IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 55, NO.
3, MARCH 2008
Heart Rate Detection From Plantar
Bioimpedance Measurements
Rafael González-Landaeta, Student Member, IEEE, Oscar Casas*, Member, IEEE, and
Ramon Pallàs-Areny, Fellow, IEEE
Abstract—In this paper, a novel technique for heart rate mea-
surement on a standing subject is proposed that relies on electrical
impedance variations detected by a plantar interface with booth
feet, such as those in some bathroom weighting scales for body
composition analysis. Heart-related impedance variations in the
legs come from arterial blood circulation and are below 500 m .
vasive bioimpedance measurements, for example, those applied
To detect them, we have implemented a system with a gain in
excess of 600, and whose fully differential AC input amplifier has
a gain of 4.5 and a common-mode rejection ratio (CMRR) higher
than 90 dB at 10 kHz. Differential coherent demodulation based
on synchronous sampling yields a signal-to-noise ratio (SNR) of
about 54 dB. The system sensitivity is 610 mV/ . The technique
has been demonstrated on 18 volunteers, whose bioimpedance
signal and ECG were simultaneously recorded. A Bland–Altman
plot shows a mean bias of 0.2 ms between the RR time intervals
obtained from these two signals, which is negligible. The technique
is simple and user friendly and does not require any additional
sensors or electrodes attached to the body, hence no conductive gel
or skin preparation.
Index Terms—Coherent demodulation, heart rate detection,
plantar bioimpedance, synchronous sampling.
I. INTRODUCTION
B IOIMPEDANCE measurements with various types of
electrodes and measuring systems have long been used
to study the physiology of the cardiovascular and respiratory
systems, tissue properties, body fluids compartments [1],
and for tissue and organ characterization by imaging [2]. In
cardiovascular studies, bioimpedance measurements allow us
to noninvasively obtain information about the stroke volume,
cardiac output [3], venous circulation [4], arterial compliance
[5], and heart rate. Usually, these measurements are performed
on the chest and/or limbs by placing surface electrodes, with
cumbersome leads, after skin preparation, which asks for a
Manuscript received January 10, 2007; revised July 9, 2007. This work was
supported in part by the Spanish Ministry of Education and Science under Con-
tract TEC2004-05520 and in part by the European Regional Development Fund.
R. González-Landaeta was supported in part by Francisco de Miranda Univer-
sity, Coro, Venezuela. Asterisk indicates corresponding author.
R. González-Landaeta is with the Instrumentation, Sensors and Interface
Group, Universitat Politècnica de Catalunya, Barcelona, Spain, on leave
from the Electromedicine Program, Francisco de Miranda University, Coro,
Venezuela (e-mail: rgonzalez@eel.upc.edu),
*O. Casas is with the Instrumentation, Sensors and Interface Group, Univer-
sitat Politècnica de Catalunya, Castelldefels, Barcelona, 08860, Spain (e-mail:
jocp@eel.upc.edu.
R. Pallàs-Areny is with the Instrumentation, Sensors and Interface
Group, Universitat Politècnica de Catalunya, Barcelona, Spain (e-mail:
ramon.pallas@upc.edu).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TBME.2007.906516
0018-9294/$25.00 © 2008 IEEE
1163
specialized training. As a result, heart rate detection based on
bioimpedance measurements is more involved than alternative
techniques such as photoplethysmography.
The drawbacks mentioned above are common to all nonin-
for body composition analysis [6]. To overcome them, several
systems have been developed with simpler body interfaces, such
as those in bathroom scales that offer bioimpedance analysis
(BIA) for body-fat composition measurement [7]. These sys-
tems inject a safe AC current into the patient through the soles
of both feet and measure basal impedance, which is related to
body composition. The time required for this measurement is
very short as it does not require any skin preparation, neither
any specialized skill to operate the device. However, the only
information obtained is about body composition.
