A Battery-Free Tag for Wireless Monitoring of Heart Sounds.

Full text

Page 1

A battery-free tag for wireless monitoring of heart sounds
Citation
Mandal, S., L. Turicchia, and R. Sarpeshkar. “A Battery-Free Tag
for Wireless Monitoring of Heart Sounds.” Wearable and
Implantable Body Sensor Networks, 2009. BSN 2009. Sixth
International Workshop on. 2009. 201-206. ©2009 IEEE.
As Published
http://dx.doi.org/10.1109/BSN.2009.11
Publisher
Institute of Electrical and Electronics Engineers
Version
Final published version
Accessed
Mon Apr 23 23:26:48 EDT 2012
Citable Link
http://hdl.handle.net/1721.1/59981
Terms of Use
Article is made available in accordance with the publisher's policy
and may be subject to US copyright law. Please refer to the
publisher's site for terms of use.
Detailed Terms

Page 2

A Battery-Free Tag for Wireless Monitoring of
Heart Sounds
Soumyajit Mandal, Lorenzo Turicchia, Rahul Sarpeshkar
Dept. of Electrical Engineering & Computer Science
Massachusetts Institute of Technology
Cambridge, MA 02139
Email: rahuls@mit.edu
Abstract—We have developed a wearable, battery-free tag that
monitors heart sounds. The tag powers up by harvesting ambient
RF energy, and contains a low-power integrated circuit, an
antenna and up to four microphones. The chip, which consumes
only 1.0µW of power, generates digital events when the outputs
of any of the microphones exceeds a programmable threshold
voltage, combines such events together by using a programmable
logic array, and transmits them to a base station by using
backscatter modulation. The chip can also be programmed to
trade-off microphone sensitivity for power consumption. In this
paper, we demonstrate that the tag, when attached to the chest,
can reliably measure heart rate at distances up to 7m from an
FCC-compliant RF power source. We also suggest how delays
between signals measured by microphones at the wrist and neck
can be used to provide information about relative blood-pressure
variations.
I. INTRODUCTION
Mechanical coupling causes the skin to vibrate in synchrony
with motions of the heart. Such vibrations generate acoustic
pressure waves, referred to as heart sounds, that can be picked
by microphones or by trained human ears using stethoscopes.
Such sounds provide important information about cardiac
health. For example, consider the characteristic lub/dub rhythm
present within each heart beat, referred to as the first heart
sound (S1) and second heart sound (S2), respectively. These
sounds are produced by the turbulent flow of blood against
the closed atrioventricular and semilunar valves, respectively,
and are thus sensitive to the condition of these valves. The
resting value and short-term variability of the heart rate, which
are both of clinical importance [1], can also be measured
from such acoustical recordings, known as phono-cardiograms
(PCGs).
Continuous monitoring and analysis of heart sounds is likely
to provide important health benefits because these sounds con-
tain large amounts of useful information. Most obviously, the
cessation of a regular heart beat can be used to detect Sudden
Cardiac Arrest (SCA). More than 5,000 people experience
