International Conference on Computer and Automation Engineering

A Review of Measurement and Analysis of Heart Rate Variability

Dipali Bansal Munna Khan A. K. Salhan

Electronics & Communication Electrical Engineering Department Biomedical Instrumentation
Engineering Department Jamia Millia Islamia Department
Apeejay College of Engineering (A Central University) D.I.P.A.S, D.R.D.O.
Sohna Palwal Road, Sohna, India New Delhi, India Delhi, India
dipali.bansal@yahoo.co.in khanmunna@yahoo.com ashoksalhan@yahoo.com

Abstract—Electrocardiogram (ECG) is one of the most It also discusses various time and frequency domain
important physiological parameter that gives correct parameters used for analyzing HRV.
assessment of heart function. QRS complex is a prominent
waveform in an ECG that provides the basis for analyzing
heart rate variability (HRV). HRV refers to the beat-to-beat II. QRS DETECTION METHODOLOGY
alterations in heart rate. Commercial devices these days
Various data acquisition systems capable of interface to a
provide preset computerized measurement of HRV, thus
computer are now available that accurately acquire and
providing the cardiologist a simple tool for both research and
clinical learning. To obtain meaningful data from the ECG, a process physiological signals [8]. Application software such
noise free inter-beat interval (IBI) time series is required to be as MATLAB and LabVIEW allow proficient realization of
extracted. This is realized using standard peak detection ECG processing algorithms. For acquiring ECG signal, a
algorithms packed with data acquisition hardware and three lead (RA, RL, LA - Lead I) arrangement is sufficient as
software. The aim of this paper is to describe various QRS it offers the maximum QRS complex. Once data is acquired,
detection methods used to derive HRV. It also reviews various detection of QRS complexes is done to evaluate RR interval.
time and frequency domain HRV parameters. The significance A flow chart showing steps to acquire and process ECG
and meaning of these different measures of HRV are a signal for HRV analysis is shown in Fig.1.
potential area of research.

Keywords-Heart rate, Heart Rate Variability - HRV, ECG,

QRS complex, RR interval

I. INTRODUCTION
The electrocardiograms (ECG) give cardiac functional
details and helps in analyzing heart abnormalities.
Corresponding to every heart beat in the ECG signal, a
quasi-periodic sequence of P, QRS and T- wave can be
observed. The QRS complex in this sequence has the highest
amplitude and once detected helps in calculating the intervals
between two consecutive RR peaks. The variation in these
RR intervals is referred to as the Heart Rate Variability
(HRV). QRS detection therefore provides the fundamental
for almost all automated ECG analysis algorithms. Several
algorithms for recognition of QRS complexes are reported in
literatures based on Pattern recognition [1], Hilbert transform
[2], Wavelet transform [3], Neuro-fuzzy technique[4],
Filtering method [5], Derivative based algorithms [6] etc.

For a regular cardiac system, the heart rate may vary due
to age, cardiac disease, neuropathy, respiration, maximum
inhalation and cardiac load [7]. Heart Rate Variability has
become the universally approved term to define variations of
both instantaneous heart rate and RR intervals. Therefore
HRV may be used as a clinical and cost effective method to
assess a patient’s cardiac details. This paper reviews the
basics of Heart Rate Variability and standard measurement Fig.1. Flow chart for ECG acquisition, processing and
techniques to detect QRS complexes from ECG waveforms. subsequent HRV analysis

978-0-7695-3569-2/09 $25.00 © 2009 IEEE 243

DOI 10.1109/ICCAE.2009.70
 ECG signal is susceptible to artifacts due to power-line
interference, electrode motion, baseline wander, high
frequency contents and myoelectric interference. Most of the
frequencies in the QRS complex are around 20 Hz. Thus a
band pass filter in the 10 – 40 Hz range is used to remove
low frequency changes such as baseline wander and high
frequency changes such as movement artifact. Generally
MIT/BIH arrhythmia database is used to evaluate the QRS
detection algorithm. For testing this method MATLAB’s
Simulink can be used. Commonly used QRS detection
algorithms based on Template matching and Differentiation
methods are discussed in subsequent sections.