Previous works show that in lower limbs there are low-level
heart-related impedance variations [8]. So far, these variations
are detected using cumbersome band-type electrodes attached
to the limb, to obtain hemodynamic information (and the heart
rate). In [9], we proposed a technique to detect impedance
variations in lower limbs from plantar bioimpedance mea-
surements. Here we describe how to measure heart-related
impedance changes when standing, by using four platform-type
aluminum electrodes, similar in composition to those used
in some commercial body-fat bathroom scales. Because of
the low amplitude of heart-related impedance variations, the
measurement system needs a high SNR. Hence, we have de-
signed a fully differential signal conditioner with high-gain
and high CMRR, and use synchronous sampling based on a
floating capacitor circuit as coherent demodulator. The result is
a system able to reliably detect the heart rate without attaching
any electrodes to the patient.
II. MEASUREMENT PRINCIPLE
All blood vessels are distensible. Arterial distensibility helps
in providing an adequate peripheral resistance to the arterial
blood flow [10], and this distensibility results in an arterial
volume change at each heart beat, more appreciable in those
body areas with less overlying tissue.
A model in [11] that relates impedance variations to changes
in blood volume and blood resistivity in limbs, assumed to be
cylindrical, leads to
(1)
where is the arterial volume change, is the length of the
arterial section between the voltage electrodes, is the blood
resistivity, and is the basal impedance of the nonpulsatile
1164 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 55, NO. 3, MARCH 2008

Fig. 1. Block diagram of the system for plantar bioimpedance measurement.

tissues. and are the impedance variations due to the

blood resistivity change and the impedance variations due to
the volume change, respectively. Because these two impedance
changes are heart-related, it should be possible to obtain the
heart rate from impedance measurements between two points
on the surface of a volume that encloses major blood vessels. If
an electrical current is injected between both feet soles, and the
drop in voltage is measured between the same feet, the current
path includes both legs, and these constitute part of that volume.
The current path will also include a part of the torso.

III. MATERIALS AND METHODS

Fig. 2. Fully differential AC amplifier [14].
A. Measurement System
We have measured impedance variations using the four-elec-
trode technique to minimize the effect of electrode contact
impedance (Fig. 1). We have designed a system supplied at
V, with high gain and novel signal conditioning tech-
niques. The aim was to obtain an impedance pulse signal with
a SNR high enough to allow us to later detect the heart rate by
simple signal processing methods.
1) Current Source: Bioimpedance measurements require the
injection of a known low-level current into the patient. We have
designed a single-ended current source based on a Wien bridge
and a current conveyor circuit, which generates a 10-kHz,
Fig. 3. Synchronous demodulator based on a floating capacitor circuit [15].
1-mA (rms) current, hence innocuous for external application
[12]. When a single-ended current source is used for measuring
impedance, the common mode voltage is the current times the
ground electrode impedance, and this voltage yields an offset corner frequencies were selected Hz and
error. For an error below 1%, we need dB [13]. kHz. Moreover, low noise amplifiers were used ( fA
Further, preliminary tests showed that the pulsatile impedance Hz, nV Hz).
variations when measuring between both feet were about The next stage in Fig. 1 is a floating capacitor circuit which
500 m , which means that, for a 1-mA injected current, the demodulates the signal by synchronously sampling it [15], as
expected voltage variations are about 0.5 mV. shown in Fig. 3. Because there are no connections to ground,
2) Differential Signal Processing: The first amplifier stage the capacitors charge only to the differential voltage and reject
in Fig. 1 is a fully differential AC-coupled amplifier [14]. Fig. 2 common mode voltages, which results in a very high CMRR. To
shows that this amplifier has no direct connection to ground, minimize signal attenuation, we used an analog switch with low
which results in a very high CMRR at the measurement fre- ON resistance (ADG436, ). In Fig. 3, the left-side
quency; there is no need to match any resistors or other passive switches are ON for a short sampling time in order to charge
components. Because the bias current path is through the patient the capacitor to – , then they are turned OFF for a hold
and the resistors of the input filter (1 M each), we use an op time . The right-side switches are first OFF during and
amp with low bias current (AD743, pA). then they are ON during .
