J. Biomedical Science and Engineering, 2009, 2, 471-479
doi: 10.4236/jbise.2009.27068 Published Online November 2009 (http://www.SciRP.org/journal/jbise/ JBiSE
Published Online November 2009 in SciRes. http://www.scirp.org/journal/jbise
Nadi Yantra: a robust system design to capture the signals
from the radial artery for assessment of the autonomic
nervous system non-invasively*
Abhinav1, Meghna Sareen1, Mahendra Kumar1, Jayashree Santhosh2,
Ashok Salhan3, Sneh Anand1
1Centre for Biomedical Engineering, Indian Institute of Technology, Delhi, India; 2Computer Services Centre, Indian Institute of
Technology, Delhi, India; 3Defence Institute of Physiology & Allied Science, Defence Research and Development Organisation,
Delhi, India.
Email: 1abhinav_cbme_iitd@hotmail.com; 2meghnasareen@gmail.com
Received 15 June 2009; revised 30 June 2009; accepted 1 July 2009.
Ayurvedic and other alternative medical practi-
tioners throughout the world have been using pulse
diagnosis to detect disease and the organ at distress
by feeling the palpations at three close yet precise
positions of the radial artery. This paper presents a
robust electro-mechanical system, ‘Nadi Yantra’
which uses piezoelectric based pressure sensors to
capture the signals from the radial artery. Mor-
phology of the waveforms obtained from our system
concurs with standard physiological arterial signals.
Reproducibility and stability of the system has been
verified. Signal processing techniques were applied
to obtain features such as amplitude, power spectral
density, bandpower and spectral centroid to reflect
variations in signals from the three channels. Fur-
ther, wavelet based techniques were used to process
the pressure signals and percussion peaks were
identified. The interval between the percussion
peaks was used to calculate Heart Rate Varibility
(HRV), a useful tool for assessing the status of the
autonomic nervous system of the human body non-
invasively. Time domain indices were calculated
from direct measurement of peak-peak (PP) inter-
vals and from differences between the PP intervals.
Frequency domain indices such as very low fre-
quency (VLF) power, low frequency (LF) power, high
frequency (HF) power, LF/HF ratio were also calcu-
lated. Thereafter, nonlinear Poincare analysis was
carried out. A map of consecutive PP intervals was
fitted to an ellipse using least squares method. Re-
sults from 7 datasets are depicted in this paper. A
novel pressure pulse recording instrument is deve-
loped for the objective assessment of the ancient sci-
ence of pulse diagnosis. The features calculated using
multi resolution wavelet analysis show potential in
the evaluation of the autonomic nervous system of the
human bo dy.
Keywords: Radial Artery; Pulse Diagnosis; Power
Spectral Density; Spectral Centroid; Multi Resolution
Wavelet; Autonomic Nervous System; Heart Rate Vari-
ability; Time Domain; Frequency Domain; Poincare
In ancient literatures of the Ayurveda, Chinese, Unani,
and Greek medicine, pulse based diagnosis has its own
unparalleled importance. The organ under distress is
zeroed down by feeling the palpations from the three
fingers (index, middle and ring) placed on the radial ar-
tery (Figure 1). These pulsations dictate the physio-
logical status of the entire human body [1]. This is a te-
dious and highly subjective process and takes years of
practice to master this art [2].
Pulse has been ubiquitously accepted by modern cli-
nicians as well. They examine the pulse using the
method of trisection i.e. apply pressure until the pulse is
maximal, and then vary pressure while concentrating on
the phases of the pulse. The arterial pulse variants (for
example pulsus alternans, bisferiens pulse, bigeminal
pulse) are used in detecting cardiac disorders. However,
alternative medicine practitioners carefully examine
pulses at different depths, each connected with a specific
part of the body and each believed to register even the
slightest physiological based change.
If there can be a device that can give an objective as-
sessment of the science of pulse diagnosis, it will assist
disease diagnosis noninv asively. It will be used by alter-
native medicine practiti oners as well as modern cli nicians.
*Nadi Yantra has been applied for patent (pending approval); Nadi
stands for Pulse and Yantra means Instrument.
472 A. Funda et al. / J. Biomedical Science and Engineering 2 (2009) 471-479
Analysis of the R-R time series has been commonly
used in electrocardiographic (ECG) signals. ECG signals
are electrical signals of the heart and pressure signals
from the radial artery are mechanical signals. However,
pressure signals also show significant complexes just
like the QRS of ECG waves. Therefore, a similar analy-
sis on percussion peak-peak (PP) time series can be car-
ried out to provide information regarding the status of
the autonomic nervous system noninvasively.
