedical Engineering, statistics,” in Proc. IEEE International Conference on Acoustic, Speech,
Vol. 39, No. 2, February 2011 (©2010) pp. 824–834. Signal Processing. (ICASSP 2008),LasVegas,Nevada, April 2008, pp.
[4] Zhilin Zhang, Zhouyue Pi and Benyuan Liu, “TROIKA: A General 613–616.
Framework for Heart Rate Monitoring Using Wrist-Type [13] R. Krishnan, B. Natarajan, and S. Warren, “Motion artifact reduction in
Photoplethysmographic (PPG) Signals During Intensive Physical photoplethysmography using magnitude-based frequency domain
Exercise”, IEEE Transactions on Biomedical Engineering, page independent component analysis,” in Proc. 17th International
no.:0018-9294 2013 IEEE. Conference Computer Communication Network, St. Thomas, Virgin
[5] Hamed Monkaresi, Rafael A. Calvo, and Hong Yan, “A Machine Islands, August 2008, pp. 1–5.
Learning Approach to Improve Contactless Heart Rate Monitoring [14] Z. Chair and P. Varshney, “Optimal data fusion in multiple sensor
Using a Webcam”, IEEE Journal of Biomedical and Health Informatics, detection systems,” IEEE Transaction on Aerospace Electronics System,
February 2013. AES, vol. 1, no. 1, pp. 98–101, Jan. 1986.
[6] Rasoul Yousefi, Mehrdad Nourani, Sarah Ostadabbas and Issa Panahi, “ [15] C. C. Lee and J. J. Chao, “Optimum local decision space partitioning for
A Motion-Tolerant Adaptive Algorithm for Wearable distributed detection,” IEEE Transaction on Aerospace Electronics
Photoplethysmographic Biosensors”, IEEE Journal Of Biomedical And System, AES,vol. 25, no. 7, pp. 536–544, Jul. 1989.
Health Informatics, Vol. 18, No. 2, March 2014.
[7] M. Raghu Ram, K. Venu Madhav, E. Hari Krishna, Nagarjuna Reddy
Komalla, and K. Ashoka Reddy, “A Novel Approach for Motion

1648