The DC impedance component (basal impedance, ) is The equivalent dB bandwidth for a synchronous sampler
about one thousand times the alternating component [8]. There- will depend on the succeeding stage; in our case, that stage is
fore, the gain of the first stage must be small, otherwise its a zero-order hold (ZOH) circuit, whose (low-pass) frequency
output voltage would saturate. To preserve the SNR, the signal response is
to be demodulated must be a band pass signal. To reduce noise,
without attenuating the useful signal too much, the dB (2)
GONZÁLEZ-LANDAETA et al.: HEART RATE DETECTION FROM PLANTAR BIOIMPEDANCE MEASUREMENTS
Fig. 4. Electrodes used for plantar bioimpedance measurements in both feet:
A and B are the front electrodes and C and D are the back electrodes.
Whenever is an integer, ; on the other hand, this
frequency response has many windows open to noise so that it is
necessary to filter the ZOH output to reduce noise contributions
to the output signal [16]. We implemented a fully differential
passive first-order bandpass filter with a high CMRR and corner
frequencies 0.5 and 10 Hz, apart enough from each other to pass
heart-related impedance variations. The contribution from
was eliminated, and the power line (50 Hz) interference and
high-frequency noise were reduced.
3) Output Stage and Data Acquisition System: Once the
bioimpedance signal has been demodulated and the basal
impedance has been filtered out, it is necessary to amplify the
AC component obtained. Because of the low level of these
heart-related impedance variations, we need a large enough
gain but at the same time not so high as to lead to output voltage
saturation because of voltage offsets from the amplifiers. We
selected . A passive, second-order, lowpass filter
with corner frequency 10 Hz determined the system noise
bandwidth. The system output was connected to a data acqui-
sition system (USB 30/26, from EAGLE), which has
14-bit resolution, 1-kHz sampling frequency and unity gain,
interfaced to a PC via USB. Data were processed and stored by
an algorithm designed in LabView.
4) Electrodes: The electrodes were designed to provide a
comfortable and easy-to-use interface to the subject. We used
four aluminum foil squares (Fig. 4) large enough for easily
adapting to any subject feet size during tests. Electrodes A and
B (front) were used to measure the voltage variations resulting
from the current injected through electrodes C and D (back).
The contact area was large enough to provide a low contact
impedance without requiring any conductive gel or skin prepa-
ration. Later on, the measurement technique described was also
tested when using the electrodes available in two commercial
bathroom scales.
B. Signal-to-Noise Ratio Estimation
The SNR of the output signal slightly depended on the vol-
unteer. This was because noise in measurements performed by
surface electrodes depends on two factors: electrode-skin inter-
face and electronic noise from the circuitry. Electrode noise de-
pends on the electrolyte and the skin properties of the subject;
large-area electrodes contribute less noise [17]. These two fac-
tors, however, are difficult to quantify. Electronic noise can be
readily measured and will determine the ultimate SNR when
electrode noise is low enough.
Because the impedance signal of interest is pulsatile, hence
with a high crest factor, it is convenient to define the SNR as
dB lg
1165
where is the peak-to-peak voltage of the impedance signal
and is the root-mean-square (rms) noise voltage.
To estimate , we connected a 500 resistor across the input
of the AC-coupled amplifier. Then we injected a 10 kHz, 1 mA
(rms) current to obtain a voltage close to that due to the basal re-
sistance between both feet. To minimize capacitive interference,
the electronic circuitry was battery-supplied and shielded by a
metallic box connected to the circuit reference voltage; these
conditions were also implemented during heart rate measure-
ments. A 6 1/2 digit multimeter (Agilent model 34401A) con-
trolled via GPIB by LabView measured the output voltage; the
multimeter was configured to measure DC voltage at 0.2 power
line cycles (PLC) (250-Hz bandwidth) and programmed to take
300 measurements. The noise bandwidth was determined by the
output lowpass filter of the circuit (10 Hz). The standard devia-
tion of the readings equals the rms noise voltage.