In this paper we discuss our device Nadi Yantra, the
three sensors of which simulate the human fingers. In
Section 2 the instrument has been explained in detail.
Section 3 describes the wavelet based signal processing
for calculation of heart rate variability and features for
evaluation of the autonomic nervous system noninva-
sively. Section 4 describes experiments conducted. In-
ferences are drawn in Section 5.
There is a need for the development of a quantitative
system for pulse diagnosis [3]. Investigations have been
attempted globally to develop a system that replicates
the human three-finger method of pulse based diagnosis
[4,5]. In previous attempts
1) Signals have been captured for a very small
span of time (1-2 minutes).
2) External pressure applied over the sensors var-
ies while recording.
3) Motion artifacts become a reasonable consid-
eration when the recordings are for a longer
duratio n o f time.
Advances over the earlier systems are that Nad i
Yantra allows recording for hours by an automated
external pressure on the three positions thereby
completely removing the potential for errors incurred
when pressure is applied (Figures 2, 3, 4). The
locking mechanism significantly resists the motion
artifacts as well.
Figur e 1 . A practitioner evaluating the patient’ s pulse.
Figure 2. Recording the Pulse using Nadi Yantra.
Figure 3. Signals (zoomed) as observed in the three
In the mechanical design, the system has three finger
like projections whose positions can be adjusted at the
tip region to find out the best locations to capture the
signal. Springs attached to them help in damping thus
simulating the natural damping present du e to muscles in
the tip region of the practitioner’s fingers (Figure 5).
Once the three best positions are found, they are
locked with another hard spring fitting. This lock resists
the motion artifacts as well. Discrete increments in
pressure are possible by changing the lock's position
towards the slant side. (Figure 6).
In the electrical design, we used three identical piezo
film based sensors to capture the waveform. The raw
signal was filtered, amplified, and transferred to the
computer using BioPac 150TM (Signal Acquisition Sys-
tem) operated at a sampling rate of 1000 samples per
It is found that the signal captured using Nadi Yantra
corresponds well with standard pulse from the radial
artery (Figure 7, Figure 8). A standard pulse from the
radial artery comprises of the following waves: [5]
a) Percussion Wave
b) Tidal Wave
c) Dicrotic Wave
d) Valley
Copyright © 2009 JBiSE
A. Funda et al. / J. Biomedical Science and Engineering 2 (2009) 471-479 473
Figure 4. Flexibilit y o f the system.
Figur e 5 . Simulation of fingers with the pressure sensors.
Figur e 6 . Lock mech anis m f or d isc rete press ure in creme nt.
3.1. Pressure Wave Decomposition Using
A radial pulse signal feature extraction system using
wavelet-based multi-resolution analysis was developed
and evaluated.
The choice of the wavelet function depends on how
closely the scaling function matches the shape of the
original signal [6,7]. Daubechies 9 (D9) of Daubechies is
similar in shape to the radial pulse signal complex and
was used in the decomposition.
Figure 9 shows the details of a signal captured by
Nadi Yantra. The original signal is shown at the top of
the plot. Below the signal, the details for seven wavelet
scales are shown which are scaled for better illustration.
Adding together all these details plus the signal ap-
proximation A7 returns the original signal.
Figure 7. A standard signal from the radial artery.
Figure 8. Zoomed signal as acquired by Nadi Yantra. (Note:
No digital filters have been used to capture these raw signals)
Figure 9. Wavelet decomposition of a typical radial pulse signal.
Copyright © 2009 JBiSE
474 A. Funda et al. / J. Biomedical Science and Engineering 2 (2009) 471-479
A few points can be not ed from the plot.
1) The high frequency noise content is captured at the
smallest scales, namely D1, D2 and D3.
2) Scales D4, D5 and D6 contain most of the informa-
tion that depicts the radial pulse signal.
3) Also, scales D7 (and above) correspond to very low
frequencies which do not contribute to the complex.
They account for the DC component of the signal and
baseline wander resulting from motion etc.
3.2. Preprocessing of Signal
Low frequencies (which constitute baseline wander)
appear at high scales (details D7 and above). Removal of
such details corrects baseline wander. High frequency
burst of noise is captured at the smallest scales (which
correspond to high frequencies), namely D1, D2 and D3.
Removal of such details corrects noise. Additionally, a
notch filter was applied to remove the 60 Hz power line
interference. The result is shown in Figure 10.
A frequency domain analysis of the denoised pressure
signal was carried out. A Fourier Transform of the pres-
sure signal from the radial artery is shown below. It
shows the significant frequencies present in the signal.