SCA every week in the United States alone. The only definitive
treatment for SCA is early defibrillation: no more than 6
minutes from arrest to first shock. The chance for survival
drops 10% per minute without defibrillation, and today, over
95% of SCA victims die. Since automatic defibrillators are
increasingly available, continuous monitoring of those at risk
can save many lives [2]. In addition, cardiac arrhythmias, some
of which can be life-threatening, can be detected from heart-
rate data. In the United States alone, approximately 2,000
infants die each year from Sudden Infant Death Syndrome
(SIDS). Since slow heart rate (bradycardia) is an important
indicator of SIDS [3], early detection of bradycardia in infants
may save many lives each year.
In this paper, we describe a low-cost wearable tag that
can be attached to the skin and used for continuous wireless
monitoring of heart sounds. In order to be effective, such
monitoring systems should be widely adopted, which means
that they must be unobtrusive, robust, and low-cost. One way
to lower the size, weight and cost of the system is to eliminate
batteries and rely entirely on harvested ambient energy, in
this case RF. This is the approach adopted by passive radio-
frequency identification (RFID) tags [4], which face similar
size and cost constraints. However, maintaining reliable and
robust operation is challenging for any biomedical device, but
particularly problematic when a reliable local energy source,
such as a battery, is absent. We describe algorithms that detect
malfunctioning tags, thereby improving the reliability of our
results.
In this paper, we have chosen to focus on sounds, not
electrical signals, for monitoring heart activity. There are
two main advantages of this approach. Firstly, listening to
heart sounds does not require any electrical contact to be
established with the body. Such contacts are particularly
difficult to establish when the skin is dry. As a result, long-
term acoustical recordings can be obtained with less regular
maintenance than comparable electrical recordings, such as
the electro-cardiogram or EKG. Secondly, the heart is easy
to sense acoustically as it is the loudest organ in the body.
As a result heart sounds can be monitored in a very low-
power fashion with a cheap microphone. Reducing power
consumption becomes extremely important for systems such as
ours that rely on energy-harvesting, since the available ambient
energy density is usually orders of magnitude lower than that
present within a battery.
II. SYSTEM DESIGN
A conceptual view of our tag and a photograph of the
prototype that was actually tested are shown in Figure 1. We
estimate that the area of the prototype, which contains the chip,
an antenna for RF energy harvesting, and two microphones,
2009 Body Sensor Networks2009 Body Sensor Networks2009 Body Sensor Networks2009 Body Sensor Networks
978-0-7695-3644-6/09 $25.00 © 2009 IEEE
DOI 10.1109/P3644.10DOI 10.1109/P3644.10DOI 10.1109/P3644.10DOI 10.1109/BSN.2009.11
203203201201978-0-7695-3644-6/09 $25.00 © 2009 IEEE978-0-7695-3644-6/09 $25.00 © 2009 IEEE978-0-7695-3644-6/09 $25.00 © 2009 IEEE