C. Measurement Protocol
We have measured on eighteen subjects (3 women and 15
men) with different height and physical constitution (mean
SD: age years; weight kg; height
m). The volunteers stood on their bare feet
on the platform electrodes shown in Fig. 4, and their ECG (lead
II) was simultaneously recorded with a custom-built system. No
indications were given to them about how to put their feet on the
electrodes; they only were told not to move their feet during the
measurement.
To determine the origin of the main contribution to the
impedance signal, an inflatable cuff was placed above the knee
of some volunteers and inflated up to 60 mmHg to occlude the
venous return but not the arterial blood flow [18]. While the
cuff was inflated, we looked for any amplitude and waveform
variations in the output signal.
IV. RESULTS AND DISCUSSION
Before measuring on volunteers, we characterized each
stage of the measurement system. The ac amplifier gain (first
stage) was 4.5, low enough to prevent the system output from
saturating due to the basal impedance. At 10 kHz, its CMRR
was higher than 90 dB and the differential input impedance
was 1 M , so that common-mode errors were negligible. The
CMRR of the synchronous demodulator was 99 dB when
sampling at 10 kHz. The sampling time ( ) and the holding
time ( ) were adjusted to 10 s and 90 s, respectively, which
resulted in a gain of 0.7. The output instrumentation amplifier
had a gain of 271 and a CMRR higher than 89 dB in the signal
bandwidth. Passive filters in the signal path reduced the gain,
so that the system sensitivity was 610 mV/ . The output rms
noise voltage was 1 mV, which, for a 500 m change, implied
a SNR of about 54 dB.
Fig. 5 shows the ECG and the plantar bioimpedance signal
from one of the volunteers. The results for the others subjects
were similar, regardless of how they put their feet on the elec-
trodes. The baseline of the impedance signal was stable as long
as the subject did not move his/her feet. To show a signal sim-
ilar to those common in the bioimpedance literature, we present
ours upside down; that is, an upward deflection represents a de-
(3) crease in impedance. The impedance variations detected were
1166
Fig. 5. Sample (bottom trace) plantar bioimpedance signals and (upper trace) ECG.
Fig. 6. Bland–Altman plot of each RR time interval detected from the ECG and the Bioimpedance signal of two volunteers.
below 500 m . The same results were obtained when using the
electrodes of two commercial bathroom scales with body com-
position analysis function. The signal-to-noise ratio was always
large enough to detect the heart rate with a simple threshold al-
gorithm. In fact, a visual inspection of all recorded signals did
not show any single instance where the QRS was not followed
by the highest peak in the bioimpedance signal.
In order to provide an agreement figure, we used a
Bland–Altman plot [19] for each RR time interval detected
from the ECG and the from bioimpedance signal. The RR
time interval for the ECG was estimated using a QRS de-
tection algorithm; for the bioimpedance signal that interval
was estimated by using an adaptive-threshold algorithm. Both
procedures were developed in Matlab. Fig. 6 shows the best-
and the worst-case estimations from all the volunteers. The
mean bias of RR time intervals was ms and the 95%
confidence interval was about ms. This broad scattering
is due to the low accuracy of the adaptive threshold algorithm
used for detecting the peaks of the bioimpedance signal in one
volunteer who presented a distorted signal (worst case). The
other sixteen volunteers yielded results similar to the best case
IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 55, NO. 3, MARCH 2008
shown in Fig. 6. Therefore, the heart rate calculated from the
bioimpedance signal essentially agrees with that obtained from
the ECG.
Measurements when the cuff was inflated, hence venous
return occluded, did not show any noticeable change in ampli-
tude or waveform. Therefore, the major contribution to plantar
bioimpedance changes come from arterial circulation in the
lower limbs, as initially assumed.