The signal contains low frequencies from 0-10 Hz (Fig-
ure 11).
3.3. Percussion Peak Detection
It was observed that the details D3, D4 and D5 are the
most significant and contain the information that re-
present the radial pulse signal complex, including per-
cussion wave. These details are devoid of high fre-
quency noise and low frequency baseline wander as well
as mean DC component. The signal was appropriately
squared to enhance percussion peaks which ensured bet-
ter detection. A thresholding technique was used to de-
tect the percussion peaks. Thereafter, the peak-peak (PP)
time series was obtained.
Figure 10. Clean denoised signal.
Figure 11. Fourier transform of denoised pressure signal.
Figure 12. Peak detection in pressure signals using wavelets.
3.4. Feature Extraction from Percussion
Peak-Peak (PP) Intervals
It is well known th at perturbation s to autonomic activ ity,
such as respiratory sinus arrhythmia and vasomotor os-
cillations cause corresponding fluctuations in heart rate.
The alterations of the heart beat known as heart rate
variability (HRV) is a useful tool for assessing the status
of the Autonomic Nervous System (ANS) non-inva-
sively [8,9]. Our approach for HRV calculation is based
on Percussion complex detection. HRV signal is com-
puted from the time difference between two consecutive
percussion complexes known as the PP time series. A
HRV signal is shown in Figure 13 below.
Changes in heart rate occur as a result of the autono-
mous nervous system’s actions through the parasympa-
thetic (P) and the sympathetic (S) pathways which have
opposite influences. The S stimulation leads to an in-
crease in heart rate and the P stimulation does the oppo-
site. These different actions result in fluctuations in the
Copyright © 2009 JBiSE
A. Funda et al. / J. Biomedical Science and Engineering 2 (2009) 471-479
SciRes Copyright © 2009
Figure 13. HRV signal.
heart rate, the best known being the sinus respiratory
arrhythmia modulated by th e P, and the ones related with
the baroreflex action modulated by the S.
Time domain indices were derived from direct meas-
urements of the PP intervals or from differences between
the PP intervals. The former include mean and standard
deviation (SDNN i.e. standard deviation of the normal to
normal interval i.e. square root of variance), SDANN
(standard deviation of the average NN interval calcu-
lated over short periods). The most commonly used
measures derived from interval differences include
RMSSD, the square root of the mean squared differences
of successive PP intervals, NN50, the number of interv al
differences of successive NN intervals greater than 50
milliseconds and pNN50, the proportion derived by di-
viding NN50 by the total number of NN intervals [10].
These provide information on the short-term and
long-term variability of the P-P time series.
Frequency domain indices provide information on
both total variability as well as distribution as a function
of frequency. The spectral analysis of this HRV signal
allows to quantitatively distinguish between the different
activities of the ANS such as very low frequency com-
ponent (VLF: 0-0.04 Hz), a low frequency component
(LF: 0.04-0.15 Hz) and a high frequency component (HF:
0.15-0.4 Hz). HF power is supposedly a pure measure of
arasympathetic activity and represents momentary
respiratory influences on heart rate or respiratory sinus
arrhythmia, and LF power is reflective of sympathetic
modulation and parasympathetic tone. It derives from
short term regulation of blood pressure.
A total of 40 signals were analyzed and results from 7
datasets are shown below in Table 1.
There is some evidence for the involvement of
nonlinear phenomena in the genesis of HRV. It is con-
ceived that assessment of HRV with nonlinear measures
may supply information different from and additional to
that derived through linear measures. A non-linear Poin-
care analysis was carried out for a representative P-P
time series where a Poincare plot is a scatter-plot of the
current P-P interval against the preceding P-P interval.
The PPi is plotted versus PPi-1 and the plot shown below
is obtained. An ellipse was fitted to it using least squares
method and values of the centre of the ellipse, major and
minor axis, angle with the x axis and equation of the
ellipse obtained. Th e plot prov ides summary information
as well as detailed beat-to-beat information on the be-
havior of the heart. A distinct advantage of Poincare
plots is their ability to identify beat-to-beat cycles and
patterns in data that are difficult to identify with sp ectral
analysis [11]. The Poincare plot (Figure 14) allows the
HRV researcher to measure the variability of heart rate
from different points of view such as long-term variabil-
ity, overall variability, variability on basal heart rate,
variability on accelerated heart rate, variability on decel-
erated heart rate as well as the s ym patheti c-pa rasym pat hetic
balance .
Figure 14. Poincare analysis.
Table 1. Features extracted from P-P time series.