A. Template Matching Methods

Template Cross-correlation: Fig.2. Differentiation based signal processing steps for QRS
Cross correlation is defined as the degree of similarity detection
between two or more signal waveforms. A QRS detection
technique based on template cross-correlation method is
discussed by Dobbs et al [9]. This QRS detection technique, A real-time QRS detection algorithm was developed by
includes calculation of cross correlation (Rxy ) between the Pan and Tompkins [14]. To attenuate noise in this method,
incoming ECG data sequence and a template of QRS the signal is passed through a band pass filter made of
complex. Then the maximum value of Rxy is found to locate cascaded high-pass and low-pass filters. High slopes are
the QRS complex. The incoming signal should be aligned found using differentiation, which distinguish the QRS
with the template for correlation. The first way of aligning complexes from other ECG waves. Then a nonlinear
the template and the incoming signal is by using the fiducial transformation that involves squaring of the signal samples is
points on each signal. These fiducial points have to be done to make the entire data positive before integration. It
assigned to the signal by some external process. also highlights the higher frequencies in the signal obtained
from the differentiation process. These higher frequencies
denote the QRS complex. Then the squared waveform is
Another aligning process involves continuous correlation
passed through a moving window integrator and a decision is
between a segment of the incoming signal and the template.
taken based on threshold detection.
Whenever a new signal data point arrives, the oldest data
point in time is discarded from the segment. A correlation is
Daskalov et al. [15] applied these differentiation based
performed between this signal segment and the template
algorithms to signals having baseline drift. The results were
segment that has the same number of signal points. This
unsatisfactory due to the use of fixed detection thresholds.
technique does not require processing time to assign fiducial
Adaptive amplitude threshold method is a solution to this,
points to the signal. The template is like a window that
where the band-pass filtered waveform continuously updates
moves over the incoming signal one data point at a time [10].
the estimates of the peak signal level and the peak noise.
Template Subtraction:
This is a relatively simple QRS detection technique
different from computing cross-correlation. A segment of the III. HEART RATE VARIABILITY
incoming ECG signal that corresponds to the QRS waveform
is stored as the template. This template is then compared A. Definition
with the incoming ECG signal. Each point in the incoming
signal is subtracted from the corresponding point in the The interval between adjacent QRS complexes is termed
template and results in a value close to zero. Small absolute as the normal to normal (NN) or the R to R (RR) intervals.
values detected denote the location of QRS complex. The Heart rate variability (HRV) refers to the beat-to-beat
method is however prone to noise [10]. alterations in heart rate. Under resting conditions, the ECG
of healthy individuals exhibits periodic variation in R-R
intervals. The HRV measurements are captured non-
B. Differentiation based Methods invasively from the ECG signal. The results from this HRV
Differentiation is implemented in many QRS detection data are capable of portraying physiological condition of the
algorithms. Fig. 2 shows the signal processing steps of patient and are an important indicator of cardiac disease.
differentiation based algorithm. A derivative based algorithm Variability in heart rate is clinically linked to lethal
developed by Balda et al. [11] was modified by Ahlstrom arrhythmias, hypertension, coronary artery disease,
and Tompkins [12]. Friesen et al. [13] then applied the congestive heart failure, organ transplant, tachycardia,
algorithm to compare noise performance for these methods. neuropathy, and diabetes [7].

244
B. Analysis of HRV analysis to recover an evenly sampled signal from the
irregularly sampled event series. The HRV spectrum
Time Domain Analysis: contains the high frequency (0.18 to 0.4 Hz) component,
The easiest estimation of HRV is portrayed by the time which is due to respiration and the low frequency (0.04 to
domain measures. Originally, HRV was measured manually 0.15 Hz) component that appears due to both the vagus and
from the mean R-R interval in time domain and its standard cardiac sympathetic nerves [7]. Ratio of the low-to-high
deviation measured on short-term 5 minute ECG segment. frequency spectra is used as an index of parasympathetic-
With these methods either the heart rate or each QRS sympathetic balance. Frequency domain HRV variables are
complex or the RR intervals between successive normal detailed in Table II [7].
complexes are determined. Simple time–domain variables
include the mean RR interval, the mean heart rate, the In the frequency-domain analysis power spectral density
difference between maximum and minimum heart rate, etc. (PSD) of the RR series is calculated. Methods for calculating
Table I shows in detail the various time domain HRV the PSD may be divided into Fast Fourier transform (FFT)
parameters [7]. based and autoregressive (AR) model based methods. The
PSD is analyzed by calculating powers and peak frequencies
Recordings for a longer period of 24 hours allow for different frequency bands. For the FFT based spectrum
complex statistical time-domain analysis. These statistical analysis powers are calculated by integrating the entire
parameters may be derived from direct measurements of the spectrum. The spectrum in AR model methods can be
RR intervals or from the differences between RR intervals. divided into components and the band powers are obtained
The simplest variable to calculate is square root of variance as powers of these components [16].
i.e. the standard deviation of the NN interval (SDNN). Since
variance is mathematically equal to total power under the TABLE II. FREQUENCY DOMAIN HRV PARAMETERS
curve, SDNN covers complete variability in the ECG
recording. Parameter Description Frequency