Page 3

can easily be halved by using a smaller package for the chip
and a more optimized antenna. The two microphones are used
to generate three types of alarm: disconnection from the body,
device malfunction, and patient emergency. One microphone
faces up (away from the body) and the other faces down. The
downward-facing microphone monitors heart sounds, while
the upward-facing one is usually switched off to save power.
It is turned on only when the downward-facing microphone
does not detect any heart sounds and a disconnection or
patient emergency is suspected. If both microphones now
pick up similar environmental sounds, a ‘disconnection alarm’
is generated since it is probable that the tag is no longer
in proximity to the skin. A ‘patient emergency alarm’ is
generated if the downward-facing microphone does not pick
up environmental sounds, but the upward-facing one does,
since in this case it is likely that the tag is still attached
and the heart has stopped. If neither microphone picks up any
sounds, the tag is probably malfunctioning; therefore a ‘device
malfunction alarm’ is generated.
Each tag contains a unique identification code. A fixed base
station communicates with multiple tags and decides if and
when an alarm should be triggered. Tags that were earlier
actively communicating with the base station, but have since
stopped doing so, may have left the area covered by the base
station, or entered an area with insufficient RF energy for
reliable operation. Software on the base station must be able
to distinguish between such events and alarms generated by
disconnections, patient emergencies and tag malfunctions.
Our tag is battery-less and operates by harvesting radiated
RF energy [5], [6]. We now calculate the available power PA
that can be harvested at different distances D from the RF
transmitter. Path loss models predict the fall-off of radiated
power density Pr(in W/m2) with D. A simple version com-
monly used for modeling indoor environments recognizes two
propagation zones: Pr∝ D−n1for D ≤ D0, and Pr∝ D−n2
for D > D0, where D0, n1and n2are constants [7]. Typically
n1 ≈ 2, the free-space value, and n2 varies between 2.5
and 4. The value of n2 exceeds 2 because of absorption
and reflection of the RF by environmental obstacles, such as
furniture and people. Using conservative values (D0 = 5m,
n1= 2, n2= 3.5) and combining the predicted path loss with
the rectifier model proposed in [6] gives us Figure 2. The figure
assumes that the equivalent isotropic radiated power (EIRP) is
4W, which is the maximum allowed in the United States for
radio-frequency identification (RFID) applications. It shows
PA as a function of D at two popular RFID frequencies:
900MHz and 2.4GHz. The main reason for going to higher
operating frequencies is to reduce the physical size of the
antenna.
The various curves in Figure 2 correspond to different
load resistances RL driven by the on-chip rectifier. They
decrease rapidly at large distances because the received RF
amplitude becomes smaller than the rectifier’s dead zone. The
load resistance is usually dominated by the power consumed
by off-chip sensors and not the chip itself. For example, a
microphone biased at 30µA and 0.5V (typical values used
ANTENNA
INTEGRATED
CIRCUIT
MICROPHONE
(FACE DOWN)
MICROPHONE
(FACE UP)
Fig. 1.
surface (top), and photograph of the actual 800MHz prototype that was tested
experimentally (bottom). The pins at the bottom of the prototype were for
testing purposes and will be removed in the final version.
Conceptual view of our tag, attached to a flexible, adhesive
in our experiments) dissipates 15µW, corresponding to an
effective load driven by the rectifier of RL = 16.7kΩ [6].
Figure 2 then predicts an operating range of approximately
12m at 900MHz and 3m at 2.4GHz. In practice the reliable
operating range will be somewhat smaller because some
tags will be mistuned by their proximity to conductive and
dielectric surfaces. In addition, we have to allow for transient
drops in received RF power level (fades), which are ubiquitous
in indoor environments because the received signal is the
superposition of multiple waves with time-varying amplitude
and phase. Nevertheless, a single base-station operating at
900MHz is sufficient for a moderately-sized room.
Our chip contains four independent channels that can be
used to interface to sensors such as microphones. The outputs
of these channels are thresholded, or digital signals. These sig-
nals can be combined in a flexible way using a Programmable
Logic Array (PLA) that can implement a variety of Boolean
logic functions. Our PLA is a four-input four-output design
with a 8 × 8 AND plane and a 4 × 8 OR plane. The PLA
allows us to implement any of the 224possible logic functions
of four inputs for any of its four outputs in a programmable
fashion. These outputs can be monitored individually, allowing
us to implement rudimentary sensor-fusion algorithms that
combine the outputs of multiple channels. Programmable
output selection logic multiplexes the four PLA outputs into
204204 202202

Page 4

10
0
10
1
10
−8
10
−7
10
−6
10
−5
10
−4
10
−3
10
−2
Distance, D (m)
Available power, PA (W)