V. CONCLUSION
A novel noninvasive technique for heart rate measurement
has been presented which uses platform-type electrodes and
plantar bioimpedance measurements. Heart-related impedance
variations due to arterial blood circulation can be detected by
measuring bioimpedance changes between both feet, without
any skin preparation, conductive gel, additional electrode, or
sensor attached to the body. Similar to commercial scales, it
is necessary to stand still during the measurement, otherwise
motion artifacts could appear. A fully-differential signal con-
ditioner has been implemented that achieves an overall CMRR
higher than 90 dB. Synchronous sampling demodulation yields
GONZÁLEZ-LANDAETA et al.: HEART RATE DETECTION FROM PLANTAR BIOIMPEDANCE MEASUREMENTS
a SNR of about 54 dB. This SNR level is good enough to detect
the heart rate by simple signal processing methods. The system
sensitivity is 610 mV/ and allows us to detect impedance
changes below 500 m . The technique can be applied by the
same electrodes of some commercial bathroom scales, making
it a low-cost and simple alternative for periodical heart rate
measurements.
ACKNOWLEDGMENT
The authors would like to thank the technical support of
F. López, as well as the volunteers for their help and patience.
REFERENCES
[1] M. E. Valentinuzzi, “Bioelectrical impedance techniques in medicine:
Bioimpedance measurement,” Crit. Rev. Biomed. Eng., vol. 24, pp.
223–255, 1996.
[2] B. Rigaud and J. P. Morucci, “Bioelectrical impedance techniques in
medicine: Impedance imaging,” Crit. Rev. Biomed. Eng., vol. 24, pp.
475–476, 1996.
[3] R. Lamberts, K. R. Visser, and W. G. Zijlstra, Impedance Cardiog-
raphy. Assen, The Netherlands: Van Gorcum, 1984, pp. 34–39.
[4] R. Patterson, “Bioelectric impedance measurements,” in The Biomed-
ical Engineering Handbook, J. D. Bronzino, Ed., 2nd ed. Boca Raton,
FL: CRC, 2000, vol. 1, pp. 73-4–73-5.
[5] F. Risacher, J. Jossinet, E. T. McAdams, J. McLaughlin, Y. Mann, M.
Schmitt, A. Matias, and R. Jarry, “Impedance plethysmography for the
evaluation of pulse-wave velocity in limbs,” Med. Biol. Eng. Comput.,
vol. 31, pp. 318–322, May 1993.
[6] R. Patterson, “Body fluid determinations using multiple impedance
measurements,” IEEE Eng. Med. Biol. Mag., vol. 8, no. 3, pp. 16–18,
Mar. 1989.
[7] K. Oguma, “Body weight scales equipped with body fat meter,” U.S.
patent 6370425, Apr. 29, 1999.
[8] B. H. Brown, R. H. Smallwood, D. C. Barber, L. P. V, and D. R. Hose,
Medical Physics and Biomedical Engineering. London, U.K.: IOP,
1999, pp. 613–615.
[9] R. Pallàs-Areny, O. Casas, and R. González-Landaeta, “Method and
apparatus to obtain the heart rate from electrical impedance variations
measured between the feet,” filed on Oct. 28, 2005.
[10] A. Guyton and J. Hall, Textbook of Medical Physiology, 10th ed.
Philadelphia, PA: Saunders, 2000, p. 152.
[11] T. M. R. Shankar, J. G. Webster, and S.-Y. Shao, “Contribution of
vessel volume change and blood resistivity change to the electrical
impedance pulse,” IEEE Trans. Biomed. Eng., vol. 32, no. 3, pp.
192–198, Mar. 1985.
[12] Medical electrical equipment. Part 1: General requirements for safety
and essential performance, 2000, IEC 60601-1.
[13] R. Pallàs-Areny and J. G. Webster, “AC instrumentation amplifier for
bioimpedance measurements,” IEEE Trans. Biomed. Eng., vol. 40, no.
8, pp. 830–833, Aug. 1993.
[14] E. M. Spinelli, R. Pallàs-Areny, and M. A. Mayosky, “AC-coupled
front-end for biopotential measurements,” IEEE Trans. Biomed. Eng.,
vol. 50, no. 3, pp. 391–395, Mar. 2003.