VLF (0-0.04 Hz) *105 3.89 3.45 3.82 3.67 4.09 4.09 4.03
LF (0.04-0.15 Hz) *104 1.66 1.38 2.02 1.55 1.80 1.81 1.91
HF (0.15-0.4 Hz) *103 5.47 3.59 6.90 4.71 5.71 5.53 6.93
LF/HF 3.03 3.84 2.93 3.30 3.15 3.26 2.76
Mean 70.4 61.2 68.6 65.8 73.6 73.6 72.7
SDNN 4.40 2.61 9.24 3.21 3.73 3.72 7.07
RMSSD 4.51 2.72 9.63 3.51 3.97 3.97 7.38
SDNN/RMSSD 0.97 0.96 0.96 0.91 0.94 0.94 0.96
NN50 69 84 73 79 72 72 72
pNN50 (%) 83.1 86.6 84.8 87.7 90 90 88.8
476 A. Funda et al. / J. Biomedical Science and Engineering 2 (2009) 471-479
Xc = 72.68, Yc = 72.65, A = 8.55, B = 4.54, Phi = 0.81
where X c and Yc are the x and y axis centre of th e ellipse,
A and B are the major and minor axis of the ellipse re-
spectively, Phi is the radian angle of the major axis with
respect to the x axis. The equation of the ellipse is
4.1. Description of Experiment
We chose 5 healthy volunteers to carry out a whole day
analysis. The dynamics of the physio logical signals were
examined for pre-lunch and post-lunch states.
The control used in our analysis was the 20 sets of
signals captured from the same subject over a period of
time prior to lunch. The post lunch signals were com-
pared with the control signals and the results are shown
below (Figures 15–17). Signals were recorded at 1245
hrs, 1315 hrs, 1355 hrs and 1705 hrs where lunch was
administered at 1420 hrs. Figure 16. Signals captured at 1315 hrs.
It can be observed that the amplitude of the 1st channel
rises steadily prior to lunch and then falls post lunch. It
can als o be seen that the amplitude of the signals from the
2nd and 3rd channels rises post lunch (Figures 15–20).
4.2. Signal Processing and Results
Power spectral density of the signals was calculated and
plotted in Figures 21–23.
It can be seen from the cumulative PSD plots that the
signals from the three channels have the same frequency
components in pre-lunch and post-lunch signals. The
frequency peaks are more in number in the signals be-
fore lunch as compared to that after lunch. The magnitude
Figure 17. Signals captured at 1355 hrs.
of power in the 1st channel increases as appetite in-
creases from 1315 to 1355 hrs and then falls post lunch
at 1755 hrs. Thus, the dynamic properties of the three
signals (magnitude of power and frequency information)
change before lunch and after lunch. Further, a 30 min-
ute post-lunch signal (1705 hrs) was divided into 6 seg-
ments of 5 minutes each and PSD for every segment was
calculated. It can be inferred that the magnitude of
power in the 2nd channel decreases and that of the 3rd
channel increases over time.
A post-lunch 30 minute signal was divided into 15
segments of 2 minutes each and bandpower calculated
Figure 15. Signals captured at 1245 hrs.
Copyright © 2009 JBiSE
A. Funda et al. / J. Biomedical Science and Engineering 2 (2009) 471-479 477
Figure 18. Signals captured at 1705 hrs.
Figure 19. Zoomed view of a portion of signals from Figur e 10.
Figure 21. PSD of the signal acquired before lunch at 1315 hrs.
Figure 22. PSD of the signal acquired before lunch at 1355 hrs.
Figure 23. PSD of the signal acquired after lunch at 1705 hrs.
Figure 24. PSD of a post-lunch signal divided into 6 segments.
Figure 20. Zoomed vi ew of a portion of s ignals fr om Figur e 12. Figure 25. PSD of a post-lunch signal divided into 6 segments.
Copyright © 2009 JBiSE
478 A. Funda et al. / J. Biomedical Science and Engineering 2 (2009) 471-479
Figure 26. PSD of a post-lunch signal divided into 6 segments.
Figure 27. PSD of a post-lunch signal divided into 6 segments.
Figure 28. PSD of a post-lunch signal divided into 6 segments.
for each segment. The calculated bandpower is the area
under the curve of a PSD in the bandlimits 0.5-10 Hz,
appropriate for arterial pulse signals. It can be seen from
Figure 30 that bandpower for the 2nd channel falls after
lunch whereas bandpower for the 3rd channel increases.
Bandpower for the 1st channel remains predominant.
Spectral centroid frequency commonly referred to as
median frequency was also calculated. It can be seen
from Figure 31 that the centroid frequency of the 2nd
channel is higher than the 1st and the 3rd channels.