900 MHz
2.4GHz
10kΩ
100kΩ
1MΩ
10kΩ
100kΩ
1MΩ
∞
∞
Fig. 2.
from the transmitter for different load resistances at 900MHz and 2.4GHz.
Theoretical limits on harvested RF power as a function of distance
a single signal that is transmitted to the base station as “event
packets” containing the chip identification code and time
stamps. Data is transmitted using backscatter modulation [4]:
a 100fF capacitor is added and subtracted from CLto change
the amount of RF power scattered by the tag. Backscatter
modulation is popular in passive RFID systems because all
the complexity and power consumption is pushed to the base
station; the tag remains simple and low-power.
A
HOLD
TIMER
EVENT
SELECT
LOGIC
BIDIRECTIONAL
COMPARATOR
STRETCHER
PULSE
PREAMP
IN
OUT
DAC
BIAS
IREF
CLKREF
P
N
Fig. 3.A single on-chip signal-processing channel.
A block diagram of a single channel is shown in Figure 3.
The preamplifier consists of a common source stage with
capacitive feedback. The size of the feedback capacitor can
be varied between C0and 16C0to set the gain. The amplifier
incorporates DC rejection, i.e., the transfer function is band-
pass with a very low cut-in frequency (typically <1Hz). It uses
a nominal bias current of 10nA and C0= 0.5pF resulting in
a bandwidth that decreases from 12KHz to 6KHz as the gain
increases from 1 to 16. A matched copy of the amplifier (minus
the capacitors) is used to determine the quiescent operating
point.
The comparator generates events whenever the output volt-
age of the amplifier differs from its quiescent value by more
than a fixed threshold voltage Vth = 80mV. There are two
types of events: positive-going, when the output voltage is
larger than its quiescent value by Vthor more, and negative-
going, when it is smaller. The smallest input amplitude that
triggers an event decreases from Vthto Vth/16 (80mV to 5mV)
as the preamplifier gain increases from 1 to 16.
Event selection logic is connected to the comparator output
in each channel and allows only positive or negative-going
events, both, or neither to be detected. This combinational
block is followed by a pulse-stretcher circuit that adds hys-
teresis in the time domain to prevent multiple comparator
transitions due to noise when an event is detected. It also
ensures that output events always last long enough for at
least one complete data packet to be broadcast during every
event. The pulse-stretcher circuit is a digitally-timed one-shot:
It allows an incoming event edge to set its output high, and a
delayed version of this edge to reset it low. The pulse-stretcher
is followed by a programmable hold timer circuit. This circuit
imposes a hold time Tholdafter each event, during which no
new events can be generated. By placing an upper bound of
1/Tholdon the event rate, the hold timer greatly reduces the
probability of timing collisions between different tags. The
average value of Tholdcan be varied between 94ms and 1.4s.
A programmable DC current source was designed for every
channel. This current source can be used to power up external
sensors, such as microphones, and consists of a 8-bit binary-
weighted current DAC that can supply between 0.5µA and
128µA. To reduce power consumption, the chip was designed
to operate on power supply voltages as low as 0.8V (core) and
0.5V (programmable current sources).
An on-chip serial interface allows the user to program the
PLA, channel selection logic, 16-bit chip identification code
and channel parameters such as sensor current, preamplifier
gain and hold time. The static power consumption of the
chip alone is only 1.0µW. The power consumption with
microphones connected is variable, since it depends on their
bias currents, which can be programmed by the user.
III. MICROPHONE INTERFACE
We used a microphone to detect heart sounds. Microphone
responses are bandpass, with typical lower (cut-in) frequencies
between 20Hz and 100Hz and upper (cut-off) frequencies
between 16kHz and 20kHz. Commercial electret microphones
contain built-in low-noise JFETs for buffering [8]. There are
two common configurations, as shown in Figure 4. In two-
terminal microphones, shown on the left, the drain of the n-
type JFET, which is normally a depletion-mode device, acts as
the output terminal. It is usually connected to an an external
resistive load, creating a common-source amplifier. The gate
voltage is internally tied to ground at DC with the large resistor
Rbig, while Cpar is a small, unwanted parasitic capacitance.
The incoming sound pressure wave creates the voltage source
velec in series with the electret capacitance Celec. Three-
terminal microphones, shown on the right, configure the JFET
as a source follower, and are typically more expensive.
We used a two-terminal microphone, but replaced the re-
sistive load with a programmable current source Ibiasrunning
off a very low supply voltage, VDD,MICto save power. In this
205205203203