Absolute Measures
TABLE I. TIME DOMAIN HRV PARAMETERS
Total Power Variance of all RR intervals Approximately < =
Parameter Description 0.4 Hz
ULF Power in ultra low frequency < = 0.003 Hz
NN 50 count No. of adjacent RR intervals differing by more than range
50 ms in entire ECG recording VLF Power in very low frequency 0.003 Hz – 0.04 Hz
pNN 50 NN 50 count divided by total number of all RR range
intervals LF Power in low frequency range 0.04 Hz – 0.15 Hz
MAX-MIN Difference between shortest and longest RR interval
HF Power in high frequency 0.15 Hz – 0.4 Hz
SDNN Standard deviation of all RR intervals range
Relative Measures
SDNN index Mean of the standard deviations of all RR intervals
for all 5 min segments in the entire recordings LF Norm Normalized low frequency LF / (Total power
SDANN Standard deviation of averages of RR intervals for power – VLF) X 100
all 5 min segments in the entire recordings HF Norm Normalized high frequency HF / (Total power
RMSSD Root mean square of the difference of successive power – VLF) X 100
RR intervals LF / HF Ratio of low and high --
SDSD Standard deviation of differences between adjacent frequency
RR intervals
HRV Index Total number of all RR intervals divided by
amplitude of all RR interval

IV. DISCUSSIONS
Frequency Domain Analysis:
Duration and magnitude of QRS waves are the essence of
In contrast to the time domain measures of HRV HRV. Therefore, precise RR intervals calculations are
mentioned above, recent developments in microprocessor necessary to accurately depict the physiological state. The
technology has enabled the calculation of frequency smaller the standard deviation in R-R intervals, the lower is
measures on the same ECG data. Frequency measures the HRV. To date, over 26 different types of arithmetic
involve the spectral analysis of HRV. The RR interval time manipulations of R-R intervals have been used in the
series is an irregularly time-sampled signal. This is not an literature to represent HRV [17].
issue in time domain, but in the frequency-domain it has to
be taken into account. If the spectrum estimate is calculated The accuracy of the R-wave occurrence time estimates is
from this irregularly time-sampled signal, additional often required to be about 1 ms and thus, the sampling
harmonic components appear in the spectrum. Therefore, the frequency of the ECG should be at least 500 Hz [7]. If the
RR interval signal is usually interpolated before the spectral sampling frequency of the ECG is less than 500 Hz, the