[15] M. Gasulla-Fomer, J. Jordana-Barnils, R. Pallàs-Areny, and J. M. Tor-
rents, “The floating capacitor as a differential building block,” IEEE
Trans. Instrum. Meas., vol. 47, no. 2, pp. 26–29, Feb. 1998.
[16] R. Pallàs-Areny and J. G. Webster, “Bioelectric impedance measure-
ments using synchronous sampling,” IEEE Trans. Biomed. Eng., vol.
40, no. 8, pp. 824–829, Aug. 1993.
1167
[17] E. Huigen, A. Peper, and C. A. Grimbergen, “Investigation into the
origin of the noise of surface electrodes,” Med. Biol. Eng. Comput.,
vol. 40, pp. 332–338, Mar. 2002.
[18] J. G. Webster, “Measurement of blood flow and volume of blood,” in
Medical Instrumentation. Application and Design, J. G. Webster, Ed.,
3rd ed. New York: Wiley, 1998, pp. 357–359.
[19] J. M. Bland and D. G. Altman, “Statistical methods for assessing agree-
ment between two methods of clinical measurement,” Lancet, vol. 327,
pp. 307–310, Feb. 1986.
Rafael González-Landaeta (S’07) was born in
Maracaibo, Venezuela, in 1975. He received the
Engineer in Electronics degree from the Rafael
Belloso Chacín University, Venezuela, in 1997. He is
currently pursuing the Ph.D. degree at the Electronic
Department, Universitat Politècnica de Catalunya,
Barcelona, Spain.
He has been with the Francisco de Miranda Univer-
sity, Coro, Venezuela, since 1999, teaching courses
in analog electronic and biomedical sensors. His cur-
rent research area includes analog signal processing,
noise and interference in electronic circuits, and noninvasive measurement of
physiological parameters.
Oscar Casas (S’93–M’99) received the Ingeniero de
Telecomunicación and Doctor Ingeniero de Teleco-
municación degrees from the Universitat Politècnica
de Catalunya (UPC), Barcelona, Spain, in 1994 and
1998, respectively.
He is an Associate Professor of electronic engi-
neering at the UPC and teaches courses in several
areas of electronic instrumentation. His research
includes sensor interfaces, autonomous sensors,
electronic instrumentation, noninvasive physiolog-
ical measurements, and sensors based on electrical
impedance measurements.
Ramon Pallàs-Areny (M’81–SM’88–F’98) re-
ceived the Ingeniero Industrial and Doctor Ingeniero
Industrial degrees from the Universitat Politècnica
de Catalunya, Barcelona, Spain, in 1975 and 1982,
respectively.
He is a Professor of electronic engineering at the
UPC and teaches courses in electronic instrumenta-
tion. In 1989 and 1990, he was a visiting Fulbright
Scholar, and, in 1997 and 1998, he was an Honorary
Fellow at the University of Wisconsin, Madison. In
2001, he was nominated Professor Honoris Causa
by the Faculty of Electrical Engineering of the University of Cluj-Napoca,
Romania. His research includes instrumentation methods and sensors based on
electrical impedance measurements, autonomous sensors, sensor interfaces,
noninvasive physiological measurements, and electromagnetic compatibility
in electronic systems. He is the author of six books, the leading author of five
books, and coauthor of two books on instrumentation in Spanish and Catalan.
He is also coauthor (with J. G. Webster) of Sensors and Signal Conditioning,
2nd ed. (Wiley, 2001) and Analog Signal Processing (Wiley, 1999).
Dr. Pallàs–Areny was a recipient, with J. G. Webster, of the 1991 Andrew
R. Chi Prize Paper Award from the IEEE Instrumentation and Measurement
Society. In 2000, he received the Award for Quality in Teaching granted by the
Board of Trustees of the UPC and, in 2002, the Narcís Monturiol Medal from
the Autonomous Government of Catalonia.