Page 5

OUT
Celec
Rbig
Cpar
velec
RS
Celec
Rbig
Cpar
velec
OUT
Fig. 4. Circuit diagram of common electret microphones, two-terminal (left)
and three-terminal (right).
regime the JFET is unsaturated and acts as a voltage-controlled
resistor. Using the standard long-channel FET equation,
Ibias= β?VDSATVOUT− V2
where the saturation voltage VDSAT is fixed for a given
microphone, and given approximately by VGS− VT, where
VGS is the gate-to-source voltage (usually zero) and VT is
the threshold voltage. Also, VOUT is the output voltage,
and β is a constant. Within the microphone passband the
sound-generated source velec changes VGS by an amount
velecCelec/(Celec+ Cpar). Since the current through the
JFET is fixed at Ibias, vout decreases as vgs increases, and
vice-versa. Solving the quadratic equation in (1) and some
simple small-signal analysis reveals that the transfer function
between velecand voutin the passband is
OUT/2?
(1)
vout
velec
= −
Celec
Celec+ Cpar
??
1
1 − Ibias/ISAT
− 1
?
(2)
where ISAT= βV2
specified by the manufacturer for a given microphone. We see
that gain increases with the bias current Ibias, which we can
vary by using the on-chip DAC. In reality the gain saturates
at a high value as Ibiasapproaches ISAT, instead of diverging
as predicted by (2), because of the finite output impedances
of the saturated transistor and the current source (which we
have ignored in this analysis). However, we can still trade-off
sensitivity with power consumption. In fact, in this application
we can save considerable amounts of power because heart
sounds, which occur in the range 20-250Hz, are relatively
loud and do not require large amounts of gain. We used
a Panasonic omnidirectional electret condenser microphone
(WM-63PR) in a plastic enclosure. The WM-63PR, which
has ISAT = 500µA, was selected since it is a small, thin
device (diameter = 6mm, thickness = 1.3mm), has a low cut-in
frequency (20Hz) and is also low-cost. Similar microphones
that are even less expensive can also be used since sound
quality is not important for this application.
The microphone is normally placed on the chest while
monitoring heart activity. However, the membrane of the
microphone cannot vibrate freely if it is located too close to
the skin. Therefore we built our microphone sensor so that
there is a very small cylindrical air chamber between the two.
The chamber has no vents, reducing the amount of ambient
DSAT/2, the JFET current in saturation, is
noise, but its diameter and shape have little effect on sound
pickup [9]. A photograph of the microphone sensor prior to
mounting on the tag is shown in Figure 5.
Fig. 5.Photograph of the microphone sensor used for measuring the PCG.
IV. EXPERIMENTAL RESULTS
We tested our system in a cluttered laboratory with nu-
merous barriers to RF propagation, as shown in Figure 6.
Experimentally, we were able to obtain 15µW of output
power at a distance of 3.1m from an 800MHz RF source
broadcasting 800mW EIRP, which was the maximum allowed
by our equipment. This is enough power to run the chip and
one microphone at a typical bias current of 30µA and power
supply voltage of 0.5V. Increasing the transmit power to the
allowed maximum of 4W should give us an operating range
of 5-7m, depending upon the multipath fading characteristics
of the environment. We also note that the curves in Figure 2
begin to drop off sharply with increasing distance when the
received RF voltage becomes smaller than the rectifier’s dead
zone, which causes its power conversion efficiency to decrease
rapidly. If the load resistance is such that the tag operates in
this region, range can also be increased by using a smaller,
but more expensive chip package with lower parasitic input
capacitance Cin. Decreasing Cinincreases the quality factor
of the input matching network, increasing the received RF
voltage. Specifically, a given amount of available power is
obtained at a distance that scales as 1/√Cin. For example,
the rectifier in [6] used a package with approximately half the
input capacitance, resulting in√2 ≈ 1.4 times more range.
We noticed that the operation of the tag was largely unaf-
fected by the presence of the human body up to a distance
of approximately 1cm from the skin. The received power
decreased sharply for smaller separations. The microphone
must also be separated from the skin to operate normally. Both
problems can be solved by backing the tag with dielectric foam
approximately 1cm thick. Figure 7 shows measured waveforms
produced by the chip in response to a single event. The
backscattered signal consists of an 48-bit long event packet
that is repeated for the duration of the event. Each packet
begins with a 16-bit ‘010101...’ sequence to aid synchroniza-
tion, followed by the tag’s unique 15-bit identification code, a
parity bit and a 16-bit time-stamp. In this case three complete
packets were transmitted before the event ended.
206 206204204