245
errors in R-wave occurrence times can cause critical REFERENCES
distortion to HRV analysis results, especially to spectrum [1] S. S. Mehta et al. Computer-aided interpretation of ECG for
estimates [18]. diagnostics. Int. Journal of System Science, 43-58, 1996.
[2] D. Benitez et al. The use of Hilbert transform in ECG signal analysis.
Several studies propose a link between negative emotions Comp. in Bio. and Med.31:399-406, 2001.
(such as anxiety and hostility) and reduced HRV [17]. In [3] S. C. Saxena et al. Feature extraction from ECG signals using wavelet
normal subjects, both sympathetic and parasympathetic tone transform for disease diagnostics. Int. Journal of System
fluctuates throughout the day [19]. It is also rate dependent Science.33:1073- 1085, 2002.
i.e. the HRV is more at lower heart rates [20]. HRV is [4] M. Engin. ECG beat classification using neuro-fuzzy network. Pat.
Rec. Letters. 25:1715-1722, 2004.
sensitive and responsive to acute stress. Mental load like
[5] S. A. Israel et al. ECG to identify individuals. Pat. Rec. 38:133-142,
making complex decisions have been shown to lower HRV. 2005.
Resting heart rate does not change significantly with
[6] N. M. Arzeno et al. Quantitative analysis of QRS detection
advancing age, but there is a decline in HRV and for an algorithms based on first derivative of the ECG. Proc. of 28th IEEE
individual, the HRV is highest during sleep. Regular EMBS Annual.
physical activity however has been shown to raise HRV. [7] Task force of the European society of cardiology and the North
American society of pacing and electrophysiology. Heart rate
On reducing the monitoring period, the frequency variability: standards of measurement, physiological interpretation,
domain HRV parameter SDNN takes shorter cycle lengths. and clinical use. Circulation,vol.93, no.5, 1043-1065, 1996.
The total variance of HRV however, increases with the [8] Bansal Dipali, Khan Munna and Salhan A.K. An ECG monitoring
system having simple interface with computer capable of real time
length of recording being analyzed [7]. Thus, on arbitrarily data transfer. International Conference, Los-Angeles, USA, P1.7,
selected ECG data, SDNN and similarly other frequency 2007.
domain parameters are not well defined statistical quantity [9] Dobbs S. E., Schmitt N. M., Ozemek H. S. QRS detection by
because of their dependence on the length of recording template matching using real-time correlation on a microcomputer.
period. However, short-term 5-min recordings and nominal Journal of Clinical Engineering, 9: 197–212, 1984.
24 h long-term recordings appear to be suitable options. [10] Valtino X. Afonso. ECG QRS Detection. Chapter 12, 236-264.
[11] Balda R. A., Diller, G., Deardorff, E., Doue, J., and Hsieh, P. The HP
Distinct spectral variations can be detected in human ECG analysis program. Trends in Computer-Processed
Electrocardiograms, 1977.
subjects in varying postures or body angle. HRV studies of
subjects on a tilt table can produce information about [12] Ahlstrom, M. L. and Tompkins W. J. Digital filters for real-time
ECG signal processing using microprocessors. IEEE Trans. Biomed.
sympathetic and parasympathetic regulation of cardiac cycle. Eng., BME-32: 708–13, 1985.
Short term HRV may describe increased or decreased [13] Friesen, G. M., Jannett, T. C., Jadallah, M. A., Yates, S. L., Quint, S.
autonomic adaptation to stimuli. Power spectrum analysis of R., Nagle, H. T. A comparison of the noise sensitivity of nine QRS
HRV may describe the extent of sympathetic and vagal detection algorithms. IEEE Trans. Biomed. Eng., BME-37: 85–97,
adaptation in patients. HRV data analysis provides additional 1990.
insight into cardio-pathologies, thereby assisting physicians [14] Pan J. and Tompkins W. J. A real-time QRS detection algorithm.
who desire speedy diagnosis provided by non-invasive IEEE Trans. Biomed. Eng. BME-32: 230–36, 1985.
means [7]. [15] I.K. Daskalov and I.I. Christov. Electrocardiogram signal
preprocessing for automatic detection of QRSboundaries. Med. Eng.
Phys., vol. 21, no. 1, pp. 37-44, 1999.
[16] Juha-Pekka Niskanen , Mika P. Tarvainen, Perttu O. Ranta-aho, Pasi
V. CONCLUSIONS A. Karjalainen. Software for advanced HRV analysis. Submitted to
Computer Methods and Programs in Biomedicine, 2002.
The analysis of HRV both by time-domain and spectral
approaches offer a non-invasive method of evaluating [17] John D. and Catherine T. Mac Arthur
Research Network on Socioeconomic Status and Health ; Heart Rate
cardiac functioning. The measurement of HRV is becoming Variability.
increasingly standardized. The major reason for the interest [18] M. Merri, D. Farden, J. Mottley, and E. Titlebaum. Sampling
in measuring HRV stems from its ability to predict survival frequency of the electrocardiogram for spectral analysis of the heart
after heart attack. Many studies have shown that reduced rate variability," IEEE Trans Biomed Eng, vol. 37, no. 1, pp. 99-106,
HRV predicts sudden death in patients. Although, the 1990.
assessment of HRV requires expertise, the equipment is not [19] Jeffrey J. Goldberger, “Sympathovagal balance: how should we
expensive, requiring only ECG equipment, microprocessors, measure it?” Am. J. Physiol. 276 (Heart Circ. Physiol. 4), 1999, pp.
H1273- H1280.
and relevant software for carrying out time and frequency
[20] Awdah Al-Hazimi, Nabil Al-Ama, Ahmad Syiamic, Reem Qosti, and
domain analyses. So far, experimental and simulation data Khidir Abdel-Galil, “Time domain analysis of heart rate variability in
appear to indicate that the various methods of expressing diabetic patients with and without autonomic neuropathy,” Annals of
HRV are largely equivalent, and there is no evidence that Saudi Medicine, 22 (5-6), 2002, pp. 400-402.
any one method is superior to another, provided
measurement windows are 5 minutes or longer.

246