Thus, it is seen that dynamics of the radial pulse sig-
nal change on application of stimulus for example ad-
ministration of lunch. These dynamics can be moni-
tored using the device developed.
There is enough evidence in ancient literature that there
is not a single disease in the human body which cannot
be diagnosed by examining the pulse. However, ancient
medical practitioners had to totally rely upon years of
clinical experience in order to come to any conclusive
diagnosis. Clinicians today have limited examination of
the pulse to its rate, rhythm and volume by virtue of
which they hardly co me to a concrete diagno sis. If there
could be a system by which the radial pulse could be
critically examined, it could be one of the most useful
tools in the field of non-invasive diagnosis of disease. In
this paper, such a system known as Nadi Yantra has been
developed. Further, wavelet based techniques were used
to decompose the pressure signal from the radial artery.
Multi-resolution wavelet analysis was used to detect the
percussion peaks and the P-P time series was obtained. A
HRV plot depicting function of the ANS was obtained
Figure 29. PSD of a post-lunch signal divided into 6 segments.
Figure 30. Variation in band power with segment.
Figure 31. Variation in spectral centroid with segment.
Copyright © 2009 JBiSE
A. Funda et al. / J. Biomedical Science and Engineering 2 (2009) 471-479
SciRes Copyright © 2009
from the P-P time series. Time domain and Spectral
analysis of the P-P series gave significant features.
Non-linear Poincare analysis was carried out to obtain a
relationship between consecutive P-P intervals. These
features hold significance in the study of short term and
long term variability of the PP time series. Wavelet
based method proved most effective because of its abil-
ity to filter out noise and low frequency components and
retain relevant detail. The wavelet based detector has
advantages of time-scale analysis as well as frequency
analysis. This approach to capture the peaks from the
radial artery and further, extraction of features from the
P-P series is a promising tool for studying the diagnostic
application of Nadi Yantra in the assessment of the
autonomic nervous system. The experiments described
in the paper show that Nadi Yantra has the potential
to objectively measure and display the physiological
changes occurring in the human body.
[2] Fuzzy Theory in Traditional Chinese Pulse Diagnosis,
Proceedings of 1993 International Joint Conference on
Neural Networks.
[3] Wang, H. and Cheng, Y., (2005) A quantitative system for
pulse diagnosis in traditional chinese medicine, IEEE
EMBS, 5676–5679.
[4] Xu, L. S., Wang, K. Q., and Zhang, D., (2002) Modern
researches on pulse waveform of TCPD, IEEE.
[5] Joshi, A., Kulkarni, A., Chandran, S., Jayaraman, V. K.,
and Kulkarni, B. D., Nadi tarangini: A pulse based diag-
nostic system, Proceedings of the 29th Annual Interna-
tional Conference of the IEEE EMBS.
[6] Grap, A., (1995) An introduction to wavelets, IEEE
Comp. Sc. and Engg., 2(2).
[7] Torrence, C. and Compo, G. P., (1998) A practical guide
to wavelet analysis, American Meteorological Society.
[8] Holger, G. and Adelmann, (1999) Design of a PC-based
system for time-domain and spectral analysis of heart
rate variability, Computers and Biomedical Research, 32,
[9] Task Force of the European Society of Cardiology and
The North American Society of Pacing and Electro-
physiology, (1996) Heart rate variability standards of
measurement, physiological interpretation and Clinical
use, European Heart Journal, 17, 354–381.
We would like to extend our thanks to Dr. Mitali Mukerji, Dr. Bhavana
Prasher, Dr. Shilpi Aggarwal and Dr. Tavpritesh Sethi from the Insti-
tute of Genomics and Integrative Biology, New Delhi and Mr. Nanda-
kumar Selvaraj from CBME–IIT Delhi for their support and encour-
agement. We would also like to thank Prakhar Prakash for his help
with wavelet decomposition.
[10] Malik, M., (1996) Heart rate variability–Standards of
measurement, physiological interpretation and clinical
use–Task Force–European Society of Cardiology and
The North American Society of Pacing and Electrophysi-
ologu, Eur. Heart J., 17(354), 81.
REFERENCES [11] Brennan, M., Palaniswami, M., and Kamen, P., (2002)
Poincare plot interpretation using a physiological model
of HRV based on a network of oscillators, Am. J. Physiol.
283: H1873-H1886.
[1] Yoon, Y., Lee, M., and Soh, K., (2000) Pulse type classi-
fication by varying contact pressure, IEEE Engineering
in Medicine and Biology Magazine, 19, 106–110.